In 1906, a German psychiatrist named Alois Alzheimer dissected the brain of a woman who had died after years of progressive memory loss, confusion, and paranoia. Under the microscope, he found two things he could not explain: sticky clumps between the nerve cells, and tangled fibers inside them. He presented his findings at a conference. Nobody was very interested. The case was filed, with characteristic German precision, as one of those oddities that medicine occasionally produces and cannot yet account for.
More than a century later, the sticky clumps — now called amyloid plaques — have become the most fought-over molecules in the history of medicine. The fight has consumed over fifty billion dollars. It has occupied the careers of thousands of researchers. It has shaped the design of more than a hundred clinical trials. It has produced two FDA-approved drugs that cost tens of thousands of dollars a year and slow cognitive decline by an amount that most patients cannot perceive. It has generated at least two major fraud scandals, the resignation of a senior NIH official, an SEC enforcement action, and a congressional investigation into the FDA's approval process. It has left millions of patients and their families with almost nothing.
The question at the center of this fight is not complicated. It is just very expensive: Does amyloid-beta cause Alzheimer's disease? Or is it something the brain produces in response to the actual cause — a bystander, a symptom, a red herring that happens to be present at every crime scene?
After thirty-four years, the field still does not have a definitive answer. What it has instead is something worse: a partly correct hypothesis that was treated as entirely correct for so long that it distorted the institutions meant to test it.
The amyloid cascade hypothesis was formally proposed in 1991 and 1992, independently, by several researchers — John Hardy, David Allsop, Gerald Higgins, Dennis Selkoe, and others. The idea was straightforward and had the virtue of elegance: the abnormal buildup of amyloid-beta peptide in the brain is the primary event in Alzheimer's disease. Everything else — the tau tangles, the neuroinflammation, the neuronal death, the cognitive decline — follows from that initial accumulation. Stop the amyloid, and you stop the disease.
The genetic evidence was genuinely compelling. People with Down syndrome, who carry an extra copy of chromosome 21 — where the amyloid precursor protein gene sits — develop Alzheimer's pathology almost universally. Rare families with mutations in the APP gene, or in the presenilin genes that regulate how APP is processed, develop early-onset Alzheimer's in their forties and fifties. These mutations increase amyloid-beta production. The causal chain seemed clear: more amyloid means more disease.
Mouse models reinforced the story. Genetically engineered mice carrying human APP mutations developed amyloid plaques and lost their memory. When you cleared the plaques, some of the cognitive deficits reversed. If you stopped the amyloid, the mice got better. "It was enormously persuasive," recalled Daniel Alkon, a former NIH lab director. The logic was simple, the animal data were clean, and the implication was direct: find a way to remove amyloid from human brains, and you have a treatment.
By the mid-1990s, the hypothesis had won. Not in the way that scientific ideas usually win — through accumulation of evidence, debate, revision, and grudging consensus — but in the way that institutional power consolidates. The amyloid hypothesis became the organizing principle of an entire field. It determined which grants got funded, which papers got published, which researchers got hired, and which companies attracted investment. It did not merely describe the leading theory of Alzheimer's disease. It became the infrastructure.
The problems started early. The first generation of anti-amyloid drugs entered clinical trials in the early 2000s. They failed. The second generation failed. The third generation failed. Between 2002 and 2022, fifty-one major clinical trials tested drugs designed to reduce, prevent, or clear amyloid from the brain. Almost all of them failed to produce any meaningful cognitive benefit. The patients kept declining. The plaques were removed; the disease continued.
This was not a subtle signal. In most fields, fifty consecutive failures would prompt a reassessment of the underlying premise. In Alzheimer's research, the failures were interpreted not as evidence against the hypothesis but as evidence that the drugs had been given too late, at too low a dose, to the wrong patients, or that they targeted the wrong form of amyloid. The hypothesis itself was not questioned. Instead, it was refined — a process that looked, from the outside, less like self-correction than like the addition of epicycles to a Ptolemaic model.
The few researchers who did question it found the doors closing. Ruth Itzhaki, a molecular neurobiologist at the University of Manchester, discovered in 1991 — the same year the amyloid hypothesis was proposed — that herpes simplex virus type 1 was present in the brains of elderly people with Alzheimer's. Her papers were rejected by top journals. When she applied for grants, reviewers told her that because her work hadn't been published in prestigious journals, it lacked credibility. The circularity was perfect: she could not get published because she was not funded, and she could not get funded because she was not published. When she applied for an Alzheimer's Association fellowship, one reviewer gave her a 3 out of 10, calling pathogens in Alzheimer's "a fringe topic." She applied annually to present at the Alzheimer's Association International Conference and was refused speaking slots. A colleague once used "considerable persuasion" to secure her ten minutes. "That was the only time I was allowed to give a talk," she said later. "They never included viruses in the list of meeting topics, and I was allowed only a poster. No one influential ever even came to see it."
She was not alone. George Perry, a neuroscientist at the University of Texas at San Antonio, had his NIH grants repeatedly rejected. A friend who served on an NIH study section told him the reason: his anti-amyloid stance. Perry later described alternative researchers as "roadkill on the highway to nowhere, their careers stalled or worse."
Malú Tansey, then a junior faculty member at Emory University, presented research on neuroinflammation as an alternative pathway in 2002. "A senior faculty member stood up and said neuroinflammation has nothing to do with Alzheimer's," she recalled. "It was intimidating." At Mount Sinai, a postdoctoral fellow working with Nikolaos Robakis submitted a grant proposal to study non-amyloid mechanisms. The score was so low it was not even discussed. The fellow left Alzheimer's research entirely.
Raymond Tesi, a biotech executive, tried to raise money for a neuroinflammation-based Alzheimer's treatment. He went to Eli Lilly, to Novartis, to venture capitalists. "We couldn't get past the front door," he said. "If you weren't doing amyloid, you could barely get a meeting." He identified the mechanism: "Every company goes to key opinion leaders in academia and asks, 'What should we do?' Since amyloid had a lock on such leaders, from Harvard to USC and in between, the answer was always the same: eliminate amyloid." Tesi estimated that by 2015, ninety percent of NIH, industry, and private foundation spending on Alzheimer's research was premised on amyloid elimination.
Rachael Neve, a researcher at Massachusetts General Hospital who had helped discover the APP gene mutation that originally supported the amyloid hypothesis, came to regret it. She said that NINDS program officers told her she should "at least collaborate with the amyloid people" or she would not receive further grants. "I am sick about the millions of people who have needlessly died," she said.
Zaven Khachaturian, who as director of the NIH's Office of Alzheimer's Disease had helped build the field from scratch, watched the consolidation with alarm. "It became gradually an infallible belief system," he said. "That's not healthy for science." Amyloid supporters, he observed, "came to permeate drug companies, journals, and NIH study sections. Things shifted from a scientific inquiry into an almost religious belief system."
Susan Fitzpatrick, president of the James S. McDonnell Foundation, identified the structural incentive: "Once a field commits to a particular hypothesis, the research resources — funding, experimental models, and training — all get in line. That brings accolades, awards, lucrative consulting deals, and well-paid academic appointments. Admitting doubt would be a blow to ego but also a threat to livelihood."
This was not a conspiracy. It was something more durable: a consensus that had calcified into an institution. No single person decided to suppress alternative research. The suppression emerged from the ordinary operation of science's own quality-control mechanisms — peer review, grant competition, journal prestige — when those mechanisms were captured by a dominant paradigm. The gatekeepers were not corrupt. They were orthodox. The effect was the same.
Then came the fraud.
In 2006, a neuroscientist named Sylvain Lesné, working in Karen Ashe's lab at the University of Minnesota, published a paper in Nature that would become one of the most cited in the history of Alzheimer's research. The paper claimed to have identified a specific form of amyloid-beta — a soluble oligomer called Aβ*56 — that, when injected into healthy rats, caused them to develop memory loss. This was significant because it provided something the field desperately wanted: a specific toxic species. Not just plaques in general, but a particular molecular configuration that directly caused cognitive damage. The paper seemed to close a gap in the causal chain.
Oligomer research funding from the NIH, which had been near zero in 2006, skyrocketed to more than 333 million dollars by 2021.
In 2022, Matthew Schrag, a neuroscientist at Vanderbilt University, noticed something wrong with the images in Lesné's papers. Working with image-analysis experts, he identified what appeared to be systematic manipulation — copy-pasted bands, doctored blots, figures that had been composited from different experiments. The evidence of tampering was not subtle. Charles Piller, an investigative journalist at Science, published the findings in an article titled "Blots on a Field." The analysis ultimately identified at least thirteen papers by Lesné containing apparently doctored images, all pertaining to Aβ*56.
Nature retracted the 2006 paper in 2024, noting that figures "showed signs of excessive manipulation." Every author signed the retraction statement except Lesné. He resigned from his tenured position at the University of Minnesota effective March 1, 2025.
But the Lesné case was not an isolated incident. It was more like pulling a thread that unraveled a sleeve.
Piller organized a team of image analysts to examine the work of dozens of prominent Alzheimer's researchers who had been flagged on PubPeer, a website devoted to scientific integrity. They compiled a 286-page dossier. The results were staggering: 571 papers with problematic images, from 65 researchers. Those 571 papers had received 77,655 citations and were referenced by 478 active patents.
The biggest name to fall was not Lesné. It was Eliezer Masliah, the director of the neuroscience division at the National Institute on Aging — the man who oversaw a 2.6-billion-dollar neuroscience research budget. An investigation by Science identified 132 of Masliah's papers containing apparently duplicated or doctored images. The NIH found evidence of "falsification and/or fabrication involving reuse and relabel of figure panels representing different experimental results." Masliah was fired in September 2024.
Then there was Cassava Sciences. Hoau-Yan Wang, a neuroscientist at CUNY, was indicted in June 2024 for allegedly defrauding the NIH of sixteen million dollars through fabricated data in grant applications related to Cassava's Alzheimer's drug simufilam. The SEC announced in September 2024 that Cassava's co-founder, her boss, and the company would pay over forty million dollars to settle charges of misleading investors about clinical trial results. At USC, Berislav Zlokovic was placed on indefinite leave amid evidence of manipulated images, and a thirty-million-dollar clinical trial studying his work was canceled.
The scale of the problem was not normal. Elisabeth Bik, a microbiologist who helped uncover some of the image alterations, called it "a severe breach of scientific integrity" and said the findings "led many other studies in the wrong direction," causing "false hope among patients and their families."
Charles Piller turned his reporting into a book, Doctored: Fraud, Arrogance, and Tragedy in the Quest to Cure Alzheimer's, published in February 2025. The Economist called it "a gripping story of medical groupthink and warped incentives." One reader described it as "one of the most important and distressing books I have read."
Here is the uncomfortable part. After all of this — the institutional capture, the suppression of alternatives, the fraud, the fifty-one failed trials — the amyloid hypothesis turned out to be at least partly right.
In July 2023, the FDA granted full approval to lecanemab (sold as Leqembi), an anti-amyloid antibody developed by Eisai and Biogen. In July 2024, donanemab (developed by Eli Lilly) followed. Both drugs remove amyloid plaques from the brain. Both showed statistically significant slowing of cognitive decline in large, well-designed phase 3 trials. For the first time in the history of Alzheimer's research, there were drugs that appeared to modify the disease's trajectory rather than merely masking its symptoms.
The celebrations were cautious, and the caution was warranted.
Lecanemab, in the Clarity AD trial, slowed cognitive and functional decline by 27 percent compared with placebo over eighteen months — measured on the CDR-SB scale, the standard composite of cognition and daily function. In absolute terms, this amounted to a difference of 0.45 points on a scale where the minimum observable change is 0.5 points. Markku Kurkinen, a pharmacologist, recalculated the benefit using final scores rather than change-from-baseline and arrived at a figure of 9.3 percent less cognitive impairment — not the 27 percent reported. Donanemab showed similar absolute differences.
What does this mean for a patient? Sharon Straus, a physician and clinical epidemiologist, put it plainly: "Clinical benefits are modest at best, with uncertain impacts on quality of life, independence, and caregiver burden." Patients on lecanemab "lost their ability to remember, think clearly, and live normal lives only slightly less rapidly" than those on placebo. The drug did not stop the disease. It did not reverse it. It slowed it, by an amount that was statistically real and clinically ambiguous.
The drugs also came with significant risks. About one in four patients on lecanemab — 24.3 percent — developed amyloid-related imaging abnormalities, or ARIA, a term that encompasses brain swelling and small hemorrhages. Most cases were mild. Some were not. At least five deaths were documented. Patients with two copies of the APOE ε4 gene — the strongest known genetic risk factor for Alzheimer's — were four times more likely to develop ARIA. Treatment required biweekly intravenous infusions, regular MRI monitoring, and PET scans or lumbar punctures for diagnosis. Annual costs ran between 26,000 and 40,000 dollars for the drug alone, plus as much as 90,000 dollars for monitoring. In July 2026, the FDA approved a subcutaneous formulation, easing the infusion burden but not the cost or the risk.
The UK's National Institute for Health and Care Excellence declined to recommend lecanemab, citing concerns about both cost and the translation of trial results into clinically meaningful improvement. Real-world data, published in August 2026 from a Duke Health study of 230 patients, found that 79 percent continued treatment for approximately a year, with 21 percent discontinuing, mostly due to side effects. Among those who stayed on the drug, 75.9 percent remained cognitively stable, 6.6 percent actually improved slightly, and 17.6 percent declined despite treatment.
Thomas Südhof, a Stanford neuroscientist and Nobel laureate, said he "wouldn't call it a breakthrough" and worried the drugs "might slow down the field" by reinforcing the amyloid paradigm just when the field was beginning to diversify.
And the field is diversifying, though whether fast enough is another question.
The 2026 Alzheimer's drug pipeline comprises 158 agents in 192 active trials. Only about a third target amyloid and tau. The rest pursue neuroinflammation, metabolic pathways, the gut-brain axis, synaptic dysfunction, epigenetics, and vascular factors. Semaglutide, the GLP-1 receptor agonist best known as the active ingredient in Ozempic, entered phase 3 trials for Alzheimer's — the logic being that its anti-inflammatory and metabolic effects might protect neurons. The results, presented in late 2025, were disappointing: biomarker improvements of about 10 percent in markers of neuroinflammation and neurodegeneration, but no clinical benefit. Statistically significant changes in blood chemistry that did not translate to changes in cognition or daily function.
The neuroinflammation hypothesis — the one that Malú Tansey was told had "nothing to do with Alzheimer's" in 2002 — is now one of the most active areas of research. Microglia, the brain's resident immune cells, are increasingly understood to play a central role in neurodegeneration. Genetic studies have identified dozens of Alzheimer's risk genes involved in innate immunity, complement activation, and lipid metabolism — pathways that have nothing to do with amyloid production but everything to do with the brain's inflammatory response. The APOE ε4 gene, the strongest risk factor for late-onset Alzheimer's, is fundamentally a gene about lipid transport and immune regulation, not about amyloid.
The gut microbiome, which would have been laughed out of an Alzheimer's conference a decade ago, now has its own research stream. Fecal microbiota transplants from mice with Alzheimer's pathology can induce tau phosphorylation and glial activation in healthy recipient mice. A Chinese drug called GV-971, derived from seaweed, was approved in China in 2019 on the basis of its effects on gut microbiota, though its efficacy remains controversial and it has not gained approval elsewhere.
The most fundamental critique of the amyloid hypothesis may be the simplest: roughly 30 to 40 percent of cognitively normal elderly people have amyloid deposits that meet the histological criteria for Alzheimer's disease. They have the plaques. They do not have the disease. This suggests that amyloid, whatever role it plays, is not sufficient. The hypothesis explains the rarest form of Alzheimer's — familial early-onset, caused by specific gene mutations, accounting for less than five percent of cases — quite well. It explains the common form — sporadic, late-onset, arising from a tangle of genetic, environmental, inflammatory, vascular, and metabolic factors — much less well. The field spent three decades treating the rare form as a model for the common one. Christian Behl, in a 2024 assessment, traced this error to a 1976 editorial by Robert Katzman proposing that familial and sporadic Alzheimer's be considered "a single disease." That editorial, Behl argued, allowed the genetics of a rare condition to dominate the research agenda for a disease that affects tens of millions.
John Hardy, one of the original authors of the amyloid cascade hypothesis, has maintained a position of measured defense. He acknowledges that the anti-amyloid antibodies "may not stop disease," that "there are other factors at play in clinical decline beyond amyloid," and that the field needs to understand "residual decline" — the cognitive deterioration that continues even after the plaques are gone. But he views lecanemab and donanemab as validation: they target amyloid, they slow decline, therefore amyloid matters. The hypothesis, he has said, "cannot be proved, only disproved: and it will only have been truly useful if it leads to the development of a treatment based on its therapeutic implementation."
This is, in one sense, a reasonable position. The drugs do something. Amyloid is not irrelevant. But it is also a position that defines success very broadly. A treatment that slows decline by 27 percent (or 9 percent, depending on how you calculate) while causing brain bleeding in a quarter of recipients and costing 50,000 to 130,000 dollars a year with monitoring — is that what "therapeutic implementation" looks like when a hypothesis has been "truly useful"?
The critic Alberto Espay, a neurologist at the University of Cincinnati, offered a less generous assessment. Among the fifty-one anti-amyloid trials from 2002 to 2022, only one — Aduhelm — outperformed placebo, and it was later pulled from the market after physicians refused to prescribe it and insurers refused to cover it. The FDA had approved it against the recommendation of its own advisory panel, in a process that a congressional investigation later found involved "inappropriate collaboration" with Biogen, unreported meetings, and failure to gain internal consensus. Biogen discontinued Aduhelm in January 2024, less than three years after its approval, marking one of the most expensive drug launch failures in pharmaceutical history.
Nikolaos Robakis, the Mount Sinai neuroscientist, summarized the field's trajectory: "Critical thinking gave way to dogma. Very big egos involved and they couldn't stand to be wrong. It wasn't science anymore."
Daniel Alkon, the former NIH lab director who left the agency to develop bryostatin — a drug that protected neurons and preserved synapses but did not target amyloid — recalled the response when he approached pharmaceutical companies with his data. They were interested, he said, until he explained the mechanism. "The only way they would consider bryostatin was if we could show it was superb at stopping amyloid. They were just interested in that, not its effect on synapses." Subsequent companies repeated the same condition. Alkon's estimate: "We would be ten or fifteen years ahead of where we are now" without amyloid's dominance.
The amyloid story is not, finally, a story about fraud. The fraud — Lesné's doctored images, Masliah's 132 compromised papers, Cassava Sciences' fabricated data — is dramatic and consequential, but it is a symptom, not the disease. The Aβ*56 paper bolstered a specific claim about a specific toxic species, but the amyloid hypothesis would have dominated the field without it. The hypothesis was already entrenched by 2006, when the Nature paper was published. It had been entrenched since the mid-1990s, when the combination of genetic evidence, mouse models, and institutional momentum created a feedback loop that became self-sustaining.
Nor is it, exactly, a story about a hypothesis being wrong. The amyloid hypothesis, in its weakest form — that amyloid-beta plays some role in Alzheimer's disease — is almost certainly correct. The genetics are real. The drugs do something. The question was never whether amyloid mattered. The question was whether it was the whole answer, and whether treating it as the whole answer was worth the opportunity cost of not pursuing the rest.
What the story is about is something harder to fix than fraud and harder to see than a wrong hypothesis. It is about what happens when a field organizes itself around a single idea so completely that the mechanisms of self-correction — peer review, grant competition, replication, clinical testing — become mechanisms of self-reinforcement. Every failed trial generated a refinement of the hypothesis rather than a challenge to it. Every critic was marginalized not by explicit suppression but by the ordinary operation of prestige, funding, and publication norms. Every fraud went undetected for years because the results it fabricated were the results the field expected to see.
There is a name for this in the philosophy of science. Thomas Kuhn called it "normal science" — the period in which a paradigm is accepted and research consists of puzzle-solving within its framework. Anomalies accumulate but are explained away. The paradigm holds not because it is correct but because no alternative has achieved sufficient institutional support to displace it. The shift, when it comes, is not gradual. It is a revolution.
Behl, writing in 2024, applied Kuhn's framework directly: "We are not quite there yet. We have not reached the final steps of the necessary paradigm shift." The amyloid hypothesis is weakening but has not collapsed, in part because the drugs — modest as they are — provide just enough evidence to keep it alive. It is, in a sense, the worst possible outcome for the field: a hypothesis that is too right to abandon and too wrong to be sufficient. A trap.
NIH funding for Alzheimer's research has reached approximately 3.9 billion dollars annually, up from under 500 million when the National Alzheimer's Project Act was passed. The 2026 pipeline is the most diverse it has ever been. But diversity in the pipeline does not mean diversity in the power structure. The key opinion leaders, the grant reviewers, the journal editors — many of the same people who built the amyloid orthodoxy over three decades — remain in positions of influence. The infrastructure of capture does not dismantle itself when the science shifts. It adjusts, accommodates, absorbs. Amyloid researchers now speak of "amyloid-plus" approaches — combination therapies that target amyloid and tau and neuroinflammation. The framework expands, but it does not cede the center.
There are currently 6.9 million Americans living with Alzheimer's disease. By 2050, projections suggest 12.7 million. Most of them will have the sporadic, late-onset form that the amyloid hypothesis explains least well. For them, the most expensive partly-right answer in the history of medicine offers a treatment that costs more than many of them earn in a year, slows their decline by an amount they may not notice, and carries a one-in-four chance of brain bleeding. When their families ask why there isn't something better, the answer is not that the science is hard — though it is — but that for thirty years, the institutions that were supposed to be exploring every possible answer were, instead, exploring mostly one.
George Perry, the Texas neuroscientist whose grants were killed for his anti-amyloid stance, offered the epitaph: "We could have gotten to this point many, many years sooner."
That is the cost. Not the fifty billion dollars. Not the failed trials. Not the fraud. The cost is the years. The cost is what might have been found if the field had been open to finding it.
Process note. I went into this expecting a fraud story — Lesné doctors his images, the field builds on fabricated data, the whole thing collapses. The material rejected that narrative within the first hour of research. The amyloid hypothesis was not built on fraud. It was built on real genetics that were genuinely compelling, then reinforced by institutional capture that did not require fraud to operate. The fraud accelerated money flows into a specific sub-hypothesis (Aβ*56 oligomers) but the broader amyloid orthodoxy predated it by a decade. The harder finding was this: the drugs work. Modestly, expensively, dangerously — but they work. That meant I could not write a clean story of collapse. Instead the material pushed toward something more uncomfortable: a hypothesis that is too right to dismiss and too wrong to be enough. A trap that catches both the field and the patients. The most surprising single fact was the scale of image manipulation — 571 papers, 65 researchers, 77,655 citations, 478 patents. That number suggests the fraud is not aberrant but endemic, which changes the story from "a few bad actors" to "a system that incentivized and failed to detect fabrication at scale." The researcher quotes — Itzhaki, Perry, Tansey, Alkon, Tesi, Neve, Khachaturian — were the strongest material, because they describe the mechanism of suppression in concrete, personal terms. The plan changed. Whether the piece is better for it, I cannot say, but the material insisted.