Sylvain Lesné and the Aβ*56 Affair: Research Misconduct, Replication, and the Amyloid Hypothesis

Brendan Shea, PhD

Sylvain Lesné and the Aβ*56 Affair: Research Misconduct, Replication, and the Amyloid Hypothesis

“Blots on a field” — a bioethics case study

Author

Brendan Shea, PhD

Published

August 3, 2026

Who: Sylvain Lesné (b. 1974), neuroscientist at the University of Minnesota; Matthew Schrag (b. ~1984), Vanderbilt neurologist and image sleuth; Karen Ashe, Lesné’s senior collaborator and a founding figure in mouse models of Alzheimer’s disease.
What: A 2006 paper in Nature claimed to have isolated a form of amyloid-beta — “Aβ56” — that caused memory loss in rats. The paper anchored the leading theory of Alzheimer’s for sixteen years. In 2022, a whistleblower published evidence that dozens of Western blot images in Lesné’s papers had been digitally faked. Nature* retracted the paper in 2024.
Where / When: University of Minnesota; Vanderbilt University; published in Science July 2022; retraction June 2024; Lesné resigned February 2025.
Why it matters: The case sits at the crossroads of research misconduct, the replication crisis, and subject protection. If a basic preclinical result is faked, every drug trial built on it enrolls human volunteers under false clinical equipoise — and the field’s billions in public funding rest on a mirage.
Pair with: Research Ethics (Protecting Research Subjects)

The Case

In August 2021, Matthew Schrag got a phone call. He was a junior neurologist at Vanderbilt, thirty-seven years old, soft-spoken, with a small sign on his desk reading “Everything is figureoutable.” A colleague wanted to connect him with an attorney. The attorney represented two short-sellers — investors betting that a small biotech, Cassava Sciences, would lose value. Cassava was developing an Alzheimer’s drug called Simufilam. The short-sellers thought the underlying research was fake. Would Schrag, who had already criticized the FDA’s approval of another anti-amyloid drug, agree to examine the published images? He said yes. Over the next months, the attorney paid him about $18,000 (Piller 2022).

What began as a paid side job became something larger. Schrag’s method was unglamorous. He downloaded published figures, zoomed in, and looked for telltale signs that an image had been built in Photoshop rather than captured from a real experiment:

Patches of pixels repeated in places they should not be.
Bands that looked identical when they should be unique.
Edges too clean to come from a working gel.

His targets were Western blots. A Western blot uses electricity to drag protein samples through a gel, separating them by size into stacked bands that are stained and photographed. Two blots from the same true experiment can never look pixel-identical. If they do, someone copied and pasted.

Western blot. A Western blot detects a specific protein in a tissue or fluid sample. The sample runs through a gel that separates proteins by size. The proteins are then moved to a membrane and tagged with antibodies that bind only to the target protein. The result is a photograph of dark bands at certain positions — a visual fingerprint. The exact pattern of background noise and band shape is random, so two real blots should never match.

While studying Cassava, Schrag noticed something far bigger. The same patterns of image trouble appeared in an older and much more famous paper: a 2006 article in Nature, first-authored by Sylvain Lesné. Lesné was then a postdoc at the University of Minnesota, working with the well-known Alzheimer’s scientist Karen Ashe (Lesné et al. 2006). The paper reported the isolation of a specific form of amyloid-beta. The authors called it Aβ56 (pronounced “A-beta star fifty-six”). They claimed that injecting purified Aβ56 into young rats made them fail memory tests they had passed before. In the language of biomedical research, this was a smoking gun. It was a direct claim that a particular amyloid species causes dementia, not merely tracks it.

The amyloid hypothesis. Since the early 1990s, the leading theory of Alzheimer’s has held that sticky protein deposits called amyloid-beta plaques in the brain are a main cause of memory loss. The hypothesis has shaped drug development for thirty years. Hundreds of clinical trials have tested antibodies and small molecules meant to prevent, dissolve, or clear amyloid. Most have failed. The few approved drugs — aducanumab (Aduhelm), lecanemab, donanemab — produce small, disputed benefits and can cause brain swelling or bleeding.

The 2006 paper was cited more than 2,300 times. It launched a generation of follow-up studies. It helped Lesné win his own NIH-funded lab at Minnesota in 2009. And, according to Schrag, dozens of the Western blots in it and in Lesné’s later papers showed signs of digital editing: bands repeated where they should be unique, backgrounds spliced together, edges that looked drawn rather than photographed (Piller 2022).

Schrag was careful. He refused to use the word “fraud” in his public critiques. He could not prove intent. He could not see the original, unedited image files. “I focus on what we can see in the published images, and describe them as red flags, not final conclusions,” he told the journalist Charles Piller. “The data should speak for itself.” But when Science sent his findings to top independent image analysts and senior Alzheimer’s researchers, they agreed with him. The reviewers included:

Elisabeth Bik, a microbiologist and the world’s best-known forensic image consultant.
Donna Wilcock, an Alzheimer’s researcher at the University of Kentucky, who called some of the images “shockingly blatant.”
Thomas Südhof, a Nobel laureate at Stanford.

Bik concluded that the authors “appeared to have composed figures by piecing together parts of photos from different experiments… The obtained experimental results might not have been the desired results, and that data might have been changed to better fit a hypothesis” (Piller 2022).

Piller’s investigation appeared in Science in July 2022 under the title “Blots on a Field?”. It named Schrag publicly for the first time. The fallout came in stages:

Nature issued an expression of concern within days of the article.
In June 2024, the editors and most of the original authors agreed to retract the 2006 paper.
In February 2025, the University of Minnesota concluded its inquiry. It found that Lesné had committed research misconduct, and he resigned his faculty post.

The deeper damage is harder to count. The U.S. National Institutes of Health spent roughly $1.6 billion on amyloid-related Alzheimer’s research in fiscal year 2022 alone — about half the federal Alzheimer’s budget. Südhof’s verdict was blunt: “The immediate, obvious damage is wasted NIH funding and wasted thinking in the field because people are using these results as a starting point for their own experiments” (Piller 2022).

The Aβ*56 affair turned out not to be alone. Over the next three years, Piller and his collaborators documented apparent image manipulation in the work of several other senior figures:

Eliezer Masliah, hired in 2016 to lead the National Institute on Aging’s expanded Alzheimer’s program.
Berislav Zlokovic, a stroke researcher at USC whose work supported a federally funded trial that was later suspended.
Marc Tessier-Lavigne, the president of Stanford University, who resigned in 2023 after a student journalist found altered images in papers from his lab.

Across forty-six leading Alzheimer’s researchers, imaging experts flagged nearly six hundred suspect papers. Those papers had collectively been cited some 80,000 times (Piller 2025).

The question the Aβ*56 affair raises is not only whether Sylvain Lesné faked data. That question is now largely settled. The harder question is what it means for a field, and for the human volunteers downstream of it, when a basic result turns out to have been an “elaborate mirage.”

The Argument Aβ*56 Started

Long before Lesné, two quarrels were already running in Alzheimer’s research: whether the amyloid hypothesis had become a paradigm too dominant to question, and whether fraud that fakes no patient data can still harm patients. The Lesné affair forced the two together. The argument it reignited is about what counts as harm, what counts as evidence, and what a research enterprise owes the people who enroll in its trials.

Schrag’s argument: fraud in preclinical research is a research-ethics violation

Schrag’s public framing is careful, but it contains a sharper ethical claim than his prose makes obvious. He worried from the start that volunteers in Cassava’s Simufilam trials “faced risks of side effects with no chance of benefit.” That worry generalizes. If the preclinical foundation of a drug-development program is faked, the trials built on it cannot meet clinical equipoise. Equipoise is the rule that a community of skilled clinicians must be in honest, good-faith doubt about whether the experimental treatment beats the alternative. Equipoise is what makes it acceptable to expose trial subjects to risk. Without it, the trial is not research. It is human experimentation in the older, condemned sense.

Schrag’s Argument

A clinical trial is ethically permissible only if there is genuine community equipoise about whether the experimental intervention will benefit participants.
Equipoise requires that the preclinical evidence supporting the intervention be honestly produced.
Several anti-amyloid drug programs rely, directly or by citation chain, on preclinical results that appear to have been faked.
Therefore, those trials enroll participants under a false equipoise and violate the conditions for ethical research.

The disputed premise is the second one. Does equipoise depend on the honesty of the preclinical evidence in particular, or on the honesty of the broader scientific community at any given moment? If the field still has other good reasons to believe amyloid matters, perhaps a single faked study does not poison the well. That is where the main reply starts.

The principal reply: science is robust to individual fraud

A defender of the amyloid hypothesis can grant most of Schrag’s facts and still resist his conclusion. The hypothesis does not stand or fall with Aβ*56. The link between amyloid and Alzheimer’s was established long before 2006. Three independent lines of evidence pointed at it:

Pathology. Amyloid plaques were identified in Alois Alzheimer’s original 1906 case.
Genetics. The amyloid-precursor-protein gene was tied to inherited Alzheimer’s in 1991.
Down syndrome. People with an extra copy of that gene develop early Alzheimer’s pathology with striking reliability.

Aβ*56 was one proposed mechanism among many. Losing it does not collapse the hypothesis.

The “Robustness” Reply

The ethics of a clinical trial depends on the overall state of evidence supporting the intervention, not the integrity of any single paper.
The amyloid hypothesis is supported by multiple independent lines of evidence (genetics, pathology, imaging) that do not depend on Lesné’s work.
Therefore, even if Lesné’s specific findings are faked, current anti-amyloid trials need not violate equipoise.

The reply’s soft spot is empirical. How much of the field’s confidence in amyloid actually traveled — by citation and by influence on grant decisions — through the Aβ*56 result and the broader network of suspect papers? If the answer is “most of it,” the reply is weak. If “very little,” it is strong. Piller’s team estimates that roughly six hundred questionable Alzheimer’s papers have been cited around 80,000 times (Piller 2025). That is not noise. It also does not, on its own, prove that the field has been bent.

The counter-reply: structural harm and the therapeutic misconception

The counter-reply does not argue that the field has been bent. It argues that the participants have been wronged regardless. Trial volunteers — especially in early-stage Alzheimer’s, where cognitive decline is the condition being treated — are vulnerable subjects. They suffer from what the bioethicist Paul Appelbaum named the therapeutic misconception. This is the tendency of seriously ill research participants to believe that what is being done to them is being chosen for their personal benefit, the way a doctor’s prescription would be.

Therapeutic misconception. The therapeutic misconception is the false but natural belief that a research protocol is a form of personalized care. Patients in trials commonly do three things at once:

They underestimate the chance of being randomized to placebo.
They overestimate the chance of personal benefit.
They fail to register that the trial’s main purpose is general knowledge, not their own treatment.

The misconception is strongest in conditions like Alzheimer’s, where standard care offers little hope.

The structural harm, in this telling, is not that the trials produced bad medical outcomes. Many anti-amyloid trials are honestly run by investigators who had no idea anything was wrong with the preclinical literature. The harm is that the consent the participants gave was undermined upstream. They were told, in effect, that skilled independent researchers had reason to believe the drug might help. That claim turned out, in part, to be false. The wrong is a wrong of disclosure and of respect for persons. It is not erased by the fact that the trialists were themselves deceived. Here is where Kant’s idea of treating persons as ends, never merely as means, earns its keep. A research subject used to test a drug whose preclinical basis was faked has been used.

A reframing move: Stegenga and medical nihilism

A different kind of critic accepts most of the exchange and then shifts the frame. The philosopher of medicine Jacob Stegenga argues, in his book Medical Nihilism, that we should “have low confidence in the effectiveness of medical interventions” in general (Stegenga 2018). The reason is not that researchers cheat. It is that even honest clinical evidence is shaped by structural problems:

Small effect sizes.
Selective publication of positive results.
p-hacking — running analyses until something looks significant.
Surrogate endpoints that may not track real benefit.
Conflicts of interest from industry funding.
Career pressure to find positive results.

On Stegenga’s view, the Lesné case is not an outlier in an otherwise healthy science. It is a dramatic version of the field’s normal pathology.

The Medical Nihilism Argument

The honest evidence base for most medical interventions is distorted by small effect sizes, surrogate endpoints, selective publication, industry funding, and pressure to find positive results.
Once those distortions are properly discounted, the residual evidence shows that most medical interventions deliver small benefits relative to their harms and costs.
Therefore, we should have low confidence in the effectiveness of most medical interventions.

The disputed premise is the second. Skeptics of Stegenga grant that the distortions are real but deny that discounting for them leaves most interventions looking marginal.

The two anti-amyloid drugs to reach the U.S. market are a test case. The first was Biogen’s aducanumab, sold as Aduhelm. The FDA approved it in June 2021 over the near-unanimous objection of its own advisory committee. Three committee members resigned in protest. Key facts:

The approval used the accelerated approval pathway. This pathway lets a drug reach patients based on a surrogate endpoint — here, the clearance of amyloid plaques from PET scans — rather than direct evidence that patients live longer or think more clearly.
The two pivotal trials had been stopped early for futility.
A post hoc re-analysis of a subset of patients given the highest dose produced a slim positive signal. The FDA accepted it.
Biogen priced the drug at $56,000 per patient per year. The Medicare program raised Part B premiums for every beneficiary in the country to cover the projected cost.
By early 2024, Biogen had withdrawn Aduhelm from the market for commercial failure.

Accelerated approval and surrogate endpoints. Accelerated approval is an FDA pathway that lets a drug for a serious illness reach patients based on a surrogate endpoint — a measurable proxy (a blood test, a scan finding, a biomarker) “reasonably likely” to predict real clinical benefit. The maker must complete confirmatory trials after approval. The pathway speeds access in conditions like cancer and HIV. Its risk is that the surrogate sometimes fails to track what patients actually care about: surviving longer, thinking more clearly, walking without help.

Lecanemab (sold as Leqembi) and donanemab (Kisunla) are the successors. The FDA approved lecanemab in January 2023 and donanemab in July 2024. Their trials were not stopped early. Their data are real. The headline numbers:

They slow cognitive decline by about half a point on the eighteen-point Clinical Dementia Rating Sum of Boxes scale over eighteen months. Many neurologists openly doubt whether that difference matters in daily life.
They cause amyloid-related imaging abnormalities (ARIA) — brain swelling or small bleeds visible on MRI — in roughly one in five treated patients, with rare but documented deaths.
Annual list price runs to roughly $26,500 per patient. That figure does not include infusion, monitoring MRI, or clinic fees. Medicare Part B now covers the drug for patients meeting its criteria, leaving a 20% coinsurance of about $5,300 a year for anyone without supplemental coverage — a reminder that “covered” and “affordable” are different claims.

If Stegenga is right, the bioethical question is not only “how do we catch the next Lesné?” It is “what should we do when even our non-fraudulent medicine routinely overpromises?” The amyloid program looks, on this view, like a field that absorbed faked data so smoothly because the rest of its evidence base was, even at its best, weaker than its advocates claimed. Aduhelm did not require fraud. It required an honest field willing to count amyloid clearance as good enough.

Stegenga’s prescription is what he calls gentle medicine. The default should lean toward less intervention. Aggressive treatment should be reserved for cases where the evidence base clears a high bar:

The Gentle Medicine Argument

When confidence in an intervention’s benefit is low and its harms are non-trivial, prescribing it does more to violate non-maleficence than withholding it does to violate beneficence.
For most medical interventions in their current evidence base, confidence in benefit is low and harms are non-trivial.
Therefore, medicine should default to less intervention, reserving aggressive treatment for cases where the evidence is unusually strong.

A defender of standard practice will press on the second premise. “Most” is doing a great deal of work. And the alternative to prescribing a marginal drug is sometimes prescribing nothing — itself a choice with costs. Some patients would have benefited. Family caregivers carry a heavy burden the drug was meant, however slightly, to ease. The argument does not, on its own, tell a memory-clinic nurse what to say to a frightened sixty-eight-year-old. It does suggest the conversation should start somewhere other than the brochure.

This is not a comforting frame. It implies that better misconduct policing — sharper image-forensics software, mandatory raw-data deposits, stronger journal retraction rules — addresses a real problem but not the deepest one. The deepest problem is the gap between what biomedical research can reliably deliver and what patients, regulators, and journalists are induced to expect from it.

The move that hasn’t ended

The institutional fallout has been steady:

June 2024. The original authors of the Aβ*56 paper agreed to retract it.
September 2024. The NIH found that Eliezer Masliah — until that month the leader of the agency’s Alzheimer’s program — had committed research misconduct. He was removed from his post (Piller 2025).
February 2025. The University of Minnesota concluded its inquiry. Lesné resigned.

None of these moves answers the live question.

The live question is what biomedical research should do now that the tools of image faking and the tools of image forensics are both becoming much more powerful. Generative AI can produce Western blots that look real to the eye. Forensic AI can scan an entire journal’s back catalog for duplications in minutes — Elisabeth Bik did her early work by hand. The arms race is real. It is being run on top of an incentive structure — publish-or-perish, novelty-or-irrelevance, single-laboratory secrecy — that the field has not seriously revised since the 1970s.

The parallel to Tuskegee is instructive but limited. Tuskegee produced direct, identifiable victims and a clear set of failures: no consent, no treatment, no public oversight. The Aβ*56 affair produces diffuse, statistical victims:

Alzheimer’s patients who enrolled in trials that should have been redesigned.
Families who paid out of pocket for drugs that should have been studied differently.
A generation of postdocs whose careers were built on a hypothesis that was overinvested in.

There is no single wronged person to apologize to. There is, instead, a field-level question about what the public is owed when it funds science. Not perfect results, but a system honest enough that its mistakes are recoverable. The question is whether we now have one.

Discussion Questions

Write out Schrag’s Argument and the Robustness Reply in your own words — premises first, then the conclusion. The two arguments share one premise about what makes a clinical trial ethically OK to run. Identify it. Then say which side has the better read on what that premise requires.
The counter-reply says that patients in anti-amyloid trials were wronged even when their own trials were honestly run — because the science behind those trials had been corrupted upstream by Lesné’s fraud. Is this idea of “being wronged” too broad? If you accept it, where does it stop? If you reject it, why?
Stegenga’s medical nihilism says the Lesné case is just an extreme version of how biomedical research normally fails. Push back on this claim. Name one part of biomedical research that you think does work well, and explain why it survives Stegenga’s critique.
You are a nurse in a memory clinic. A patient with mild cognitive impairment asks you whether she should enroll in a new Alzheimer’s drug trial. She mentions news coverage of the “new Alzheimer’s drugs.” She has not read the Science article on the fraud. What do you say to her? What do you not say?
The 2006 Nature paper was retracted in 2024 — eighteen years after it was published, and only after a whistleblower (paid in part by stock-market short-sellers) brought it to public attention. Design one specific reform — at the journal, the university, the NIH, or the FDA — that would have caught the problem earlier. What trade-off would your reform force on the system?

References

Lesné, Sylvain, Ming Teng Koh, Linda Kotilinek, et al. 2006. “A Specific Amyloid-Beta Protein Assembly in the Brain Impairs Memory.” Nature 440 (7082): 352–57. https://doi.org/10.1038/nature04533.

Piller, Charles. 2022. “Blots on a Field?” Science 377 (6604): 358–63. https://doi.org/10.1126/science.add9993.

Piller, Charles. 2025. Doctored: Fraud, Arrogance, and Tragedy in the Quest to Cure Alzheimer’s. Atria/One Signal Publishers.

Stegenga, Jacob. 2018. Medical Nihilism. Oxford University Press.