That result is in. It is negative. And in the months since it was published, the paper reporting it has been placed under formal review.

The hope that led here

Exenatide had one of the better stories in Parkinson's research.

A phase 2 trial published in 2017 followed about 60 people for 48 weeks and found that those on exenatide were, on average, around 4 points better on the motor scale than those on placebo — and, importantly, the gap persisted 12 weeks after everyone stopped the drug. For a field that had watched candidate after candidate fail, this was a real result, and it drove a decade of follow-on work.

It was still a small trial. That is what phase 2 means: a signal worth chasing, not a conclusion. Chasing it properly is exactly what this new trial did.

What they did

The trial enrolled 194 people with Parkinson's across six research hospitals in the UK. Participants were randomly assigned to a weekly injection of 2 mg exenatide or a matching placebo, double-blind, and followed for 96 weeks — close to two years, roughly twice the length of the earlier trial.

The primary endpoint was the motor examination score on the MDS-UPDRS, measured in the practically-defined off state: participants withheld their usual Parkinson's medication overnight before assessment. This is a deliberately demanding choice. It strips out the effect of levodopa and tries to reveal the underlying state of the disease.

What they found

Both groups got worse, at almost exactly the same rate.

The numbers slightly favoured placebo, but nowhere near enough to mean anything — a p-value of 0.47 is what you expect when nothing is happening. The confidence interval is the more useful number here: it comfortably includes zero, and its narrowness means the trial was large enough to rule out the sort of benefit the 2017 study had suggested.

The absence of an effect held across the secondary measures too — non-motor symptoms, quality of life, medication requirements, and imaging. There was no subgroup that clearly benefited. The drug was safe and well tolerated. It simply did not work.

Then the paper was flagged

On 18 May 2026, more than a year after publication, The Lancet issued an expression of concern about the paper.

The reason was not the statistics. Regulators had inspected King's College Hospital NHS Foundation Trust, one of the six sites, and found department-wide problems with how clinical trials there were being conducted, overseen and governed. Some findings were graded critical and others major — the two most serious categories in a good clinical practice inspection.

The Lancet asked the corresponding author's institution to investigate, and attached the notice while that work goes on.

It is worth being precise about what this does and does not mean.

It does not mean the result is wrong. An expression of concern is a flag, not a verdict. It says a question has been raised that has not yet been answered.

It does not mean anyone falsified anything. Inspection findings of this kind more often concern record-keeping, consent documentation, staff training, and oversight than deliberate misconduct.

It does mean the conclusion is provisional. One of six sites is under question. Until the investigation reports, nobody — including the trial's own authors — can say with certainty how much the affected data influenced the result.

Why it matters

A decade-long hypothesis was tested properly. That is how the process is meant to work, and it is worth saying plainly: a negative trial from a well-run study is a real contribution. It stops other groups spending years on the same idea, and it stops patients enrolling in studies of something that does not help.

It shows what phase 2 results are worth. The 2017 trial was not fraudulent or incompetent. It was small, and small trials produce unstable estimates. Chance, a strong placebo response, and the enthusiasm that surrounds an early positive result can combine to produce a promising number that evaporates under scrutiny. This happens constantly, in every area of medicine.

And it shows the correction machinery working in public. An inspection found problems. A journal attached a warning. An investigation is under way, and its outcome will be published. This is slow and unglamorous, and it is the part of science that most rarely gets reported — which is exactly why it is worth reporting.

Reading the two trials together

Several explanations are live, and the honest position is that nobody yet knows which is right.

It may be that assessing people on their usual medication, as the French trial did, is more forgiving — capturing a modest symptomatic effect rather than true disease modification. It may be that 12 months is short enough for a small early difference to appear and 96 weeks long enough for it to wash out. It may be that the two drugs genuinely differ in how much reaches the brain. Or the French result may simply have been a false positive, of the same kind the 2017 exenatide trial now appears to have been.

Until the investigation into the exenatide paper concludes, there is one more possibility that cannot be dismissed: that the negative result is itself unreliable. It is unlikely — one site out of six, in a trial whose result was resoundingly flat — but "unlikely" is not "ruled out".

What this doesn't tell us

Whether the exenatide result will stand. That depends on an investigation that has not reported. If a substantial share of data came from the affected site, the analysis may need redoing.

Whether GLP-1 drugs as a class are finished in Parkinson's. Two drugs, two trials, two opposite answers. That is not enough to close a question.

Whether either drug might work in people at an earlier stage. Both trials enrolled people who already had a diagnosis, by which point a great deal of damage is done.

What to watch next

The investigation's outcome. The Lancet will publish it, and it will either lift the notice, correct the paper, or retract it. That is the single most consequential thing pending in this story.

Whether the lixisenatide finding replicates in a larger trial. That is now the decisive question for this line of research.

And whether trial design shifts earlier. The argument that treatments must start before diagnosis — in the prodromal stage, identified by tests like the seed amplification assay we covered two weeks ago — gets stronger every time a trial in diagnosed patients comes back empty.

This trial tested an unlikely candidate: a drug developed for type 2 diabetes.

Why a diabetes drug?

Lixisenatide belongs to a family called GLP-1 drugs — the same class as semaglutide, now familiar from weight-loss coverage. They were designed to act on the pancreas, but their receptors also appear on nerve cells in the brain.

Over the past fifteen years, laboratory work has suggested these drugs might do something useful there: reduce inflammation, improve how cells manage energy, and help nerve cells survive stress. In animal models of Parkinson's, they protected dopamine-producing cells. An earlier small trial of a related drug, exenatide, produced encouraging results that the field had been arguing about ever since.

The logic of repurposing is straightforward: these drugs are already approved, their safety profile is known from millions of patients, and they are relatively inexpensive. If one of them slowed Parkinson's, it could reach people quickly.

What they did

The trial enrolled 156 people with Parkinson's diagnosed within the previous three years, across a network of hospitals in France, with 78 assigned to each group. Participants were randomly assigned to a daily injection of lixisenatide or a matching placebo injection, and neither they nor the assessing doctors knew which was which — a double-blind design.

Everyone stayed on their normal Parkinson's medication throughout. The trial ran for 12 months, followed by a two-month washout period in which everyone stopped the trial drug.

The primary endpoint was the change in the motor examination portion of the MDS-UPDRS at 12 months, scored while participants were on their usual medication. Higher scores mean worse symptoms.

What they found

The headline is the first row. The placebo group followed the expected trajectory for early Parkinson's: gradual worsening over the year. The lixisenatide group did not measurably worsen at all.

The difference of 3.08 points was statistically significant (95% confidence interval 0.86 to 5.30, p=0.007).

After the two-month washout, when nobody had taken the study drug for eight weeks, the gap was still visible: motor scores off medication were 17.7 in the lixisenatide group against 20.6 on placebo. That detail matters, because it hints the drug had done something lasting rather than simply masking symptoms while it was in the bloodstream. But the confidence interval on that washout difference runs from 0.1 to 5.8 — it only just excludes zero, and a range that wide is consistent with an effect that is real and large, or real and negligible.

The side effects were not incidental. Nearly half the treatment group experienced nausea against 12% on placebo, and 13% vomited against 3%. Some participants needed their dose reduced.

Why it matters

A 3-point difference on a scale that runs past 100 is not something a person would necessarily feel after one year. Its importance is as a proof of principle.

If the effect is real and it continues to accumulate, three points a year compounds. Over five years, that is the difference between independence and dependence for some people. The reason to take this seriously is not the size of the gap at 12 months but the possibility that the underlying process was slowed.

There is also a strategic argument. GLP-1 drugs are manufactured at enormous scale and are already in the world's pharmacies. A repurposed treatment does not need the decade a novel compound requires.

What this doesn't tell us

Whether the drug slowed the disease or improved symptoms. This is the crux. A drug that makes symptoms better looks identical, on a rating scale, to a drug that slows the damage — for a while. The washout result leans toward genuine modification, but two months is a short washout for a disease that moves over decades.

Whether it works in people further along. Everyone here was within three years of diagnosis. Results in early disease often do not transfer to people who have had Parkinson's for fifteen years.

Whether the tolerability is workable. A trial can support participants through nausea with close monitoring. Sustaining a daily injection with that side-effect profile for years, in ordinary life, is a different proposition.

Whether it holds up at scale. 156 participants is a reasonable phase 2 trial. It is not a phase 3 trial. The history of Parkinson's research is full of promising mid-size trials that did not survive a larger one — as next week's issue shows.

What to watch next

The obvious next step is a larger, longer trial — more participants, more sites, several years rather than one.

The wider question is whether this is a class effect. Lixisenatide is one of several GLP-1 drugs. Exenatide has been tested in Parkinson's for years, and results from a larger trial of it are due. If the whole class helps, the case becomes much stronger. If lixisenatide works and exenatide does not, the field has a puzzle.

This study is a serious attempt to change that. Researchers ran a laboratory test on cerebrospinal fluid from more than a thousand people and asked a simple question: can we detect the disease itself, rather than the symptoms it causes?

What the researchers were trying to find out

Parkinson's is defined biologically by clumps of misfolded alpha-synuclein protein inside nerve cells. Until recently, the only way to confirm those clumps were present was to examine the brain after death.

A newer laboratory technique, the seed amplification assay, offers a way around that. The principle is borrowed from COVID-era PCR testing. You take a sample of a person's spinal fluid, add a supply of normal alpha-synuclein protein, and wait. If even a trace of the misfolded form is present in the sample, it acts as a seed: it converts the added protein into the same wrong shape, over and over, until the clumping becomes detectable by an instrument. A tiny signal is amplified into an unmistakable one.

The question was whether this worked reliably enough, in enough people, to be useful.

What they did

The team used samples from the Parkinson's Progression Markers Initiative, a long-running international study that has been collecting fluid, scans, and clinical data from the same participants for years. That backdrop matters: these were not anonymous samples, but people whose diagnoses and symptoms were documented in detail.

They ran the assay on 1,123 participants, of whom 1,104 gave a clear result, across several groups:

• people with a clinical diagnosis of Parkinson's
• healthy volunteers of similar age
• people carrying Parkinson's-linked gene variants such as LRRK2 and GBA1, some with symptoms and some without
• people in the prodromal stage — no diagnosis, but warning signs such as REM sleep behaviour disorder or a lost sense of smell

Each sample was scored positive or negative, and the results were compared against what was already known about each person.

What they found

Three things stand out.

The test is accurate in typical Parkinson's. For people with the usual presentation — particularly those who had also lost their sense of smell — it was positive 98.6% of the time. Fewer than four in a hundred healthy volunteers tested positive.

It sees the disease before diagnosis, but only once warning signs have started. Of 51 people who had REM sleep behaviour disorder or a lost sense of smell without a diagnosis, 44 tested positive. Their brains already carried the biological signature.

That is different from simply carrying a risky gene. Among 310 people who had a Parkinson's-linked gene variant but no symptoms of any kind, only 25 — 8% — tested positive. The assay is not a crystal ball for anyone with a family history. It appears to turn positive somewhere between carrying risk and showing the first subtle signs, which makes when it flips the interesting question.

Genetic Parkinson's behaves differently. Roughly a third of people with LRRK2-related Parkinson's tested negative despite having the disease. That is not a flaw in the test so much as a finding in its own right: it suggests some people arrive at Parkinson's through a route that does not involve the same alpha-synuclein clumping. "Parkinson's disease" may turn out to be several biologically distinct conditions wearing similar clothes.

Why it matters

The most immediate consequence is for research rather than for the clinic.

Every trial of a disease-modifying treatment faces the same problem: by the time someone is diagnosed, a large share of their dopamine-producing cells in the substantia nigra is already gone. Protecting cells that have already died is not possible. If a biological test can identify people years earlier, trials can start earlier, when there is more left to save.

It also helps with a quieter problem. Some people enrolled in Parkinson's trials do not actually have Parkinson's — they have a condition that resembles it. Those participants dilute the results and can make a working drug look useless. A biological entry test would tighten the enrolment.

What this doesn't tell us

This was a cohort study carried out at a single point in time, not a trial. Several limits are worth holding onto.

It does not measure severity or predict speed. A positive result says misfolded protein is present. It does not say how fast anyone will progress, or how badly they will be affected.

It does not prove the prodromal participants will all develop Parkinson's. They have the biological marker and known warning signs. Following them forward over years is the only way to learn what share actually convert, and how long it takes.

And that group was small. Fifty-one people is enough to be interesting and not enough to be certain: shift a handful of results either way and 86% becomes 78% or 94%. The headline number deserves less weight than the direction it points in.

And it does not yet work outside a research setting. The people running these assays were specialists using standardised samples. Reproducing that in ordinary hospital laboratories is a separate undertaking.

What to watch next

Three threads are worth following.

The first is the move away from lumbar punctures. Groups are working on versions of this assay that read skin biopsies, blood, and nasal swabs. A blood-based version would change the picture entirely.

The second is longitudinal follow-up. The prodromal participants who tested positive are being tracked. What happens to them over the next five to ten years is the real test of whether this predicts anything.

The third is how trials adapt. Expect to see new studies that require a positive assay to enrol — and expect some earlier failed trials to be reconsidered in light of who was actually in them.