GLP-1 and the Brain: What Does the Evidence Actually Say About Neurodegeneration?
GLP-1 and the Brain: What Does the Evidence Actually Say About Neurodegeneration?
For several years, GLP-1 receptor agonists looked like one of the more plausible ways to slow neurodegeneration. They were already approved, broadly safe, and backed by preclinical work and large observational datasets. By 2026, that hypothesis has been tested in phase 3 trials for both Parkinson's and Alzheimer's disease. In both, the drugs did not slow progression. The observational signals have not gone away, though. Making sense of the field means keeping three kinds of evidence apart: what the population data show, what the trials in established disease show, and what we still do not know about how much of these drugs reaches the brain.
Where the optimism came from
The early human signal came mostly from people with type 2 diabetes. One analysis pooled three double-blind, placebo-controlled cardiovascular outcome trials covering 15,820 patients and found a lower rate of dementia with GLP-1 receptor agonist treatment than with placebo (HR 0.47, 95% CI 0.25–0.86). In the same paper, a Danish nationwide registry cohort of 120,054 patients with type 2 diabetes showed a hazard ratio of 0.89 (0.86–0.93) for each additional year of exposure (Nørgaard et al., 2022). Because the pooled trials were designed around cardiovascular outcomes rather than dementia, the result generates a hypothesis rather than confirming one.
Semaglutide attracted the most attention. A target trial emulation in US electronic health records (TriNetX) compared 17,104 new semaglutide users with new users of other diabetes drugs, all with type 2 diabetes. Over up to three years, semaglutide was associated with fewer first-time Alzheimer's diagnoses: HR 0.33 (0.21–0.51) against insulin and 0.59 (0.37–0.95) against other GLP-1 receptor agonists (Wang et al., 2024). The authors list the limitations themselves. These are retrospective data that cannot establish causation, Alzheimer's was identified by an ICD-10 code rather than biomarkers, follow-up was short, and adherence was not recorded.
The observational studies all have the same features: the participants had metabolic disease, the outcome was a new diagnosis, and treatment was not randomised. Each of those turns out to matter.
Parkinson's disease: a promising phase 2, a negative phase 3
The Parkinson's story begins with exenatide. In a single-centre phase 2 trial, 62 patients were randomised to weekly exenatide or placebo for 48 weeks, followed by a 12-week washout. At 60 weeks, off-medication motor scores (MDS-UPDRS part III) favoured exenatide by 3.5 points (95% CI −6.7 to −0.3; p=0.0318). The authors could not tell whether this reflected disease modification or a lasting symptomatic effect (Athauda et al., 2017).
Lixisenatide followed. In a French phase 2 trial of 156 people with early Parkinson's, MDS-UPDRS III scores at 12 months were essentially unchanged on lixisenatide (−0.04 points) but worsened by 3.04 points on placebo, a difference of 3.08 (95% CI 0.86–5.30; P=0.007) (Meissner et al., 2024). The authors called the result preliminary. The primary outcome was measured on medication, the secondary endpoints did not clearly support it, and gastrointestinal effects were common: nausea occurred in 46% of the lixisenatide group compared with 12% on placebo.
The definitive test was Exenatide-PD3. In this UK trial, 194 people with Parkinson's were randomised to weekly extended-release exenatide 2 mg or placebo for 96 weeks. Off-medication MDS-UPDRS III scores worsened by 5.7 points on exenatide and 4.5 points on placebo (adjusted difference 0.92, 95% CI −1.56 to 3.39; p=0.47). The investigators concluded that they "found no evidence to support exenatide as a disease-modifying treatment" and called for trials of agents with stronger target engagement (Vijiaratnam et al., 2025).
Alzheimer's disease: biomarker movement, no clinical benefit
In Alzheimer's, the phase 2b ELAD trial randomised 204 people with mild-to-moderate disease, excluding anyone on diabetes treatment, to daily liraglutide or placebo for 52 weeks. It missed its primary endpoint, cerebral glucose metabolism on FDG-PET (P=0.14). Exploratory MRI analyses showed less loss of total grey matter volume with liraglutide (7,274 mm³, 95% CI 2,704–11,845), and an executive-function score favoured treatment. These analyses were not adjusted for multiple comparisons, and CDR-SB did not differ between groups. The authors say these findings need to be confirmed in larger studies (Edison et al., 2025).
The phase 3 answer came from evoke and evoke+. Together the two trials randomised 3,808 people aged 55–85 with amyloid-confirmed mild cognitive impairment or mild Alzheimer's dementia to oral semaglutide (up to 14 mg daily) or placebo. At week 104, the change in CDR-SB was effectively identical between arms in both trials. The differences were −0.08 (95% CI −0.35 to 0.20) in evoke and 0.10 (−0.17 to 0.38) in evoke+ (Cummings et al., 2026).
The biomarker data are what make the result interesting rather than simply negative. In a CSF substudy of 199 participants, semaglutide lowered several markers by up to about 10% compared with placebo, including p-tau181, p-tau217, total tau, the astrocyte marker YKL-40 and the synaptic marker neurogranin. These analyses were not adjusted for multiplicity. Systemic hsCRP fell in both trials, although baseline levels were already normal. Plasma GFAP, however, rose with treatment, and plasma p-tau did not change. None of this translated into slower clinical decline. Treatment was also harder to tolerate: about 17% of the semaglutide group stopped because of adverse events, compared with about 8% on placebo (Cummings et al., 2026).
Why the observational and trial results disagree
The following are explanations rather than established findings, but each is testable.
Prevention is not treatment. The observational studies measured new diagnoses in people without dementia, followed over years. The trials enrolled people who were already symptomatic and had confirmed amyloid pathology. A drug could lower risk years before diagnosis and still do nothing once neurodegeneration is well underway.
The populations differ. The positive signals came from people with type 2 diabetes, where some of a GLP-1 drug's effect on dementia risk could plausibly run through better glycaemic, vascular or weight control. ELAD excluded people treated for diabetes, and evoke recruited on the basis of amyloid status, not metabolic disease.
It is unclear how much drug reaches the brain. In rodents, labelled semaglutide reached the hypothalamus, brainstem and septal nucleus through the circumventricular organs and periventricular regions, but did not cross the blood–brain barrier (Gabery et al., 2021). That is enough for appetite control. It does not show that hippocampal or nigral neurons see meaningful drug concentrations, and this is a rodent finding that has not been quantified in humans. The Exenatide-PD3 authors' call for agents with stronger target engagement points at the same gap.
Residual confounding. Even well-designed target trial emulations compare people whose clinicians chose different drugs for them. Access to newer drugs, healthcare use and diagnostic practice can all differ between those groups, and propensity matching can only adjust for what was recorded.
What would move the field forward
In our view, the negative trials narrow the GLP-1 hypothesis rather than end it. Four questions now seem to matter most:
- Does GLP-1 receptor agonism reduce incident dementia when tested prospectively as prevention, ideally in people at metabolic risk, where the observational signal is strongest?
- Can central target engagement be measured in humans, so a negative trial can be read as "the mechanism doesn't work" rather than "the drug didn't get there"?
- Are there biologically defined subgroups, for example by metabolic status or inflammatory profile, in whom an effect exists?
- Why did CSF tau and glial markers move in evoke without clinical benefit? Is the effect too small, too late, or unrelated to the processes driving decline?
The bottom line
Here is the current state of the evidence. Randomised trials in established Parkinson's and early symptomatic Alzheimer's have not shown that GLP-1 receptor agonists slow progression. Large observational studies in people with type 2 diabetes still link these drugs to lower rates of dementia diagnosis, but they cannot establish cause. Biomarker shifts in the semaglutide trials show that the drug does something measurable in the central nervous system. What that is, and whether it matters clinically, has not been established. Prevention remains an open and testable question.
Explore Purna's Molecular Intelligence Platform
AI-powered workspace for biology teams to accelerate drug discovery from target identification to lead optimization.
Try Purna AI →