Liraglutide cut Alzheimer's cognitive decline 18% — and the oral version failed the same week.

The headline numbers, accurately

The ELAD trial (Evaluating Liraglutide in Alzheimer's Disease) reported its phase 2b results in Nature Medicine in December 2025, with broader coverage following into 2026 [Edison 2026]. The trial randomized 204 patients with mild-to-moderate Alzheimer's disease across 24 UK clinical sites to once-daily subcutaneous liraglutide (1.8 mg) or matched placebo for 52 weeks, on top of standard of care.

The primary endpoint was the change in cerebral glucose metabolic rate — measured by FDG-PET in pre-specified brain regions — which did not reach statistical significance on its own. The headline that travelled the field came from the secondary and exploratory endpoints:

• Cognition: Liraglutide-treated patients showed an 18% slower decline in cognitive function at 52 weeks versus placebo, measured by the Alzheimer's Disease Assessment Scale — executive function composite (ADAS-Exec z-score) [Edison 2026].
• Brain volume: Voxel-based morphometry analysis showed nearly 50% less reduction in whole cortical grey matter, frontal lobe, temporal lobe, and parietal lobe volume in the liraglutide arm versus placebo [NeurologyLive 2025].
• Tolerability: Liraglutide was well tolerated in the Alzheimer's population, with an adverse-event profile consistent with the molecule's prior diabetes and obesity use.

Those are real numbers from a registered, double-blind, placebo-controlled trial — not a press-release effect size. The primary endpoint did not hit, which is genuinely the most honest single sentence anyone can write about this readout. The cognitive and structural-MRI signals from the secondaries are what is generating attention, and they are the part of the story that may or may not survive replication.

Inside the ELAD trial design

ELAD was led by Professor Paul Edison at Imperial College London, with funding from Alzheimer's Society UK, Alzheimer's Drug Discovery Foundation, and the John and Lucille van Geest Foundation. It ran for 52 weeks across 24 clinical sites, with patients receiving daily subcutaneous injections — the same delivery format used in liraglutide's approved diabetes (Victoza, 1.8 mg) and weight-loss (Saxenda, 3.0 mg) indications [Edison 2026].

The trial used FDG-PET (fluorodeoxyglucose positron emission tomography) for the primary endpoint because cerebral glucose hypometabolism is a well-characterized Alzheimer's pathology — neurons in affected regions progressively lose their ability to take up and use glucose, and the rate of that loss correlates with disease progression. The hypothesis was that liraglutide, which improves peripheral insulin signaling, would translate that effect into neuronal energy metabolism. The signal was directionally consistent with the hypothesis but did not clear the statistical threshold the trial was powered for.

The cognitive endpoint used the ADAS-Exec composite, which weights executive function tasks more heavily than the older ADAS-Cog scale. This matters because executive function is one of the domains that declines earliest and most steeply in mild-to-moderate Alzheimer's, and where small absolute changes translate into meaningful real-world differences in independence. An 18% reduction in slope over 52 weeks is not a cure — but it is the difference between a patient being able to manage their own medication schedule for another year and not.

Why oral semaglutide failed at the same moment

The contrast that makes the ELAD readout meaningful is what happened two months earlier. Novo Nordisk had two phase 3 trials — EVOKE and EVOKE+ — testing once-daily oral semaglutide (14 mg) against placebo in early Alzheimer's disease, enrolling 3,808 patients combined [Novo Nordisk 2025]. The trials measured change in the Clinical Dementia Rating — Sum of Boxes (CDR-SB) at 156 weeks. Topline readout came in November 2025; Novo Nordisk announced the program's discontinuation alongside the results.

Both trials failed to demonstrate superiority over placebo on the primary clinical endpoint. The biomarker secondary endpoints did improve — plasma p-tau217 and neurofilament light chain (NfL) both moved in the direction one would expect if semaglutide were affecting the underlying pathology. But that biomarker signal did not translate into clinical benefit on the CDR-SB [Alzheimer Europe 2026]. The biomarker–clinical disconnect is the structurally interesting finding from EVOKE, and it raises the same question that haunts the anti-amyloid antibody class: which biomarkers actually predict slowing of decline, and which are noise?

Two GLP-1 receptor agonist readouts. Same drug class. Same indication. One hit the cognitive secondary; one missed the cognitive primary. What differs between them is not the receptor — it is the molecule, the route, and the trial design.

Two GLP-1 readouts. Same class, same indication. One hit a secondary; one missed. The difference is the molecule, the route, and the dose getting into brain tissue.

The mechanism — what GLP-1 actually does in the brain

GLP-1 (glucagon-like peptide-1) receptors are widely expressed in the central nervous system, including the hippocampus, hypothalamus, cortex, and brainstem. Activating those receptors does several things at once, and the longevity-of-brain hypothesis is that the combination matters more than any single mechanism in isolation [Holscher 2024]:

• Neuronal insulin signaling. Alzheimer's brains show progressive insulin resistance in neurons — sometimes informally called "Type 3 diabetes." GLP-1 receptor activation restores downstream insulin signaling (IRS-1 / Akt / GSK-3β) in preclinical models, including in non-human primates [Batista 2018].
• Neuroinflammation dampening. GLP-1 receptor activation modulates microglial polarization, reducing pro-inflammatory cytokine release (TNF-α, IL-1β, IL-6) and shifting the local environment toward repair rather than ongoing damage.
• Amyloid and tau effects. In rodent and primate models of Alzheimer's, chronic liraglutide reduces both amyloid plaque burden and tau hyperphosphorylation. The translation to human pathology is still being worked out.
• Synaptic plasticity. Liraglutide preserves long-term potentiation (LTP) in the hippocampus of disease models — the cellular correlate of learning and memory.
• Cerebral glucose handling. The ELAD trial's primary endpoint rested on this mechanism. Even though FDG-PET didn't hit significance, the structural and cognitive signals suggest something downstream of glucose handling is being preserved.

The mechanism story is plausible, multi-layered, and supported by decades of preclinical work — Christian Hölscher's lab in particular has been developing it since the late 2000s [Holscher 2024]. What was missing until ELAD was a human readout that showed any of this mattered clinically. The trial doesn't close the question, but it does provide the first signal that the preclinical pharmacology wasn't pure mouse fantasy.

Why delivery route is not a footnote

It is tempting to read the ELAD-vs-EVOKE split as a story about molecules — liraglutide vs semaglutide. That reading misses what is probably the more important structural variable: route of administration.

Oral semaglutide (Rybelsus, the formulation tested in EVOKE) achieves only ~1% oral bioavailability, even with the SNAC (sodium N-(8-(2-hydroxybenzoyl)amino)caprylate) absorption enhancer that allows the peptide to survive the gut and partially cross the gastric epithelium. The 14 mg oral dose used in EVOKE produces steady-state plasma exposure roughly comparable to once-weekly 1.0 mg injectable semaglutide — i.e., enough for glycemic and weight endpoints, but the variability between patients is much higher than with injectable forms.

Injectable liraglutide bypasses the gut entirely. It produces more predictable plasma exposure, less inter-patient variability, and — importantly for a CNS indication — a more stable concentration at the blood-brain barrier over the day. The ELAD protocol's once-daily 1.8 mg subcutaneous dose delivers a saturating exposure that the oral version, by formulation chemistry, cannot match.

This is the same lesson that runs across the peptide field: the delivery route isn't a convenience trade-off. It's part of why the molecule works at all. The "needle problem" is, in the relevant physiological sense, why these peptides actually get to where they need to act.

Blood-brain barrier penetration: liraglutide vs semaglutide

Here the literature splits into two readings, and they're both worth knowing. The preclinical CNS-pharmacology work, summarized in a 2025 narrative review in Neurology and Therapy, concluded that semaglutide actually crosses the blood-brain barrier more efficiently than liraglutide on a molar basis — semaglutide is larger but its albumin-binding chemistry gives it better CNS penetration in rodent and non-human primate models [Salameh 2025].

That makes the EVOKE failure harder to attribute to "the drug didn't get into the brain." A more defensible reading is that EVOKE's oral formulation didn't deliver enough drug into circulation to saturate CNS receptors at the right time of day — and that the trial's primary endpoint (CDR-SB at 156 weeks) was structurally harder to move than the secondary endpoints ELAD hit. CDR-SB requires a clinically meaningful change in functional staging, which most disease-modifying Alzheimer's trials have failed on regardless of mechanism.

The cleanest version of the comparison: ELAD demonstrated that a well-characterized injectable GLP-1 agonist, in a 52-week trial powered for biomarker endpoints, can produce cognitive and structural signals consistent with disease modification. EVOKE demonstrated that an oral GLP-1 agonist, in a 156-week trial powered for clinical endpoints, could not. Both are true. They are not contradictory.

What this readout does and doesn't say

ELAD is a phase 2b trial, not a phase 3. The cognitive and structural endpoints that produced the headline numbers were secondaries — they were not the prespecified primary the trial was statistically powered on. Secondary endpoints in trials whose primary missed are hypothesis-generating, not confirmatory. The honest position is that the readout justifies a properly powered phase 3, not regulatory approval.

The patient population matters. ELAD enrolled mild-to-moderate Alzheimer's, which is a different question from early-stage prodromal disease (the population EVOKE targeted) and from cognitive complaint without imaging-confirmed pathology. Any extrapolation across these groups is doing more work than the data supports.

The trial does not establish that injectable liraglutide should be prescribed off-label for Alzheimer's prevention. The risk-benefit calculus in non-demented patients with no diabetes or obesity indication is not what ELAD measured. The patients in ELAD met clinical criteria for active disease and received structured monitoring across a 52-week protocol — that's the population the 18% number is from.

What it means for the field

Three things, all worth holding at once.

First: the GLP-1 class is not dead in neurodegeneration, but the oral pivot probably is — at least for Alzheimer's. The EVOKE failure should be read as a failure of the oral formulation paradigm for a CNS indication, not as a failure of the receptor. The pharma incentive will likely move toward injectable once-weekly molecules (semaglutide subcutaneous, tirzepatide) and toward dual or triple agonists (GLP-1/GIP, GLP-1/GIP/glucagon) that the preclinical work suggests cross the blood-brain barrier more efficiently than liraglutide.

Second: the disease-modification framing for Alzheimer's now has a non-antibody candidate worth taking seriously. Until ELAD, the only readouts producing positive clinical signals in Alzheimer's were the anti-amyloid antibodies — lecanemab, donanemab — with their attendant ARIA risk, IV infusion logistics, and patient selection requirements. A subcutaneous, well-characterized peptide with a 20-year safety record in millions of diabetic and obese patients is a structurally different proposition.

Third: the mechanism-versus-clinical gap that EVOKE surfaced — biomarker improvement without clinical benefit — is going to remain a problem for this field. Plasma p-tau217, NfL, and amyloid PET signals are useful trial-enrollment tools, but their predictive value for clinical slowing is still being worked out. Future GLP-1 Alzheimer's trials will probably need to pre-register both mechanistic and clinical primaries, not pick one and hope the other follows.