Beyond Magic Bullets: Alzheimer’s Needs the Right Cocktail and Timing
Modest GLP-1 signals, negative phase 3s, and amyloid drugs all point toward combination therapy. Trigger warning for antivaxxers: an Alzheimer's vaccine! Read on...
Notes: this is probably the last of these longer discursive pieces I’m going to do here for a while - I am going to change the focus in favour of shorter pieces revolving around interesting (to me, at least) research papers, given the wealth of research being published at present. My cadence here will be about three or so times per month.
The reason for this change is that I am undertaking a ‘writing a book in public’ project, with a new identity and focus: it’s called Cognitive Republic: Building Societies That Experiment, Learn, and Evolve (click this link for more). The Cognitive Republic reimagines democracy as an evolving intelligent system: one that perceives, remembers, learns, and adapts, where policy is treated as hypothesis; failure is feedback; civic life becomes collective inquiry.
The point is not to add another political ideology to the pile. The point is to build a new method—an operating system—for political self-correction at scale. Please consider subscribing - just click here.
Alzheimer’s Disease
Alzheimer’s has finally entered the era of drugs that do something. Anti-amyloid antibodies such as lecanemab and donanemab clear amyloid from the brain and slow decline on clinical scales by a very modest but real amount - at least in a certain fraction of patients.
However, based on recent data, I’m thinking we should not be looking for a single miracle drug for Alzheimer’s, and we should probably admit we’re dealing with a cocktail-and-timing problem: several partially effective treatments, aimed at different parts of the disease process, used early and in combination. And we should consider other broader approaches - to make the anti-vaxxers even more nuts, there are now good data pointing towards a possibly efficacious Alzheimer’s vaccine - discussed here also.
Alzheimer’s, GLP-1 drugs, and why a small win still matters
Drugs such as lecanemab and donanemab have one target: amyloid. The new Nature Medicine published ELAD trial and Novo Nordisk’s evoke/evoke+ programme shine a spotlight on a completely different pathway: metabolism and insulin signalling in the brain, via GLP-1 receptor agonists originally developed for diabetes and obesity.
Punchline:
Liraglutide: a mid-sized trial, no spectacular win, but a modest hint of cognitive benefit.
Semaglutide: large trial, negative on clinical outcomes, despite nice-looking biomarkers.
That combination tells us something about why some drugs work, some don’t – and why a small positive signal from a non-amyloid drug is still worth paying attention to.
what ELAD found
The ELAD trial (“Evaluating liraglutide in Alzheimer’s disease”) enrolled 204 people with mild to moderate Alzheimer’s, all without diabetes. Participants were randomised to daily liraglutide or placebo injections for 52 weeks, with brain imaging and a detailed cognitive test battery.
Key results:
Primary endpoint – brain metabolism: No significant differences in change in cerebral glucose metabolic rate on FDG-PET between liraglutide and placebo (estimated difference −0.17; 95% CI −0.39 to 0.06; P = 0.14).
Secondary cognition – executive function: On the Alzheimer’s Disease Assessment Scale – Executive domain (ADAS-Exec), patients on liraglutide did better than placebo (effect estimate 0.15; 95% CI 0.03–0.28; unadjusted P = 0.01).
Functional outcomes: No significant differences on activities of daily living (ADCS-ADL) or global Clinical Dementia Rating (CDR).
Safety: Liraglutide is generally safe and well tolerated in this non-diabetic Alzheimer’s population.
The primary brain imaging endpoint was negative, and the cognitive benefit appears on an executive-function composite, not the global clinical scales regulators care most about. The signal is small and sits in secondary outcomes. But it is a real, pre-specified outcome, consistent with what you might expect from a drug that tweaks insulin signalling, synaptic function, and neuroinflammation in frontal brain networks rather than directly dissolving amyloid plaques.
what evoke/evoke+ showed for semaglutide
Novo Nordisk’s evoke and evoke+ trials were much bigger and much more definitive, and the news was clearly negative.
Population: 3,808 people aged 55–85 with either mild cognitive impairment or mild dementia due to Alzheimer’s, all with confirmed amyloid positivity.
Design: Global, randomised, double-blind, placebo-controlled phase 3 trials of once-daily oral semaglutide 14 mg versus placebo, on top of standard care, for 156 weeks (104-week main phase plus planned 52-week extension).
Primary endpoint: Change in CDR-SB from baseline to week 104.
Top-line result: No statistically significant reduction in progression of Alzheimer’s disease on CDR-SB for semaglutide versus placebo.
Biomarkers and safety: Semaglutide improved Alzheimer’s-related biomarkers, but this did not translate into clinical benefit; safety was in line with existing diabetes and obesity trials.
Semaglutide failed the one test that matters – how slowly people decline.
same family of drugs, different behaviour: why?
Liraglutide and semaglutide both activate the GLP-1 receptor. They both have strong metabolic effects. They both have encouraging preclinical data in neurodegeneration models.
So why might one show a small cognitive signal while the other fails outright?
pharmacology and brain access
GLP-1 drugs are not all the same from the brain’s point of view.
Semaglutide is a long-acting drug with a half-life of about a week and which stays in the circulation for several weeks after the last dose. Preclinical work suggests that GLP-1 analogues like semaglutide have minimal blood–brain barrier penetration, with effects mediated via regions where the barrier is leaky (hypothalamus, brainstem, periventricular structures), rather than uniform exposure across cortex and hippocampus.
Liraglutide is also long-acting compared to native GLP-1, but it is given once daily and reaches different peak/trough patterns. Imaging and experimental work in animals and humans suggest modest but measurable CNS exposure, and there are data supporting direct neurotrophic and neuroprotective actions in hippocampal and cortical neurons.
From that perspective, it is not entirely surprising that:
A drug optimised to sit in the bloodstream for a week at a time for diabetes and obesity might have excellent peripheral effects but limited cortical exposure.
A shorter-acting analogue with somewhat better brain entry might have just enough central engagement to nudge executive function, without being strong or widespread enough to move global disability scales in a one-year trial.
target engagement vs. clinical effect
Semaglutide appears to have done something biologically relevant – Alzheimer’s-related biomarkers shifted in the right direction – yet CDR did not improve.
The semaglutide data simply underline a rule: shifting a biomarker is not enough. For amyloid drugs, we now know that large, sustained shifts in plaque burden do correlate with slower decline, but with effect sizes smaller than hoped for. And for non-amyloid mechanisms, we don’t know biomarkers truly predict meaningful outcomes and which do not.
stage of disease and length of treatment
ELAD enrolled mild–to-moderate Alzheimer’s; evoke/evoke+ enrolled people earlier, with MCI or mild dementia. You might think earlier treatment should be better. However:
In more advanced disease, small symptomatic effects can sometimes show up more clearly because there is more room to worsen; stabilising or slightly improving can look impressive against a steep downhill trajectory.
In very early AD, global scales like the CDR change slowly; detecting a modest drug effect over two years demands enormous samples and very low noise.
Finally, in both programmes, the treatment windows may still be too short to see the full impact of metabolic tinkering in a neurodegenerative disease that unfolds over decades.
why a non-amyloid signal still matters
It would be easy to look at semaglutide’s failure and shrug off the whole GLP-1 idea. That would be a mistake.
A few points:
Different biology, different therapeutic bets. Amyloid drugs target the “classic” pathology. GLP-1 agonists act on insulin signalling, mitochondrial function, synaptic plasticity, and neuroinflammation – a very different slice of Alzheimer’s biology.
The ELAD signal is modest, but it is in the right ballpark. An executive-function benefit, in a trial powered primarily for imaging, is exactly the sort of weak but intriguing result that often precedes more carefully designed follow-ups. It’s not clinically practice-changing; it is hypothesis-sustaining.
Non-amyloid wins would matter disproportionately. If any non-amyloid drug demonstrates a clear, replicated clinical benefit, it will:
Validate alternative pathways as genuine levers on disease expression.
Open the door to combination strategies: modest amyloid slowing plus modest metabolic or inflammatory modulation could add up to something more.
The next wave is already in motion. Dual GLP-1/GIP agonists and other “multi-incretin” compounds have been engineered with neuroprotection in mind and show stronger effects in animal models of Alzheimer’s and Parkinson’s, partly by improving brain access and partly by engaging more than one receptor system.
A final point: it may be (probably is) imho that these GLP-1 drugs are administered too late - like taking a statin or antihypertensive after the onset of heart problems. Better to take them long ahead of time.
where this leaves Alzheimer’s treatment
Right now, the Alzheimer’s landscape looks something like this:
Amyloid-targeting antibodies: clear amyloid, modestly slow decline in early disease, with meaningful side-effects and access questions.
GLP-1 drugs: strong biological plausibility, promising animal data, and now:
One mid-sized trial (liraglutide) with a small domain-specific cognitive signal and good safety.
A definitive trial (semaglutide) that improved biomarkers but did not slow clinical progression.
ELAD is a hint, not a breakthrough.
evoke/evoke+ is a boundary condition on what one long-acting GLP-1 analogue can do in early Alzheimer’s.
If we want non-amyloid therapies to work, getting the pharmacology and brain access right is not optional. If the next generation of brain-penetrant, multi-target drugs can turn metabolic and inflammatory biology into robust clinical benefits – especially in combination with amyloid and tau therapies – that will be the real test of whether this line of thinking is up to patient treatment.
beyond single targets: Alzheimer’s will almost certainly need combination therapy
There’s also a more basic point we keep forgetting: almost no complex disease in medicine is ultimately treated with a single magic bullet.
Hypertension, HIV, tuberculosis, many cancers: in all of these, we ended up with polydrug regimens that hit different parts of the system at once. One drug handles viral load, another blocks resistance pathways; one drug lowers blood pressure via the kidneys, another via vascular tone. The whole is greater than the sum of the parts.
Alzheimer’s is at least as complicated as any of those.
Instead of asking whether one amyloid drug, or one tau drug, or one GLP-1–type metabolic drug can carry the entire load, the more realistic question is:
What combination of partially effective levers gets us the largest overall benefit, especially in early disease?
In practice, that might mean targeting several fronts at once:
Amyloid (to shift the upstream pathology that starts the cascade).
Tau (to stabilise structures tracking clinical decline).
A metabolic/GLP-1-type lever (to improve insulin signalling, energy use, inflammation, and synaptic resilience in vulnerable networks).
None of these has to be a miracle on its own. If each moves the slope of decline a little, and if they do so via partly independent mechanisms, the combined effect on a person’s lived trajectory could be substantial, especially if started early, before extensive, irreversible brain damage has accumulated.
Alzheimer’s might turn out not to be a “find the right target” problem at all, but a “get the right cocktail and timing” problem.
Vaccination as an upstream lever: shingles and dementia across the disease course
A second non-amyloid, non-metabolic lever has come from an unexpected direction: shingles vaccination (trigger warning: antivaxxers look away now).
In a large natural experiment in Wales, Eyting and colleagues exploited a sharp date-of-birth eligibility rule for the live-attenuated herpes zoster (HZ) vaccine Zostavax. People born just before 2 September 1933 were ineligible for the vaccine; those born just after were eligible for one year, even though the two groups differed in age by only a few days. Using this quasi-randomisation, the authors found that receiving the shingles vaccine reduced the probability of a new dementia diagnosis over seven years by about 20% relative risk (3.5 percentage points absolute; 95% CI 0.6–7.1), with stronger effects in women.
A new Cell paper by Xie, Eyting, Bommer, Ahmed, Geldsetzer and colleagues extends that story along the entire clinical course of dementia. Using similar natural-experiment logic around date-of-birth thresholds, the authors report that live-attenuated HZ vaccination is associated with: (i) fewer new diagnoses of mild cognitive impairment (MCI) among people without prior cognitive problems, and (ii) fewer deaths due to dementia among people already living with dementia. The effect does not appear confined to a single dementia subtype, and the benefits are seen at both ends of the disease spectrum: entry into the cognitive-impairment pathway and mortality later on.
This work avoids the classic “healthy vaccinee” problem of standard observational studies. Individuals who choose vaccination typically differ from those who decline it in all sorts of ways: health status, health-seeking behaviour, education, social networks, biologically-implausible beliefs. By contrast, people born just before and just after an arbitrary date-of-birth cutoff are plausibly similar on these dimensions, apart from very small age differences and a large difference in vaccine eligibility. That quasi-randomisation makes the observed reductions in MCI, dementia diagnoses and dementia deaths far less likely to be artefacts of confounding.
Biologically, shingles vaccination fits naturally into a multi-target view of dementia. Varicella zoster virus establishes lifelong latency in the nervous system, reactivates more often with age, and can drive both direct neurotropic damage and chronic immune activation. The broader herpesvirus family has been linked to amyloid seeding, tau phosphorylation and vascular injury patterns resembling those seen in Alzheimer’s disease. By reducing clinical and subclinical reactivations, an HZ vaccine may lower this chronic inflammatory load; it may also exert more general “off-target” immune effects that partially counter immunosenescence in later life. That these effects appear in both incident MCI and dementia mortality suggests the vaccine is not acting on a single tight window, but modulating risk and progression throughout the clinical course.
Against that backdrop, shingles vaccination looks less like a curiosity and more like one component of the emerging “cocktail and timing” model. Amyloid-directed antibodies, tau-targeting agents, metabolic modulators such as GLP-1 analogues, and upstream immune interventions like HZ vaccination may each offer only modest, stage-dependent effects on their own. Used together, at the right time points—early enough to alter risk and transition rates, but long enough to influence progression—they may add up to a more substantial reshaping of the dementia trajectory than any single mechanism is likely to deliver.
Anyway - if you’ve read this far, are worried about our democracies, and would like to know more about the Cognitive Republic: Building Societies That Experiment, Learn, and Evolve (just click this link for more). The Cognitive Republic reimagines democracy as an evolving intelligent system: one that perceives, remembers, learns, and adapts, where policy is treated as hypothesis; failure is feedback; civic life becomes collective inquiry.
The point is not to add another political ideology to the pile. The point is to build a new method—an operating system—for political self-correction at scale. Please consider subscribing - just click here.
References
Xie, M., Eyting, M., Bommer, C., Ahmed, H., Michalik, F., Chung, S., & Geldsetzer, P. (2025). The effect of shingles vaccination at different stages of the dementia disease course. Cell. Advance online publication. https://doi.org/10.1016/j.cell.2025.11.007
Eyting, M., Xie, M., Michalik, F., Heß, S., Chung, S., & Geldsetzer, P. (2025). A natural experiment on the effect of herpes zoster vaccination on dementia. Nature, 641(8062), 438–446. https://doi.org/10.1038/s41586-025-08800-x
Pomirchy, M., Bommer, C., Pradella, F., Michalik, F., Peters, R., & Geldsetzer, P. (2025). Herpes zoster vaccination and dementia occurrence. JAMA, 333(23), 2083–2092. https://doi.org/10.1001/jama.2025.5013
Bjornevik, K., Cortese, M., Healy, B. C., Kuhle, J., Mina, M. J., Leng, Y., … Ascherio, A. (2022). Longitudinal analysis reveals high prevalence of Epstein–Barr virus associated with multiple sclerosis. Science, 375(6578), 296–301. https://doi.org/10.1126/science.abj8222
Hou, Y., Chen, M., Bian, Y., Hu, Y., Chuan, J., Zhong, L., … Tong, R. (2024). Insights into vaccines for elderly individuals: From the impacts of immunosenescence to delivery strategies. npj Vaccines, 9, 77. https://doi.org/10.1038/s41541-024-00874-4