She carries two copies of APOE4 and has a family history of dementia. She is also an avid long-distance cyclist. Over the past 10 years, she has fallen from her bike several times. On one occasion, she lost consciousness and was diagnosed with a traumatic brain injury (TBI).

For APOE4 carriers, traumatic brain injury is an important potential contributor to later neurodegeneration. But what counts as a meaningful traumatic brain injury? How should repeated impacts be considered? And once an injury has occurred, can anything be done to reduce its long-term consequences?

These questions are especially relevant to APOE4 because APOE is not simply an Alzheimer risk protein. It is also part of the nervous system’s response to injury.

What actually qualifies as a traumatic brain injury?

A traumatic brain injury, or TBI, occurs when an external force disrupts normal brain function. A concussion is considered a mild TBI.

According to the CDC, concussion can result from a bump, blow, or jolt to the head, or from a hit to the body that causes the head and brain to move rapidly back and forth.[1]

This is different from simply bumping the head.

A minor impact followed by no neurological symptoms would not ordinarily be considered a concussion. A TBI becomes more likely when trauma produces a temporary change in brain function, such as confusion, amnesia, feeling dazed, dizziness, imbalance, slowed thinking, difficulty concentrating, headache, visual symptoms, or loss of consciousness.[1]

Importantly, loss of consciousness is not required. A normal CT scan also does not exclude concussion because mild TBI is primarily a clinical diagnosis.[1]

There is another category that matters particularly in sports: repetitive head impacts. These include both impacts that cause recognized concussions and smaller impacts that produce no obvious symptoms.[2]

The biological burden of head trauma may therefore be greater than the number of diagnosed concussions.

What makes a TBI meaningful?

There is no single threshold separating an important injury from an unimportant one.

A few seconds of unconsciousness is not a validated cutoff. Nor is there a known number of concussions above which later dementia becomes inevitable.

The response to TBI is highly heterogeneous. Outcomes are influenced by the magnitude and direction of the impact, rotational forces, previous injuries, the interval between injuries, age, vascular and metabolic health, sleep, alcohol exposure, and genetics.

Repeated injury is particularly relevant because another impact may occur while the brain is still undergoing metabolic, vascular, and inflammatory recovery.

For this reason, cumulative head-impact exposure may be more informative than concussion count alone.

APOE is part of the injury response

APOE is usually discussed in the context of Alzheimer disease, but its normal biology is broader.

APOE transports lipids. The nervous system contains large amounts of cholesterol and phospholipid in neuronal membranes, axons, synapses, and myelin. Following injury, damaged membranes must be cleared while lipids are redistributed to cells rebuilding axons and membranes.

APOE is strongly recruited into this process.

In work we published with Ted Price and colleagues, injury to the sciatic nerve produced a marked increase in APOE. Proteomic analysis showed little detectable APOE in uninjured nerve, followed by strong induction after injury, with evidence of new APOE synthesis within the damaged nerve.[3]

The biological explanation is intuitive:

neural injury creates a sudden demand for lipid recycling and membrane repair, and APOE is part of the machinery mobilized to meet that demand.

This provides an important framework for understanding APOE4 after trauma.

Why APOE4 may reduce resilience after injury

APOE4 differs structurally and functionally from APOE3. These differences affect lipid handling, protein stability, receptor interactions, inflammatory signaling, vascular biology, and communication among neurons and glial cells.

After injury, the nervous system becomes highly dependent on processes in which APOE participates.

A plausible sequence is therefore:

injury → increased demand for lipid transport and membrane repair → altered APOE4-mediated repair → greater persistence of secondary injury

APOE4 does not determine outcome by itself. Rather, it may contribute to why apparently similar injuries have different consequences in different people.

Two other biological pathways strengthen this model: the blood-brain barrier and tau.

APOE4, TBI, and the blood-brain barrier

The blood-brain barrier is the specialized interface separating circulating blood from brain tissue.

Montagne and colleagues demonstrated greater blood-brain-barrier breakdown in cognitively normal APOE4 carriers, particularly in the hippocampus and medial temporal lobe. This was associated with pericyte injury and predicted cognitive decline independently of measured amyloid and tau pathology.[4]

TBI can damage the same neurovascular system. Mechanical forces strain small cerebral vessels and increase blood-brain-barrier permeability, allowing blood proteins and inflammatory signals to enter brain tissue.

The two processes therefore converge:

APOE4-associated vascular vulnerability + trauma-induced vascular injury

This may be one mechanism by which APOE4 alters resilience after head trauma.

Tau provides another connection

Tau normally helps stabilize axons. Mechanical injury stretches axons and can cause tau to detach from microtubules, become abnormally phosphorylated, misfold, and accumulate.

Ojo and colleagues studied repetitive mild TBI in mice expressing human tau. Repeated injury produced persistent axonal damage, gliosis, altered cerebral blood flow, and accumulation of tau and tau oligomers.[5]

APOE4 also appears to influence what happens once pathological tau develops. Experimental models show that APOE4 worsens tau-associated inflammation and neurodegeneration, while removal of neuronal APOE4 reduces tau-related gliosis, neurodegeneration, and propagation of pathology through connected regions.[6]

Human imaging studies point in the same direction. APOE4 carriers appear to enter a phase of accelerated tau accumulation at lower amyloid levels than noncarriers.[7]

Together, these findings support a model in which:

head trauma → axonal damage and abnormal tau → APOE4-associated abnormalities in repair, inflammation, and vascular function → greater persistence or propagation of tau pathology

This pathway has not yet been demonstrated prospectively from concussion to dementia in humans, but several independent lines of evidence converge on the same biology.

What is CTE?

Chronic traumatic encephalopathy, or CTE, is a neurodegenerative disease associated with prolonged exposure to repetitive head impacts. Its defining pathology is abnormal phosphorylated tau deposited around small blood vessels, particularly at the depths of cortical sulci.

CTE is not synonymous with concussion. The strongest association is with long-term cumulative exposure to repetitive head impacts, not with one isolated concussion.[2]

Atherton and colleagues studied 364 brain donors with substantial repetitive-head-impact exposure. APOE4 did not simply determine who had CTE. Among older donors, however, APOE4 was associated with more advanced CTE stage and greater phosphorylated-tau burden. In people aged 65 or older, APOE4 carriage was associated with approximately twice the odds of advanced stage III-IV CTE.[8]

This is consistent with APOE4 acting as a disease modifier: trauma supplies the exposure, while APOE4 may influence the biological response to that exposure.

cPLA2 and the secondary injury response

Another pathway linking neural trauma to neurodegeneration is cytosolic phospholipase A2, or cPLA2.

cPLA2 cleaves membrane phospholipids and releases arachidonic acid. After neural trauma, calcium and inflammatory signaling can activate cPLA2, increasing arachidonic-acid-derived signaling and downstream production of prostaglandins, leukotrienes, and other bioactive lipid mediators.[9]

Arachidonic acid itself is a normal and important component of cell membranes. The concern is not that dietary arachidonic acid is harmful.

Rather, excessive cPLA2 activation after injury may produce an abnormal burst of arachidonic acid signaling that shifts the tissue environment toward persistent or unresolved inflammation.

Experimental studies of TBI and spinal cord injury place cPLA2 within the secondary injury cascade, with effects on membrane integrity, oxidative stress, inflammation, lysosomal biology, autophagy, and neuronal survival.[9]

This has particular relevance to our laboratory because we are developing cPLA2 inhibitors for Alzheimer disease. Whether acute cPLA2 inhibition after TBI can reduce subsequent axonal injury, tau pathology, or neurodegeneration is a testable therapeutic question.

Where omega-3 fatty acids fit

DHA is a major structural fatty acid in neuronal membranes.

Following trauma, membrane lipids undergo rapid remodeling. DHA and EPA also serve as precursors for specialized lipid mediators, including resolvins and protectins, involved in the resolution of inflammation.

This creates an interesting intersection with cPLA2 biology. Excessive cPLA2 activation may increase arachidonic-acid-derived inflammatory signaling, while greater DHA and EPA availability may favor pathways involved in resolution and repair.

Omega-3 fatty acids are not cPLA2 inhibitors, but both approaches influence the lipid environment in which secondary neural injury develops.

Landmark human omega-3 studies

In 2016, Oliver and colleagues randomized 81 NCAA Division I football players to placebo or 2, 4, or 6 g/day of DHA over a 189-day season. They measured neurofilament light, or NfL, a blood marker of axonal injury.

NfL increased during the season, particularly among athletes with greater playing exposure. DHA attenuated that rise when the supplemented groups were considered together.[10]

This provided proof of principle that modifying fatty-acid status might alter a biological marker associated with repetitive head impacts. It did not establish that DHA prevents concussion, tau pathology, cognitive decline, or dementia.

A second randomized study in 2022 tested approximately 2.4 g DHA plus 1.0 g EPA per day in collegiate football players. NfL increased in both groups, and supplementation did not significantly reduce NfL or the inflammatory markers IL-6 and TNF-α.[11]

The individual studies therefore produced heterogeneous results.

A 2024 individual-participant meta-analysis combined three studies involving 179 football players. No significant treatment difference was detected at midseason, but omega-3 supplementation was associated with a smaller postseason increase in NfL.[12]

The overall signal is therefore biologically interesting but not yet clinically definitive.

What about DHA after an actual concussion?

This may be the more relevant therapeutic question.

Miller and colleagues randomized 40 adolescents with sport-related concussion to 2 g/day DHA or placebo, beginning within four days of injury.[13]

The DHA group became symptom-free at a median of approximately 11 days compared with 16 days with placebo and began return-to-sport progression at about 14 versus 19.5 days.

Those differences did not reach conventional statistical significance, but this was a small pilot trial. A roughly five-day numerical separation in recovery provides a rational signal for a properly powered study rather than evidence that the intervention is ineffective.

Why spinal cord injury studies matter

Spinal cord injury provides a useful model because injury timing and treatment can be controlled precisely.

In a landmark experiment, King and colleagues found that omega-3 fatty acids administered shortly after spinal cord injury improved neuronal and oligodendrocyte survival and functional recovery.[14] Subsequent preclinical studies and meta-analyses have generally supported neuroprotective effects of omega-3 treatment, although study quality and publication bias remain limitations.[15]

Some experiments also found that manipulating arachidonic-acid availability altered recovery. This should not be interpreted as evidence that arachidonic acid in the diet is harmful. A more relevant interpretation is that the balance of lipid signaling after injury matters, and excessive cPLA2-driven arachidonic-acid signaling may contribute to unresolved inflammation.

The timing of treatment may be particularly important. In these experiments, omega-3 treatment is not simply improving long-term nutrition. It is being delivered while membrane disruption, inflammation, oxidative stress, and secondary tissue injury are actively evolving.

The missing clinical trial

The major gap is now specific.

A sufficiently powered trial should enroll patients soon after a documented TBI, measure baseline omega-3 status, and randomize them to a defined DHA-rich intervention or placebo.

Clinical recovery should be paired with biological measures such as NfL, GFAP, phosphorylated tau, inflammatory lipid mediators, markers of blood-brain-barrier injury, and advanced MRI.

APOE genotype should be prespecified.

Such a trial could determine whether early DHA modifies secondary axonal injury, whether baseline omega-3 status matters, whether timing affects efficacy, and whether APOE4 carriers respond differently.

The same framework could eventually be used to test cPLA2 inhibition.

What does this mean for APOE4 carriers involved in sports?

A useful distinction is whether head impacts are intrinsic to the sport or occur primarily through accidents.

In American football, rugby, boxing, kickboxing, and mixed martial arts, repeated head impacts are part of normal participation, including many that never cause recognizable concussion symptoms. Cumulative exposure is therefore relevant even when diagnosed concussions are uncommon.

There is no established guideline excluding APOE4 carriers from these sports solely on the basis of genotype. However, reducing repetitive head-impact exposure is biologically consistent with what is known about APOE4, neural repair, vascular vulnerability, and tau pathology.

Cycling, surfing, skiing, snowboarding, and horseback riding are different. Head injury is not intrinsic to participation. It occurs mainly through falls or collisions.

In these activities, individual history becomes especially important.

A cyclist who has ridden for 20 years without a significant crash has a very different exposure history from someone who has fallen four times in a decade and sustained a TBI with loss of consciousness.

The relevant question is therefore not simply whether an activity is classified as a contact sport, but how much meaningful head-impact exposure that individual is actually accumulating.

What can be done after a TBI?

The initial mechanical injury cannot be reversed, but the biological response continues for hours to days.

During this period, the brain undergoes changes in blood flow, energy metabolism, membrane integrity, mitochondrial function, inflammation, oxidative stress, and axonal stability. This secondary-injury window provides a biologically plausible opportunity for intervention.

Current management emphasizes appropriate medical evaluation, avoidance of another injury during recovery, and gradual return to physical and cognitive activity rather than prolonged complete rest.[16]

Long-term risk reduction should also address vascular and metabolic health, including blood pressure, diabetes, smoking, sleep apnea, excessive alcohol use, and physical inactivity.

A Mediterranean-style dietary pattern is reasonable for overall vascular and brain health. Omega-3 fatty acids deserve particular study because they combine a plausible mechanism, substantial preclinical evidence, a human biomarker signal, and an encouraging but underpowered post-concussion trial.

What remains to be established is the optimal dose, timing, duration, baseline omega-3 status, and whether APOE genotype modifies treatment response.

Take-home messages

• TBI is a disruption of brain function after trauma, not simply any bump to the head.

• APOE participates in neural repair, and APOE4 may reduce resilience through effects on lipid handling, vascular integrity, inflammation, and tau biology.

• Cumulative head-impact exposure matters, particularly in football, rugby, boxing, and other sports where impacts are intrinsic to participation.

• cPLA2 activation and altered lipid signaling may contribute to secondary injury, while omega-3 fatty acids are a plausible intervention that deserves better acute-TBI trials.

• After TBI, the priorities are appropriate recovery, prevention of another injury, and aggressive management of broader vascular and metabolic risk.

References

1. Centers for Disease Control and Prevention. About Mild TBI and Concussion. CDC.

2. Centers for Disease Control and Prevention. About Repeated Head Impacts. CDC.

3. Melemedjian OK, Yassine HN, Shy A, Price TJ. Proteomic and functional annotation analysis of injured peripheral nerves reveals ApoE as a protein upregulated by injury that is modulated by metformin treatment. Mol Pain. 2013;9:14. doi:10.1186/1744-8069-9-14.

4. Montagne A, Nation DA, Sagare AP, et al. APOE4 leads to blood-brain barrier dysfunction predicting cognitive decline. Nature. 2020;581:71-76. doi:10.1038/s41586-020-2247-3.

5. Ojo JO, Mouzon B, Algamal M, et al. Chronic repetitive mild traumatic brain injury results in reduced cerebral blood flow, axonal injury, gliosis, and increased T-tau and tau oligomers. J Neuropathol Exp Neurol. 2016;75:636-655. doi:10.1093/jnen/nlw035.

6. Knoferle J, Yoon SY, Walker D, et al. Neuronal APOE4 removal protects against tau-mediated gliosis, neurodegeneration and myelin deficits. Nat Aging. 2023;3:275-296.

7. Steward A, Biel D, Dewenter A, et al. ApoE4 and connectivity-mediated spreading of tau pathology at lower amyloid levels. JAMA Neurol. 2023;80:1295-1306.

8. Atherton K, Han X, Chung J, et al. Association of APOE genotypes and chronic traumatic encephalopathy. JAMA Neurol. 2022;79:787-796. doi:10.1001/jamaneurol.2022.1634.

9. Zhang HJ, Chen YT, Hu XL, et al. Functions and mechanisms of cytosolic phospholipase A2 in central nervous system trauma. Neural Regen Res. 2023;18:258-266. doi:10.4103/1673-5374.346460.

10. Oliver JM, Jones MT, Kirk KM, et al. Effect of docosahexaenoic acid on a biomarker of head trauma in American football. Med Sci Sports Exerc. 2016;48:974-982. doi:10.1249/MSS.0000000000000875.

11. Mullins VA, et al. Effects of fish oil on biomarkers of axonal injury and inflammation in American football players: a placebo-controlled randomized controlled trial. Nutrients. 2022;14:2139.

12. Heileson JL, Macartney MJ, Watson NL, et al. Nutritional optimization for brain health in contact sports: a systematic review and meta-analysis on long-chain ω-3 fatty acids and neurofilament light. Curr Dev Nutr. 2024;8:104454.

13. Miller SM, Zynda AJ, Sabatino MJ, Jo C, Ellis HB, Dimeff RJ. A pilot randomized controlled trial of docosahexaenoic acid for the treatment of sport-related concussion in adolescents. Clin Pediatr (Phila). 2022;61:785-794.

14. King VR, Huang WL, Dyall SC, Curran OE, Priestley JV, Michael-Titus AT. Omega-3 fatty acids improve recovery, whereas omega-6 fatty acids worsen outcome, after spinal cord injury in the adult rat. J Neurosci. 2006;26:4672-4680.

15. MacIntosh-Smith WAC, Abdallah A, Cunningham CJ. The potential effects of polyunsaturated ω-3 fatty acids on spinal cord injury: a systematic review and meta-analysis of preclinical evidence. Prostaglandins Leukot Essent Fatty Acids. 2023;191:102554.

16. Patricios JS, Schneider KJ, Dvorak J, et al. Consensus statement on concussion in sport: the 6th International Conference on Concussion in Sport, Amsterdam 2022. Br J Sports Med. 2023;57:695-711.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

She carries two copies of APOE4, and her mother developed dementia. Since learning her genotype, she has been doing many of the things we would normally recommend for brain health: exercising regularly, eating a healthy diet, sleeping well, and paying attention to cardiovascular risk.

But she was understandably anxious.

Her neurologist had suggested lithium orotate, 5 mg per day, increasing the dose until she developed side effects. She was hesitant to start it.

Her question was simple:

Should a healthy APOE4/4 carrier take lithium to prevent Alzheimer disease?

Recent work from Liviu Aron, Bruce Yankner, and colleagues has made this an important question to study. Their experiments suggest that naturally occurring lithium may play a role in normal brain aging and that reducing lithium availability can worsen several Alzheimer-related changes in mice.[1]

The biology is intriguing.

But there is still a large gap between discovering an interesting biological pathway and knowing what a healthy 60-year-old should take for the next 20 years.

Lithium is already part of our environment

Most people think of lithium as a medication for bipolar disorder. At those doses, it is a prescription drug, blood levels are monitored, and kidney and thyroid function require surveillance.

But lithium is also a naturally occurring element found in soil, food, and drinking water. Almost everyone consumes small amounts.

That raises an interesting possibility: perhaps lithium has useful physiological effects at concentrations far below those used in psychiatry.

The problem is that we do not know what “enough” lithium means.

Lithium is not officially considered an essential nutrient. There is no recognized human lithium-deficiency syndrome, no recommended daily allowance, and no validated blood test that tells us a healthy person is deficient.[2]

An intake of roughly 1 mg of elemental lithium per day has been proposed as a possible nutritional target, but that was a hypothesis, not a dose proven in clinical trials to prevent disease.[2]

The first clue came from drinking water

One of the best-known observations came from Denmark.

Researchers compared more than 73,000 people with dementia with over 733,000 controls and estimated their long-term exposure to naturally occurring lithium in drinking water.[3]

People exposed to the highest concentrations had lower dementia rates than those in the lowest-exposure group.

But there was a catch.

The relationship was not a simple straight line. Intermediate exposure was not consistently better.

This raises the possibility of a Goldilocks effect. In the children’s story, Goldilocks tastes three bowls of porridge: one is too hot, one is too cold, and one is “just right.” Lithium may eventually behave similarly: too little could be unfavorable, but more is not necessarily better, and very high doses clearly carry risks.

That is important because the epidemiology does not support the idea that progressively increasing lithium should produce progressively greater protection.

The Danish study also cannot prove that lithium caused the lower dementia rate. People living in different areas differ in diet, socioeconomic conditions, healthcare, other minerals in the water, and many other environmental exposures.

So drinking-water studies give us a clue, not a treatment dose.

Where do we get lithium from naturally?

Lithium is present in small amounts in many foods, particularly vegetables and grains. Leafy vegetables appear to be relatively good sources, and older studies found especially high levels in spinach and eggplant.

But lithium content varies greatly with local soil and water, so no reliable food table compares with what we have for calcium or potassium. Drinking water can also make an important contribution in regions where groundwater naturally contains more lithium.

Is U.S. drinking water low in lithium?

There is no single American exposure.

Lithium concentrations vary considerably across U.S. water supplies. Groundwater generally contains more lithium than surface water, and concentrations tend to be higher in some western and central regions.[4,5]

So it would be misleading to say that American water is generally “lithium deficient.”

We do not even know what concentration would define deficiency.

A more recent U.S. population-level analysis also reported an association between higher drinking-water lithium and somewhat lower dementia rates.[6] But again, it compared communities rather than individual people, and the relationship was not simply “more lithium, less dementia.”

The signal is interesting because it has appeared in different populations.

But it remains epidemiology, not proof that lithium prevents Alzheimer disease.

What about reverse-osmosis water?

This is a reasonable practical question.

Conventional water treatment generally leaves much of the lithium intact. Reverse osmosis can remove more than 90% of it.[7]

Someone drinking almost exclusively reverse-osmosis or distilled water could therefore receive substantially less lithium from water.

Does that matter for brain health?

We do not know.

Food also contains lithium, and there is no evidence that people drinking mineral-free water should automatically take lithium supplements.

The numbers are also helpful.

If water contains 10 micrograms of lithium per liter and someone drinks two liters per day, that provides approximately:

0.02 mg of lithium per day.

A 1 mg elemental lithium supplement provides about 50 times that amount.

So we cannot take the drinking-water studies and simply conclude that taking 1 mg, much less 5 mg, reproduces the exposure associated with lower dementia rates.

What Aron, Yankner, and colleagues found in human brains

The Harvard group measured 27 metals in brain tissue from older adults who were cognitively normal, had mild cognitive impairment, or had Alzheimer disease.

Lithium stood out.

It was the only metal significantly reduced in the prefrontal cortex in both MCI and Alzheimer disease. Importantly, blood lithium was not significantly different between the groups.[1]

Then came perhaps the most interesting finding.

Lithium accumulated inside amyloid plaques, while lithium outside those plaques was reduced.

This suggests a possible feedback loop:

amyloid accumulates → lithium becomes trapped → available lithium falls → Alzheimer-related biology may worsen.

But this immediately raises the problem of reverse causation.

Did low lithium contribute to Alzheimer disease?

Or did Alzheimer disease itself lower available lithium by trapping it inside amyloid?

The human data cannot answer that.

And because blood lithium was not reduced, the findings do not support the simple conclusion that people with MCI or Alzheimer disease were consuming too little lithium.

The mouse experiment was striking

The researchers then did something that cannot reasonably be done in people: they deliberately depleted lithium.

They reduced dietary lithium in mice by 92%. Blood lithium fell by about 89%, while available lithium in the cortex fell by roughly half.[1]

The consequences were broad.

Lithium depletion increased amyloid and abnormal tau, activated inflammatory microglia, reduced amyloid clearance, damaged synapses and myelin, and worsened memory.

That provides stronger evidence that lithium itself can influence Alzheimer-related biology.

But there is an important question:

Does a 92% lithium-deficient mouse diet resemble anything that commonly occurs in humans?

We do not know.

The investigators produced a fall in brain lithium similar in magnitude to what they observed in Alzheimer brains, but the cause was different.

In the mice, lithium was removed from food.

In human Alzheimer brains, amyloid itself may have been trapping lithium.

That is a major gap between the animal experiment and clinical practice.

How might lithium affect the brain?

Several mechanisms are plausible.

One involves an enzyme called GSK3β. This enzyme can contribute to abnormal changes in tau, one of the major proteins involved in Alzheimer disease.

Lithium deficiency increased GSK3β activity in the mice. Blocking that enzyme reversed several of the harmful effects, including abnormal tau, amyloid accumulation, and inflammation.[1]

Lithium also affected microglia, the immune cells of the brain. When lithium was depleted, microglia became more inflammatory and less effective at clearing amyloid. Lithium deficiency also changed the activity of several genes associated with Alzheimer risk, including APOE and TREM2.

Another possible pathway involves cPLA2, an enzyme that helps release arachidonic acid from cell membranes. Arachidonic acid can then be converted into inflammatory signaling molecules. Animal studies suggest that chronic lithium can dampen this pathway.[8]

These findings help explain why lithium could matter biologically.

They do not tell us what dose a human should take.

Why lithium orotate?

Aron, Yankner, and colleagues compared several lithium salts.

Lithium orotate showed less binding to amyloid than lithium carbonate in their experimental system. At very low doses, it increased lithium outside plaques and improved amyloid, tau, inflammation, synaptic changes, and memory in mouse models.[1]

The dose is worth emphasizing.

Their lowest exposure corresponded to only 0.03 mg of elemental lithium per liter of drinking water.

The experiment therefore does not show that 5 mg is better than 1 mg, or that higher doses provide more protection.

If anything, the remarkable part of the experiment is that very small exposures appeared biologically active in mice.

What have human trials shown?

The human evidence is much less convincing.

Earlier small trials in people with mild cognitive impairment, or MCI, produced some encouraging signals involving cognition and Alzheimer-related biomarkers.[9,10]

More recently, the LATTICE trial randomized older adults with MCI to lithium carbonate or placebo and followed them for two years.[11]

LATTICE tested low-dose lithium carbonate, but “low dose” here still meant a prescription lithium regimen rather than trace supplementation.

Participants took 150 or 300 mg of lithium carbonate, and among those completing the trial the average dose was about 195 mg/day, producing an average serum lithium concentration of 0.17 mEq/L. This is well below typical bipolar-treatment levels, but it is still a very different exposure from taking 1 mg of elemental lithium as a supplement.

The main result was straightforward:

Lithium did not significantly improve any of the six prespecified primary outcomes.

Those outcomes included memory, overall cognition, measures of brain structure, and a blood biomarker.[11]

There was an interesting signal suggesting slower decline on one verbal-memory measure, but it did not meet the study’s stricter statistical threshold after accounting for the multiple outcomes being tested.

So LATTICE was an important negative pilot trial with a possible signal, not evidence that lithium improves cognition.

It also tested a different hypothesis from the new lithium-orotate work.

LATTICE used lithium carbonate and targeted blood concentrations derived largely from psychiatric use. It was testing pharmacological lithium in people who already had MCI.

That is different from asking whether maintaining very low physiological lithium exposure decades earlier might prevent disease.

What do decades of bipolar treatment tell us?

Lithium has been used in bipolar disorder for decades, giving us extensive experience with higher doses.

If lithium were a powerful memory enhancer, we probably would have noticed.

Reviews of bipolar studies show no consistent improvement in memory or overall cognition with lithium.[12]

Some observational studies suggest lower rates of dementia among people treated with lithium, and some imaging studies suggest possible neuroprotective effects.

But neuroprotection is not the same as improving memory, and observational studies can be influenced by differences between patients who do and do not receive lithium.

Again, the evidence does not support the idea that more lithium must be better for the brain.

Why can’t we simply run the definitive prevention trial?

Because we still do not know what dose to test.

Folic acid and neural-tube defects provide a useful contrast.

With folate, supplementation could be given around conception, and the outcome was clear within months: did the baby develop a neural-tube defect or not?[13]

Alzheimer prevention is much harder.

A cognitively healthy 60-year-old APOE4/4 carrier might not develop symptoms for 10 or 20 years.

And before launching a huge trial, we need answers to more basic questions.

How much lithium reaches the human brain after 0.3 mg, 1 mg, or 5 mg?

Does lithium orotate behave differently from lithium carbonate in people?

What dose actually changes the biology we care about?

And how would we know that it is working?

Should an effective dose change amyloid? Tau? Plasma p-tau217? Microglial activity? GSK3β?

At present, we do not have a validated short-term marker showing that trace-dose lithium has engaged the relevant target in a human brain.

This is a PK/PD problem.

PK, or pharmacokinetics, means what the body does to lithium: how much is absorbed, how long it stays in the blood, and how much reaches the brain.

PD, or pharmacodynamics, means what lithium does once it gets there.

Those relationships are well understood for prescription lithium.

They are not established for trace-dose lithium or lithium orotate.

Without that information, even a randomized trial could be difficult to interpret.

If 1 mg failed, was lithium ineffective, or was the dose simply too low?

If 5 mg changed a biomarker, would that biomarker actually predict less dementia 15 years later?

The logical sequence is therefore:

first understand dose and brain exposure, then demonstrate a biological effect, and only then test long-term prevention.

Early human lithium-orotate studies are being developed, including work involving Yankner and collaborators.[14] But these studies will not immediately answer whether a healthy APOE4/4 carrier should take lithium for decades.

Do you need a lithium blood level?

Not to determine whether you are “lithium deficient.”

There is no validated nutritional lithium range.

And in the Aron-Yankner study, brain lithium differed even though blood lithium did not.[1]

So a trace serum lithium level currently cannot tell us whether the brain has enough lithium.

Blood monitoring becomes important at pharmacological doses, where lithium levels are used mainly to reduce the risk of toxicity.

What happens as the dose gets higher?

Here we have decades of experience.

At doses used for bipolar disorder, lithium can cause tremor, nausea, diarrhea, thirst, frequent urination, fatigue, and sometimes cognitive slowing.

Long-term treatment can affect the thyroid, parathyroid glands, and kidneys.

More serious toxicity can cause worsening tremor, persistent vomiting or diarrhea, marked weakness, unusual sleepiness, confusion, slurred speech, poor coordination, seizures, and in severe cases coma.

Dehydration, impaired kidney function, and medications such as NSAIDs, some diuretics, ACE inhibitors, and ARBs can raise lithium levels.

Someone taking lithium who develops significant tremor, persistent gastrointestinal symptoms, confusion, unusual drowsiness, slurred speech, or difficulty walking should stop taking it and seek medical evaluation.

Larger lithium doses should not be used without medical supervision.

Most importantly:

Side effects are not evidence that an Alzheimer-prevention dose has been reached.

There is no scientific basis for increasing lithium until toxicity appears.

Where the uncertainty remains

We still do not know whether low brain lithium is a cause of Alzheimer disease, a consequence of amyloid accumulation, or both.

We do not know whether the severe lithium deficiency created experimentally in mice commonly occurs in humans.

We do not know whether lithium orotate has an important advantage over other lithium salts in people.

We do not know the lowest dose that meaningfully changes Alzheimer-related biology in the human brain.

We do not have a validated short-term marker telling us that low-dose lithium is working.

And most importantly, we do not know whether taking lithium for years prevents cognitive decline in cognitively normal people at genetic risk.

What would I tell an APOE4 carrier who is scared of dementia and wants to take lithium?

This may be the most important part.

Someone who carries APOE4, especially two copies, may reasonably say:

“I understand the evidence is incomplete. But I am frightened of dementia. If a very small dose of lithium seems safe, why shouldn’t I take it?”

That is a fair question.

The first thing I would emphasize is that APOE4 is a risk factor, not a diagnosis. Even APOE4/4 does not mean that dementia is inevitable.

The goal should be to reduce risk without allowing fear of a future disease to create harm in the present.

Lithium is also not an unreasonable molecule to be interested in. There is intriguing epidemiology, plausible biology, strong animal data, and decades of human experience with the drug.

But we still do not know whether lithium prevents Alzheimer disease in cognitively healthy APOE4 carriers.

A 1 mg elemental lithium dose is very different from prescription lithium used for bipolar disorder. The expected risk is likely substantially lower.

But “probably low risk” is not the same as “proven safe and effective for decades.”

And we have no evidence that 1 mg is the correct dose.

For someone who understands those uncertainties and still wishes to try a very-low-dose supplement, I would view that as an individual experimental choice, preferably discussed with a clinician who knows the person’s kidney function, medications, and medical history.

What I would not recommend is progressively increasing the dose in search of greater protection.

Conventional lithium has not produced convincing cognitive benefit in MCI or bipolar disorder.

And higher doses clearly produce more toxicity.

Lithium should also remain in perspective. For an APOE4 carrier, the best-supported strategies remain regular exercise, control of blood pressure and other vascular risks, avoiding smoking, treating hearing loss, maintaining healthy sleep, and addressing diabetes and other metabolic risks when present.

Lithium should not displace those measures simply because it feels more directly connected to Alzheimer biology.

Take-home messages

• The lithium-Alzheimer connection is scientifically interesting, but it is not yet clinically proven.

• Drinking-water studies suggest an association with dementia, but the relationship may be Goldilocks-like: too little may be unfavorable, while more is not necessarily better.

• Low lithium in Alzheimer brain tissue may partly be a consequence of the disease itself, because amyloid appears to trap lithium.

• Lithium deficiency worsens Alzheimer-like biology in mice, but the experimental deficiency may not resemble ordinary human exposure.

• Lithium orotate performed impressively in mice, but mouse efficacy does not establish a human dose.

• LATTICE did not meet any of its six primary outcomes, arguing against assuming that conventional lithium treatment improves cognition in MCI.

• A 1 mg elemental lithium supplement is far below psychiatric dosing, but it has not been shown to prevent MCI or dementia.

• Blood lithium testing cannot currently diagnose nutritional lithium deficiency.

• Higher-dose lithium can cause kidney, thyroid, gastrointestinal, and neurological toxicity.

• Escalating lithium until side effects occur is not an evidence-based Alzheimer-prevention strategy.

• The next step is to identify the lowest dose that reaches the human brain and demonstrably changes relevant biology, and only then determine whether it prevents cognitive decline.

For a healthy APOE4/4 carrier, lithium orotate remains an intriguing possibility worth studying.

It is not yet a proven prevention therapy, and an individual experimental choice.

References

1. Aron L, Ngian ZK, Qiu C, et al. Lithium deficiency and the onset of Alzheimer’s disease. Nature. 2025;645:712-720. doi:10.1038/s41586-025-09335-x.

2. Schrauzer GN. Lithium: occurrence, dietary intakes, nutritional essentiality. J Am Coll Nutr. 2002;21:14-21.

3. Kessing LV, Gerds TA, Knudsen NN, et al. Association of lithium in drinking water with the incidence of dementia. JAMA Psychiatry. 2017;74:1005-1010.

4. Lindsey BD, Belitz K, Cravotta CA III, Toccalino PL, Dubrovsky NM. Lithium in groundwater used for drinking-water supply in the United States. Sci Total Environ. 2021;767:144691.

5. Lithium occurrence in drinking water sources of the United States. Chemosphere. 2022;305:135458.

6. Hwang SH, Jang H, Park SK, Shin HM. Associations between low-dose drinking water lithium exposure and dementia: a nationwide U.S. ecological study. Environ Pollut. 2026;394:127766.

7. Keithley AE, Jordan P, Muhlen C, et al. Lithium removal from drinking water. AWWA Water Sci. 2025;7:e70016.

8. Rao JS, Lee HJ, Rapoport SI, Bazinet RP. Mode of action of mood stabilizers: is the arachidonic acid cascade a common target? Mol Psychiatry. 2008;13:585-596. doi:10.1038/mp.2008.31.

9. Forlenza OV, Diniz BS, Radanovic M, et al. Disease-modifying properties of long-term lithium treatment for amnestic mild cognitive impairment: randomised controlled trial. Br J Psychiatry. 2011;198:351-356.

10. Forlenza OV, Radanovic M, Talib LL, Gattaz WF. Clinical and biological effects of long-term lithium treatment in older adults with amnestic mild cognitive impairment: randomised clinical trial. Br J Psychiatry. 2019;215:668-674.

11. Gildengers AG, Ibrahim TS, Anderson SJ, et al. Low-dose lithium for mild cognitive impairment: a pilot randomized clinical trial. JAMA Neurol. 2026;83:310-319.

12. Deng R, Shargorodsky A, Teopiz K, et al. Lithium and long-term cognitive outcomes in bipolar disorder and early dementia: a systematic review. CNS Spectr. 2026;31:e18.

13. Czeizel AE, Dudás I. Prevention of the first occurrence of neural-tube defects by periconceptional vitamin supplementation. N Engl J Med. 1992;327:1832-1835.

14. Boles S. An Alzheimer’s breakthrough 10 years in the making. Harvard Gazette. January 27, 2026.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

But hearing loss has also emerged as one of the most consistent potentially modifiable risk factors associated with dementia.

The 2024 Lancet Commission included hearing loss among 14 potentially modifiable dementia risk factors and estimated that, under the assumptions of its model, hearing loss accounts for roughly 7% of dementia cases at the population level.[1]-Figure below:

That number needs careful interpretation. It does not mean that hearing loss causes 7% of dementia or that treating hearing loss would prevent 7% of cases. A population-attributable fraction depends both on how common a risk factor is and how strongly it is associated with disease, while also making assumptions about causality.

The more interesting question is biological:

What do years of degraded hearing do to the brain, and could that burden matter more in someone already vulnerable because of APOE4?

Special thanks to Dr. Janet Choi, an ENT physician-scientist and expert in hearing loss and dementia, for taking the time to review this post and share her thoughtful perspective.

Hearing is a brain task

The cochlea in the inner ear converts sound into neural signals, but hearing does not end at the ear.

In a noisy room, the brain must separate one voice from competing sounds, direct attention to that voice, reconstruct incomplete information, and connect what is heard with language and memory.

When the auditory signal becomes degraded, the brain has to work harder.

This is the cognitive-load hypothesis: more mental resources are used to understand speech, potentially leaving fewer resources available for other cognitive tasks.[2]

That explains why hearing loss can affect performance in the moment. It does not fully explain why hearing loss might be associated with dementia many years later.

Long-term hearing loss may also alter brain networks and reduce social and cognitive stimulation. Communication becomes harder, social interactions require more effort, and some people gradually withdraw from activities they previously enjoyed.

But there may also be more direct effects on memory systems.

Hearing and memory are connected

The auditory cortex, which processes sound, is connected with structures in the medial temporal lobe involved in memory, including the entorhinal cortex and hippocampus.

The hippocampus is central to forming and retrieving memories. The entorhinal cortex acts as an important interface between the hippocampus and the rest of the cerebral cortex.

These systems communicate with each other.

Experimental studies have shown that the entorhinal cortex can influence how auditory memories are formed in the auditory cortex.[11,12] Human studies also show coordinated activity between auditory cortex and hippocampus when people hold sounds in memory.[13]

More direct recordings from the human brain suggest that this communication can run in both directions: information travels from the auditory cortex toward the hippocampus during encoding and can flow back toward the auditory cortex during memory replay.[14]

This makes hearing and memory parts of an interacting network rather than completely separate systems.

Animal studies also show that auditory deprivation can alter synaptic plasticity, meaning the ability of connections between neurons to strengthen or weaken with experience, within the hippocampus.[15]

A 2024 study by Pan and colleagues pushed this further. In an Alzheimer mouse model, experimentally induced hearing loss worsened memory, hippocampal synaptic function, and amyloid pathology. Restoring GDF1 signaling improved several of these abnormalities.[3]

These findings do not prove that hearing loss causes Alzheimer disease in humans. But they provide a plausible biological route through which chronic hearing loss could affect brain systems involved in memory.

Not all hearing problems are the same

A key distinction is whether the problem is mainly in the ear or in the brain’s processing of sound.

First, let’s introduce these two hearing tests:

A pure-tone audiogram is the standard hearing test in which tones of different pitches and loudness are played through headphones. It measures the quietest sounds a person can detect and helps identify peripheral hearing loss, which usually arises from the ear and cochlea.

A speech-in-noise test asks whether someone can understand speech when background noise is present. That task depends not only on the ear, but also on attention and higher-order auditory processing.

This distinction matters because typical age-related high-frequency hearing loss is usually peripheral, while difficulty understanding speech in a noisy environment can also reflect changes in central brain networks.

A 2024 review led by Suji Hong and colleagues notes that Alzheimer-related changes may affect higher-order sensory association regions before basic sensory detection is severely impaired. In other words, a person may still detect sounds reasonably well but have increasing difficulty interpreting complex auditory information.

That means poor speech-in-noise performance can have two possible interpretations.

It may be the consequence of long-standing hearing loss.

Or it may sometimes be an early sign that neurodegeneration is already affecting auditory processing.

Both possibilities need to be considered.

What do human studies show?

One of the landmark studies came from the Baltimore Longitudinal Study of Aging.

Lin and colleagues studied 639 adults aged 36 to 90 who were free of dementia when their hearing was measured. Over almost 12 years, 58 developed dementia.[4]

Risk increased as hearing worsened. Each 10-decibel deterioration was associated with about a 27% increase in dementia risk. Mild hearing loss was associated with roughly twice the risk of dementia, and moderate hearing loss with about three times the risk.[4]

The dose-response pattern was important.

But this was an observational study. Hearing loss could contribute to dementia, both conditions could share common causes, or hearing difficulty could sometimes reflect disease already developing in the brain.

The Health ABC Study later found a similar association in adults aged 70 to 79 followed for about nine years.[5]

A much larger UK Biobank study of 82,039 participants used a speech-in-noise test and again found that poorer performance predicted later dementia.[6]

The 2024 Lancet Commission combined six studies totaling approximately 666,000 people. Overall, hearing loss was associated with about a 37% higher rate of subsequent dementia.[1]

But the size of that association varied considerably between studies. So the direction of the relationship is fairly consistent, while the exact magnitude is not.

Timing matters

Hearing loss measured at age 50 and followed by dementia at 75 is different from hearing difficulty first appearing at 78, shortly before dementia is diagnosed.

The first pattern is more consistent with hearing loss potentially contributing to later disease.

The second raises the possibility of reverse causation: early neurodegeneration may already be impairing the brain’s ability to process sound.

This is particularly relevant because Alzheimer pathology begins many years before dementia becomes clinically obvious.

Broome, Dening, and colleagues therefore emphasize that late-life auditory difficulty can sometimes be a harbinger of neurodegeneration, rather than purely a cause of it.[2]

We also do not yet know precisely how risk varies with the severity or duration of hearing loss. Twenty years of mild hearing loss may not be biologically equivalent to several years of severe hearing loss.

Severity, duration, and age at onset may all matter.

Could tau be part of the connection?

Some human biomarker studies suggest that auditory dysfunction may be more closely associated with tau pathology than with amyloid.

Tau is a protein that normally helps stabilize neurons but accumulates abnormally in Alzheimer's disease.

Hong and colleagues summarize studies linking poor hearing performance with higher tau levels in cerebrospinal fluid or greater tau deposition in medial temporal regions involved in both memory and auditory processing.

One possibility is that altered neural activity during difficult listening interacts with brain regions already vulnerable to tau-related damage.

This is an interesting hypothesis, but it remains associative. We do not know that hearing loss causes tau accumulation in humans.

Where does APOE4 fit?

A useful way to think about APOE4 is not that it creates a continuously damaged brain, but that it may reduce resilience when an additional stressor appears.

We have discussed a similar idea in previous posts with inflammation, alcohol, high cholesterol, sleep, and other modifiable risk factors.

Hearing loss could represent another potential second hit.

APOE4 affects lipid transport, synaptic repair, inflammatory responses, vascular biology, and resilience to Alzheimer pathology. Hearing loss may simultaneously increase neural workload and alter activity in auditory and memory networks.

The question is therefore not only whether APOE4 and hearing loss independently increase dementia risk.

The biology is plausible.

The epidemiology is less clear.

Does APOE4 actually strengthen the association?

A recent Framingham Heart Study analysis included 935 adults aged 60 or older, followed for 15 years.[7]

Hearing loss was associated with increased dementia risk overall. Among APOE4 carriers, at least slight hearing loss was associated with an estimated 2.86-fold greater dementia risk compared with normal hearing.[7]

That result is consistent with a second-hit model.

But the APOE4 subgroup was relatively small, and the confidence interval around the estimate was wide, so the result needs replication.

A much larger 2026 UK Biobank study examined 80,287 adults aged 40 to 69 with genetic data and speech-in-noise testing. During follow-up, 1,086 dementia cases were identified.[8]

Hearing impairment predicted dementia in people both with and without APOE4, but there was no clear evidence that APOE4 carrier status made the link stronger.

UK Biobank has much greater statistical power, but dementia is identified largely from routine medical records rather than repeated specialist cognitive assessments. Framingham is smaller but provides deeper longitudinal characterization.

The most defensible conclusion today is:

Hearing loss is associated with dementia risk regardless of APOE4 status. APOE4 may amplify that association in some populations, but this has not yet been consistently demonstrated.

Can treating hearing loss change cognition?

The most important randomized trial is ACHIEVE.

ACHIEVE randomized 977 adults aged 70 to 84 with untreated mild-to-moderately severe hearing loss to either a comprehensive hearing intervention or a health-education control program and followed them for three years.[9]

Across the full study population, hearing treatment did not significantly slow cognitive decline.

But ACHIEVE included a prespecified subgroup of participants from the ARIC cardiovascular cohort who were older and had more dementia risk factors. In that higher-risk group, hearing intervention was associated with slower cognitive decline.[9]

That result is interesting because it raises the possibility that hearing treatment may matter more in people who are already biologically vulnerable.

But the overall trial was negative, so we cannot say that hearing aids prevent dementia.

There is also an important timing issue. Treatment began between ages 70 and 84. If hearing loss affects dementia risk through decades of altered auditory input, treating it at 75 may not answer what would happen if treatment began at 50 or 60.

What should an APOE4 carrier do?

There is currently no APOE4-specific hearing-screening guideline.

But establishing hearing status in midlife is reasonable.

WHO guidance recommends hearing screening beginning at age 50, approximately every five years through age 64, and every one to three years from age 65 onward.[10]

Testing should happen sooner if there is difficulty following conversation, trouble understanding speech in noisy environments, increasing television volume, tinnitus, significant noise exposure, or a noticeable change in hearing.

If symptoms mainly involve difficulty hearing in noise despite a relatively normal audiogram, speech-in-noise testing may be particularly useful.

An abnormal hearing test does not always mean hearing aids

The next step depends on the cause.

Some forms of hearing loss are reversible. Impacted earwax, middle-ear fluid or infection, eustachian-tube dysfunction, and other middle-ear disorders can impair hearing and may improve with treatment.

Medication history and noise exposure also matter because some drugs and repeated excessive noise can damage hearing.

Sudden hearing loss, particularly in one ear, requires prompt medical assessment. Asymmetric or rapidly progressive hearing loss also deserves further evaluation.

Typical age-related hearing loss is usually sensorineural, involving the inner ear or auditory pathway. This generally cannot be medically reversed, but its functional consequences can often be reduced with hearing aids, assistive listening technologies, communication strategies, and audiologic follow-up.

What remains unanswered?

The major questions are now fairly clear.

Which matters most: peripheral hearing loss, central auditory processing difficulty, or both?

How do severity, duration, and age at onset combine to influence risk?

Does APOE4 truly make hearing loss more harmful?

Does tau biology help connect hearing loss with Alzheimer's disease?

And does treating hearing loss in midlife alter long-term brain aging?

What we do not yet have is proof that treating hearing loss prevents Alzheimer's disease.

Take-home messages

• Hearing loss is consistently associated with greater dementia risk, but association does not prove causation.

• Hearing and memory systems are biologically connected, and chronic auditory deprivation may affect networks involved in memory and plasticity.

• Peripheral hearing loss and central auditory dysfunction are different. Difficulty understanding speech in noise may sometimes reflect early brain changes as well as hearing loss.

• Timing matters. Midlife hearing loss is less vulnerable to reverse causation than hearing difficulty emerging shortly before dementia.

• APOE4 may increase vulnerability to hearing loss as a second hit

• ACHIEVE did not show an overall cognitive benefit from hearing treatment, although a higher-risk subgroup appeared to benefit.

• For APOE4 carriers, establishing hearing status in midlife is reasonable.

• An abnormal hearing test should prompt evaluation of the cause, because some forms of hearing loss are reversible.

For an APOE4 carrier, we cannot yet say that treating hearing loss prevents Alzheimer's dementia.

But hearing loss is measurable, often treatable, and worth addressing for its own clinical consequences.

References

1. Livingston G, Huntley J, Liu KY, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. Lancet. 2024;404:572-628. doi:10.1016/S0140-6736(24)01296-0.

2. Broome EE, Calvert S, Heffernan E, Henshaw H, Khan A, Pelekanos V, Sollini J, Stancel-Lewis J, Dening T. Dementia and hearing loss: from risk to mechanisms and management. Front Dement. 2026;5:1736003. doi:10.3389/frdem.2026.1736003.

3. Pan L, Li C, Meng L, et al. GDF1 ameliorates cognitive impairment induced by hearing loss. Nat Aging. 2024;4:568-583. doi:10.1038/s43587-024-00592-5.

4. Lin FR, Metter EJ, O’Brien RJ, Resnick SM, Zonderman AB, Ferrucci L. Hearing loss and incident dementia. Arch Neurol. 2011;68:214-220. doi:10.1001/archneurol.2010.362.

5. Deal JA, Betz J, Yaffe K, et al. Hearing impairment and incident dementia and cognitive decline in older adults: the Health ABC Study. J Gerontol A Biol Sci Med Sci. 2017;72:703-709.

6. Stevenson JS, Clifton L, Kuźma E, Littlejohns TJ. Speech-in-noise hearing impairment is associated with an increased risk of incident dementia in 82,039 UK Biobank participants. Alzheimers Dement. 2022;18:445-456.

7. Kolo FB, Lu S, Beiser AS, et al. Hearing loss, brain structure, cognition, and dementia risk in the Framingham Heart Study. JAMA Netw Open. 2025;8(11):e2539209. doi:10.1001/jamanetworkopen.2025.39209.

8. Peng C, Gong G, Chen Z, et al. Speech-in-noise hearing impairment and incident dementia: a UK Biobank cohort study across polygenic, APOE ε4, and familial risk strata. Alzheimer Dis Assoc Disord. 2026;40(3):192-201. doi:10.1097/WAD.0000000000000747.

9. Lin FR, Pike JR, Albert MS, et al. Hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss in the USA (ACHIEVE): a multicentre, randomised controlled trial. Lancet. 2023;402:786-797. doi:10.1016/S0140-6736(23)01406-X.

10. World Health Organization. Hearing screening: considerations for implementation. Geneva: WHO; 2021.

11. Chen X, Guo Y, Feng J, et al. Encoding and retrieval of artificial visuoauditory memory traces in the auditory cortex requires the entorhinal cortex. J Neurosci. 2013;33:9963-9974. doi:10.1523/JNEUROSCI.4078-12.2013.

12. Li X, Yu K, Zhang Z, et al. Cholecystokinin from the entorhinal cortex enables neural plasticity in the auditory cortex. Cell Res. 2014;24:307-330. doi:10.1038/cr.2013.164.

13. Kumar S, Joseph S, Gander PE, Barascud N, Halpern AR, Griffiths TD. A brain system for auditory working memory. J Neurosci. 2016;36:4492-4505. doi:10.1523/JNEUROSCI.4341-14.2016.

14. Dimakopoulos V, Mégevand P, Stieglitz LH, Imbach L, Sarnthein J. Information flows from hippocampus to auditory cortex during replay of verbal working memory items. eLife. 2022;11:e78677. doi:10.7554/eLife.78677.

15. Temporary conductive hearing loss in early life impairs spatial memory of rats in adulthood. 2018.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

Both of her grandparents had developed dementia.

She was already doing many of the things we associate with healthy aging. She eats a healthy diet, exercises regularly, does Pilates, and stays socially engaged. Once a week, she meets friends to play games, talk, laugh, and catch up.

At the end of the evening, they often enjoy a glass of wine.

After learning her APOE genotype, she decided to stop drinking alcohol for three months.

Something changed.

Her sleep improved. She could now sleep through the night.

Then she asked me a very practical question:

Can I still have that occasional glass of wine with my friends?

It would be convenient to answer with a number: one drink is fine, two are not, or stay below some number of drinks per week.

Unfortunately, the science is not that precise, particularly for APOE4 carriers.

But that does not mean we have nothing useful to tell her.

We know quite a lot about what alcohol does to the brain. We know that people respond differently to the same amount. We have experimental evidence about alcohol and sleep, and several prospective studies have asked whether APOE4 changes the relationship between alcohol use and subsequent dementia.

Perhaps the most useful question is therefore not simply:

How much alcohol is allowed?

It is:

What does this amount of alcohol actually do to me?

What happens in the brain after a drink?

Alcohol reaches the brain rapidly and changes communication between neurons.

Its actions are complicated, but two neurotransmitter systems help explain many of its immediate effects: GABA and glutamate.

GABA is the major inhibitory neurotransmitter in the brain. Glutamate is the major excitatory neurotransmitter and plays an essential role in learning, memory, and synaptic plasticity.

Alcohol generally shifts this balance toward inhibition. It enhances aspects of GABAergic signaling while suppressing glutamatergic signaling, including signaling through NMDA receptors. It also modifies dopamine and several other neurotransmitter systems involved in reward, attention, memory, and motor control.[1,2]

This helps explain why alcohol can initially produce relaxation and reduced inhibition, followed at increasing doses by slower reaction time, impaired judgment, poorer coordination, and memory problems.

Importantly, the biological effect is not identical from person to person.

Some people notice surprisingly strong sedation or cognitive changes after a small amount of alcohol. Others notice very little.

And feeling different after a drink is not meaningless. It tells us that the central nervous system is responding to that dose, even if the effect would not necessarily be detected by a standard cognitive test.

Alcohol can make us sleepy without helping us sleep well

This may be the most relevant part of the story for this woman.

A recent systematic review and meta-analysis combined 27 experimental studies examining alcohol and subsequent sleep.[3] Alcohol delayed the onset of REM sleep and reduced the amount of REM sleep. These effects became progressively larger as alcohol dose increased.

Changes in REM sleep were detectable even in the investigators’ lower-dose category.

At higher doses, alcohol can shorten the amount of time needed to fall asleep. This probably contributes to the common perception that a drink helps with sleep.

But sedation is not the same as normal restorative sleep.

The review was also appropriately cautious. Results for total sleep time, sleep efficiency, and waking after sleep onset were less consistent.[3] We therefore should not say that one glass of wine inevitably causes nighttime awakening.

That is where individual response becomes important.

If someone sleeps normally after a glass of wine, that tells us one thing.

If someone repeatedly wakes during the night after drinking and sleeps through the night when alcohol is removed, that tells us something different.

The woman who asked me this question had already conducted a useful experiment on herself.

She stopped drinking alcohol for three months.

Her sleep improved substantially.

That does not establish causality. Many things can change during three months. But suppose she reintroduces one glass of wine and again begins waking during the night. Then she avoids wine and sleeps normally. If the pattern repeats, the evidence becomes considerably more compelling.

It is essentially a simple N-of-1 experiment.

The same glass of wine is not necessarily the same biological exposure

Imagine two people who each drink one glass of wine.

One sleeps normally, wakes refreshed, and notices no difference in concentration, coordination, heart rate, or energy.

The other wakes at 2 a.m., sleeps restlessly, feels cognitively foggy the following morning, or notices that her heart is beating faster during the night.

They consumed the same nominal dose.

But their bodies did not respond in the same way.

Population recommendations necessarily average together people who differ in age, sex, body size, alcohol metabolism, medications, liver function, sleep physiology, vascular health, and genetics.

For someone particularly concerned about brain health, I think several responses are worth paying attention to.

Sleep disruption is probably the most useful. If even a small amount of alcohol reproducibly produces fragmented sleep or early awakening, that exposure is clearly not physiologically neutral for that individual.[3]

Feeling cognitively different also matters. Unusual sedation, slowed thinking, poorer concentration, or imbalance tell us that the nervous system is responding to the dose.

Prolonged recovery deserves attention as well. If someone still feels below baseline the following morning after a relatively small exposure, I would not dismiss that simply because the amount consumed falls within a conventional definition of moderate drinking.

There may also be autonomic signals. In a controlled laboratory study of 26 adults aged 30 to 60, evening alcohol increased nighttime heart rate and altered autonomic cardiovascular regulation, with larger effects at higher doses.[4]

None of these responses is a biomarker of Alzheimer's disease.

That distinction is important.

But they do tell us whether an exposure that seems small on paper is actually disrupting normal physiology in that person.

There is also an important asymmetry:

The presence of a reproducible adverse response tells us that the exposure is not neutral. Feeling completely normal after a drink cannot prove that it carries zero long-term risk.

Where does APOE4 fit?

This is where I think APOE4 biology becomes particularly interesting.

I do not think the most useful model is that an APOE4 carrier has a continuously damaged or inflamed brain.

Instead, APOE4 may prime the brain toward greater vulnerability when another stressor is added, as I discussed in my last week's post.

We see hints of this pattern elsewhere.

Last week, I presented data from the Framingham Offspring cohort where APOE4 carriers had lower circulating CRP concentrations than noncarriers, yet when signs of systemic inflammation were present, this was associated with substantially greater Alzheimer risk among APOE4 carriers.[5]

That is an intriguing observation.

APOE4 did not simply produce more baseline inflammation. Rather, the consequences of an inflammatory state appeared different in people carrying APOE4.

This fits a second-hit model.

APOE4 creates susceptibility. Another insult or stress reveals it.

The second hit could be chronic inflammation, vascular disease, metabolic dysfunction, poor sleep, head injury, or potentially repeated alcohol insults.

Alcohol is interesting in this context because it intersects with several pathways affected by APOE4. APOE4 alters lipid handling and membrane biology and has been associated with differences in oxidative and inflammatory responses. Alcohol can perturb neurotransmission, sleep, lipid metabolism, oxidative balance, and inflammatory signaling.[1,2]

The hypothesis, therefore, is not simply that alcohol is inherently toxic to every APOE4 carrier.

It is that an APOE4 brain may respond differently to an additional physiological stressor.

And if alcohol repeatedly disrupts sleep or recovery in a particular APOE4 carrier, that may be especially worth paying attention to.

The epidemiologic question is whether we see evidence of this interaction in humans.

We do, although not consistently.

CAIDE: starting in midlife and looking 23 years later

One of the most informative studies comes from the Finnish CAIDE cohort.[6]

Researchers studied 1,018 people who had originally undergone assessment in midlife. Their mean age at baseline was approximately 48 years. They were reassessed at a mean age of about 72, after approximately 23 years of follow-up.

That long interval is a major strength.

Studies beginning in old age face an important problem: Alzheimer disease develops over many years before dementia is diagnosed. Early disease or declining health could already be causing people to change how much they drink.

Measuring drinking behavior around age 48 reduces, although does not eliminate, that concern.

In CAIDE, the association between drinking frequency and later dementia differed according to APOE4 status. Among APOE4 carriers, dementia risk increased with greater drinking frequency, and the statistical interaction between alcohol use and APOE4 was significant.[6]

There are important limitations.

Alcohol exposure was self-reported. The number of people became much smaller after dividing the cohort by both genotype and drinking category. And after more than two decades, the people who survived and returned for reassessment were necessarily a selected group.

There is another important detail: “frequent” drinking in CAIDE did not usually mean daily drinking. Most people in that category drank considerably less frequently.

So CAIDE raises an intriguing possibility that APOE4 may modify the consequences of alcohol beginning in midlife.

It does not tell us that one weekly glass causes dementia.

Higher habitual intake raises more concern

Another important study came from the Cardiovascular Health Study.[7]

Mukamal and colleagues compared 373 older adults who developed dementia with 373 matched controls, with approximately six years of follow-up. Participants were largely in their 70s and 80s.

One advantage was that alcohol consumption had been assessed on repeated occasions, and analyses accounted for many demographic and vascular factors.

Among APOE4 carriers, participants reporting 14 or more drinks per week had more than threefold higher odds of dementia compared with abstainers.[7]

But the subgroup was relatively small, which makes the estimate imprecise.

This study therefore tells us more about the potential concern surrounding higher habitual intake than about an occasional glass of wine.

A larger study did not confirm the APOE4 interaction

The story becomes less tidy when we look at another prospective cohort.

Koch and colleagues studied 3,021 dementia-free adults aged 72 or older in the Ginkgo Evaluation of Memory Study.[8] Their median age was 78 years.

During a median follow-up of approximately six years, 512 participants developed dementia.

Participants who already had mild cognitive impairment and consumed more than 14 drinks per week showed greater cognitive decline.

But importantly, APOE4 did not significantly modify the relationship between alcohol and dementia.[8]

That negative result matters.

It means that the APOE4-alcohol interaction seen in some cohorts has not been consistently replicated.

GEMS also has limitations. Alcohol intake was assessed at baseline rather than repeatedly over decades, and a cohort with a median age of 78 is a selected population of older survivors. Its findings may not apply directly to a healthy 56-year-old.

Taken together, these studies support a careful conclusion:

There are reasons to suspect that APOE4 may increase vulnerability to alcohol, but the evidence is not consistent enough to define an APOE4-specific threshold.

So how much is too much?

We do not know.

There is no scientifically established number of drinks per week that is known to be safe specifically for APOE4 carriers.

The literature becomes much more concerning with heavier habitual exposure, including intake of around 14 or more drinks per week in some cohorts.[7,8]

But 14 is not a biological cliff.

Thirteen drinks are not thereby safe and fourteen dangerous.

More importantly for our patient, those studies tell us very little about the difference between zero and one glass of wine per week.

That is exactly where her question sits.

Perhaps recovery matters as much as dose

There is another dimension that weekly totals obscure.

One glass once a week, one glass every night, and seven drinks on Saturday night can all produce very different biological exposures.

Peak alcohol concentration differs.

Effects on sleep differ.

And the time available for physiological recovery differs.

We do not yet know whether recovery from alcohol is systematically different in APOE4 carriers.

But I think the concept is useful because it shifts attention from the drink itself to the person’s response.

Did you sleep normally?

Do you feel cognitively normal?

Are you back to your baseline the next morning?

Those questions do not diagnose brain injury.

They help identify whether the exposure is producing an observable physiological cost.

So what did I tell her?

She had already discovered something important.

She stopped drinking alcohol, and her sleep improved enough that she began sleeping through the night.

I cannot tell her that one glass of wine would accelerate Alzheimer's disease.

We do not have evidence to support that statement.

But if she reintroduces a glass of wine and her nighttime awakenings reliably return, I would view that as meaningful.

For her, the exposure is disrupting sleep.

Given her elevated genetic susceptibility, I see little reason to repeatedly accept a physiological cost when there is no necessity to do so.

And I would strongly encourage her to preserve everything else about that weekly evening:

the friends, the games, the conversation, the laughter, and the social connection.

But maybe the wine is optional? Ok, one glass is ok… who knows!

What remains uncertain

We do not know whether APOE4 homozygotes are more sensitive to a single small dose of alcohol, and at what age or disease stage this becomes detrimental.

We do not know whether alcohol-related sleep disruption translates into greater Alzheimer risk.

We do not know whether once-in-a-while glass changes dementia risk in APOE4 carriers.

And although some prospective studies support an APOE4-alcohol interaction, others do not.[6-8]

Those uncertainties are real and should not be replaced with false precision.

But uncertainty does not mean that an individual’s response is irrelevant.

If removing alcohol produces a substantial improvement in sleep, and reintroducing it reproducibly reverses that improvement, that is useful biological information even if it is not an Alzheimer biomarker.

Take-home messages

• Alcohol affects brain function immediately, altering inhibitory and excitatory neurotransmission.[1,2]

• Feeling sleepy after alcohol is not the same as sleeping well. Experimental studies show that alcohol alters REM sleep, with greater disruption as dose increases.[3]

• People respond differently to the same amount of alcohol. Sleep disruption, cognitive changes, autonomic symptoms, or prolonged recovery are reasonable signals that a dose is not physiologically neutral.[3,4]

• APOE4 may be better understood as a vulnerability state than as a fixed abnormal state. Other human data suggest that a second stressor, such as persistent inflammation, can have different consequences in APOE4 carriers.[5]

• Alcohol could represent another second hit, but this remains a hypothesis rather than an established mechanism.

• Human studies suggest a possible interaction between APOE4 and alcohol, but the results are inconsistent. The strongest studies differ in age, drinking pattern, follow-up, and study design.[6-8]

• There is no established safe alcohol threshold specifically for APOE4 carriers.

• Higher habitual intake is much more concerning than an occasional single drink. Current epidemiology cannot tell us whether one weekly glass changes dementia risk in an otherwise healthy middle-aged APOE4 homozygote.[6-8]

• Individual response can still guide decisions. If stopping alcohol clearly improves sleep and restarting it reliably worsens sleep, that is meaningful information.

For an APOE4 carrier, the question may therefore be less:

“How many drinks am I allowed?”

and more:

“What does this amount of alcohol do to me, and how well do I recover?”

For this woman, her three months without alcohol may already have provided part of the answer.

She sleeps better. That is worth paying attention to.

References

1. Ron D, Barak S. Molecular mechanisms underlying alcohol-drinking behaviours. Nat Rev Neurosci. 2016;17:576-591. doi:10.1038/nrn.2016.85.

2. Roberto M, Varodayan FP. Synaptic targets: chronic alcohol actions. Neuropharmacology. 2017;122:85-99. doi:10.1016/j.neuropharm.2017.01.013.

3. Gardiner C, Weakley J, Burke LM, et al. The effect of alcohol on subsequent sleep in healthy adults: a systematic review and meta-analysis. Sleep Med Rev. 2025;80:102030. doi:10.1016/j.smrv.2024.102030.

4. de Zambotti M, Willoughby AR, Franzen PL, et al. Impact of evening alcohol consumption on nocturnal autonomic and cardiovascular function in adult men and women: a dose-response laboratory investigation. Sleep. 2021;44(1):zsaa135.

5. Tao Q, Ang TFA, DeCarli C, et al. Association of chronic low-grade inflammation with risk of Alzheimer disease in ApoE4 carriers. JAMA Netw Open. 2018;1(6):e183597. doi:10.1001/jamanetworkopen.2018.3597.

6. Anttila T, Helkala EL, Viitanen M, et al. Alcohol drinking in middle age and subsequent risk of mild cognitive impairment and dementia in old age: a prospective population based study. BMJ. 2004;329:539. doi:10.1136/bmj.38181.418958.BE.

7. Mukamal KJ, Kuller LH, Fitzpatrick AL, Longstreth WT Jr, Mittleman MA, Siscovick DS. Prospective study of alcohol consumption and risk of dementia in older adults. JAMA. 2003;289:1405-1413. doi:10.1001/jama.289.11.1405.

8. Koch M, Fitzpatrick AL, Rapp SR, et al. Alcohol consumption and risk of dementia and cognitive decline among older adults with or without mild cognitive impairment. JAMA Netw Open. 2019;2(9):e1910319. doi:10.1001/jamanetworkopen.2019.10319.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

Higher brain EPA relative to arachidonic acid was associated with better cognition and slower cognitive decline. In a complementary transcriptomic analysis, greater activation of the arachidonic acid pathway was associated with worse cognition. [1]

These findings fit with a pathway that our group has studied for years. But before interpreting them as evidence that inflammation causes cognitive decline, it is worth considering a major cardiovascular trial that recently produced a sobering result.

It illustrates one of the central problems in inflammation research:

An inflammatory pathway can be strongly associated with disease, a drug can successfully suppress that pathway, and the clinical disease may still not improve.

That lesson applies to our own work as much as it does to cardiovascular disease.

A major test of the inflammation hypothesis

For decades, higher circulating inflammatory markers have been associated with cardiovascular disease. Among the most prominent are interleukin-6, or IL-6, and C-reactive protein, or CRP.

One important test of whether inflammation was more than an association came from CANTOS.

CANTOS enrolled more than 10,000 people who had experienced a myocardial infarction and continued to have elevated hsCRP despite standard treatment. Participants received canakinumab, an antibody targeting IL-1β, or placebo.

The 150 mg dose reduced the primary cardiovascular endpoint by about 15 percent, with a hazard ratio of 0.85. This was statistically significant, but the clinical benefit was modest. All cause mortality was not reduced, and fatal infections were more common. Canakinumab was ultimately not developed as a cardiovascular prevention therapy. [2]

CANTOS was therefore important proof of concept. Manipulating an inflammatory pathway could affect cardiovascular events. But it was far from demonstrating that broadly targeting inflammation would transform cardiovascular prevention.

Then came ZEUS.

ZEUS: the biomarker moved, the disease did not

The phase 3 ZEUS trial studied ziltivekimab, an antibody targeting the IL-6 ligand.

The population seemed particularly well suited to an anti-inflammatory strategy. More than 6,300 participants had established atherosclerotic cardiovascular disease, chronic kidney disease, and evidence of systemic inflammation, defined by hsCRP of at least 2 mg/L.

The rationale was strengthened by the earlier phase 2 RESCUE trial, in which ziltivekimab produced striking reductions in hsCRP, reaching approximately 77 to 92 percent across tested doses, along with reductions in several other inflammatory biomarkers.

ZEUS asked the question that ultimately matters: would those biological effects translate into fewer cardiovascular events?

They did not.

According to headline results released in July 2026, ziltivekimab produced the expected inhibition of IL-6 signaling and reductions in free IL-6 and hsCRP. Yet the hazard ratio for cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke was 0.99, with a 95 percent confidence interval of 0.88 to 1.11.

In other words, cardiovascular outcomes were essentially identical to placebo. Serious infections were also more frequent with ziltivekimab. [3]

An important caveat is that, as of this writing, these are headline results. The complete ZEUS results have not yet been published, so mechanistic and subgroup conclusions should wait for the full dataset.

Still, the central result is difficult to ignore.

The inflammatory pathway was engaged. The biomarkers fell. The cardiovascular endpoint did not.

Association, target engagement, and causation are different things

ZEUS provides an unusually clear illustration of a problem that runs through biomedical research.

High IL-6 and CRP predict cardiovascular events.

That does not necessarily mean that IL-6 itself is the process that directly causes a vulnerable plaque to rupture and form a thrombus.

IL-6 participates in a large inflammatory network. It may contribute to atherosclerosis. It may amplify other responses. But circulating IL-6 may also reflect macrophage activity, renal inflammation, adipose inflammation, infection, or other processes occurring alongside vascular disease.

A biomarker can therefore be highly predictive without being the critical therapeutic lever.

There is another important limitation to how trials such as CANTOS and ZEUS approached inflammation.

They targeted defined molecular pathways, so the drugs themselves were not nonspecific. But patient selection was relatively blunt from the perspective of tissue biology.

Participants were selected because they had cardiovascular disease and evidence of systemic inflammation, primarily elevated hsCRP. That does not demonstrate that the IL-1β or IL-6 pathway is particularly active in the coronary plaque that will later rupture. Nor does lowering CRP tell us whether the vascular wall actually became biologically more stable.

This distinction may be crucial.

The question is not simply:

Is this person inflamed?

It is:

Which inflammatory pathway is active, where is it active, what tissue process is it affecting, and is that process actually causing the clinical event?

Alzheimer disease has the same problem

We frequently describe Alzheimer disease as having a major inflammatory component.

That is true, but “neuroinflammation” is an extraordinarily broad label.

Microglia, astrocytes, endothelial cells, pericytes, neurons, infiltrating immune cells, cytokines, complement proteins, reactive oxygen species, eicosanoids, and many other pathways can all participate.

More importantly, inflammation is not inherently pathological.

An inflammatory response may be:

A driver, actively producing tissue injury.

A response, appearing because amyloid, tau, vascular injury, infection, or dying neurons have triggered it.

Or compensatory, representing an attempt to clear damaged tissue and restore homeostasis.

The same pathway might even move between these roles over time.

This is especially important in a disease that evolves over decades.

It is difficult to imagine that the brain simply enters a continuously destructive inflammatory state in middle age and remains there unchanged for 30 years.

A more plausible model is dynamic.

Inflammatory pathways may turn on in response to an insult and subsequently resolve. Repeated activation may become increasingly difficult to terminate. With aging, amyloid accumulation, vascular disease, oxidative stress, or metabolic dysfunction, an initially adaptive response may become persistent and damaging.

And once tissue injury begins, the damaged tissue itself can generate new inflammatory stimuli.

The critical issue may therefore be not merely how much inflammation exists, but whether an inflammatory response resolves appropriately.

APOE4 does not simply mean “more inflammation”

This is where APOE4 becomes particularly interesting.

APOE4 is the strongest common genetic risk factor for late-onset Alzheimer disease, but it does not inevitably lead to dementia.

If APOE4 produced lifelong chronic inflammation, one might expect cognitively healthy APOE4 carriers to have consistently greater inflammation.

Human data suggest a more complicated picture.

In the Framingham Heart Study, APOE4 carriers actually had lower circulating CRP concentrations than APOE2 and APOE3 carriers across repeated examinations (Table).

Yet when persistent systemic inflammation was present, its association with Alzheimer disease was much stronger among APOE4 carriers. [4] Fig 1

That combination is important.

APOE4 itself was not associated with higher systemic inflammation in those observations. But inflammation occurring on the background of APOE4 appeared more consequential.

Our previous brain lipidomics work pointed in a similar direction.

Cognitively unimpaired APOE4 carriers did not show the exaggerated lipid inflammatory phenotype seen in APOE4 carriers with dementia. As a matter of the fact, the opposite. APOE4 carriers who died without a diagnosis of dementia had less inflammatory lipids in their brain. In the dementia brains, however, APOE4 was associated with larger elevations in both inflammatory and pro-resolving lipid mediators. [5]

This has led us toward a different model:

APOE4 may create an inflammatory vulnerability or fingerprint rather than a state of continuously elevated inflammation.

A second insult may be required.

That insult might be amyloid accumulation, vascular injury, metabolic stress, infection, trauma, oxidative stress, or some combination.

Once activated, the APOE4 brain may have greater difficulty bringing the response back under control.

Inflammation has to resolve

Resolution is not simply what happens when pro-inflammatory molecules disappear.

It is an active biological process.

Lipid mediators participate in switching inflammatory responses off, clearing damaged material, and restoring tissue homeostasis. Lipoxins can be derived from arachidonic acid, while DHA and EPA provide substrates for several families of specialized pro-resolving mediators.

This is why the simultaneous presence of inflammatory and pro-resolving lipid mediators in Alzheimer brain tissue is so interesting.

A high pro-resolving signal does not necessarily mean that inflammation is successfully resolving. It could mean that the tissue is mounting a compensatory response that is being overwhelmed.

The biological state may therefore be better described as unresolved inflammation.

cPLA2 and arachidonic acid provide a more specific pathway

Our recent study focused on one pathway that may help us move beyond the generic label of inflammation.

cPLA2 releases arachidonic acid, or AA, from membrane phospholipids.

AA can subsequently enter multiple metabolic pathways producing prostaglandins, leukotrienes, thromboxanes, lipoxins, and other oxylipins.

Some of these products promote inflammation. Others participate in resolution.

This is why AA itself should not simply be described as a “bad” or pro-inflammatory fatty acid.

The important question is how actively the pathway is turning over and what metabolic products are being generated.

cPLA2 can be activated by intracellular calcium, MAP kinase signaling, amyloid related stress, oxidative stress, and other cellular stimuli. Experimental APOE4 models show increased activation through pathways that include p38 MAP kinase.

And this pathway has another feature that may be just as important as inflammation.

It can amplify oxidative stress.

Our new study: linking the pathway to cognitive trajectories

In our newly published study, we analyzed postmortem dorsolateral prefrontal cortex tissue from participants who had undergone annual cognitive testing during life.

Brain lipidomics were available in 191 individuals, while transcriptomic data were analyzed in 411. [1]

Rather than simply comparing brains from people with and without dementia, we asked whether these lipid pathways were associated with the trajectory of cognition over time.

One of the clearest findings involved EPA relative to AA.

Higher brain EPA/AA was associated with better global cognition close to death and a slower rate of cognitive decline. A one standard deviation higher EPA/AA was associated with better late-life cognitive performance and slower terminal decline.

We then examined transcriptomic signatures related to AA metabolism.

Greater AA pathway activation was associated with worse cognition, including lower global cognition near death and impairment across several cognitive domains.

These observations support an association between the balance of cPLA2-related lipid metabolism and cognitive decline.

But the ZEUS lesson applies directly to us.

Our associations are not proof of causation either

Our findings do not prove that increased cPLA2 activity caused cognitive decline.

cPLA2 could contribute directly to neuronal or vascular injury.

But it could also be activated because amyloid, tau, oxidative stress, vascular disease, or neurodegeneration is already occurring.

And some lipid products may represent compensatory efforts to repair that injury.

Our study therefore identifies a candidate pathway, not a proven therapeutic target.

This distinction is not a weakness. It tells us what experiment needs to come next.

Looking beyond inflammation to oxidative and vascular injury

There is an additional reason cPLA2 interests us.

Lowering cPLA2 activation may do more than decrease inflammatory signaling.

Activated cPLA2 increases AA metabolism and downstream oxylipin production. PUFA metabolism and lipid peroxidation can generate reactive oxygen species and oxidized lipid products. Oxidative stress can, in turn, activate cPLA2 further.

This creates the possibility of a feed-forward cycle:

cellular stress → cPLA2 activation → PUFA release and metabolism → oxylipins and lipid oxidation → oxidative stress → further cPLA2 activation

Experimental work shows that inhibiting cPLA2 can reduce both inflammatory signaling and oxidative stress markers.

That may be particularly important at the neurovascular unit, the integrated system of endothelial cells, pericytes, astrocytic endfeet, neurons, and extracellular matrix that maintains the blood brain barrier and regulates delivery of nutrients to brain tissue.

Our recent Perspective led by Ambreen Kanwal and Bilal Kerman proposed that cPLA2 may sit at the intersection of AA signaling, oxylipin production, oxidative stress, membrane lipid depletion, cerebral amyloid angiopathy, and blood brain barrier dysfunction (Figure 5). [6] The paper also shows activated cPLA2 in and around cerebral vessels containing amyloid, providing a tissue based rationale for considering cPLA2 as part of a neurovascular inflammatory phenotype, rather than simply a generic brain inflammatory marker.

The proposed biology therefore extends beyond:

cPLA2 → inflammation

to something closer to:

cPLA2 → abnormal lipid signaling + oxidative stress + membrane vulnerability + neurovascular dysfunction

That distinction could matter therapeutically.

Can we see the pathway in the living brain?

This brings us to one of the biggest differences between the type of precision trial we envision and previous inflammation trials.

Blood CRP can tell us that systemic inflammation exists.

It cannot tell us whether cPLA2 is abnormally active in the brain or where that activity is greatest.

Fatty acid PET imaging offers a potential way to begin answering that question.

When cPLA2 releases AA from membrane phospholipids, the released fatty acid eventually needs to be replaced. Greater cPLA2-mediated AA turnover therefore creates increased demand for AA incorporation from the circulation.

Dynamic PET imaging can measure this incorporation.

Studies using [11C]AA PET have reported higher brain AA incorporation in Alzheimer disease, and the incorporation coefficient has been used as an indirect surrogate for cPLA2 activity. [7]

Carbon-11 has a short radioactive half-life of about 20 minutes. Newer 18F-labeled AA probes, with a half-life of approximately 110 minutes, may make this approach more practical for multicenter studies.

DHA PET provides complementary information about brain omega-3 uptake and metabolism. We have reported that young cognitively normal APOE4 carriers have shown higher DHA incorporation coefficients than noncarriers, possibly reflecting greater utilization or compensatory uptake.

PET will not give us microscopic cellular resolution. An AA PET signal cannot, by itself, tell us whether cPLA2 is active specifically in an endothelial cell, pericyte, astrocyte, microglial cell, or neuron.

But it can potentially provide something that a blood inflammatory marker cannot:

regional evidence that an abnormal lipid metabolic pathway is active in the organ we intend to treat. Tissue studies can then help us understand which cells and structures contribute to that signal.

We are actively developing 18-F DHA PET tracers to measure this inflammation in the living brain.

From inflammation trials to precision inflammation trials

This may be the larger lesson.

CANTOS and ZEUS targeted specific cytokines, but patients were selected using clinical disease and circulating evidence of inflammation. They did not establish that the targeted inflammatory pathway was active in the particular arterial lesion responsible for the future event.

For Alzheimer disease, we should try to go further.

We imagine a trial in which we:

1. identify a biologically susceptible population, potentially including APOE4 carriers;

2. demonstrate evidence of excessive cPLA2-related AA metabolism in the living brain;

3. determine where in the brain the abnormality is occurring;

4. administer a selective, brain-penetrant cPLA2 inhibitor;

5. use PET to establish that brain pathway activity actually changed;

6. measure independent consequences of that intervention, including oxidative stress and neurovascular function;

7. and only then ask whether cognitive decline slows.

For me, this is a precision inflammation trial.

The critical difference is that we would not treat someone simply because they carry APOE4 or have an elevated systemic inflammatory biomarker.

We would attempt to show that the specific pathway we intend to inhibit is abnormal in the target organ before treatment begins.

Even that may fail.

We could lower DHA/AA PET uptake, reduce biochemical markers of pathway activity, and still discover that cognition does not improve. If so, we would have our own version of ZEUS. That is how science goes!!

Perhaps cPLA2 activation is a response to injury rather than the driver.

Perhaps we intervened too late.

Perhaps another pathway is more important.

But that would still be a far more informative than a blunt, expensive negative large trial because we would know that the drug reached the relevant organ and changed the biology we intended to change.

Take Home Messages

• Association is not causation. ZEUS demonstrated that profound suppression of an inflammatory pathway and its biomarkers does not guarantee improvement in clinical outcomes.

• Inflammation is not one biological state. An inflammatory signal may cause injury, respond to injury, or represent an attempt to repair it. Its role may also change over the course of disease.

• APOE4 may confer an inflammatory vulnerability rather than simply causing lifelong inflammation. Human data suggest that the relevant phenotype may emerge when APOE4 is combined with additional inflammatory, metabolic, vascular, or neurodegenerative stress.

• Our new study links cPLA2-related AA metabolism to cognitive trajectories, but it does not prove that the pathway causes cognitive decline. That is the hypothesis that now needs experimental testing.

• cPLA2 may matter for more than inflammation. Its activation connects lipid signaling with PUFA oxidation, oxidative stress, membrane injury, and potentially dysfunction of the neurovascular unit.

• Where a pathway is active matters. Blood inflammatory markers cannot tell us whether the pathway being targeted is active in the brain. AA and DHA PET imaging may provide regional metabolic information that helps bridge this gap.

• Future trials should become precision trials. The ideal study would select patients with evidence that the targeted pathway is active in the relevant organ, demonstrate that treatment changes that pathway, measure downstream tissue biology, and then determine whether the clinical disease changes.

The goal is therefore not simply to develop a stronger anti-inflammatory drug.

It is to identify the right pathway, in the right patient, in the right tissue, at the right stage of disease, and then prove that changing that pathway changes the disease.

That is a much harder experiment. It is also the experiment that inflammation research increasingly needs.

Full reference

Yaskolka Meir A, Wang X, Li B, Wilson RS, Bennett DA, Louie SG, Yassine HN, Arvanitakis Z. Brain pro-resolving-to-pro-inflammatory lipids and cPLA2 activation associations with late-life cognitive trajectories. eBioMedicine. 2026;131:106421. doi:10.1016/j.ebiom.2026.106421.

Acknowledgements

We thank the ROS and MAP participants for their tireless altruism towards the pursuit of scientific knowledge about ageing human conditions.

This research was supported by National Institutes of Health Grants: R01AG076124, R01NS084965, RF1AG059621, R01AG074549, P30AG10161, P30AG072975, R01AG15819, R01AG017917, U01AG61356, and P30AG066530. A.Y.M. was supported by the Council for Higher Education-Zuckerman support program for outstanding postdoctoral female researchers.

References

1. Yaskolka Meir A, Wang X, Li B, et al. Brain pro-resolving-to-pro-inflammatory lipids and cPLA2 activation associations with late-life cognitive trajectories. eBioMedicine. 2026;131:106421.

2. Ridker PM, Everett BM, Thuren T, et al. Antiinflammatory therapy with canakinumab for atherosclerotic disease. N Engl J Med. 2017;377:1119-1131.

3. Novo Nordisk. Update on the ZEUS phase 3 trial in people with ASCVD, CKD and inflammation. July 31, 2026. Full scientific results pending at the time of writing.

4. Tao Q, Ang TFA, DeCarli C, et al. Association of chronic low-grade inflammation with risk of Alzheimer disease in ApoE4 carriers. JAMA Netw Open. 2018;1:e183597.

5. Ebright B, Asante I, Poblete RA, et al. Eicosanoid lipidome activation in postmortem brain tissues of individuals with APOE4 and Alzheimer’s dementia. Alzheimers Res Ther. 2022;14:152.

6. Kanwal A, Kerman BE, Wang S, et al. A perspective: PLA2G4A as drug target for vascular inflammation in Alzheimer’s disease. Alzheimers Dement. 2026;22:e71320.

7. Duro MV, Ebright B, Yassine HN. Lipids and brain inflammation in APOE4-associated dementia. Curr Opin Lipidol. 2022;33:16-24.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

I realized that, although I often discuss cognitive health and dementia prevention, I had not looked closely enough at the brain training literature itself. So I went back to several landmark experiments, the major cognitive training trials, the newer long-term follow-up data, and the 2026 WHO dementia prevention guideline.

The question is increasingly relevant. Commercial brain training has grown from simple puzzles into a large digital health and wellness industry. Apps commonly promote improvements in memory, attention, processing speed, or general “brain health.” Some have historically made stronger claims about slowing cognitive decline or dementia.

There is good reason to study the idea. The adult brain remains capable of changing with learning, and cognitive training can improve some of the abilities being practiced. But improving performance on a cognitive exercise is not the same as preventing dementia.

So what does cognitive stimulation actually do, how strong is the evidence that it affects dementia risk, and where should it fit within a broader prevention strategy?

The Brain Changes With Learning

The biological rationale for cognitive stimulation begins with neuroplasticity, the brain’s ability to change in response to experience.

One of the best-known demonstrations comes from London taxi drivers. To earn a black cab license, drivers historically had to master “The Knowledge,” an extraordinarily detailed mental map of London involving thousands of streets and landmarks (Fig 1).

This was not simply learning the names of streets. They had to know thousands of roads, landmarks, and routes well enough to mentally calculate efficient journeys between locations without relying on a map.

In a landmark MRI study, Maguire and colleagues found that experienced taxi drivers had more gray matter in the posterior hippocampus, a brain region important for spatial navigation. The longer they had worked as taxi drivers, the greater this difference tended to be. [1]

There was an obvious limitation. Perhaps people with a particular hippocampal anatomy were simply more likely to become successful taxi drivers.

A later longitudinal study addressed this by scanning trainees before they had mastered The Knowledge and following them for several years. Those who eventually qualified showed increased posterior hippocampal gray matter. Trainees who did not qualify and control participants did not show the same change. [2]

The lesson is important:

The brain adapts to what it repeatedly practices.

That does not mean every aspect of cognition improves. The taxi drivers developed changes in brain systems related to the unusually demanding spatial skill they had spent years acquiring.

Other experiments have found structural brain changes after people learned new motor and visual skills, including in older adults. Neuroplasticity therefore persists into later life.

But neuroplasticity is not the same as dementia prevention. Showing that learning changes the brain does not establish that it reduces amyloid, tau, vascular injury, or neurodegeneration.

Cognitive Reserve

A related concept is cognitive reserve, the capacity to maintain cognitive function despite brain aging or pathology.

The classic Nun Study helped illustrate this idea. Researchers analyzed autobiographies written by Catholic sisters when they were young adults and followed them many decades later. Greater linguistic complexity in early adulthood was associated with better cognition in later life and with differences in Alzheimer related neuropathology at autopsy. [3]

The study did not show that writing complex sentences prevents dementia. Early linguistic ability could reflect education, development, baseline cognitive ability, and other lifelong influences.

Its importance was conceptual: people can differ substantially in how much brain pathology they tolerate before cognitive impairment becomes clinically apparent.

Cognitive stimulation may therefore contribute to resilience without necessarily preventing Alzheimer pathology itself. WHO similarly defines cognitive reserve as the brain’s ability to cope with or compensate for neuropathology or damage and notes that greater education and occupational complexity are associated with lower dementia risk. [4]

Cognitive Stimulation Is Broader Than a Brain Game

WHO distinguishes cognitive training from cognitive stimulation.

Cognitive training generally means repeated exercises targeting specific abilities such as memory, attention, reasoning, or processing speed. Cognitive stimulation is broader and includes reading, learning, games, storytelling, and other intellectually engaging activities. [4]

Social interaction belongs in this discussion too. Conversation requires attention, language, memory, interpretation, and rapid response. Playing a complex game with other people adds strategy, prediction, inhibition, and decision making.

Observational studies consistently associate greater social engagement with lower dementia risk. WHO reports a relative risk of about 0.81, meaning roughly 19% lower observed dementia risk among people with greater social engagement. Randomized social interventions show small cognitive benefits, but they have not demonstrated that increasing social activity prevents dementia. [4]

There is also the problem of reverse causation. Early neurodegenerative disease can cause people to withdraw socially years before dementia is diagnosed. Some of the association between isolation and dementia may therefore reflect early disease rather than isolation causing the disease.

What Do Brain Training Studies Actually Show?

Across the evidence reviewed for its 2026 guideline, WHO concluded that cognitive training produces a small average improvement in cognitive performance, with an effect size called Hedges’ g of approximately 0.25. [4]

Hedges’ g expresses the difference between two groups in standard deviation units. A value of 0.25 means that, on average, the trained group performed about one quarter of a standard deviation better than the comparison group. This is generally considered a small effect.

So brain training can “work” if we mean improving cognitive performance.

The more difficult question is whether improvement extends beyond what was practiced.

Improvement on closely related tasks is sometimes called near transfer. Improvement in more distant abilities, such as driving, managing finances, or maintaining independence, is called far transfer. Evidence for far transfer is less consistent.

The most important question is harder still: does cognitive training actually reduce dementia?

Most cognitive training studies cannot answer that. They typically last weeks or months and measure cognitive test scores rather than dementia incidence. WHO therefore says cognitive training may be offered to older adults with normal cognition or mild cognitive impairment, but the recommendation is conditional and the certainty of evidence is low. [4]

A Long Term Test of Cognitive Training

The ACTIVE trial provides one of the most informative examples. It enrolled 2,802 adults aged 65 and older and randomly assigned them to memory training, reasoning training, processing speed training, or no training. [5]

Memory and reasoning programs taught strategies for those particular abilities.

The processing speed intervention required participants to identify visual information rapidly while also noticing information elsewhere in the visual field.

Processing speed refers to how quickly the brain can take in information and respond.

Divided attention means attending to more than one source of information at the same time.

The training was also adaptive. As someone improved, the exercise became faster or more difficult. Participants therefore continued working close to the limits of their current ability rather than simply repeating a task that had become easy.

Initial training consisted of up to ten 60 to 75 minute sessions over about 5 to 6 weeks. Participants who completed at least eight sessions became eligible for later refresher, or booster, sessions at approximately 11 and 35 months. [6]

There is a reasonable neuroscience rationale for this kind of training. Processing speed depends on distributed systems involved in visual attention, rapid information selection, and cognitive control. Repeated adaptive training may improve the efficiency of these systems.

There was also evidence of some transfer to everyday function. Earlier ACTIVE analyses found that speed training was associated with fewer at fault automobile collisions, an outcome requiring rapid visual processing and attention outside the laboratory. [6]

That makes the intervention interesting. It still does not establish that the training alters Alzheimer pathology.

What Happened Over 20 Years?

A 2026 analysis linked ACTIVE participants to Medicare records through 2019 and examined diagnosed Alzheimer disease and related dementias over approximately 20 years. The analysis included 2,021 participants. [6]

The first finding is important: none of the three original randomized training groups had a statistically significant reduction in dementia diagnoses.

The hazard ratios were 0.85 for memory training, 0.87 for speed training, and 0.88 for reasoning training. In each case, the 95% confidence interval included 1.0. [6]

A hazard ratio, or HR, compares the rate at which an event occurs over time. An HR of 1.0 means no difference between groups. Values below 1.0 favor the intervention.

The 95% confidence interval gives a range of effect sizes reasonably compatible with the observed data. If that interval includes 1.0, the study cannot rule out the possibility that there was actually no difference between the groups and that the observed difference arose from random variation in the sample. By convention, that result is not considered statistically significant.

The notable finding appeared among participants who received speed training plus booster sessions (Table below). Their HR for a dementia diagnosis was 0.75, with a 95% confidence interval of 0.59 to 0.95. In practical terms, the rate of dementia diagnosis during follow-up was about 25% lower than in controls. Because the confidence interval remained below 1.0, the result met the conventional threshold for statistical significance. Speed training without boosters had an HR of 1.01, essentially no difference from controls. [6]

The finding is encouraging, but it needs qualification.

Participants first had to complete most of their initial training before becoming eligible for boosters. This could select people who were healthier, more adherent, or less likely to have subtle early cognitive impairment. Booster assignment among eligible participants was randomized, which reduces this concern but does not eliminate the selection that occurred beforehand. [6]

Dementia was also identified from Medicare diagnostic codes rather than detailed neurological examinations or Alzheimer biomarkers.

The appropriate conclusion is therefore fairly narrow: repeated adaptive processing speed and divided attention training has encouraging long term evidence, but this does not establish that cognitive training as a whole prevents dementia.

Should I Pay for a Brain Training App?

For most people, the evidence does not support saying that they need to.

The long-term processing speed findings do not establish that Sudoku, crossword puzzles, memory matching, word games, or collections of unrelated cognitive exercises prevent dementia.

If someone enjoys a brain training app, finds it challenging, and uses it consistently, that is reasonable. But there are many other ways to obtain substantial cognitive stimulation without purchasing a subscription.

What Can I Actually Do?

A useful principle is to choose activities that require continued learning rather than comfortable repetition.

Repetition is important while a skill is being learned. But once an activity becomes highly practiced and automatic, the cognitive demand falls. The evidence points more toward continued challenge than novelty for its own sake.

This is one reason adaptive training is interesting: as performance improves, the difficulty increases. WHO similarly notes that everyday cognitively stimulating activities should ideally remain new or challenging, while acknowledging that the optimal type, frequency, intensity, and duration are still unknown. [4]

A familiar activity can therefore remain cognitively stimulating if it continues to become harder.

Mahjong is one example. It combines visual search, memory, pattern recognition, planning, inhibition, decision-making, and usually social interaction. A randomized study in older adults with mild cognitive impairment found improvements in several cognitive measures after regular Mahjong play. [7] The study was small and short and did not test dementia prevention.

Comparable activities vary across cultures and interests. Chess, Go, bridge, card games, and traditional strategy games can all require sustained attention, planning, memory, and adaptation to other players.

A lifelong chess player does not necessarily need to abandon chess and take up something completely different. Playing stronger opponents, learning unfamiliar openings, or studying more difficult positions can restore the learning challenge.

Learning another language is another example. In a landmark observational study, Bialystok and colleagues reported that lifelong bilingual patients developed dementia symptoms about four years later than monolingual patients. [8] This contributed to the hypothesis that decades of managing two languages might build cognitive reserve through repeated demands on attention and executive control.

But lifelong bilingualism is not randomly assigned, and it is not equivalent to beginning an 11-week language course at age 70. Short randomized language interventions in older adults have generally shown much more modest and inconsistent transfer to other cognitive abilities.

Still, a language course can provide sustained learning, memory retrieval, auditory processing, and social interaction. Progressing from vocabulary exercises to conversation, reading, and increasingly difficult material keeps the task challenging.

Other possibilities include:

• learning an instrument or progressing to harder music

• taking a class in history, science, art, computing, or another unfamiliar subject

• joining a discussion or book group that tackles demanding material

• learning photography, painting, woodworking, sewing, coding, or another complex skill

• dancing, particularly when learning new sequences and coordinating with others

• volunteering in a role that requires planning, communication, and learning new responsibilities

• cooking unfamiliar cuisines or learning increasingly complex techniques

The specific activity probably matters less than several common features: it should require attention, learning, problem solving, and enough progression that it does not become entirely automatic.

Activities that involve other people may also combine cognitive stimulation with social engagement.

What We Still Do Not Know

Important uncertainties remain.

We do not know whether cognitive training alters the underlying biology of Alzheimer disease or primarily improves neural efficiency and cognitive reserve.

We do not know how much improvement transfers from a practiced task to everyday function.

We also do not know the optimal type, intensity, or duration of cognitive stimulation, or whether decades of cognitive activity have effects that cannot be reproduced by beginning training later in life.

And no long-term trial has established that one particular game, language, musical activity, educational program, or commercial app is the preferred strategy for preventing dementia.

Take Home Messages

• The adult and aging brain remains plastic. Learning can produce measurable changes in brain structure and function.

• The brain adapts to what it practices. Improvement on a trained task does not necessarily translate into broad cognitive improvement.

• Cognitive training produces small average improvements in cognitive performance, but evidence for dementia prevention remains limited.

• A specific form of repeated adaptive processing speed and divided attention training has encouraging long-term evidence, but this cannot be generalized to all brain games.

• Most people do not need to pay for a commercial brain training program.

• Meaningful cognitive stimulation can come from continued learning, strategy, problem solving, and progressively challenging skills.

• Cognitive stimulation is best viewed as one component of a broader dementia risk reduction strategy, alongside physical activity, social engagement, hearing and vision care, and management of vascular and metabolic risk factors.

The practical goal is not to identify the best brain game. It is to maintain a cognitively active life while addressing the other modifiable factors that influence how the brain ages.

References

1. Maguire EA, Gadian DG, Johnsrude IS, et al. Navigation related structural change in the hippocampi of taxi drivers. Proc Natl Acad Sci USA. 2000;97:4398-4403.

2. Woollett K, Maguire EA. Acquiring “the Knowledge” of London’s layout drives structural brain changes. Curr Biol. 2011;21:2109-2114.

3. Snowdon DA, Kemper SJ, Mortimer JA, et al. Linguistic ability in early life and cognitive function and Alzheimer’s disease in late life: findings from the Nun Study. JAMA. 1996;275:528-532.

4. World Health Organization. Risk Reduction of Cognitive Decline and Dementia: WHO Guidelines. 2nd ed. Geneva: WHO; 2026.

5. Ball K, Berch DB, Helmers KF, et al. Effects of cognitive training interventions with older adults: a randomized controlled trial. JAMA. 2002;288:2271-2281.

6. Coe NB, Miller KEM, Sun C, et al. Impact of cognitive training on claims based diagnosed dementia over 20 years: evidence from the ACTIVE study. Alzheimer’s Dement. 2026;12:e70197.

7. Zhang H, Peng Y, Li C, et al. Playing Mahjong for 12 weeks improved executive function in elderly people with mild cognitive impairment. Front Neurol. 2020;11:178.

8. Bialystok E, Craik FIM, Freedman M. Bilingualism as a protection against the onset of symptoms of dementia. Neuropsychologia. 2007;45:459-464.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy. I have no relevant conflicts of interest. These opinions are mine and do not reflect the opinions of my employer.

She had undergone genetic testing and learned that she carries one copy of APOE4, the most common genetic risk factor for late-onset Alzheimer’s disease. She wanted to be proactive about prevention.

Her primary care physician had started her on what she called a low-dose “GLP3” drug to lower inflammation and reduce her future risk of dementia.

But her metabolic profile was already favorable. Her BMI was below 25. Her hemoglobin A1c was 4.6%. Her high-sensitivity C-reactive protein, or hsCRP, was undetectable. She did not have diabetes, obesity, or an obvious metabolic abnormality.

Her question was simple:

Is this actually good for my brain?

GLP-1–based drugs have substantial benefits when they treat a biological problem that is present. In obesity, diabetes, insulin resistance, and cardiovascular disease, there is a measurable target.

But what happens when the target is missing?

For a lean, normoglycemic APOE4 carrier, we have no evidence that GLP-1 therapy prevents dementia. At the same time, these drugs suppress appetite and energy intake, cause gastrointestinal side effects, and can reduce both lean mass and fat.

That changes the risk-benefit calculation.

First: What Drug Are We Talking About?

There is no standard medication class called “GLP3.”

Patients may mean a GLP-1 receptor agonist such as semaglutide, a dual GIP/GLP-1 agonist such as tirzepatide, or one of the newer multi-receptor agonists.

The exact molecule, dose, formulation, and source matter, especially with compounded products. But even with an FDA-approved product, the more important question is the indication.

What Problem Are We Treating?

The major GLP-1 trials were conducted in people with identifiable metabolic or cardiovascular risk.

The SELECT trial enrolled 17,604 adults age 45 or older with BMI ≥27 and established cardiovascular disease, but without diabetes. Semaglutide reduced major cardiovascular events in this high-risk population. [1]

STEP 1 enrolled adults with BMI ≥30, or BMI ≥27 with at least one weight-related condition, and demonstrated substantial weight loss with semaglutide. [2]

These are important results, but neither trial studied healthy, normal-weight adults taking GLP-1 therapy for longevity or dementia prevention.

The evidence supports a narrower conclusion:

GLP-1 drugs improve important outcomes when there is metabolic or cardiovascular pathology to treat.

They should not automatically be extrapolated into longevity drugs for metabolically healthy people.

Inflammation Is Not One Biological Process

The argument that GLP-1 drugs might prevent dementia because they “lower inflammation” needs more precision.

Inflammation is not a single biological state. Acute immune responses, autoimmune inflammation, infection-related inflammation, neuroinflammation, and the chronic low-grade inflammation associated with obesity and insulin resistance involve overlapping but distinct pathways.

The inflammatory state most closely linked to the metabolic effects of GLP-1 therapy is metabolic inflammation, sometimes called metaflammation: chronic, low-grade immune activation associated with excess adiposity, insulin resistance, type 2 diabetes, and metabolic syndrome.

In that setting, dysfunctional adipose tissue, altered adipokine signaling, and immune-cell activation contribute to insulin resistance and cardiometabolic disease. Reducing excess adiposity and improving insulin sensitivity can therefore reduce a real inflammatory burden. [10]

But that is different from assuming that any reduction in systemic inflammation is beneficial to the brain.

Neuroinflammation in Alzheimer’s disease is not synonymous with metabolic inflammation, and circulating CRP is not a direct measure of inflammatory activity in the brain.

That distinction is especially relevant in someone who is lean, normoglycemic, and has undetectable hsCRP.

EVOKE Helps Separate Systemic Inflammation From Brain Benefit

The EVOKE and EVOKE+ trials are informative because participants were selected for early symptomatic Alzheimer’s disease—not for obesity, diabetes, insulin resistance, or cardiovascular disease.

Participants were 55 to 85 years old with mild cognitive impairment or mild dementia due to Alzheimer’s disease, objective cognitive impairment, and confirmed amyloid pathology. Mean BMI was about 25.7 kg/m², nearly half had a normal BMI, and only a minority had obesity or type 2 diabetes. [3]

This was very different from the populations in SELECT and STEP 1.

EVOKE and EVOKE+ randomized 3,808 participants to oral semaglutide or placebo. Semaglutide did not slow decline on the primary clinical endpoint, the Clinical Dementia Rating–Sum of Boxes. Figure 1 [3]

Yet semaglutide lowered hsCRP, Figure 2.

That combination is important:

Systemic inflammation decreased, but clinical Alzheimer’s progression did not.

This does not mean inflammation is unimportant in Alzheimer’s disease. Nor does it exclude the possibility that treating obesity, insulin resistance, diabetes, or vascular disease earlier in life could indirectly reduce dementia risk.

But it shows that lowering a systemic inflammatory marker with a GLP-1 receptor agonist, in a population not selected for major metabolic disease, was not sufficient to produce cognitive benefit.

For a lean patient whose hsCRP is already undetectable, lowering CRP further has no demonstrated meaning for brain protection yet.

But the more relevant question is:

What inflammatory process is present, and is that process actually being treated?

The Lean-Mass Problem

For a lean APOE4 carrier, the concern is not simply weight loss. It is catabolism.

GLP-1 drugs reduce appetite and energy intake. In a person with obesity, that can be beneficial because much of the lost weight is excess fat. But the weight lost is not entirely fat.

In the STEP 1 body-composition substudy, semaglutide was associated with reductions in both fat and lean mass. DXA-measured lean mass is not identical to skeletal muscle—it also includes water, organs, connective tissue, and glycogen-associated water—but the finding remains relevant when there is little excess weight to lose.

In the SURMOUNT-1 DXA substudy, approximately 75% of weight lost with tirzepatide was fat mass and about 25% was lean mass, Figure 3. [4]

For someone with obesity, the net change may still be favorable.

For someone who is already lean, there is less excess adipose tissue to remove and less margin for avoidable loss of lean tissue.

Regaining Lost Muscle May Be More Difficult With Aging

A loss of lean mass with aging is problematic. A 2025 study by Jiang and colleagues provides important insights on why it is difficult to regain lean mass as we age. The investigators used UM-HET3 mice, the genetically heterogeneous strain used extensively in the NIA Interventions Testing Program. Importantly, these mice were first made obese and metabolically unhealthy with a high-fat diet. [5]

Continuous liraglutide improved weight and metabolic health. Again, the drug was acting in a system with clear metabolic pathology to correct.

Repeated treatment and withdrawal produced a different pattern. Fat mass returned more readily than lean mass, while visceral adiposity and hyperleptinemia increased and some metabolic benefits disappeared.

This was not a lifespan or cognition experiment, and mouse findings should not be assumed to occur in humans. But it illustrates an important concept:

Weight regain after stopping GLP1s is not necessarily muscle recovery.

That may matter more with aging, when skeletal muscle is already harder to rebuild because of anabolic resistance and progressive loss of muscle mass and strength.

Returning to the same number on the scale may not mean returning to the same body composition.

Why APOE4 and Catabolism Matter

APOE4 affects lipid transport, glucose metabolism, inflammation, and brain energy metabolism. This makes weight loss more complicated than simply reducing fat mass.

Adipose tissue and skeletal muscle are also endocrine organs. They release adipokines and myokines that participate in signaling between peripheral tissues and the brain and can influence metabolism, vascular function, inflammation, and neuronal health.

When obesity, insulin resistance, diabetes, or metabolic syndrome are present, reducing dysfunctional adipose tissue may improve metabolic inflammation and vascular risk enough to outweigh some loss of lean mass.

But the balance may be different in a lean APOE4 carrier. Loss of adipose tissue and muscle can also reduce metabolic reserve and alter adipokine and myokine signaling.

In our review of APOE alleles and diet, Caleb Finch and I discussed evidence that APOE4 alters brain energy metabolism and that the cognitive consequences of weight loss may differ by genotype. We highlighted Look AHEAD, where favorable cognitive effects of intensive weight loss were seen primarily in APOE4 non-carriers. [6]

These findings do not prove that GLP-1–induced weight loss harms APOE4 carriers. They suggest a more nuanced principle:

When metabolic disease is present, correcting it may provide substantial brain benefit. When metabolic health is already favorable, loss of muscle, adipose-derived signals, and metabolic reserve may carry greater importance.

Side Effects Matter

The most common adverse effects of GLP-1–based therapies are gastrointestinal.

Jalleh and colleagues reviewed nausea, vomiting, diarrhea, constipation, delayed gastric emptying, and related gastrointestinal effects of GLP-1 receptor agonists in The Lancet Gastroenterology & Hepatology. [7]

A large systematic review and meta-analysis of randomized trials also found increased risks of cholelithiasis and probably gastroesophageal reflux disease. [8]

For a patient with obesity, diabetes, or substantial cardiovascular risk, these adverse effects may be an acceptable tradeoff for demonstrated benefit.

For a healthy lean person, the calculation is different.

Reduced appetite may also reduce total energy, protein, and micronutrient intake. In someone with little excess weight to lose, that reinforces the concern about catabolism and preservation of lean tissue.

Does “Microdosing” Solve the Problem?

An increasingly common proposal is to use a very small dose of a GLP-1 drug—often called microdosing—with the hope of obtaining metabolic or anti-inflammatory effects while avoiding substantial weight loss or side effects.

The problem is that this strategy has not been validated for dementia prevention.

Komé, Chandran, Buse, Klein and colleagues discussed fractional semaglutide dosing in Diabetes Care in 2025, largely in the context of individualized diabetes care, tolerability, and cost. It was not evidence that microdosed semaglutide prevents Alzheimer’s disease or extends lifespan. [9]

The major clinical benefits of GLP-1 drugs were demonstrated at specific studied doses in defined disease populations.

Lower doses may reduce adverse effects, but they may also reduce efficacy. A patient may still experience early satiety, nausea, constipation, reflux, or reduced food intake at a dose that has never been shown to reproduce the clinical benefits demonstrated in outcome trials.

So microdosing introduces a different uncertainty:

A person may receive enough drug to alter normal physiology without receiving a dose that has demonstrated clinical benefit.

When GLP-1 Therapy Is Indicated, Protect Muscle

None of this argues against GLP-1 therapy when there is an appropriate indication.

If a patient has obesity, diabetes, insulin resistance, cardiovascular disease, or another established indication, the potential benefits may substantially outweigh the risks.

But muscle preservation deserves particular attention whenever these drugs are used in older adults or anyone vulnerable to age-related muscle loss.

Aging itself is associated with progressive loss of muscle mass and strength and reduced anabolic responsiveness. Adding pharmacologic weight loss to that process makes muscle an important treatment outcome.

Resistance training should be incorporated into the therapeutic plan, together with adequate protein and total energy intake. In older adults, monitoring weight alone is insufficient. Strength, functional performance, dietary intake, and, where appropriate, body composition should also be followed. [11]

The pharmaceutical pipeline is moving in the same direction. Investigational drugs targeting myostatin and activin signaling are being combined with GLP-1–based therapies to preserve lean mass while maintaining fat loss. Bimagrumab has been studied with semaglutide, and apitegromab with tirzepatide, with early trials showing improved preservation of lean tissue. [12,13]

These approaches remain experimental, but they reinforce an important shift:

The goal should not simply be maximum weight loss. It should be loss of excess fat while preserving functional muscle.

What We Still Do Not Know

The uncertainty cuts in both directions.

EVOKE and EVOKE+ studied people with symptomatic Alzheimer’s disease. They cannot tell us whether treatment decades earlier might alter dementia risk.

APOE4 carriers who also have obesity, diabetes, insulin resistance, metabolic syndrome, or substantial vascular disease may benefit when those conditions are effectively treated.

Conversely, we do not have randomized trials showing that GLP-1 therapy causes cognitive harm in lean APOE4 carriers.

The concern about catabolism is therefore biologically plausible, not clinically proven.

A definitive prevention trial would require metabolically healthy participants, APOE genotyping, careful measurement of muscle and body composition, nutritional status, inflammatory phenotype, drug exposure, and long-term cognitive outcomes.

That study is not likely to be done, given the complexity of long-term cognitive outcome trials.

For preventive pharmacology in someone who is currently healthy, absence of demonstrated harm should not be mistaken for evidence of benefit.

What I Would Tell This Patient

I would tell her that GLP-1 drugs are effective medications when there is a clear metabolic or vascular indication.

But we currently do not have evidence that either standard-dose or microdose GLP-1 therapy prevents dementia in a lean, normoglycemic APOE4 carrier with undetectable hsCRP and no clear evidence of metabolic or vascular inflammation.

I would return to one question:

What abnormality are we treating?

If the answer is obesity, diabetes, insulin resistance, metabolic syndrome, cardiovascular disease, fatty liver disease, or another established indication, there is a biological target and an evidence base. Treat it—and, particularly with aging, actively protect muscle.

If the answer is simply APOE4 carrier status and dementia prevention, the evidence is not there.

For APOE4 carriers seeking prevention, we already have measurable targets: blood pressure, apoB and LDL cholesterol, glycemic health, fitness, muscle strength, sleep apnea, smoking, diet quality, hearing, and other vascular and metabolic risk factors.

The goal should be to identify abnormalities and correct them—not assume that altering normal metabolism is beneficial simply because the same drug helps people with metabolic disease.

Take-Home Messages

• GLP-1 drugs have proven benefits when there is metabolic or cardiovascular disease to treat.

• They are not proven longevity or dementia-prevention drugs in healthy, lean adults.

• Inflammation is heterogeneous. Metabolic inflammation associated with obesity, insulin resistance, diabetes, and metabolic syndrome is not the same as neuroinflammation.

• In EVOKE/EVOKE+, semaglutide lowered systemic inflammation in an Alzheimer’s population not selected for obesity or diabetes, yet did not slow cognitive decline.

• GLP-1–associated weight loss includes lean-mass loss, which matters increasingly with aging.

• In APOE4 carriers, treat harmful metabolic disease when it is present, but do not assume that loss of muscle, adipose tissue, and metabolic reserve is inherently brain-protective when metabolic health is already favorable.

• Withdrawal deserves attention because fat may return more readily than lean tissue.

• Microdosing has not been validated for dementia prevention and may produce pharmacologic effects without proven clinical benefit.

• When GLP-1 therapy is indicated, particularly during aging, resistance training, adequate nutrition, and monitoring of muscle and function should be part of treatment.

• The central clinical question remains: What biological abnormality are we treating?

References

1. Lincoff AM, Brown-Frandsen K, Colhoun HM, et al. Semaglutide and cardiovascular outcomes in obesity without diabetes. New England Journal of Medicine. 2023.

2. Wilding JPH, et al. Once-weekly semaglutide in adults with overweight or obesity. New England Journal of Medicine. 2021.

3. Johannsen P, Cummings J, et al. Efficacy and safety of oral semaglutide in early-stage symptomatic Alzheimer’s disease: EVOKE and EVOKE+. Lancet. 2026.

4. Look M, Dunn JP, Kushner RF, et al. Body composition changes during weight reduction with tirzepatide in the SURMOUNT-1 study of adults with obesity or overweight. Diabetes, Obesity and Metabolism. 2025;27(5):2720-2729. doi:10.1111/dom.16275.

5. Jiang N, Yin J, Lawrence N, et al. Repeated withdrawal of a GLP-1 receptor agonist induces hyperleptinemia and deteriorates metabolic health in obese aging UM-HET3 mice. Aging Cell. 2025.

6. Yassine HN, Finch CE. APOE alleles and diet in brain aging and Alzheimer’s disease. Frontiers in Aging Neuroscience. 2020;12:150.

7. Jalleh RJ, Rayner CK, Hausken T, Jones KL, Camilleri M, Horowitz M. Gastrointestinal effects of GLP-1 receptor agonists: mechanisms, management, and future directions. Lancet Gastroenterology & Hepatology. 2024;9:957–964.

8. Chiang C-H, et al. Glucagon-like peptide-1 receptor agonists and gastrointestinal adverse events: a systematic review and meta-analysis. Gastroenterology. 2025.

9. Komé AM, Chandran MM, Tungate Lopez SS, Buse JB, Klein KR. One size does not fit all: understanding microdosing semaglutide for diabetes in multidose pens. Diabetes Care. 2025;48:e25–e27.

10. Tilg H, Ianiro G, Gasbarrini A, et al. Adipokines: masterminds of metabolic inflammation. Nature Reviews Immunology. 2025.

11. McKendry J, Coletta G, Nunes EA, Lim C, Phillips SM. Mitigating disuse-induced skeletal muscle atrophy in ageing: resistance exercise as a critical countermeasure. Experimental Physiology. 2024.

12. Heymsfield SB, Aronne LJ, Montgomery P, et al. Bimagrumab plus semaglutide alone or in combination for the treatment of obesity: a randomized phase 2 trial. Nature Medicine. 2026.

13. Pratley RE, Denham DS, Trivedi R, et al. Apitegromab for lean mass preservation during tirzepatide-induced weight loss: a randomized, double-blind, placebo-controlled phase 2 trial. Nature Medicine. 2026.

Disclosure: This post contains partially AI-assisted text. All references were reviewed for accuracy.

Amyloid plaque burden imaged using PET shows that it can accumulate years before symptoms. Tau PET, by contrast, tends to align more closely with where the brain is failing and how cognition is declining. This observation has shaped the field. Based on findings that tau burden and distribution are more closely associated with cognitive impairment and disease stage than amyloid plaque burden [1], why don’t we directly target tau?

That is the assumption the ceperognastat trial forces us to examine.

Ceperognastat is an oral drug designed to inhibit O-GlcNAcase, also called OGA. OGA is an enzyme that removes a small sugar-like chemical tag, called O-GlcNAc, from proteins inside cells. The idea was that blocking OGA would keep more O-GlcNAc on tau, reduce tau phosphorylation and aggregation, and slow Alzheimer’s disease progression. [2]

The rationale was plausible. But plausibility is not proof.

PROSPECT-ALZ

PROSPECT-ALZ was a large, expensive Phase 2 program. It enrolled 327 participants with early symptomatic Alzheimer’s disease, ran at 72 sites in 5 countries, used tau PET screening/enrichment, MRI, plasma biomarkers, repeated clinical assessments, randomized to placebo, low-dose ceperognastat, or high-dose ceperognastat, and followed participants for up to 76–124 weeks, with an extension period.

The trial did not show clinical benefit. In the primary outcome population, the low-dose group had no clinically meaningful improvement over placebo, and the high-dose group showed 32% greater clinical progression than placebo on the Integrated Alzheimer’s Disease Rating Scale at 100 weeks (Figure 1).

Some biomarker measures moved in directions that could have appeared favorable, including less whole-brain volume loss and a smaller increase in tau PET signal in one brain region Figure 2. [3]

This is the central lesson: a biomarker can track disease without proving that modifying it will improve cognition.

The Correlation–Causation Problem

The strongest argument for tau as a therapeutic target comes from human observational data. Tau PET burden and distribution correlate more closely with cognitive impairment and clinical stage than amyloid plaque burden. That makes tau PET valuable for staging disease, selecting participants for trials, and estimating risk of progression. [1]

But correlation is not causation.

A marker of disease severity is not automatically the mechanism that must be modified to improve the disease. In Alzheimer’s disease, tau pathology could play several roles. It may directly injure neurons. It may amplify stress initiated by amyloid, inflammation, vascular injury, or endolysosomal dysfunction. It may reflect a cellular response to injury. Or it may mark neurons that are already failing.

These possibilities lead to different therapeutic predictions.

If tau is a direct and reversible driver of neuronal death, lowering tau should help preserve neurons and cognition. If tau is downstream of a broader cellular failure, lowering tau may improve tau biomarkers without restoring the injured system that produced the tau pathology in the first place.

Ceperognastat does not prove that tau is irrelevant. It shows that changing tau-related biology through broad OGA inhibition did not produce clinical benefit in early symptomatic Alzheimer’s disease. [3]

Genetic Tauopathies Are Not the Same as Alzheimer’s Disease

There are diseases where tau has a stronger causal role.

MAPT is the gene that encodes the tau protein. In some inherited frontotemporal dementias and related primary tauopathies, mutations in MAPT alter tau itself. Some mutations change the tau protein sequence. Others alter tau splicing, which changes the balance of tau isoforms. These changes can make tau more prone to misfolding, aggregation, or toxic dysfunction. In these diseases, tau biology is closer to the root cause of neurodegeneration. [4]

Alzheimer’s disease is different.

Most patients with late-onset Alzheimer’s disease do not have MAPT mutations. AD is not usually a disease of genetically overexpressed tau. It is a disease in which tau pathology emerges in the setting of amyloid, endolysosomal stress, immune activation, vascular injury, metabolic dysfunction, synaptic failure, and aging.

This distinction matters because many preclinical tau models rely on high expression of mutant human tau. One common model, rTg4510, overexpresses human tau carrying the P301L mutation. P301L means that one amino acid in tau — proline at position 301 — is replaced by leucine. This mutation pushes the system toward tau pathology. It is useful for testing whether a drug changes tau aggregation, but it does not reproduce the full biology of sporadic Alzheimer’s disease. [4,5]

A mutant tau overexpression model asks one question:

Can this drug change tau aggregation biology?

A human Alzheimer’s trial asks a broader question:

Can this drug preserve neurons, synapses, networks, and cognition in a multifactorial disease of aging?

Those are not the same question.

Tau Production, Tau Aggregation, and Tau Spread Are Different Targets

One reason tau drug development is difficult is that “targeting tau” can mean several different things.

Tau production means how much tau protein a cell makes. The MAPT gene is copied into MAPT mRNA, and that RNA message is used to produce tau protein.

Tau aggregation means tau proteins begin to misfold and clump together into abnormal species, including oligomers, fibrils, and eventually neurofibrillary tangles.

Tau spread refers to the observation that tau pathology appears to move through connected brain networks as Alzheimer’s disease progresses.

A drug can target one step without fixing the others.

An antisense oligonucleotide, or ASO, is a short synthetic strand of nucleic acid designed to bind a specific RNA message. A tau ASO binds MAPT mRNA, the message used to make tau protein, and promotes its degradation or blocks its translation. In plain English, a tau ASO reduces tau production. [8,9]

OGA inhibitors do something different. They do not reduce tau production. They try to change tau’s chemical state after tau has already been made by increasing O-GlcNAcylation, with the hope that tau becomes less likely to phosphorylate, aggregate, or spread. [2]

This distinction matters. A tau ASO tests whether reducing total tau production helps. An OGA inhibitor tests whether changing tau modification helps. These are related but not identical hypotheses.

Why the Ceperognastat Rationale Looked Appealing

The rationale for ceperognastat followed a clear biological sequence.

Tau is regulated by post-translational modifications. One of these is O-GlcNAcylation, the addition of a small O-GlcNAc tag to serine or threonine residues on proteins. OGA removes this tag. Inhibiting OGA increases O-GlcNAcylation. [2,6]

Because tau phosphorylation and O-GlcNAcylation can interact, OGA inhibition was proposed as a way to shift tau away from pathological phosphorylation and aggregation. In animal tauopathy models, OGA inhibition increased tau O-GlcNAcylation and reduced some measures of tau pathology. [2,4,5]

Ceperognastat was then optimized as a potent, orally available, CNS-penetrant OGA inhibitor. Phase 1 studies showed high brain OGA occupancy in humans. OGA occupancy means the percentage of OGA enzyme molecules bound by the drug. High occupancy shows that the drug entered the brain and engaged its target. In ceperognastat studies, brain OGA occupancy above 95% was reported. [2]

But there is a gap between target engagement and disease modification.

Disease modification, explained in greater detail before, means changing the underlying biology of the disease in a way that slows future decline. In Alzheimer’s disease, a disease-modifying therapy should ultimately preserve cognition, daily function, synapses, neurons, or brain networks over time.

Showing that a drug reaches the brain and binds OGA is important. It does not prove that the intervention improves neuronal function. It does not prove that the dose is safe for synapses. It does not prove that modifying tau-related chemistry will alter the clinical trajectory of Alzheimer’s disease.

This is where the evidentiary foundation appears limited.

The Preclinical Data Answered a Narrow Question

The preclinical work showed that OGA inhibition could reduce tau pathology in tauopathy mouse models. That is meaningful, but narrow.

In rTg4510 mice, chronic OGA inhibition reduced aggregated tau, several phosphorylated tau species in insoluble brain fractions, and CSF total tau. These are pharmacodynamic effects on tau biology (Figure 4). But rTg4510 mice overexpress mutant human tau. They do not reproduce the full biology of late-onset Alzheimer’s disease. [4]. And the investigators missed behavioral readouts.

This is a recurring problem in Alzheimer’s drug development. Animal models are often interpreted beyond the question they were designed to answer. A model can be useful for testing target biology and still be limited for predicting clinical benefit.

The relevant limitation is not that the tau mouse data were wrong. The limitation is that they did not establish that broad OGA inhibition would preserve synaptic function and cognition in cell and animal models before they went to human AD.

That should have been a higher bar before moving into a large and expensive clinical program. The unfortunate implications of null trials are decreased investment.

OGA Is Not a Tau-Specific Target

The failure of ceperognastat may also reflect the biology of the target.

OGA does not act only on tau. O-GlcNAcylation regulates many nuclear and cytoplasmic proteins involved in metabolism, stress responses, cell signaling, survival, protein turnover, transcription, and synaptic function. [2,6]

This matters because cognition depends on synaptic physiology, not only on pathology markers.

If a drug reduces a tau signal but disrupts synaptic transmission or plasticity, the net clinical effect may be neutral or harmful. The phase 2 trial raises this concern because high-dose ceperognastat was associated with worse clinical progression despite some biomarker changes that did not look unfavorable. [3]

A later brain-slice study came too late for the Lilly-sponsored trial. Investigators tested three structurally distinct OGA inhibitors — ceperognastat, ASN90, and MK8719 — in adult mouse hippocampal slices. All three suppressed short-term synaptic plasticity and long-term potentiation, or LTP (Figure 5). LTP is one of the experimental measures scientists use to study how synapses strengthen during learning. The same study also reported altered synaptic protein markers, including PSD-95 and synaptophysin 1. [7]

This experiment does not prove that synaptic toxicity caused the clinical outcome in PROSPECT-ALZ. But it identifies exactly the kind of functional risk that should be assessed early in CNS drug development, before moving to expensive Phase II or Phase III trials.

For Alzheimer’s disease, the key preclinical question should not be limited to:

Does the drug lower tau?

It should also include:

Does the drug preserve synaptic transmission, plasticity, and network function?

What the Trial Actually Tells Us

The PROSPECT-ALZ trial should not be interpreted as proof that tau is irrelevant.

It tells us something more specific: broad OGA inhibition with ceperognastat did not slow clinical progression in early symptomatic Alzheimer’s disease, and the high dose was associated with worse clinical outcomes. [3]

It also highlights a biomarker–clinical disconnect. Some imaging outcomes appeared to move in directions that might otherwise be interpreted as favorable, while cognition and function did not improve. [1,3]

That is important because Alzheimer’s trials increasingly rely on biomarkers for enrollment, staging, target engagement, and early decision-making. Biomarkers are essential. But they are not complete substitutes for clinical outcomes.

A biomarker can identify disease without proving that it is the best therapeutic lever. Tau PET may be a strong marker of neurodegeneration and clinical stage. That does not mean every tau-lowering or tau-modifying strategy will be beneficial.

What About Tau ASOs?

Tau ASOs should be considered separately from ceperognastat.

Unlike OGA inhibitors, tau ASOs target MAPT mRNA to reduce tau production more directly. This makes them a cleaner test of the tau-lowering hypothesis than broad O-GlcNAc manipulation. [8,9]

The leading example is diranersen, also known as BIIB080, an investigational tau-targeting ASO. Early clinical studies showed reductions in CSF tau measures and suggested effects on tau PET in smaller datasets. [8] In 2026, Biogen reported phase 2 CELIA results showing robust reductions in CSF total tau and tau PET pathology, with prespecified cognitive signals. However, the study did not meet its primary endpoint assessing dose response, and the strongest clinical signals were not simply dose-dependent. [9]

This is encouraging, but it still requires careful interpretation.

If tau ASOs show reproducible clinical benefit in larger trials, that would support tau as a therapeutically relevant pathway in selected Alzheimer’s populations or disease stages. If they lower tau biomarkers without preserving cognition, that would strengthen the concern that tau may often be a marker or amplifier rather than the primary reversible cause of neuronal loss.

The key question for tau ASOs is therefore not only whether tau can be lowered.

The question is whether lowering tau preserves synapses, neurons, networks, and cognition.

A Challenge to the Linear Amyloid Cascade

Ceperognastat also fits into a broader problem with the linear amyloid cascade model.

The classic sequence is familiar: amyloid accumulates, tau spreads, neurons die, cognition declines. This model explains important parts of Alzheimer’s disease biology, especially in familial AD and in amyloid-positive disease trajectories. It has also supported amyloid-lowering therapies that modestly slow decline.

But late-onset Alzheimer’s disease is not a pure amyloid disease and not a pure tauopathy. I discussed problems with the linear amyloid cascade in a previous post.

Briefly, late-onset AD genetics implicate endolysosomal trafficking, immune activation, lipid metabolism, vascular injury, mitochondrial stress, proteostasis, and synaptic vulnerability. In this broader framework, amyloid and tau remain important. They may be toxic. They may amplify injury. They may stage disease. But they may not always be the root cause.

This is where targeting endolysosomal and immune pathways becomes relevant. If upstream cellular clearance and immune pathways are failing, then amyloid and tau may partly reflect the visible output of a deeper biological process.

Lowering one output may not restore the system.

The Uncertainty We Need to Acknowledge

The field should be careful here.

We do not yet have definitive proof that lowering tau improves cognition in Alzheimer’s disease. Ceperognastat modified tau-related biology but did not improve clinical outcomes. [3] Diranersen has reported encouraging phase 2 biomarker and clinical signals, but it has not yet provided replicated registrational evidence that tau lowering slows AD progression. [9]. Cell or Animal models that genetically overexpress tau (or amyloid) create environments that do not mimic human AD pathology and complicate distinguishing causation from association. The effect of targeting excessive tau expression in animal models may not translate to a mild or moderate increase in tau pathology in chronic diseases like aging.

This uncertainty does not mean tau is unimportant. It means the therapeutic role of tau in AD remains unresolved.

Tau may be causal in some contexts, amplifying in others, and downstream in others. The task is to identify when tau is a driver, when it is a marker, and when lowering it meaningfully preserves brain function.

Take-Home Messages

1. Tau PET is a strong disease marker, but not yet a validated therapeutic surrogate.
Tau correlates with cognition better than amyloid, but this does not prove that lowering tau without understanding the mechanism of action will improve cognition.

2. Correlation is not causation.
Tau may drive neuronal injury, amplify other disease processes, or mark neurons already under stress. These possibilities require different treatments.

3. Alzheimer’s disease is not the same as a genetic tauopathy.
MAPT mutation diseases show that tau can cause neurodegeneration, but late-onset AD usually does not arise from MAPT mutations or tau overexpression.

4. Tau production, aggregation, and spread are different therapeutic targets.
Tau ASOs reduce tau production by targeting MAPT mRNA. OGA inhibitors try to change tau modification after tau has already been made.

5. Target engagement is not disease modification.
High OGA occupancy showed that ceperognastat reached and bound its target in the brain. It did not prove that synapses, neurons, or cognition would be protected.

6. OGA inhibition is not tau-specific.
OGA affects O-GlcNAcylation on many proteins involved in metabolism, signaling, stress responses, and synaptic function.

7. Synaptic function should be part of preclinical screening.
For Alzheimer’s drugs, moving a biomarker is not enough. A therapy should preserve or improve synaptic transmission, plasticity, and network function before advancing to large trials.

8. Tau ASOs remain important, but they still must prove clinical benefit.
They are a cleaner test of tau lowering than OGA inhibition, but the key endpoint remains cognition and function, not tau reduction alone.

9. Alzheimer’s disease likely requires systems-level thinking.
Amyloid and tau are important, but late-onset AD also involves endolysosomal, immune, vascular, metabolic, and synaptic biology.

10. The goal is not to treat the biomarker.
The goal is to preserve the living system the biomarker is supposed to represent: neurons, synapses, networks, and cognition.

References

1. Rabinovici GD, Grill JD. Ceperognastat in Alzheimer disease: lessons from a negative clinical trial. JAMA. 2026.

2. Kielbasa W, Goldsmith P, Donnelly KB, et al. Discovery and clinical translation of ceperognastat, an O-GlcNAcase inhibitor, for the treatment of Alzheimer’s disease. Alzheimer’s & Dementia: Translational Research & Clinical Interventions. 2024.

3. Fleisher AS, Munsie L, Mancini M, et al. Ceperognastat in early symptomatic Alzheimer disease: a randomized clinical trial. JAMA. 2026.

4. Hastings NB, Wang X, Song L, et al. Inhibition of O-GlcNAcase leads to elevation of O-GlcNAc tau and reduction of tauopathy and cerebrospinal fluid tau in rTg4510 mice. Molecular Neurodegeneration. 2017;12:39.

5. Graham DL, Gray AJ, Joyce JA, et al. Increased O-GlcNAcylation reduces pathological tau without affecting its normal phosphorylation in a mouse model of tauopathy. Neuropharmacology. 2014;79:307–313.

6. Pratt MR, Vocadlo DJ. Understanding and exploiting the roles of O-GlcNAc in neurodegenerative diseases. Journal of Biological Chemistry. 2023;299:105411.

7. Meade J, Mesa H, Alamgir S, Bieniecka I, Liu L, Zhang Q. Synaptic toxicity of OGA inhibitors and the failure of ceperognastat. The Journal of Prevention of Alzheimer’s Disease. 2026;13:100456.

8. Mummery CJ, Börjesson-Hanson A, Blackburn DJ, et al. Tau-targeting antisense oligonucleotide MAPTRx in mild Alzheimer’s disease: a phase 1b randomized, placebo-controlled trial. Nature Medicine. 2023;29:1437–1447.

9. Biogen. Phase 2 CELIA results for diranersen (BIIB080) in early Alzheimer’s disease. 2026.

APOE4 is not simply a “bad gene,” and APOE2 is not simply a “good gene.” Many APOE4 carriers do not develop dementia, and some APOE2 carriers have a lipid handling disorder.

APOE alleles reflect biological tradeoffs. A variant that helps in one environment, tissue, or life stage may harm in another.

That is why the question sounds easier than it is:

If APOE2 protects against Alzheimer’s disease, why don’t we give APOE2 to APOE4 carriers?

The answer is that APOE2 is not a vitamin, supplement, or simple replacement part. APOE is a lipid transport protein, receptor ligand, immune modulator, and brain-cell signaling molecule. Its effects depend on where it is made, which cell makes it, how it is lipidated, when it is delivered, and what disease stage is already present.

Not a Good or a Bad Gene Copy

APOE is a lipid transport gene with three common alleles: APOE2, APOE3, and APOE4. In many large datasets, APOE3 is the most common allele, roughly 75–80%, while APOE4 is about 10–15% and APOE2 about 5–10%, with variation by ancestry and population. In genotype terms, APOE3/3 is the most common, often present in about two-thirds of people [1].

APOE4 increases late-life Alzheimer’s risk, but human data suggest it may have carried advantages in environments where infection, undernutrition, and reproduction shaped survival more strongly than late-life dementia.

In the Tsimane, a natural fertility population with high infectious burden, a study of 795 women found that APOE4 carriers had about 0.3 to 0.5 more children than APOE3 homozygotes, while APOE4 homozygotes had about 1.4 to 2.1 more children [2]. In a Brazilian shantytown cohort of 72 children followed from birth, APOE4 carriers appeared more resilient to the long-term cognitive effects of heavy early-childhood diarrhea [3]. Several studies also describe amyloid-beta as an antimicrobial-like peptide that can inhibit microbial growth through membrane disruption or microbial entrapment [4].

These studies do not prove that APOE4 is beneficial. They support a more careful idea: what helps early-life survival may not help the aging brain.

APOE2 also has tradeoffs. APOE2 is associated with lower Alzheimer’s risk and, in some studies, longevity. Key to how it works, APOE2 binds poorly to the LDL receptor. This property may explain why about 5–10% of APOE2 homozygotes develop type III hyperlipoproteinemia, with accumulation of triglyceride-rich remnant particles [4]. APOE2 has also been associated with increased risk of cerebral amyloid angiopathy and selected neurological disorders [5].

So the goal is not to replace a “bad” allele with a “good” allele. The goal is to identify which APOE2 functions protect the aging brain — lipidation, amyloid handling, synaptic support, vascular effects, or microglial signaling — and reproduce those functions without creating APOE2-related complications.

What APOE2 Does Differently from APOE4

APOE is best known as a lipid carrier. In the brain, APOE helps move cholesterol and phospholipids between cells. This is essential because neurons depend on glial cells to help maintain membranes, synapses, repair programs, and lipid balance.

APOE2 differs from APOE3 and APOE4 at key amino acid positions. One important consequence is weaker binding to the LDL receptor (LRP1). In some lipid-handling contexts, APOE2 can therefore behave like a partial loss-of-function variant.

But in the brain, APOE2-containing lipoprotein particles may be more lipidated than APOE3 or APOE4 particles. More lipidated particles may better support lipid transport, synaptic maintenance, amyloid plaque seeding, and amyloid clearance [4].

But there is the APOE2 paradox:

In the periphery, APOE2 can impair remnant clearance. In the brain, APOE2 may create a more protective lipidated particle.

That is why therapeutic design has to be precise.

APOE2-Associated Longevity Involves Lipid Handling, Not Just Alzheimer’s Pathology

APOE2’s link to longevity may not simply reflect lower Alzheimer’s pathology. In an analysis of NACC clinical records, carrying APOE4 had a similar effect on survival independent of AD pathology. Similarly, APOE2 had a protective effect on survival independent of amyloid and tau pathology-Figure 2 [6]. I have discussed before why amyloid and tau do not fully explain the clinical syndrome of AD, and here is where lipids become interesting.

In the same eLife study, APOE2-targeted replacement mice lived longer in a model that did not overexpress mutant APP or tau and did not show Alzheimer-like amyloid or tau deposition. In 118 mice, median survival was 911 days for APOE2, 825 days for APOE3, 753 days for APOE4, and 738 days for Apoe knockout [6].

This same study linked APOE2 to preserved activity and lipid metabolism during aging. The authors reported associations among activity, apoE protein, HDL-cholesterol, and triglycerides, and suggested that higher HDL-cholesterol and triglyceride levels regulated by apoE2 might contribute to APOE2-mediated longevity [6]. These findings may offer interesting insights into how Obicetrapib may affect AD biology.

This matters because APOE2 may influence aging through lipid transport and brain-body metabolism, not only through amyloid and tau. An APOE2 therapy should therefore ask not just whether plaques go down, but whether lipid handling, vascular integrity, brain metabolism, learning, and resilience move in the right direction.

The Brain Has Its Own APOE System

One major reason we cannot simply give APOE2 by mouth or IV is that the body and brain have largely separate APOE pools.

A classic liver transplantation study showed that after liver transplant, the APOE phenotype in plasma changed almost completely to match the donor liver. But the APOE phenotype in cerebrospinal fluid did not change [7].

Liver APOE does not simply become brain APOE.

The liver supplies most plasma APOE. The brain makes APOE locally, mainly through astrocytes, with contributions from microglia, vascular mural cells, choroid plexus, and stressed neurons [4].

So oral APOE would likely be digested like other proteins. IV APOE may mostly act in the periphery, where it could affect triglycerides, remnant lipoproteins, vascular biology, or inflammation. But it may not meaningfully change APOE biology inside the brain.

The therapeutic problem is not just what to deliver.

It is where to deliver it, and whether APOE is properly lipidated.

AAV Gene Therapy: Silence and Replace

One promising approach is CNS gene therapy.

In APOE4 carriers, the ideal intervention may not be adding APOE2 alone. It may require two steps:

Silence APOE4. Replace with APOE2.

This makes biological sense because APOE4 may not simply represent loss of protection. APOE4 may also have toxic gain-of-function effects involving lipidation, amyloid aggregation, amyloid clearance, synaptic injury, mitochondrial function, blood-brain barrier integrity, and neuroinflammation.

A recent AAV “silence-and-replace” strategy used microRNAs to suppress endogenous APOE4 while adding a modified APOE2 gene designed to resist that silencing [8].

The idea is elegant: reduce the high-risk APOE4 signal and shift the local CNS APOE environment toward APOE2.

But delivery remains difficult. In mice, APOE AAV studies often bypass the blood-brain barrier by direct stereotaxic brain injection. In the APOE4 silence-and-replace study, investigators drilled burr holes and injected AAV.S2 vectors bilaterally into the hippocampus. That is very different from giving a drug by mouth or IV, and it highlights the central challenge for human translation: how to deliver APOE2 safely, broadly, and to the right CNS cells without direct regional injections into every vulnerable brain area.

Many AAV vectors preferentially transduce neurons. Yet most endogenous brain APOE is made by astrocytes. That mismatch matters. If a therapy causes neurons to make APOE2, is that equivalent to astrocytes or microglia making APOE2?

Likely not.

Cell Type Matters

This may be the most important concept.

APOE2 made by a neuron may not be the same as APOE2 made by an astrocyte or microglia.

Astrocyte-derived APOE participates in lipid transport, synaptic maintenance, cholesterol handling, and neuronal support.

Microglial APOE may be more closely linked to immune activation, plaque-associated responses, TREM2 biology, phagocytosis, damaged-neuron clearance, and disease-associated microglial states.

Cell type also affects APOE processing. APOE made in different cells may differ in lipidation, secretion, receptor interactions, and post-translational modification [8].

So simply measuring “more APOE2 in the brain” may not tell us whether the right biology has been restored.

The future may require cell-specific APOE2 expression.

Current Clinical Translation

The leading clinical example is LX1001, an investigational AAVrh.10-based gene therapy designed to deliver human APOE2 into the central nervous system of APOE4 homozygotes with Alzheimer’s disease [9].

The Phase 1/2 study tests one-time intrathecal administration of AAVrh.10hAPOE2 in APOE4 homozygotes with mild cognitive impairment or dementia due to Alzheimer’s disease. A long-term follow-up study is also underway [9,10].

Intrathecal delivery means placing the therapy into the cerebrospinal fluid, the fluid that surrounds the brain and spinal cord. This brings an AAV vector closer to the CNS than an IV infusion, but it does not guarantee that the right brain cells will express APOE2. After intrathecal delivery, APOE2 may be expressed by cells reached from the CSF space — including meningeal, ependymal, choroid plexus, spinal, or some brain parenchymal cells depending on the vector. The key question is whether this reproduces the biology that matters most for Alzheimer’s disease: APOE made by astrocytes and microglia, properly lipidated, secreted at the right level, and active in the brain regions most vulnerable to disease

Interim topline results reported 15 dosed participants, dose-dependent APOE2 detection in CSF, general tolerability, and favorable movement in some amyloid and tau biomarkers [11].

These findings support continued study, but they are not proof of clinical efficacy.

The core translational questions remain:

Can APOE2 be delivered safely to the CNS?
Can APOE2 be expressed durably at meaningful levels?
Does it improve amyloid, tau, lipid, vascular, or inflammatory biology?
Does it avoid vascular, inflammatory, lipid, lysosomal, or immune toxicity?
Does it slow cognitive and functional decline?

What Would Success Look Like?

The benchmark for success cannot be “we detected APOE2.” That is target engagement, not disease modification.

A successful APOE2 therapy would need several layers of evidence.

First: safety. No unacceptable inflammation, immune response, edema, hemorrhage, vascular injury, or accelerated neurodegeneration.

Second: target engagement. APOE2 should appear in CSF or other CNS-linked measures in a dose-responsive and durable way.

Third: biological correction. The therapy should improve APOE lipidation, lipid transport, amyloid handling, tau-related injury, synaptic biology, glial function, or vascular integrity.

Fourth: clinical effect. It should slow worsening on cognition, function, MRI atrophy, tau PET, amyloid PET, CSF biomarkers, or plasma biomarkers.

The hardest endpoint is the one patients care about most:

Do people think, remember, function, and live independently longer?

That is the real bar.

The Liposome Cautionary Tale

A recent liposome-based APOE2 study is important because it complicates the simple story.

The study used brain-targeting liposomes to deliver APOE2 plasmid material in an amyloid mouse model. APOE2 delivery produced some potentially favorable transcriptional signals related to neuronal development and synaptic pathways [12].

But it also increased soluble amyloid-beta, including oligomeric forms, and worsened neurite dystrophy [12].

That distinction matters.

Lowering plaque burden is not automatically the same as lowering toxicity. Some soluble amyloid species may be more synaptotoxic than deposited plaque.

So “APOE2 reduces plaques” is not enough.

We need to ask:

What happens to soluble amyloid?
What happens to synapses?
What happens to neurites?
What happens to microglia?
What happens over time?

Prevention Is Not Treatment

A person born with APOE2 has decades of APOE2 biology. Their brain develops and ages with different lipidation, receptor binding, amyloid handling, synaptic support, vascular biology, and immune tone.

That is not the same as giving APOE2 at age 70 after amyloid plaques, tau pathology, vascular injury, or neuroinflammation are already present.

APOE2 may be most powerful as preventive biology. It may reduce amyloid accumulation, improve lipid transport, support synapses, or shape microglial responses before pathology is advanced.

But once amyloid burden is high, adding APOE2 could have different effects. It may mobilize amyloid into soluble species, alter glial lipid handling, or interact with tau and vascular disease in ways we do not yet understand.

This does not mean APOE2 therapy cannot work in treatment.

It means disease stage matters.

One APOE2 strategy may not fit young APOE4 carriers, preclinical amyloid-positive carriers, mild cognitive impairment, established dementia, vascular comorbidity, amyloid-dominant disease, or tau-dominant disease.

The Ethical Question

APOE2 gene therapy also raises some ethical issues.

AAV-based gene therapy is designed to be long-lasting. That makes the risk-benefit calculation different from taking a pill that can be stopped.

The prevention question is harder. What level of risk would justify CNS gene therapy in a cognitively normal APOE4 homozygote? Amyloid positivity? Tau positivity? Plasma p-tau? Family history? Age?

Equity also matters. Genetic testing, biomarker screening, lumbar puncture, gene therapy delivery, and long-term monitoring may not be equally accessible.

The ethical standard should be high:

Clear risk disclosure, careful genetic counseling, long-term follow-up, transparent uncertainty, fair access, and a strict distinction between prevention and treatment.

So Why Aren’t We Giving APOE2 to APOE4 Carriers?

Because APOE2 is not simply good, and APOE4 is not simply bad.

We are not replacing a missing vitamin.

We are trying to reprogram a complex lipid-immune-neurodegenerative pathway inside the brain.

APOE genotypes teach us that Alzheimer’s risk is modifiable. It points toward protective mechanisms involving lipidation, receptor binding, amyloid handling, synaptic support, vascular biology, microglial function, and lipid metabolism.

But APOE2 is not a simple replacement part for APOE4.

The future is likely precision APOE biology: the right APOE function, in the right brain compartment, in the right cell type, at the right dose, and at the right disease stage. This is where downstream mechanisms become interesting. Can we lipidate APOE3 and APOE4 to behave like APOE2? Can we block APOE4-accentuated neuroinflammation, like APOE2?

APOE2 may or may not be the answer by itself.

It will offer us important clues.

Key Takeaways

• APOE4 and APOE2 are not simply bad and good alleles. They likely reflect biological tradeoffs.

• APOE2 is associated with lower Alzheimer’s risk and longevity, but it is not universally benign.

• APOE2 longevity may involve lipid metabolism, not only amyloid and tau.

• Peripheral APOE and brain APOE are largely separate pools, and oral or IV APOE2 is unlikely to reproduce brain APOE2 biology.

• Cell type matters. Astrocyte, microglial, and neuronal APOE2 may not have the same effects.

• AAV silence-and-replace strategies are promising, but delivery and cell targeting remain major challenges.

• Reducing plaques is not enough if soluble amyloid oligomers or neurite injury increase.

• Prevention and treatment may require different APOE2 strategies.

• The future is likely precision APOE biology, not simply “give APOE2.”

References

1. Farrer LA, Cupples LA, Haines JL, et al. Effects of age, sex, and ethnicity on the association between apolipoprotein E genotype and Alzheimer disease. JAMA. 1997;278(16):1349–1356.

2. Trumble BC, Stieglitz J, Blackwell AD, et al. Apolipoprotein-ε4 is associated with higher fecundity in a natural fertility population. Science Advances. 2023.

3. Oria RB, Patrick PD, Zhang H, et al. APOE4 protects the cognitive development in children with heavy diarrhea burdens in Northeast Brazil. Pediatric Research. 2005.

4. Moir RD, Lathe R, Tanzi RE. The antimicrobial protection hypothesis of Alzheimer’s disease. Alzheimer’s & Dementia. 2018.

5. Li Z, Shue F, Zhao N, Shinohara M, Bu G. APOE2: protective mechanism and therapeutic implications for Alzheimer’s disease. Molecular Neurodegeneration. 2020;15:63.

6. Shinohara M, Kanekiyo T, Tachibana M, et al. APOE2 is associated with longevity independent of Alzheimer’s disease. eLife. 2020;9:e62199.

7. Linton MF, Gish R, Hubl ST, et al. Phenotypes of apolipoprotein B and apolipoprotein E after liver transplantation. Journal of Clinical Investigation. 1991;88(1):270–281.

8. Karan KR, Hackett NR, Crystal RG. Adeno-associated virus-mediated central nervous system gene transfer to suppress Alzheimer’s disease high-risk APOE4 variant and replace with protective APOE2. Human Gene Therapy. 2026.

9. ClinicalTrials.gov. Gene Therapy for APOE4 Homozygote of Alzheimer’s Disease. NCT03634007.

10. ClinicalTrials.gov. Long-Term Follow-up of Gene Therapy for APOE4 Homozygote Alzheimer’s Disease. NCT05400330.

11. Johnson K, et al. Topline results from Phase 1/2 AAV gene therapy (LX1001) in APOE4/4 homozygotes with Alzheimer’s disease. Alzheimer’s & Dementia. 2025.

12. Wang N, Parsons TM, Ren Y, et al. Brain-targeting liposome-based APOE2 gene delivery exacerbates soluble amyloid-β accumulation in AppNL-G-F mice. Heliyon. 2024;10:e39607.

She was not trying to become a bodybuilder. She was scared of something very real: losing muscle mass as she aged.

She told me she had been aiming for 100 grams of protein per day. She had a relatively small muscle build, but since she started using protein bars — including Maui and David bars — a DEXA scan showed that she had gained muscle mass. She was excited. Understandably so.

Then she asked me the more important question.

Is this actually good for my health?

She admitted that maintaining that level of protein intake was hard. The bars helped her reach the number. But she wanted to know whether chasing 100 grams of protein every day was the right goal for aging.

That is the question this post is really about.

Protein matters. Muscle matters. But aging well is not just about pushing protein intake higher until the number looks right. It is about preserving strength, mobility, metabolic health, gut health, resilience, and independence.

The protein bar may help someone hit a target.

But the bigger question is whether the entire dietary pattern helps the body age well.

Among those in the image above, who will age better?

The answer is not simply either/or. Strength training matters. Protein matters. But the protein craze often overlooks the bigger question and ends up nutrient-poor.

The Protein Halo

Protein has become the “super nutrient.”

It is added to bars, yogurts, cereals, smoothies, cookies, chips, coffee drinks, and pancake mixes. Once a package says “20 grams of protein,” many people assume it is healthy.

That is the protein halo.

But a high-protein cookie is still a cookie. A protein bar can be useful in a pinch, but it is not the same as a meal built from beans, lentils, yogurt, tofu, fish, eggs, vegetables, nuts, seeds, and whole grains.

The danger is not protein itself.

The danger is using protein as a shortcut for health.

Do We Really Need 100 Grams a Day?

The woman at the conference had absorbed one of the most common messages in wellness culture on social media: aim for 100 grams of protein per day.

But that number is not magic.

Protein needs depend on body size, age, health, appetite, physical activity, total calories, and training goals. The current U.S. Dietary Guidelines for Americans, 2025–2030, now list a protein target of 1.2 to 1.6 grams per kilogram of body weight per day, adjusting as needed for individual caloric requirements. That is a major shift from the older adult RDA of 0.8 grams per kilogram per day, which was designed mainly to prevent deficiency rather than optimize muscle or metabolic function. [1,2].

For a 60 kg person, the current U.S. guideline range of 1.2 to 1.6 g/kg/day translates to about 72 to 96 grams per day. So 100 grams is not wildly outside the new federal target, but it is still not a magic number. For a smaller person, it may be at the high end or above it; for a larger or very active person, it may be reasonable.

Her DEXA scan showed more muscle mass, and that is encouraging. But a DEXA scan does not answer every health question. It does not tell us whether the diet improved insulin sensitivity, vascular health, gut microbial diversity, inflammation, appetite regulation, or long-term resilience.

So the question is not only:

Did the protein target increase lean mass?

It is also:

What did it take to get there, and what did it replace?

If reaching 100 grams requires multiple protein bars, fewer legumes, fewer whole grains, fewer fruits and vegetables, and less dietary variety, then the number may be distracting us from the larger goal.

A protein bar may deliver 20 grams of protein.

But it may deliver little fiber, plant diversity, healthy fat, slow carbohydrate, or microbial fuel.

Whole foods deliver protein inside a broader matrix.

That matrix matters.

The Real Tension: Muscle Versus Longevity Signals

The protein debate in aging has a real tension.

On one side, older adults are often told to eat more protein to preserve muscle and prevent frailty. Some studies support this. In the Health ABC Study, older adults with higher protein intake lost less lean mass over three years than those with lower intake [3].

On the other side, aging biology raises a different concern.

In The Hallmarks of Protein and Amino Acid Restriction in Aging and Longevity, Bailey Knopf and Dudley Lamming review evidence that protein restriction, and restriction of specific amino acids such as methionine, isoleucine, and valine, can improve metabolic health, reduce frailty, extend healthspan, and lengthen lifespan in several model organisms [4].

So which is it?

Do we need more protein for muscle, or less protein for longevity?

The answer is that protein should not be treated as a single lever. Protein affects muscle, but amino acids also act as nutrient signals. They influence pathways that regulate growth, repair, metabolism, inflammation, cellular cleanup, mitochondrial function, and aging.

Protein is not just building material.

Protein is also information that affects metabolism.

That means more is not automatically better. It depends on the person, the dose, the source, the exercise pattern, and the whole diet.

Protein Restriction Biology: Why the Science Is Interesting

The science of protein restriction is not about starving the body of protein.

It is about how the body responds when amino acid availability changes while basic nutritional needs are still met.

Knopf and Lamming describe several hallmarks of protein restriction that overlap with major pathways in aging biology.

One is nutrient sensing. Amino acids regulate mTORC1, a growth-related pathway that helps drive protein synthesis and cell growth when nutrients are abundant. That is useful for building muscle after resistance exercise. But chronically high nutrient signaling may also reduce cellular cleanup processes such as autophagy, which help remove damaged proteins and organelles.

Protein restriction can also activate GCN2, a stress-responsive pathway that helps cells adapt to amino acid scarcity. This is part of a broader shift away from constant growth and toward maintenance, repair, and metabolic adaptation.

Another important signal is FGF21, a hormone induced by protein restriction and some amino acid restrictions. In animal and human studies, FGF21 is linked to changes in energy expenditure, glucose metabolism, lipid handling, and insulin sensitivity.

Protein restriction also intersects with mitochondrial function, cellular senescence, and inflammation. In several models, lower protein or selective amino acid restriction improves metabolic health and reduces markers of aging-related dysfunction. Restriction of specific amino acids, especially methionine and some branched-chain amino acids such as isoleucine and valine, can reproduce some of the benefits of overall protein restriction.

This does not mean people should randomly restrict amino acids.

It means the biology of protein is not simple.

Protein is a muscle nutrient, but amino acids are also signals that tell the body whether to grow, repair, conserve, recycle, or adapt.

Healthspan Is Not Just Muscle Mass

The “more protein” argument often assumes that preserving muscle mass is the same as preventing frailty.

It is not that simple.

Muscle matters, but aging well is about function: strength, balance, walking speed, glucose control, recovery from stress, immune resilience, and independence.

Knopf and Lamming point out that higher protein can promote muscle growth, especially with resistance exercise, but protein restriction can blunt age-related increases in frailty in animal studies. In some late-life models, protein restriction reduced lean mass while also reducing or slowing further declines in frailty [4].

That sounds paradoxical only if muscle size is the only outcome.

The better goal is not maximum muscle at any cost.

The better goal is durable function.

Resilience: The Missing Word

A useful concept in the review is resilience.

Resilience means the ability to bounce back from stress: infection, injury, fasting, heat, cold, hospitalization, inflammation, or metabolic strain.

In humans, loss of resilience contributes to chronic disease, multimorbidity, disability, and death. In animal studies, protein restriction has improved some forms of resilience during aging, including responses to starvation, bacterial infection, and heat stress in fruit flies [4].

We do not yet know how much of this translates to humans.

But the concept is important.

Aging nutrition should not only ask:

How much muscle can I build?

It should also ask:

Can this body recover, adapt, resist disease, and remain independent?

Protein may be part of that answer.

But it is not the whole answer.

Okinawa: A Low-Protein Diet Was Not a Frailty Diet

Okinawa is important because it challenges the assumption that aging well requires protein-maxing.

The traditional Okinawan diet was not built around protein bars, powders, or large portions of animal protein. It was largely plant-based, rich in complex carbohydrates, and centered on foods such as sweet potatoes, vegetables, legumes, and soy. Knopf and Lamming note that the traditional Okinawan diet has been estimated at 9% protein, similar to some protein-restriction laboratory diets, and about 80% of calories from plant sources [4].

This does not prove that low protein caused Okinawan longevity.

The Okinawan pattern also included lower calorie intake, daily movement, social connection, and a broader cultural environment that supported aging.

But that is exactly the point.

Okinawa was not a protein-maximizing culture. It was a dietary and lifestyle pattern. The population became famous for longevity, not frailty.

That should make us cautious about reducing aging nutrition to a single protein target.

The Mediterranean Diet: Muscle Without Protein Obsession

The Mediterranean diet tells a similar story in a more contemporary human study.

Mazza and colleagues studied 528 adults aged 50 and older living in southern Italy and examined dietary patterns in relation to handgrip strength, muscle mass, and sarcopenia [5].

The average participant was not protein deficient. Mean protein intake was about 1.0 g/kg/day.

The researchers identified several dietary patterns. The Mediterranean pattern was characterized by higher intake of carbohydrates, plant proteins, and fiber. People with higher adherence consumed more legumes, cereals, vegetables, and fruit, and lower amounts of meat, fish, and eggs.

This pattern was associated with better handgrip strength and lower prevalence of sarcopenia. Low adherence to the Mediterranean pattern was associated with higher odds of low muscle strength and sarcopenia [5].

This matters because the analyses adjusted for daily protein intake per kilogram of body weight.

So the result was not simply “more protein equals better muscle.”

It was a pattern effect.

The authors make the key point directly: the Mediterranean diet is primarily plant-based, about 20% lower in total protein than a typical Western diet, and relies more on legumes, nuts, seeds, and whole grains as protein sources. They argue that protein quality may be more important than quantity in explaining the benefits of the Mediterranean diet for muscle mass [5].

This helps resolve the protein paradox.

A diet can support muscle without making protein the star.

It can provide adequate amino acids while also delivering fiber, unsaturated fats, polyphenols, minerals, antioxidants, and plant diversity.

The Mediterranean diet is not a low-muscle diet.

It is a food-pattern diet.

Why Protein Bars Miss the Point

A protein bar can deliver 20 grams of protein.

But it cannot deliver a dietary pattern.

It cannot reproduce a meal built from legumes, greens, whole grains, fruit, nuts, seeds, olive oil, yogurt, fish, tofu, herbs, and fermented foods. It cannot compress plant diversity, fiber architecture, resistant starches, polyphenols, minerals, food texture, satiety signals, and microbial substrates into a bar.

This matters for the gut microbiome.

The gut microbiome is shaped by repeated exposure to diverse plant foods. Fibers and polyphenols from legumes, vegetables, fruits, whole grains, nuts, and seeds feed microbial communities that produce metabolites such as short-chain fatty acids. These metabolites influence inflammation, glucose handling, gut barrier function, immune tone, and possibly aging biology.

Protein bars are built around density: more protein in less space.

Healthy aging may depend on the opposite: diversity, structure, and food complexity.

The real question is not, “Did I hit my protein target?”

It is:

What kind of internal ecosystem is this diet building?

The Real Deficiency Is Often Fiber

The modern protein craze has a hidden cost.

When protein goes up, something else often goes down.

For many people, that “something else” is carbohydrate. But carbohydrates are not one thing. Soda and lentils are both carbohydrates. White bread and barley are both carbohydrates. Candy and carrots both contain carbohydrates.

When people cut carbohydrates aggressively to make room for protein, they often cut the foods that feed the gut microbiome: beans, oats, whole grains, fruit, vegetables, potatoes, nuts, and seeds.

That is how a high-protein diet becomes a low-fiber diet.

Fiber supports bowel health, gut microbial diversity, short-chain fatty acid production, cholesterol metabolism, glucose control, satiety, and inflammation.

It is not glamorous. It is not a gym supplement. It does not look sexy on TikTok.

But it may be one of the most important missing nutrients in modern diets.

Muscle Still Needs Training

None of this means protein is irrelevant.

Protein supports muscle repair and adaptation. But resistance exercise is the main signal that tells the body to keep muscle.

Someone can eat a high-protein diet and still lose strength if they do not train. Someone can eat adequate protein from whole foods and preserve muscle if they lift, carry, climb, walk, and remain active.

For aging, muscle protection is a system:

Resistance training + adequate protein + enough calories + fiber-rich plants + sleep + metabolic health + recovery.

Protein is one part of that system.

It should not take over the whole conversation.

The Better Question

Instead of asking:

How do I get more protein?

Ask:

Am I getting enough protein from real foods while also getting enough fiber, plant diversity, healthy fats, and exercise?

That changes the meal.

Greek yogurt becomes Greek yogurt with berries, walnuts, and chia.
Eggs become eggs with greens, mushrooms, and whole-grain toast.
Fish becomes fish with lentils, roasted vegetables, and olive oil.
Tofu becomes tofu with vegetables, sesame, edamame, and brown rice.

Protein stays in the meal.

It just stops being the whole story.

Take-Home Messages

• Protein is essential, but more is not always better.

• A 100-gram target is not magic. Needs depend on body size, age, health, appetite, activity, and goals.

• Protein is both building material and biological signal. Amino acids influence mTOR, GCN2, FGF21, autophagy, mitochondria, senescence, and metabolism.

• The frailty argument is more complicated than “eat more protein.” Muscle mass matters, but aging well is about function and resilience.

• Okinawa and the Mediterranean diet challenge protein-maxing. These patterns are plant-rich, fiber-rich, and not built around protein density.

• Protein quality may matter more than protein quantity.

• The gut microbiome needs plant diversity. Fiber, polyphenols, and resistant starches cannot be fully compressed into a bar.

• Muscle needs training, not just protein. Resistance exercise is the main signal.

• The best target is protein adequacy inside a high-quality dietary pattern.

References

1. U.S. Department of Agriculture and U.S. Department of Health and Human Services. Dietary Guidelines for Americans, 2025–2030. 10th Edition. January 2026. Available at RealFood.gov.

2. Bauer J, Biolo G, Cederholm T, et al. Evidence-based recommendations for optimal dietary protein intake in older people: a position paper from the PROT-AGE Study Group. Journal of the American Medical Directors Association. 2013;14(8):542–559.

3. Houston DK, Nicklas BJ, Ding J, et al. Dietary protein intake is associated with lean mass change in older, community-dwelling adults: the Health ABC Study. American Journal of Clinical Nutrition. 2008;87(1):150–155.

4. Knopf BA, Lamming DW. The hallmarks of protein and amino acid restriction in aging and longevity. Cell Press Blue. 2026. doi:10.1016/j.cpblue.2026.100079.

5. Mazza E, Ferro Y, Maurotti S, et al. Association of dietary patterns with sarcopenia in adults aged 50 years and older. European Journal of Nutrition. 2024;63:1651–1662. doi:10.1007/s00394-024-03370-6.

An umbrella review is a review of reviews. Instead of asking what one study found, it asks what the field is showing across systematic reviews, meta-analyses, and the primary studies inside them.

Our question was simple:

Does physical activity appear to matter differently for people who carry APOE4?

The answer was encouraging, but not final.

The strongest favorable signals came from observational studies of younger or cognitively unimpaired adults. In several studies, physical activity was linked to better cognition, lower dementia risk, better brain imaging markers, or better vascular and metabolic health — sometimes more clearly in APOE4 carriers.

But clinical trials have been less convincing. Many enrolled older adults with mild cognitive impairment or dementia. APOE4 subgroups were often small. Interventions were short. Outcomes varied. Many trials asked whether exercise could improve cognition after disease was already underway.

That may be the wrong question.

Exercise may be most powerful before symptoms begin, when vascular, metabolic, inflammatory, and brain-resilience pathways are still modifiable.

Why APOE4 Carriers Should Care

If you carry APOE4, especially two copies, “exercise is good for you” is probably not motivating enough.

You already know that.

The better question is:

Can movement change the biology that makes APOE4 risky?

APOE4 is not only an Alzheimer’s risk gene. It is involved in lipid transport, cholesterol metabolism, vascular function, inflammation, glucose handling, and brain energy use. These are systems that exercise can influence.

Exercise does not erase genetic risk.

But it may improve the biological environment in which that risk is expressed.

For APOE4 carriers, physical activity may not be just about weight or fitness. It may be a way to train the vascular and metabolic systems that support the brain.

What the Positive Studies Suggest

Several studies help explain why physical activity remains interesting for APOE4 carriers.

In the CAIDE cohort, Rovio and colleagues followed adults from midlife into later life. Leisure-time physical activity at least twice per week was associated with lower dementia and Alzheimer’s risk, with more pronounced benefits reported among APOE4 carriers [2].

Krell-Roesch and colleagues found that inactive APOE4 carriers had increased risk of incident mild cognitive impairment compared with active non-carriers. Physical activity in midlife and late life was linked to lower MCI risk [3].

Shih and colleagues reported that APOE4 carriers with low physical activity had more than a threefold higher risk of dementia or cognitive impairment compared with highly active non-carriers. When low activity, APOE4, and diabetes occurred together, risk was nearly tenfold higher [4].

These studies lacked the sample size to directly test whether exercise works better for APOE4 carriers.

They show a simpler pattern: people with APOE4 who are inactive often have higher cognitive risk, so we need better trials to test whether exercise changes that risk.

That finding is important because it points to the system, not just the gene.

APOE4 risk may interact with diabetes, blood pressure, insulin resistance, lipids, inflammation, and vascular health. Exercise is one of the few interventions that touches many of these pathways at once.

Other studies point to brain markers. Physical activity or fitness in APOE4 carriers has been linked to lower amyloid burden, preserved hippocampal volume, better brain activation during memory tasks, and greater cerebral glucose uptake [5–8].

Figure 1. A hypothetical model of protective effects of physical activity (PA) on age-related pathology in APOE ε4 carriers across the lifespan, from Ali et al, 2026 (1).

None of this proves that exercise prevents Alzheimer’s disease in APOE4 carriers.

But together, the studies suggest something worth taking seriously:

Inactivity may be a particularly bad environment for an APOE4 brain.

Why Trials Have Been Disappointing

The trial evidence has not yet matched the observational signal.

That does not mean exercise does not work.

It may mean we have not tested it in the right way.

Many trials enrolled people who already had MCI or dementia. At that stage, the brain may already have substantial synaptic loss, hippocampal atrophy, vascular injury, amyloid, tau, and inflammation. Exercise may still help function, sleep, mood, mobility, falls, and vascular health. But it may be harder to show cognitive rescue.

Some trials were also too small to test APOE4 subgroups properly. Others used broad cognitive tests that may be too blunt. Some measured physical activity by self-report. Some did not prove that fitness actually improved.

A better trial would enroll low-active APOE4 carriers before dementia begins. It would measure movement objectively. It would test whether fitness improves. It would focus on outcomes that exercise is likely to change first: blood pressure, insulin resistance, lipids, inflammation, sleep, cerebral blood flow, hippocampal structure, and blood biomarkers such as p-tau217.

Dementia prevention trials are hard because dementia takes years to develop.

So we need smarter intermediate outcomes.

What Does Sedentary Mean?

Sedentary does not mean lazy.

Sedentary means spending too much waking time sitting, reclining, or lying down with very low energy use.

A practical definition is this:

You sit most of the day, get little intentional movement, and rarely reach moderate-intensity activity.

Many people under about 5,000 steps per day are functionally low-active. But step count does not capture everything. You can reach a decent number of steps and still sit for long blocks.

That matters.

My colleague at USC, David Raichlen, led an analysis of a large cohort that linked sedentary behaviors and dementia risk. This UK Biobank study used wrist accelerometers to measure sedentary behavior in nearly 50,000 older adults. Dementia risk began to rise after about 10 hours per day of sedentary time. Compared with about 9.3 hours per day, risk was higher at 10 hours, much higher at 12 hours, and highest at 15 hours [9].

This does not prove that sitting alone causes dementia.

But it tells us something practical:

The total time spent sedentary matters.

For people who work behind a computer, 10 hours can happen easily: eight hours of work, commuting, meals, and evening screen time.

So the first exercise prescription may not be “join a gym.”

It may be:

Stop letting the chair win the whole day.

The Benefit Is Not Linear

One of the most important ideas in exercise science is also one of the most encouraging:

The benefit is not linear.

Going from sedentary to somewhat active may matter more than going from active to perfectly optimized.

This is good news.

It means the first steps count. The first walk counts. The first week of interrupting long sitting counts. The goal is not to become an athlete. The goal is to stop giving the body a signal of stillness all day long.

For a sedentary person, the first meaningful intervention is often simple:

Move out of the sedentary category.

A person who goes from 3,000 to 6,000 steps per day may gain more biologically than a person who goes from 10,000 to 14,000 steps per day. A person who interrupts sitting every hour may change their daily metabolic pattern even without joining a gym.

This does not mean higher fitness is useless. Fitness matters. Strength matters. Vigorous activity can be valuable.

But the priority is not optimization.

The priority is restoring movement to the day.

A Gym Session Does Not Cancel a Sedentary Day

There is another trap.

Someone may exercise for 45 minutes in the morning and then sit behind a computer for 8 to 10 hours.

That is better than not exercising.

But it does not fully solve the problem.

We were not built to move once and then remain still all day. Human physiology expects repeated muscle contraction, posture change, walking, lifting, bending, carrying, and light movement across the day.

A workout is a dose.

A movement pattern is an environment.

For brain health, the environment may matter as much as the dose.

So the question is not only:

Did I exercise today?

It is also:

How long did I sit without interruption?
How often did I move during the day?
Did my muscles wake up after meals?
Did I spend the whole day in a chair?

This is especially relevant for computer work. A desk job can quietly create 10 hours of sedentary time: work, meetings, commuting, meals, and evening screens.

That is why the practical goal is not only more exercise.

It is less uninterrupted sitting.

Exercise Is Not Just Exercise

When people hear “exercise,” they picture a gym.

But physical activity is broader.

It includes walking to the store, gardening, climbing stairs, cleaning, carrying groceries, dancing, walking after meals, playing with grandchildren, standing from the floor, taking a sunset stroll, and breaking up long sitting time.

Structured exercise matters.

Habitual movement matters too.

The body does not care whether movement came from a gym or from daily lifestyle. It cares that the muscles worked, the heart responded, the blood moved, and the pattern repeated.

The best plan is usually both:

Daily movement plus structured training.

The First Target: Interrupt Sitting

If you are sedentary, the first goal is not VO2 max.

The first goal is to break up long sitting.

Try this:

Every 30 to 60 minutes, stand up and move for 2 to 5 minutes.

Walk around the room.
Climb one flight of stairs.
Do 10 body-weight squats.
Stretch your hips and calves.
Carry laundry.
Water plants.
Walk outside for sunlight.
Take a phone call standing or walking.

These small breaks look unimpressive.

But they repeatedly activate muscle, blood flow, glucose handling, and posture throughout the day.

A single workout is good. A day with repeated movement is better.

Ordinary Movement Counts

Physical activity does not have to look like exercise.

A good hike counts.

Walking the extra distance from the parking lot to the store counts.

Taking the stairs counts.

A morning stroll counts.

A sunset walk counts.

Stretching during the workday counts.

Gardening, cleaning, carrying groceries, walking the dog, dancing, and playing with grandchildren count.

These are not inferior forms of movement. They are the kind of movement human lifestyle used to contain by default.

The body does not care whether movement came from a gym, a trail, a sidewalk, a garden, or a grocery store parking lot.

It cares that the muscles worked, the heart responded, the blood moved, and the pattern repeated.

The goal is not to optimize a workout. The goal is to build a lifestyle the brain can live in.

Steps: Useful, Not Sacred

Steps are a simple way to measure daily movement.

You do not need a wearable, but a phone, watch, ring, or simple pedometer can help you see your baseline.

Measure one normal week.

If you average 3,000 steps per day, aim for 3,500 to 4,000.
If you average 5,000, aim for 6,000.
If you average 7,000, make it consistent.

The first goal is not 10,000 steps.

The first goal is more than your current normal.

For many adults, 7,000 to 10,000 steps per day is a reasonable long-term range. But the biggest gain may come from moving out of the lowest activity category.

A useful rule:

Increase gradually, then make it boringly consistent.

The Walk After Meals

One of the simplest habits is a 10-minute walk after meals.

This is especially useful after lunch or dinner, when blood sugar often rises, and people are likely to sit.

Walking after meals does not require a gym. It does not require special clothes. It does not require motivation for a full workout.

It just requires standing up and going outside.

For APOE4 carriers, this habit makes biological sense because exercise may act through vascular and metabolic pathways: glucose handling, insulin sensitivity, triglycerides, blood pressure, inflammation, and blood flow.

A walk after meals is small enough to repeat.

That is why it works.

The Morning or Sunset Stroll

Not every walk has to be a workout.

An early morning or sunset stroll can combine movement, light exposure, stress reduction, and rhythm.

Morning light helps anchor circadian timing. Evening walks can reduce stress and prevent the drift into hours of sitting after dinner.

These walks may not raise VO2 max dramatically.

That is not the point.

They help build the daily pattern.

And the pattern is the intervention.

Stretching Has a Role

Stretching is not a substitute for aerobic exercise or strength training.

But it helps people move.

Tight hips, stiff ankles, back pain, and poor balance all make activity harder. A few minutes of stretching during the workday can reduce stiffness and make the next walk more likely.

Try simple movements:

Calf stretch
Hip flexor stretch
Hamstring stretch
Chest opener
Thoracic rotation
Neck and shoulder mobility
Gentle yoga or Tai Chi

The goal is not flexibility for its own sake.

The goal is to keep the body available for movement.

Then Add Structured Exercise

Once daily movement is improving, add structured exercise.

Aim for at least 150 minutes per week of moderate aerobic activity, or 75 minutes per week of vigorous activity, plus two days per week of strength training [10].

Moderate means you can talk, but not sing.

Examples include brisk walking, cycling, swimming, hiking, dancing, rowing, or elliptical training.

The simplest plan:

30 minutes, 5 days per week.

If you are starting low, begin with 10 minutes and build.

A good hike once a week can be a powerful anchor. It adds longer aerobic work, hills, sunlight, balance, and often social connection.

That may be better than thinking of exercise only as something done indoors on a machine.

What About VO2 Max?

VO2 max measures how well your body uses oxygen during hard exercise. It is useful in research, but most people do not need a lab test.

You can track fitness in simpler ways:

Can you walk the same route faster?
Can you climb stairs with less effort?
Can you walk uphill without stopping?
Can you cover more distance in six minutes?
Is your resting heart rate lower than before?

Those are practical fitness markers.

If safe, add short intervals once or twice per week:

Warm up for 5 minutes.
Walk fast for 1 minute.
Walk easy for 2 minutes.
Repeat 4 to 6 times.
Cool down.

You do not need to suffer.

You need to progress.

If you have chest pain, severe shortness of breath, dizziness, heart disease, or uncontrolled blood pressure, talk with a clinician before vigorous exercise.

Strength and Balance

Strength training is not optional after midlife.

Muscle is a metabolic organ. It helps glucose control, balance, bone health, mobility, and independence.

Do strength training two days per week.

Include:

Squat or sit-to-stand
Hip hinge
Push
Pull
Carry
Core stability

Use body weight, resistance bands, machines, dumbbells, or household objects.

You do not need to become a bodybuilder.

You need enough strength to keep moving through daily activities.

For older adults, balance also matters. Falls can change the course of aging. Tai Chi, yoga, heel-to-toe walking, step-ups, single-leg stands, and supervised balance work all count [11].

How Do You Know If It Is Working?

Do not expect your memory test to change after 12 weeks.

Do not expect plasma Alzheimer’s biomarkers to clearly improve after a few months of exercise.

Blood biomarkers such as p-tau217 are promising research tools. Future trials should use them. But they are not yet practical personal feedback tools for exercise.

Look for nearer signals.

Your sitting time is lower.
Your steps are higher.
You walk after meals more often.
Your blood pressure improves.
Your resting heart rate falls.
Your walking pace improves.
Stairs feel easier.
Your sleep becomes more regular.
Your waist changes, even if weight does not.
Your triglycerides fall.
Your HDL rises.
Your glucose or A1c improves.
Your strength improves.
Your balance improves.
Your mood and energy improve.

For APOE4 carriers, LDL cholesterol may or may not fall with exercise alone. LDL is often influenced by genetics, diet, saturated fat intake, body weight, and medication.

That does not mean exercise failed.

It means exercise is one part of the system.

The first win is not a biomarker. The first win is changing the daily pattern.

Does Exercise Work Alone?

Exercise helps, but it does not live alone.

A person who exercises but sleeps poorly, eats a highly processed diet, has untreated sleep apnea, uncontrolled blood pressure, high stress, social isolation, or insulin resistance is still pushing the brain in the wrong direction.

This is not because people need a perfect protocol.

They do not.

It is because biology is a network.

Movement improves glucose handling, blood pressure, sleep, mood, inflammation, and muscle metabolism. Better sleep improves appetite, stress hormones, amyloid clearance, and immune regulation. A healthier diet supports the microbiome, lipids, and inflammation. Social connection lowers chronic stress signaling.

These are not separate boxes.

They reinforce each other.

Exercise may be the best first domino because it often makes the next healthy behavior easier.

Uncertainty

We should be honest.

We do not yet have definitive proof that exercise prevents dementia specifically in APOE4 carriers.

Most favorable evidence comes from observational studies. Active people may differ from inactive people in diet, education, sleep, income, medical care, vascular risk, and social connection.

The trials so far have not settled the question.

But the absence of definitive proof is not the same as evidence of no benefit.

Exercise is biologically plausible, low-cost, and strongly beneficial for systems that matter to the brain: blood vessels, metabolism, muscle, sleep, mood, inflammation, and independence.

For APOE4 carriers, that is enough reason to take it seriously while still asking for better trials.

Take-Home Messages

• Exercise for APOE4 is not just gym advice. The goal is to improve the vascular, metabolic, inflammatory, and brain-resilience systems that shape how genetic risk is expressed.

• Timing may matter. The most encouraging signals appear earlier in life, before dementia, especially in cognitively unimpaired adults.

• The benefit is not linear. Moving from sedentary to somewhat active may matter more than moving from active to highly optimized.

• Sitting is its own problem. A morning workout is helpful, but it does not fully erase 8 to 10 hours of uninterrupted sitting.

• We are built to move repeatedly. Walking, standing, stretching, carrying groceries, taking stairs, gardening, hiking, and short movement breaks all count.

• Steps are useful, not sacred. Measure your baseline if it helps, then increase gradually. The first goal is not 10,000 steps. It is more than your current normal.

• Look for near-term wins. Better blood pressure, sleep, glucose control, waist size, triglycerides, HDL, strength, balance, mood, and energy may appear long before any change in memory or Alzheimer’s biomarkers.

• Exercise works best as part of a lifestyle pattern. Diet, sleep, blood pressure control, metabolic health, stress reduction, social connection, and cognitive engagement reinforce each other.

• Do not overpromise. Exercise is not a guaranteed shield against Alzheimer’s disease.

• Do not underuse it either. For APOE4 carriers, becoming a person who moves throughout the day may be one of the most rational places to start.

Citation and Acknowledgment

Ali N, Chakbazof N, Ghasem Pour S, Contreras L, Estrada J, Alexander GE, Raichlen DA, Yassine HN. APOE ε4, physical activity, and the brain: a review of systematic reviews. Frontiers in Aging Neuroscience. 2026;18:1798639. doi:10.3389/fnagi.2026.1798639.

The work was supported in part by the National Institutes of Health/National Institute on Aging, the Alzheimer’s Drug Discovery Foundation, donations from the Vranos and Tiny Foundations and Lynne Nauss, and support to Gene E. Alexander from the Arizona Department of Health Services and the McKnight Brain Research Foundation.

The authors have no conflicts of interest.

References

1. Ali N, Chakbazof N, Ghasem Pour S, et al. APOE ε4, physical activity, and the brain: a review of systematic reviews. Frontiers in Aging Neuroscience. 2026.

2. Rovio S, et al. Leisure-time physical activity at midlife and the risk of dementia and Alzheimer’s disease. Lancet Neurology. 2005.

3. Krell-Roesch J, et al. Physical activity and incident cognitive impairment: the Mayo Clinic Study of Aging.

4. Shih IF, et al. Physical activity modifies the influence of APOE ε4 and type 2 diabetes on dementia and cognitive impairment. Alzheimer’s & Dementia. 2018.

5. Head D, et al. Exercise engagement as a moderator of APOE effects on amyloid deposition.

6. Smith JC, et al. Physical activity, APOE ε4, and brain activation during memory processing.

7. Woodard JL, et al. Physical activity, APOE ε4, hippocampal volume, and cognitive decline.

8. Deeny SP, et al. Exercise, APOE ε4, and cerebral glucose metabolism.

9. Raichlen DA, Aslan DH, Sayre MK, et al. Sedentary behavior and incident dementia among older adults. JAMA. 2023;330(10):934–940.

10. CDC. Adult physical activity recommendations.

11. CDC. Physical activity recommendations for older adults.

12. del Pozo Cruz B, Gallardo-Gómez D, del Pozo-Cruz J, Ding D. How many steps a day to reduce the risk of all-cause mortality? A dose-response meta-analysis. Journal of Internal Medicine. 2022;291(4):519–521.

13. Ekelund U, Tarp J, Steene-Johannessen J, et al. Dose-response associations between accelerometry measured physical activity and sedentary time and all-cause mortality. BMJ. 2019;366:l4570.

The short version: landing in the highest p-tau217 group is an important finding, and it’s also, like most things in medicine, probabilistic rather than deterministic. It meaningfully shifts the odds of decline. It does not function like a positive or negative diagnostic result, and it does not tell a person what their future holds. Most people in that highest group, in every dataset discussed below, did not go on to develop dementia within the years these studies were able to follow them. That distinction, informative versus determinative, is the throughline of everything that follows. (More on how to weigh that balance, with a real-world comparison, at the end.) Today, p-tau217’s clearest use is identifying good candidates for prevention trials, replacing expensive and complex PET scans; but it is not yet ready for clinical use, and current guidelines recommend against using it to screen healthy people outside of research.

What does p-tau217 actually measure?

p-tau217 is a small piece of tau protein that leaks out of brain cells into blood. Its release is triggered mostly by amyloid, the sticky protein plaques that build up early in Alzheimer’s disease. So for most of its range, a rising p-tau217 mainly means “more amyloid,” not “more tau tangles.” Only later in the disease does it start reflecting actual tau buildup too, the kind seen on a specialized brain scan called a tau PET. That scan lights up wherever tangled tau protein has physically accumulated inside neurons; tau doesn’t spread evenly, it tends to start in one small area and then move into connected regions over time, the way an outbreak spreads along a transit line rather than raining down on a whole city at once.

Three more things are worth knowing in plain terms. A Centiloid is just a 0-to-100-plus scale for how much amyloid shows up on a brain scan, built so different scanners and tracers can be compared fairly; 0 is a young, healthy brain, 100 is roughly what’s typical in mild-to-moderate Alzheimer’s dementia. Harmonizing means converting each lab’s raw number onto a shared scale (usually “how many standard deviations from average”), since different test brands don’t report the same raw numbers for the same biology. And logistic regression, a term that shows up later, is simply the standard math for turning a measurement into a probability, like “what’s the chance this person’s tau has started accelerating,” producing a curve rather than a hard yes/no line.

How results get labeled, and why that label isn’t the whole story

One widely used commercial test (Lumipulse, made by Fujirebio) sorts results into three bands:

Category Lumipulse p-tau217 (pg/mL) Interpretation Negative ≤ 0.18–0.20 Alzheimer’s-type pathology unlikely Indeterminate / gray zone 0.18–0.32 Borderline; further evaluation recommended Positive ≥ 0.32–0.33 High probability of Alzheimer’s-type pathology

What these numbers actually mean: these specific cutoffs were derived by comparing symptomatic memory-clinic patients with confirmed Alzheimer’s pathology (via PET scan or spinal fluid) against those without it. They mark where, in that particular symptomatic population, the test best tells the two groups apart, not a universal biological line in the sand. Other manufacturers (Eli Lilly, Roche Elecsys, C2N, and others) use different chemistry and validate their own, different pg/mL cutoffs. There is no single, agreed-upon p-tau217 cutoff across labs today. That lack of a uniform cutoff is one of the main reasons this test cannot yet support a specific clinical recommendation, and it’s worth remembering as a take-home point on its own.

Just as important regardless of which band a result falls in: no p-tau217 result should replace an actual cognitive evaluation. A “low” number doesn’t mean skip a checkup if something feels off, and a “high” number doesn’t mean a diagnosis has been made. The blood test is one input, not a verdict, and it hasn’t been validated for screening healthy people outside of research studies.

Research studies mostly sidestep these fixed bands. Instead they rank people against each other, low, intermediate, high, “very high,” based on their own study population. So “very high” in a study (roughly the top fifth of people, corresponding to substantial amyloid buildup on a brain scan) is a relative ranking within that study, not a number you could match against a personal lab report using the commercial bands above.

The JAMA study: what nearly 2,700 people showed

This study pooled six research groups (ADNI, the A4 and LEARN studies, Harvard Aging Brain Study, Wisconsin Registry for Alzheimer’s Prevention, and HABS-HD), for 2,684 cognitively healthy older adults followed a median of about 5.4 years, up to 13.5 years. Everyone had both a blood test and a brain amyloid scan at the start.

One term needs defining clearly: “progression.” Here it meant reaching a diagnosis of mild cognitive impairment (MCI), dementia, or two consecutive visits showing very mild impairment on a standard staging scale. A separate, second measure tracked gradual decline on cognitive testing, without requiring a formal diagnosis; it told the same story.

The results, in plain numbers:

• Low or intermediate p-tau217 (roughly under 25 Centiloids of amyloid): just a 12–15% chance of progressing over 5 years. Genuinely reassuring.

• High p-tau217: 24% over 5 years, 62% over 10.

• Very high p-tau217 (roughly the top fifth, over 60 Centiloids): 38% over 5 years, 78% over 10, about four times the lowest group’s risk.

Here’s the point worth sitting with: even in that highest-risk group, nearly two-thirds had not progressed after 5 years, and even at 10 years, about 1 in 5 still hadn’t. A high result shifts the odds substantially. It doesn’t seal anyone’s fate.

Some honest limits: these six groups were mostly white, well-educated research volunteers, not a random slice of the public, so results may not generalize perfectly. The four risk groups also overlap somewhat rather than being sharply distinct categories, deaths weren’t tracked as a competing outcome, and only 5% of people were followed past 10 years, so the longest-term numbers are least certain. Risk also ran a bit higher in men and older participants at the same p-tau217 level, a reminder that one number doesn’t mean the same thing for everyone.

A quick note on sources, because three different tools appear in this piece and they shouldn’t be treated as one. The JAMA study above used its own six research groups and its own definition of progression. A separate paper from the Mayo Clinic Study of Aging, cited by the JAMA authors, used a different measure (brain amyloid scans rather than blood tests) to estimate risk specifically of MCI. And the Mayo calculator used below is a third, distinct tool, built on blood p-tau217, projecting decades into the future using its own statistical model. All three are useful. None of their numbers should be lined up against each other as if they were measuring the exact same thing.

Seeing it with real amyloid numbers

A related analysis (from the A4 and LEARN studies) split people into thirds by p-tau217 and matched each third to actual brain amyloid levels: roughly 47, 61, and 84 Centiloids for the low, middle, and top thirds. The top third showed a clearly steeper rise in confirmed decline over six years than the bottom third, whose amyloid levels averaged only about 4 Centiloids.

A worked example: the Mayo calculator

The Mayo Clinic Study of Aging built a public tool (the Mayo CACR calculator) that turns an actual p-tau217 number, plus age, sex, and genetics, into a personalized curve. It’s a different, separate model from the JAMA study above, using its own long-term projections, so treat the two as complementary rather than directly interchangeable; its multi-decade curves are computed projections, not observed follow-up. Readers curious to see how the numbers shift for a different age, sex, or p-tau217 value can try it directly at the link above.

Running it for a 60-year-old woman carrying the APOE4 gene (more on that below), at three “positive-range” p-tau217 values:

By age 80 (20 years out) p-tau217 0.2 p-tau217 0.3 p-tau217 0.4 Chance of MCI 26% 34% 40% Chance of dementia 11% 17% 24%

Even within this one “positive” range, risk more than doubles. That’s the main lesson: a positive result isn’t one fixed risk level; it behaves like a sliding scale.

And here’s the projection worth remembering most: even at 0.4 pg/mL, the highest value tested here, the model estimates about three in four people (76%) may not develop dementia within 20 years. This example is not exactly representative; it is based on cognitively normal individuals with AD-level brain amyloid deposition (rare occurrence). This is also a simulation, not an observed outcome, so treat it as a projection rather than a fact about any real person’s future.

Picture 100 women exactly like this one, modeled forward: somewhere between 11 and 24 are projected to develop dementia by their early 80s, and the rest are projected not to, within that stretch of time, though real life includes things this kind of model can’t fully capture, other illnesses, earlier death from unrelated causes, or simply better or worse luck than average. A high biomarker is a real signal. It is not a countdown clock.

A useful way to weigh how important this all is: a classic analysis validating the Framingham risk model found men with the highest total cholesterol (280 mg/dL or above) were about twice as likely to have a coronary heart disease event as men with a more typical level (160–199 mg/dL). That twofold gap is in the same range as the two- to fourfold difference this JAMA study found between its highest and lowest p-tau217 groups. High cholesterol built an entire field of preventive cardiology on a risk difference of that size, and nobody dismisses it as minor. Yet no one calls a high cholesterol reading a diagnosis: most people with high cholesterol never have a heart attack. Both things are true of p-tau217 too. A high result is a real, medically meaningful risk signal, and it still isn’t a prediction about any one person’s future.

Genetics changes the math: the Brain study on APOE4

The APOE4 gene is the strongest common genetic risk factor for Alzheimer’s. A study in Brain, using two independent groups of people (one from ADNI, one from the A4-LEARN study), asked a sharp question: does carrying APOE4 change what a given p-tau217 number means?

The answer split in an interesting way. APOE4 did not make amyloid trigger more p-tau217 release. But it did make a given p-tau217 level more likely to trigger actual tau spread across the brain, and the effect grew with each copy of the gene. Put simply: carrying APOE4 seems to lower the bar for tau pathology to start actively spreading, at the very same blood test number.

Using statistical modeling, the researchers estimated the p-tau217 level marking a 50% chance of that tau-spread milestone: in one group, about 0.62 pg/mL for non-carriers, 0.34 for one copy of APOE4, 0.15 for two copies. A second, independent group confirmed the same pattern using a different lab test, though the specific numbers from different assay brands shouldn’t be compared to each other directly, they’re not on the same scale.

A separate comparison using the Mayo calculator points the same direction:

p-tau217 (pg/mL) Carrier: dementia risk by 80 Non-carrier Gap 0.2 11% 9% 2 points 0.3 17% 14% 3 points 0.4 24% 20% 4 points

The gap between carriers and non-carriers widens as p-tau217 rises. That’s consistent with the Brain study’s finding, though this specific comparison comes from a separate tool and isn’t direct proof of the same mechanism.

One nuance worth flagging: does this hold at every level, or only some? The Brain study’s effect was clearer in the group with more advanced disease and murkier in the earlier-stage group, so it’s not yet settled whether the genetic effect is constant from the very earliest biology all the way through, or whether it grows stronger over time. That distinction matters for anyone hoping to use a single p-tau217 cutoff for everyone: the same number may carry meaningfully different risk depending on genotype, and possibly depending on how far along someone already is.

And the thresholds identified for “tau starting to spread” are low in absolute terms, especially for APOE4 carriers, yet the resulting dementia risk, as shown above, still stays fairly modest for years afterward. That’s not a contradiction: crossing that biological threshold marks the quiet start of a process that typically takes additional years to turn into noticeable decline. Think of it as an early warning light coming on in a car with a lot of road left before the engine actually stalls.

Some real limits here too: a key brain region (the hippocampus) had to be excluded from the tau scans for technical reasons, the “spread” maps rely on estimated brain wiring rather than confirmed anatomy, and blood p-tau217 reflects tau leaking out of cells, not the actual tangles forming inside them. The two research groups also used different lab-test brands, so their exact pg/mL cutoff numbers, and the Mayo calculator’s numbers above, shouldn’t be treated as the same scale even where the figures happen to look similar.

What’s coming next: chasing a “goldilocks” window

That gap between early biology and later symptoms is exactly what the newest prevention trials are trying to exploit. The idea: treat people early enough that amyloid is clearly present (worth the risks of a drug) but before tau pathology has spread very far, since once tau is well underway, removing amyloid may not change the trajectory much anymore. Call it a goldilocks window, not too early, not too late.

Three trials are testing this now. TRAILBLAZER-ALZ 3 (donanemab) screened over 63,000 people with plasma p-tau217 to find about 2,200 eligible participants meant to have amyloid with only early tau. AHEAD 3-45 (lecanemab) splits people into two amyloid-level groups, an earlier group with moderate amyloid buildup and a more advanced group with higher amyloid, and uses blood tests mainly to cut down on unnecessary brain scans rather than to gatekeep tau status directly. PrevenTRON, a newly announced trial of a drug called trontinemab, plans to enroll around 1,600 people selected largely by an elevated blood result, with the main goal of measuring time until noticeable decline begins.

Here’s the catch, and it connects right back to the earlier point that p-tau217 partly reflects tau, not just amyloid: using it alone to find “amyloid present, tau still limited” people risks letting some people in who already have more tau than intended. TRAILBLAZER-ALZ 3’s own data show this plainly, roughly 1 in 6 to 1 in 4 of its p-tau217-selected participants already had elevated tau on a brain scan. That’s less a flaw than an unavoidable trade-off: scanning 63,000 brains isn’t realistic, so a blood test is used instead, at some cost to precision.

It’s also worth saying plainly that this whole “goldilocks” idea comes mostly from looking back at subgroups of older trials, not from a trial designed around it from the start. Ideas like that often look weaker, or disappear, once tested directly instead of spotted after the fact. These three trials are a real test of whether the idea holds up.

Research tool, not a clinical test

To repeat the core message: this blood test is not ready for ordering on a healthy person outside of research. Official guidelines recommend against it in routine care, and even the JAMA study’s own authors describe their findings as useful for research and trial design, “not yet precise enough to guide individual prognosis.”

What is a very high p-tau217 good for today? Mainly, identifying good candidates for prevention trials, exactly how it’s used in the trials described above. It is not a trigger for a specific medical decision, because there’s no approved treatment yet that a healthy person’s blood result would unlock. The general advice for someone with a high result, manage vascular risk factors, stay physically and mentally active, sleep well, avoid excess alcohol, is the same advice anyone at elevated risk should already be getting, biomarker or not. Being “eligible for a trial” and “needing a different care plan” are not the same thing, and it’s easy to blur the two. Larger, more representative studies, and clearer genetics- and lab-specific calibration, are still needed before that changes.

Take-home messages

• A very high p-tau217 result is a genuinely important finding. It meaningfully raises the odds of future cognitive decline in a healthy older adult, comparable in scale to how a marker like high cholesterol raises heart disease risk.

• Important is not the same as certain. Most people, even in the highest-risk group, do not progress within the years these studies could follow. A strong signal shifts the odds; it doesn’t decide any one person’s fate.

• There’s no universal cutoff yet. Different labs and assays define “positive” differently, largely because those cutoffs were built for diagnosing symptomatic patients, not screening healthy ones.

• A blood result never replaces a clinical evaluation, whatever the number says.

• Genetics changes what a given number means. Carrying APOE4 appears to shift the biology, so the same result may not carry the same meaning for everyone.

• Long-range projections are models, not facts. Multi-decade risk curves are useful for intuition, not for predicting any individual’s future.

• New prevention trials are testing an appealing but unproven idea: treating people early enough to matter, before too much tau has spread. Whether that strategy actually works is still an open question.

• This is a research tool, not a clinical test. Current guidelines recommend against using it to screen healthy people outside of research studies and trials.

References

1. Buckley RF, Townsend DL, Birkenbihl CJ, et al. Prognostic value of blood-based p-tau217 levels for progression to cognitive impairment. JAMA. Published online July 14, 2026. doi:10.1001/jama.2026.12556

2. Sperling RA, Donohue M, Rissman R, et al. Amyloid and tau prediction of cognitive and functional decline in unimpaired older individuals: longitudinal data from the A4 and LEARN studies. J Prev Alzheimers Dis. 2024;11(4):802-813.

3. Jack CR Jr, Hu M, Wiste HJ, et al. Lifetime and 10-year absolute risk of cognitive impairment in relation to amyloid PET severity: a retrospective, longitudinal cohort study. Lancet Neurol. 2025;24(12):1016-1025.

4. Mayo Clinic Study of Aging. CACR p-tau217 calculator (Fujirebio Lumipulse Research Use Only assay, version 2). Available at: rtools.mayo.edu/CACR. Accessed July 2026.

5. Steward A, Dewenter A, Hirsch F, et al. ApoE4 lowers the ptau217 threshold for tau aggregation and spread in an allele dose-dependent manner. Brain. 2026;149(6):1902-1914. doi:10.1093/brain/awaf463

6. Palmqvist S, Whitson HE, Allen LA, et al. Alzheimer’s Association Clinical Practice Guideline on the use of blood-based biomarkers in the diagnostic workup of suspected Alzheimer’s disease within specialized care settings. Alzheimers Dement. 2025;21(7):e70535.

7. Yaari R, Holdridge KC, Williamson M, et al. Donanemab in preclinical Alzheimer’s disease: screening and baseline data from TRAILBLAZER-ALZ 3. Alzheimers Dement. 2025;21(9):e70662. doi:10.1002/alz.70662

8. Rafii MS, Sperling RA, Donohue MC, et al. The AHEAD 3-45 study: design of a prevention trial for Alzheimer’s disease. Alzheimers Dement. 2023;19(4):1227-1233. doi:10.1002/alz.12748

9. Roche/Genentech. PrevenTRON: rationale and design of a Phase III trial of trontinemab in cognitively unimpaired individuals with biomarkers of Alzheimer’s disease at high risk of clinical decline. Presented at the Alzheimer’s Association International Conference (AAIC), London, July 2026.

10. Gander J, Sui X, Hazlett LJ, Cai B, Hébert JR, Blair SN. Factors related to coronary heart disease risk among men: validation of the Framingham Risk Score. Prev Chronic Dis. 2014;11:E140. doi:10.5888/pcd11.140045

Finding the right population

The first problem for lifestyle trials is deciding who to enroll. Enroll people already impaired, and you’re fighting neurodegeneration built up over a decade. Enroll young and healthy people, and you may need twenty years before enough develop dementia to measure anything. FINGER recruited adults aged 60–77 with an elevated CAIDE dementia risk score (vascular risk factor score) and cognition at or just below what’s expected for their age [1]. U.S. POINTER cast a wider net: about 2,100 older adults, 78% with a family history of memory loss and 30% carrying the APOE-e4 gene copy, recruited to include groups historically underrepresented in dementia research [2]. MIND enrolled people with a family history of dementia, a BMI over 25, and — this is worth being precise about — a MIND diet screener score of 8 or less out of 14, meaning eligibility required eating poorly by that specific measure [3]. That’s a genuinely hard recruitment target, and credit is due for the effort: of roughly 1,900 people screened, more than 500 were excluded for already eating too well on the MIND scale. This was not a casual “self-reported suboptimal diet” — investigators actively went looking for a diet-deficient population.

There’s a further wrinkle worth naming, though: advantaged and diet-poor are not the same thing. FINGER and POINTER recruited educated, engaged older adults with reasonable baseline access to exercise, social stimulation, and health care — capable of sustaining a multi-year program. MIND, despite deliberately selecting for a poor baseline diet, still enrolled a demographically similar cohort on nearly every other axis: 87.7% white and an average of 16.9 years of education, recruited through two academic medical centers. A low MIND-diet score and genuine socioeconomic disadvantage are different problems, and it’s easy to conflate them. Layering an elaborate program onto an already-favorable baseline — of education, health-system access, and time to participate — has a much smaller ceiling to push against than aiming the same idea at a population without stable food access, safe places to exercise, or equivalent support.

That reframes where the bigger opportunity might sit. A far cheaper intervention — reliable access to real food, a free community fitness membership, social infrastructure as a safety net — delivered to a population currently missing all of that could plausibly produce a larger absolute effect than an elaborate protocol layered onto people already doing reasonably well.

The FINGER investigators noticed this population gap. World-Wide FINGERS has expanded to more than 40 countries, including Africa-FINGERS, now piloting structured multidomain coaching across several African nations with genuine community co-design [4]. But what’s being exported is still the FINGER template — coached diet counseling, structured exercise, cognitive training, vascular monitoring — adapted in delivery, not in kind. A different, more urgent question: does closing the gap itself — food security, safe places to exercise, a functioning safety net — produce more benefit, more cheaply, than a coached protocol layered on scarcity? That trial, as far as I can tell, hasn’t been run.

None of that is meant to sound easy. Recruiting and retaining a genuinely underserved population in a multi-year lifestyle trial is its own serious problem — transportation, work schedules, trust in research institutions, and the competing demands of poverty all erode compliance in ways a motivated, retired, academic-medical-center-adjacent volunteer pool doesn’t face. That’s precisely why a standard multi-year RCT of the FINGER or POINTER kind may be the wrong instrument here. Pragmatic and decentralized trial designs — community-embedded cluster randomization through existing food-assistance or primary-care infrastructure, stepped-wedge rollouts, or remote and digital monitoring that cuts down on repeated clinic visits — are better suited to testing a basic-access intervention in a population a conventional trial would struggle to retain.

Choosing the intervention — and the control

FINGER and POINTER both bundled diet, exercise, cognitive training, and vascular risk management together. MIND isolated one variable, testing a hybrid Mediterranean-DASH diet against a mild-calorie-restriction control [3]. But the more revealing design choice, across all three, is what the comparison group actually received — and in every case, it wasn’t nothing. FINGER’s control got “regular health advice” on diet, exercise, and vascular risk. POINTER’s self-guided arm attended six peer group meetings, got general health education across every domain, and received small incentives [2]. MIND’s control diet produced almost exactly as much weight loss as the MIND diet itself — about five kilograms in both arms over three years, with real protective effects on blood pressure, lipids, and insulin sensitivity baked into both arms before the specific food pattern is even considered.

MIND’s design raises a further question, and it’s baked into the score itself, not just the intervention. The MIND diet index sums 15 food groups — 10 to eat more of (leafy greens, other vegetables, nuts, berries, beans, whole grains, fish, poultry, olive oil, wine) and 5 to eat less of (red meat, butter, cheese, pastries and sweets, fried food). It’s an optimization checklist, built to reward a specific idealized eating pattern, not a measure targeted at ultra-processed food intake, added sugar, or fiber as such. Someone can score low on MIND simply by skipping berries, nuts, and olive oil, whether or not their diet is high in ultra-processed food; someone else could hit enough of the positive categories to score reasonably while still eating a lot of packaged, sugar-heavy, low-fiber food the index was never built to catch. The typical U.S. diet MIND’s participants started from derives roughly 55% of calories from ultra-processed foods [5], and both low fiber and high ultra-processed intake are independently linked to worse cognitive trajectories via gut-brain and inflammatory pathways [6,7] — pathways the MIND score, by design, wasn’t built to target directly. Cutting calories, ultra-processed foods, and raising fiber may be the bigger lever than a curated food list of berries, olive oil, and nuts layered on a diet still built around what it should avoid.

That matters for how to read the results. The real question isn’t “does lifestyle change help, compared with nothing.” It’s narrower: does an intensive, expert-coached optimization protocol add much on top of the active ingredient the comparison group already shared — guidance in FINGER, structured support in POINTER, weight loss in MIND? In all three, the honest answer is: a little, but not a lot — a very different message than “breakthrough lifestyle program prevents cognitive decline,” which is closer to how these trials were actually presented.

Reading the effect sizes honestly

This is where the field’s self-presentation runs ahead of the data. FINGER’s primary outcome, a composite neuropsychological z-score, improved 0.20 in the intervention group and 0.16 in control over two years [1] — an absolute difference of about 0.04 standard deviations (see illustration below). Reported as “25% greater improvement than control,” that’s true, but it’s a relative figure built on a very small absolute one, and the same arithmetic produced the widely quoted 83% gain in executive function and 150% gain in processing speed.

A more honest yardstick is Cohen’s d, the standard way to express an effect on a common scale, alongside its plain-language twin, the common language effect size: the odds a random person from the structured group scored higher than a random person from the comparison group. Using FINGER’s own model-based d ≈ 0.13, that’s roughly a 54-in-100 chance — barely better than a coin flip. Given the comparison group wasn’t null but a lighter-touch version of the same idea, that’s about what you’d expect: the marginal value of intensive structuring on top of ordinary guidance, not lifestyle change measured against nothing. (This figure comes from the trial’s own supplementary analyses; check the appendix directly.)

What’s notable is less the size of that number than how it’s been communicated: as a landmark breakthrough, driving funding to replicate the FINGER model in more than 60 countries through World-Wide FINGERS. A modest, real, worth-pursuing signal is not a breakthrough.

U.S. POINTER’s own translation — “equivalent to one to two years of slowed cognitive aging” — deserves similar scrutiny. That figure is a model-based extrapolation against an assumed normative decline rate, not a directly observed outcome — a generous way to make a small number sound tangible.

Set against the wider record, FINGER and POINTER look like the best case, not the norm. MAPT [8] found no significant effect on cognitive decline; preDIVA [9] — the rare trial actually powered to detect a difference in dementia incidence itself — found none either; nor did HATICE [10] (see table below). “Mixed and modest” is a fair summary of this literature.

preDIVA’s null result also points to a structural problem: incident dementia is an extraordinarily hard outcome to power a trial on, requiring thousands of participants and years of follow-up — why most trials fall back on cognitive composites as a proxy, the very proxies producing the small, hard-to-interpret effect sizes above. This is a core reason nutrition and lifestyle trials are structurally harder to design than drug trials: drugs can often show a biomarker signal, while lifestyle change needs years before it plausibly shows up in diagnosable disease.

Effect sizes across the major prevention trials, at a glance:

Trial Comparison Primary effect size Plain-language interpretation FINGER (2015) [1] Coaching vs. general health advice, 2 yrs Composite z: 0.20 vs 0.16 (Δ≈0.04 SD); d ≈ 0.13 ~54-in-100 chance a treated person outscores a control person — barely above a coin flip U.S. POINTER (2025) [2] Structured vs. self-guided coaching, 2 yrs 0.029 SD/yr greater, structured arm Marketed as “1–2 years of slowed cognitive aging” — a model-based extrapolation, not a measured outcome MIND (2023) [3] Diet vs. calorie-restriction control, 3 yrs No significant difference in cognition or MRI Both arms improved and lost ~5 kg; weight loss, not the food pattern, may be the active ingredient MAPT (2017) [8] Coaching + omega-3 vs. placebo, 3 yrs No significant effect on cognitive decline Null on the primary outcome preDIVA (2016) [9] Nurse-led vascular care vs. usual care, 6 yrs Dementia incidence: 7% vs 7%; HR 0.92 Null; the rare trial actually powered on dementia diagnosis, not a proxy HATICE (2019) [10] Internet-coached vascular care vs. control, 18 mo No significant cognitive effect Modest improvement in cardiovascular risk factors only

The two columns look like a verdict — "significant" versus "null" — but the underlying effect sizes are small enough on both sides that the line between them is closer to a statistical threshold than a true difference in what these interventions accomplished. A slightly larger sample, or a slightly different outcome measure, could plausibly have moved several of these trials across the line in either direction.

Choosing outcomes that match the mechanism

Layered on the incidence problem is a measurement problem, and it isn’t just that standard batteries are blunt — they may be tuned to the wrong domain. Look again at FINGER’s own breakdown by cognitive domain: the largest relative gains were in executive function (83%) and processing speed (150%), while memory — the domain most closely tied to classic Alzheimer’s pathology, and the one nearly every AD drug trial is built around — showed the smallest relative gain, at 40%. That pattern fits what a lifestyle intervention plausibly does: support attention, task-switching, and processing speed through cardiovascular and metabolic pathways, rather than directly reversing amyloid-and-tau-driven memory decline.

Standard batteries, though, are built around exactly the domain lifestyle interventions seem to move least. Memory subtests dominate most AD-oriented composites, because memory is what clinicians use to diagnose dementia. Executive function and processing speed are typically assessed with a handful of brief, coarse paper-and-pencil tasks — a few minutes of Trail Making or digit-symbol substitution — far less sensitive to subtle change than a memory test built and validated for detecting AD. If the domains most responsive to diet and exercise are the ones our batteries measure worst, that could make a real effect look smaller and noisier than it is, on top of the floor, ceiling, and practice effects already limiting these tools in healthy populations [11,12].

This is where digital assessment tools have a genuine edge, beyond being reliable or convenient. They can deliver dozens of processing-speed and executive-function trials in a single sitting, repeated over months from a phone, with millisecond-level precision a stopwatch and paper form can’t match — capturing graded change in reaction time and task-switching cost, not one coarse pass-fail score. Repeated at-home testing also averages out day-to-day noise and practice-effect drift, which often swamps small true effects in one-off batteries. A recent scoping review found digital tools improve reliability and ecological validity relative to a single in-clinic assessment [13]; the sharper opportunity is fitting the tool to the domain — fine-grained, repeatable measures of processing speed and executive function as primary outcomes, rather than composite scores weighted toward a memory domain a diet-and-exercise program was never especially likely to move.

What basic science suggests we might be missing

A newly published mouse study adds a mechanistic layer here, not because mouse data settle anything about human trials, but because it shows how cell-type-restricted an exercise response can be. Researchers used single-nucleus RNA sequencing to identify which brain cells respond to exercise in the hippocampal dentate gyrus of an Alzheimer’s mouse model [14]. The response was concentrated in immature neurons, in a neurovascular-associated astrocyte population depleted in Alzheimer’s mice but partly restored by exercise, and in oligodendrocyte progenitor cells, which had the largest share of disease-related genes normalized by exercise of any cell type studied. An earlier study similarly found that long-term voluntary exercise reversed cognitive impairment in an Alzheimer’s mouse model through astrocyte remodeling, without changing amyloid-beta plaque burden [15]. Exercise may not clear the disease’s hallmark pathology so much as build a parallel, cell-specific form of resilience around it.

The clearest thread connecting these findings to the human domain pattern above runs through the neurovasculature unit. Both studies converge on astrocytes, cells that couple neural activity to blood flow and glucose delivery — so an exercise-induced astrocyte response plausibly improves neurovascular efficiency rather than clearing amyloid or tau directly. That fits a well-established pattern in the vascular cognitive impairment literature: circuits depending on blood flow and white-matter integrity, largely underlying processing speed and executive function, are especially sensitive to vascular health, while memory decline is more specifically driven by amyloid and tau damage to hippocampal circuits. If exercise, weight loss, and vascular risk management work substantially through this route, that would explain why FINGER’s and POINTER’s clearest signal was in executive function and processing speed, not memory — a coherent hypothesis, not a demonstrated mechanism, but better-specified than “lifestyle helps the brain.”

Uncertainty and future directions

None of this is settled. The cell-specific effects above come from mice, and the neurovascular story linking them to the human executive-function signal is a plausible synthesis, not a demonstrated chain. Real, useful signals are also getting inflated by relative-risk framing and normative-decline translations, in a field where funding rewards breakthrough narratives over careful ones — not reason to dismiss FINGER or POINTER, but reason to ask, as Africa-FINGERS scales up, whether it’s testing the right thing: an elaborate coached protocol exported wholesale, or a cheaper, gap-filling alternative that hasn’t had its own trial.

Take-Home Messages

• Target the population where the room to improve is largest: FINGER, POINTER, and even diet-screened MIND enrolled cohorts that were educated and well-resourced, whatever their baseline diet score. A poor diet score and genuine socioeconomic disadvantage are not the same thing — the bigger opportunity likely sits in vulnerable populations with limited access to whole foods or exercise infrastructure, and even Africa-FINGERS is still exporting the expensive coached protocol rather than testing whether closing basic gaps works better alone.

• De-emphasize expensive “optimization protocols”: every comparison group already received an active ingredient — health advice, self-guided coaching, or weight loss — and the elaborate add-on produced only a small increment on top. The MIND diet score itself is an optimization checklist, not a measure built around ultra-processed food, added sugar, or fiber; cutting ultra-processed foods and raising fiber may matter more than a curated food list.

• Recruiting underserved populations is genuinely hard, and that’s not a footnote: transportation, work schedules, and trust in research institutions all erode compliance. Pragmatic, decentralized, or community-embedded trial designs may be better suited to testing basic-access interventions than a standard multi-year RCT.

• Matching outcomes to mechanism: interventions working through vascular pathways plausibly act on processing speed and executive function more than on amyloid/tau-driven memory decline — outcomes should be weighted accordingly, not defaulted to memory-heavy composites built for drug trials.

• Digital, phone-based executive-function testing could change the landscape: repeated, low-burden testing averages out day-to-day noise and practice effects, detecting small, domain-specific effects one-off batteries are too noisy to catch.

References

1. Ngandu T, Lehtisalo J, Solomon A, et al. A 2 year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial. Lancet. 2015;385(9984):2255-2263.

2. Baker LD, Espeland MA, Whitmer RA, et al. Structured vs self-guided multidomain lifestyle interventions for global cognitive function: the U.S. POINTER randomized clinical trial. JAMA. Published online July 28, 2025. doi:10.1001/jama.2025.12923.

3. Barnes LL, Dhana K, Liu X, et al. Trial of the MIND diet for prevention of cognitive decline in older persons. N Engl J Med. 2023;389(7):602-611.

4. Udeh-Momoh CT, Maina R, Anazodo UC, et al. Dementia risk reduction in the African context: multi-national implementation of multimodal strategies to promote healthy brain aging in Africa (the Africa-FINGERS project). Alzheimers Dement. 2024;20(suppl 12). doi:10.1002/alz.14344.

5. National Center for Health Statistics (CDC). Ultra-processed food consumption among adults and youth, United States, August 2021–August 2023. NCHS Data Brief No. 536. Hyattsville, MD: 2025.

6. Zhao Y, Bazan NG. Nutrition and gut-brain axis: opposing effects of dietary fiber and Western-style diets on Alzheimer’s disease. Curr Opin Clin Nutr Metab Care. 2026;29(4):421-430.

7. Gomes Gonçalves N, Vidal Ferreira N, Khandpur N, et al. Association between consumption of ultraprocessed foods and cognitive decline. JAMA Neurol. 2023;80(2):142-150.

8. Andrieu S, Guyonnet S, Coley N, et al. Effect of long-term omega 3 polyunsaturated fatty acid supplementation with or without multidomain intervention on cognitive function in elderly adults with memory complaints (MAPT): a randomised, placebo-controlled trial. Lancet Neurol. 2017;16(5):377-389.

9. Moll van Charante EP, Richard E, Eurelings LS, et al. Effectiveness of a 6-year multidomain vascular care intervention to prevent dementia (preDIVA): a cluster-randomised controlled trial. Lancet. 2016;388(10046):797-805.

10. Richard E, Van den Heuvel E, Moll van Charante EP, et al. Healthy ageing through internet counselling in the elderly (HATICE): a multinational, randomised controlled trial. Lancet Digit Health. 2019;1(8):e424-e434.

11. Cohen S, Cummings J, Knox S, Potashman M, Harrison J. Clinical trial endpoints and their clinical meaningfulness in early stages of Alzheimer’s disease. J Prev Alzheimers Dis. 2022;9(3):507-522.

12. Aschenbrenner AJ, Hassenstab J, Wang G, et al. Avoid or embrace? Practice effects in Alzheimer’s disease prevention trials. Front Aging Neurosci. 2022;14:883131.

13. Berron D, et al. A scoping review of remote and unsupervised digital cognitive assessments in preclinical Alzheimer’s disease. npj Digit Med. 2025;8:266.

14. da Rocha JF, Lance ML, Luo R, et al. Protective exercise responses in the dentate gyrus of Alzheimer’s disease mouse model revealed with single-nucleus RNA-sequencing. Nat Neurosci. 2025;28:1546-1561.

15. Belaya I, Ivanova M, Sorvari A, et al. Astrocyte remodeling in the beneficial effects of long-term voluntary exercise in Alzheimer’s disease. J Neuroinflammation. 2020;17:271.

You are reading because the stakes are personal.

You want to know what to eat, which medications to take or avoid, which supplements might help, whether you should be on anti-amyloid treatments, whether to join a trial, and what you can do now. That urgency is understandable. It also makes people vulnerable to exaggerated claims.

A supplement guru can cite a case report. An influencer can cite a breakthrough observational study. A podcast guest can quote an alarming FDA warning. A headline can promise “the truth,” “what doctors won’t tell you,” or “the end of Alzheimer’s.”

The problem is not curiosity. Patients and families should read, ask questions, debate, challenge claims, and participate in research when possible. That is how the field improves.

The problem is certainty with weak evidence, or worse, certainty without evidence.

Science is cautious because biology is complicated. Breakthroughs are rare. A patient story is not a clinical trial. An association is not causation. A statistically significant result is not always meaningful. A warning label is not proof that a drug causes disease. And one study is rarely the final word.

1. Words Designed to Catch Your Attention, but Should Make You Slow Down

The first warning sign is often the headline.

Be careful when a health claim begins with phrases like:

“The truth about…”

“Breaking news…”

“The cure”

“The end of disease”

“The number one cause…”

“Medication risks nobody warns you about”

“Why doing this changes everything”

“What doctors won’t tell you”

“Doctors have it all wrong”

“Big Pharma has been lying to you”

“The real driver of disease”

“The study they don’t want you to see”

“Clinically proven”

“Reverses Alzheimer’s”

“Targets the root cause”

“Ancient remedy validated by science”

These phrases are designed to create a sense of urgency, distrust, or certainty before the evidence is discussed.

Scientific language usually sounds more modest: “is associated with,” “may reduce,” “suggests,” “requires replication,” “hypothesis-generating,” “not powered to detect,” “exploratory analysis,” or “clinical significance remains unclear.”

That language may sound less exciting, but it is usually more honest.

The late astrophysicist Richard Feynman captured this spirit well: “I think it’s much more interesting to live not knowing than to have answers which might be wrong.”

That is the heart of science. It is better to admit uncertainty than to be certain about something false.

A good scientific claim would not give the impression of certainty without weighing the evidence. It shows you the evidence, the uncertainty, and helps you make an informed decision.

2. Case Reports and Abstracts: Useful Clues, Weak Proof

Some of the most persuasive health claims begin with a story.

A patient was declining. Then they started a protocol, supplement, diet, or program. Their memory improved. Their family noticed. Their test score looked better.

Stories matter. And Social Media platforms are filled with these stories.

Case reports can alert doctors to rare side effects, unusual symptoms, unexpected harms, or new disease mechanisms. Some major discoveries began with careful observation of one patient, one family, or one rare genetic variant.

In Alzheimer’s disease, rare families with inherited mutations helped scientists understand amyloid biology long before amyloid PET scans and blood biomarkers existed, or rare mutations that led to medications.

So the problem is not the case report.

The problem is using a case report to claim that a treatment works.

A case report can say:

“This happened.”

It cannot say:

“This works.”

To know whether something works, we need comparison groups, clear methods, fair outcome measurement, and replication.

This is especially important in cognitive disorders. Memory test scores can fluctuate. People can improve because they have taken the same test before, called a practice effect. Symptoms can improve when sleep, mood, medications, thyroid disease, vitamin B12 deficiency, or sleep apnea are treated. Others may be selected for publication because they improved, while non-responders are left out, and we never hear about the patient who did not improve.

This is also why we need to be fair about popular “Alzheimer’s reversal” programs.

Some advice in these programs is reasonable: exercise, better sleep, a healthier diet, treatment of sleep apnea, attention to vascular risk, hearing care, and checking for reversible causes of cognitive symptoms. These are not fringe ideas. Many dementia clinics already recommend them.

The concern is what gets added around them.

Helpful advice can be bundled with expensive supplements, large panels of blood tests, coaching programs, branded protocols, and a price tag that many families can barely afford. The scientific evidence for those added layers is often much weaker than the evidence for the basic lifestyle and medical care recommendations.

That distinction matters.

Exercise is not the same as an expensive supplement stack.
Sleep apnea treatment is not the same as a branded protocol.
Checking B12 or thyroid function is not the same as ordering unconventional tests on toxins or hormones without clear evidence that acting on them changes dementia outcomes.

A program can feel scientific because it includes lab tests, supplements, diet rules, coaching, and personalized reports. But complexity is not the same as evidence.

A common pattern is to market a protocol with words such as “prevent,” “reverse,” “restore,” or “end,” while the published evidence consists mainly of uncontrolled case reports or case series, or even limited trials. These reports may lack clear methods, inclusion criteria, dosing, duration, blinded outcome assessment, the correct sample size, outcome, complete reporting of non-responders, validated cognitive measures, or transparent financial disclosures.

Those omissions matter.

Without them, readers cannot know whether improvement reflects the intervention, natural fluctuation, practice effects, better sleep, treatment of another medical condition, extra attention, placebo response, or selective reporting.

There is no harm in participating in such programs as long as we are well-informed.

It is prudent to separate reasonable, low-risk health practices from expensive claims that have not yet been proven.

Conference abstracts have similar limitations.

A conference abstract is a summary presented at a scientific meeting. It may describe interesting preliminary findings, but it is not the same as a full peer-reviewed paper. Abstracts often do not provide enough detail to judge the methods, missing data, statistical plan, subgroup analyses, limitations, or conflicts of interest.

Peer review is not perfect. Flawed papers can pass peer review, and good papers can be rejected. Reviewers can miss errors. Journals can favor exciting findings. Scientists have biases. Science is a human process, and humans make mistakes.

But science also has tools for correction: replication, criticism, meta-analysis, reanalysis, corrections, and sometimes retractions.

My personal hero, Carl Sagan, captured this well: “Science is far from a perfect instrument of knowledge. It’s just the best we have.”

That is why we should not treat science as a single paper, a single p-value, or a single headline. Science is the process that allows wrong ideas to be tested, challenged, corrected, or replaced.

3. Observational Studies: Association Is Not Causation

Observational studies are essential.

They can follow large groups of people over time and identify patterns that would be impossible, unethical, or too expensive to test in a randomized trial. We cannot randomize people for decades to poverty, air pollution, poor education, diabetes, hypertension, poor access to healthy food, or social isolation.

But observational studies have limits.

They can show that two things occur together. They cannot automatically prove that one caused the other.

One problem is confounding by indication.

This means the reason someone receives a treatment is also related to the outcome being studied.

A simple example is insulin.

People who use insulin are often sicker than people with early or mild diabetes. They may have had diabetes longer, have higher blood sugar, or have more complications. If insulin users later have more kidney disease, heart disease, or dementia, we cannot conclude that insulin caused those problems.

More often, insulin use is a marker of more advanced diabetes.

The same problem can occur with statins.

People prescribed statins often have high LDL, diabetes, hypertension, obesity, vascular disease, inflammation, or prior cardiovascular events. Those same conditions also increase dementia risk. So if statin users later develop dementia at higher rates, the statin may not be the cause. The statin may identify people who were already at higher vascular and metabolic risk.

Another problem is reverse causation.

Reverse causation means the arrow points in the opposite direction from what we first assume. The early disease process may change the exposure, instead of the exposure causing the disease.

In Alzheimer’s disease, brain and metabolic changes can begin many years before dementia is diagnosed. During that period, people may lose weight, change eating patterns, become less active, start or stop medications, or show changes in cholesterol levels.

So if a study finds that people with lower cholesterol are more likely to develop dementia, one possibility is that low cholesterol contributed to dementia.

But another possibility is the reverse: early disease, frailty, weight loss, or inflammation lowered cholesterol before dementia was diagnosed.

The arrow may point in the opposite direction.

4. The Statin Example: When a Signal Spreads Faster Than the Correction

Statins are a good example of how scientific information can be distorted and how easy misinformation can become viral.

In 2012, the FDA added language to statin labels about reports of memory loss, forgetfulness, and confusion. These reports were largely based on patient complaints after statins were already in use.

Such reports matter. They can alert regulators and clinicians to possible side effects and demand clinical trials. But they are not proof of causation.

This does not mean patient symptoms are imaginary.

Some people do experience side effects from statins. Muscle symptoms, medication intolerance, and individual reactions or cognitive symptoms can be real. A large study may show no evidence of harm at the population level, but that does not mean every individual will tolerate the medication perfectly.

Population evidence answers one question: “Does this drug increase risk on average?”

A patient’s experience raises another question: “Is this drug right for this person?”

Both questions matter.

What population studies and randomized trials do not support is the stronger claim that statins cause Alzheimer’s disease or progressive dementia.

A warning label is not a randomized trial.
A complaint is not proof.
A reversible symptom report is not Alzheimer’s disease.

The warning was real, but it was heavily misused, and that may have misled many patients. Online, it was often presented as proof that statins cause dementia or Alzheimer’s disease.

A later conference abstract added fuel. It reported that among people with early mild cognitive impairment and lower baseline cholesterol, users of lipophilic statins had higher dementia conversion over eight years than non-users.

At first glance, this sounds alarming.

But this was not a randomized trial. It was an observational subgroup analysis presented as an abstract. Participants were not assigned to statins by chance. They were taking statins because clinicians had already judged them to have cardiovascular or lipid risk.

The analysis also focused on people with early mild cognitive impairment, exactly the group where reverse causation is a concern. Early Alzheimer’s biology may already be changing weight, diet, frailty, medical care, medication patterns, and cholesterol levels.

The careful conclusion is not:

“Statins cause dementia.”

The careful conclusion is:

“This subgroup analysis raised a concern that needed replication and better-controlled evidence.”

Later, more careful studies and meta-analyses did not replicate the claim that statins cause dementia. But the viral message had already spread on social media platforms.

This is how misinformation works: a signal spreads faster than the studies that later correct it.

5. Clinical Trials: Better Evidence, Still Not Perfect

Randomized clinical trials are a reasonable way to test whether a treatment works. Randomization makes groups more similar at the start. Blinding reduces expectation effects. A control group shows what might have happened without the treatment.

But even trials need careful reading.

A trial should tell us its primary outcome before it starts. This is the main question. Secondary outcomes are supportive. Exploratory outcomes and subgroup findings are clues, not proof.

This matters when the primary outcome is missed. For example, the ALZ-801/APOLLOE4 trial in APOE4 homozygotes did not meet its primary cognitive endpoint, although secondary and subgroup findings suggested possible signals in milder participants and imaging outcomes. Those signals may guide another trial, but they do not replace the missed primary endpoint.

Subgroups need the same caution. The BROADWAY obicetrapib study reported a promising p-tau217 biomarker signal, especially in APOE4 carriers. But this was a biomarker substudy, not proof that the drug prevents Alzheimer’s disease or slows cognitive decline. The right conclusion is: promising signal, needs a dedicated trial.

So when reading a trial, ask:

Was this the primary outcome?
Was the study large enough?
Was the finding replicated?
Was it a subgroup or an exploratory result?
Did the outcome matter to patients?

A signal is something worth testing.

A finding is something that survives testing.

6. P-Values, Effect Size, and the Word “Breakthrough”

A p-value tells us how surprising a result would be if there were truly no difference between two groups.

A small p-value can suggest that the result is unlikely to be due to chance alone. But a p-value does not tell us whether the effect is large, important, or meaningful to a patient.

That is why we also need the effect size.

An effect size tells us how big the difference is.

A treatment can have a statistically significant effect that is still very small. With enough people in a study, even a tiny difference can produce a low p-value.

One common effect size is Cohen’s d. It expresses the difference between two groups in standard deviation units.

A Cohen’s d of 1.0 is large.
A Cohen’s d of 0.5 is moderate.
A Cohen’s d of 0.2 is small.
A Cohen’s d of around 0.1 is very small.

There is a simple way to think about this.

If two groups are identical, and you randomly pick one person from each group, the chance that the person from the treatment group scores higher is about 50 in 100.

If Cohen’s d is 0.13, that chance rises to about 54 in 100.

That is not nothing.

But it is modest.

And modest effects need modest language.

One example to illustrate effect sizes and p-values is multidomain lifestyle trials such as FINGER and US POINTER, and the recent Anti-amyloid trials.

FINGER was an interesting experiment. It was an intensive multidomain intervention compared with a less intensive control condition that also included health advice. The intervention group did better, but the standardized effect size was small, with a reported Cohen’s d ≈ 0.13.

US POINTER (FINGER experiment in the US) also needs careful framing. It did not compare lifestyle intervention with doing nothing. It compared a structured, intensive lifestyle program with a self-guided lifestyle program. Both groups received lifestyle support. Both groups improved. The structured arm did somewhat better, but like FINGER, the effects were very modest (Cohen’s d ≈ 0.06).

Why interpreting this low effect size matters.

Participants in these trials were often motivated and highly educated. An expensive program that works modestly in a highly educated research population may not automatically translate to communities with limited access to healthy food, safe exercise spaces, preventive medical care, hearing care, sleep treatment, or social support, where cheaper “standard” programs can be good enough.

The same caution applies to anti-amyloid treatments.

Lecanemab and donanemab are scientifically important because they showed that removing amyloid can slow clinical decline in early Alzheimer’s disease. But the effect sizes are modest.

In CLARITY AD, lecanemab slowed decline on the Clinical Dementia Rating–Sum of Boxes, or CDR-SB, by 0.45 points over 18 months compared with placebo. The CDR-SB ranges from 0 to 18, with higher scores meaning worse impairment. In standardized terms, this is roughly a small effect size, around Cohen’s d = 0.18 to 0.20.

In TRAILBLAZER-ALZ 2, donanemab slowed decline by about 3.25 points on the Integrated Alzheimer’s Disease Rating Scale, or iADRS, in the primary analysis population. The iADRS ranges from 0 to 144, with lower scores meaning worse cognition and function. In standardized terms, this is also a small effect size, roughly Cohen’s d = 0.20 to 0.25. On CDR-SB, donanemab’s treatment difference was about 0.7 points, again a modest effect.

These numbers matter.

They show that anti-amyloid drugs are not “nothing.” They changed the trajectory of disease in carefully selected patients with early Alzheimer’s disease and confirmed amyloid pathology.

But they are not cures. They have not shown that they reverse Alzheimer’s disease yet. They modestly slow decline, while also carrying risks such as amyloid-related imaging abnormalities, infusion reactions, monitoring burden, cost, and access barriers.

This is the same lesson as the lifestyle trials.

A result can be real and still modest.

A result can be scientifically important and still require careful language.

Good science does not need exaggeration.

7. A Simple Checklist for Reading Health Claims

Conclusion: Ask for Better Evidence

The goal is not to be dismissive.

The goal is to ask better questions.

A patient's story can raise a question.
An observational study can show a pattern.
A clinical trial can test a treatment.
Replication tells us whether the finding holds up.

If the claim is big — “reverses Alzheimer’s,” “statins cause dementia,” “this supplement prevents disease,” or “this protocol ends cognitive decline” — the evidence needs to be big too.

Patients and families should not be told to stop asking questions.

They should be encouraged to ask sharper ones.

Challenge claims. Ask for the primary outcome. Ask about the effect size. Ask whether the finding was replicated. Ask whether someone is selling something.

Good science does not ask us to believe harder.

It asks us to test better. And embrace uncertainty over false claims.

References

1. Hellmuth J. Can we trust The End of Alzheimer’s? Lancet Neurology. 2020;19(5):389–390.

2. Ioannidis JPA. Why most published research findings are false. PLoS Medicine. 2005;2(8):e124.

3. Jager LR, Leek JT. An estimate of the science-wise false discovery rate and application to the top medical literature. Biostatistics. 2014;15(1):1–12.

4. U.S. Food and Drug Administration. FDA Drug Safety Communication: Important safety label changes to cholesterol-lowering statin drugs. 2012.

5. Swiger KJ, Manalac RJ, Blumenthal RS, Blaha MJ, Martin SS. Statins and cognition: a systematic review and meta-analysis of short- and long-term cognitive effects. Mayo Clinic Proceedings. 2013;88(11):1213–1221.

6. Journal of Nuclear Medicine. Statin use, cholesterol level, and dementia conversion in early mild cognitive impairment. J Nucl Med. 2021;62(Supplement 1):102.

7. Li G, Mayer CL, Morelli D, et al. Effect of simvastatin on CSF Alzheimer disease biomarkers in cognitively normal adults. Neurology. 2017;89:1251–1255.

8. Abushakra S, et al. Clinical efficacy, safety and imaging effects of oral valiltramiprosate in APOE4/4 homozygous individuals with early Alzheimer’s disease: the APOLLOE4 Phase III randomized controlled trial. CNS Drugs. 2025.

9. Davidson MH, Szarek M, Scheltens P, et al. Effect of obicetrapib, a potent cholesteryl ester transfer protein inhibitor, on p-tau217 levels in patients with cardiovascular disease. The Journal of Prevention of Alzheimer’s Disease. 2025. doi:10.1016/j.tjpad.2025.100394.

10. Ngandu T, Lehtisalo J, Solomon A, et al. A 2-year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people: the FINGER randomized controlled trial. Lancet. 2015;385(9984):2255–2263.

11. Baker LD, Snyder HM, et al. U.S. POINTER Study: multidomain lifestyle intervention in older adults at risk for cognitive decline. JAMA. 2025.

12. van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in early Alzheimer’s disease. New England Journal of Medicine. 2023;388:9–21.

13. Sims JR, Zimmer JA, Evans CD, et al. Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA. 2023;330(6):512–527.

14. Feynman RP. The Pleasure of Finding Things Out. Basic Books; 1999.

15. Sagan C. The Demon-Haunted World: Science as a Candle in the Dark. Random House; 1995.

A recent analysis from the BROADWAY trial reported that obicetrapib slowed the rise of plasma p-tau217 over 12 months in people with cardiovascular disease [1].

That finding deserves attention.

P-tau217 is one of the strongest blood biomarkers we have for Alzheimer’s disease biology. It often rises when amyloid and tau pathology are building in the brain. So when a drug changes p-tau217 in a randomized placebo-controlled study, the field should pay attention.

But the interpretation needs to stay proportional to the study design.

BROADWAY was designed as a cardiovascular lipid trial. The Alzheimer’s biomarker analysis was pre-specified, which makes it more meaningful than a purely post-hoc analysis. But it was still an analysis within a cohort built for another primary disease question.

That distinction matters.

A pre-specified biomarker signal can point us toward an important hypothesis. It does not, by itself, prove Alzheimer’s prevention, clinical benefit, or disease modification.

The question now is whether the signal reflects a real effect on Alzheimer’s biology, whether it is strongest in the right biological subgroup, and whether it eventually translates into cognition or function.

Why Obicetrapib Is Biologically Interesting

Obicetrapib blocks cholesterol ester transfer protein (CETP), a protein that helps transfer cholesterol between lipoprotein particles in the blood.

When CETP is inhibited, the lipid system is remodeled. LDL cholesterol falls. ApoB and non-HDL cholesterol fall. Lp(a) can fall. HDL cholesterol rises, and HDL particles may become larger or more functional.

That is the established biology.

This is also why obicetrapib worked in BROADWAY for its main lipid readouts. BROADWAY enrolled people with atherosclerotic cardiovascular disease (ASCVD) and/or heterozygous familial hypercholesterolemia whose LDL cholesterol remained inadequately controlled despite lipid-lowering therapy. In that population, obicetrapib lowered LDL-C, raised HDL-C, and lowered Lp(a) [1].

The Alzheimer’s question is different.

The question is whether this peripheral lipid remodeling can also influence Alzheimer’s-related biology.

There are several plausible pathways. The effect could be vascular, through lower LDL-C, ApoB, Lp(a), and improved vascular risk biology. It could be HDL-mediated, through larger or more functional HDL particles that support cholesterol efflux, antioxidant activity, and anti-inflammatory signaling. It could also involve amyloid handling, including a possible peripheral sink effect where remodeled HDL particles bind or help transport amyloid-related material in blood.

These mechanisms are plausible, but they are not proven.

The key point is that obicetrapib’s clearest biology is in the blood lipid system. The BROADWAY p-tau217 signal may reflect a downstream effect of that remodeling, but we do not yet know whether it is central, peripheral, vascular, HDL-mediated, or mixed.

What BROADWAY Tested

BROADWAY was a large cardiovascular lipid trial.

The parent trial evaluated obicetrapib in adults with established ASCVD, and/or heterozygous familial hypercholesterolemia. Participants were recruited from cardiology clinics and lipid specialty centers. They were already receiving maximally tolerated lipid-lowering therapy, but their LDL cholesterol remained inadequately controlled [1].

So BROADWAY did not enroll a general aging population. It enrolled people with residual lipid and cardiovascular risk.

The trial randomized 2,530 participants in a 2:1 ratio to obicetrapib 10 mg daily or placebo for 12 months. The parent trial’s main lipid readout was LDL-C lowering. At day 84, LDL-C decreased by about 30% with obicetrapib and increased slightly with placebo, producing a placebo-adjusted LDL-C difference of about 33%. Obicetrapib also increased HDL-C by more than 130% and lowered Lp(a) by about one-third [1].

The BROADWAY study materials summarize the key inclusion criteria as ASCVD or HeFH, LDL-C of at least 55 mg/dL with additional risk factors or LDL-C of at least 100 mg/dL, and maximally tolerated lipid-lowering therapy [1]. The parent trial endpoints included LDL-C at 12 weeks as the primary endpoint, with ApoB, Lp(a), non-HDL-C, and safety as key readouts [1].

That context is important.

The Alzheimer’s biomarker substudy did not arise from a general Alzheimer’s prevention cohort. It arose from a lipid-risk population, which may be exactly where a lipid-modifying drug is most likely to show a signal.

What the Biomarker Substudy Found

The Alzheimer’s biomarker analysis used stored blood samples from participants who had known APOE status and p-tau217 measured at baseline and after 12 months. The analysis included 1,535 of the 2,530 randomized BROADWAY participants, or 61% of the randomized population [1].

In the overall biomarker group, p-tau217 rose less in people who received obicetrapib than in people who received placebo. The signal appeared stronger in APOE4 carriers and strongest in APOE4/E4 participants.

Among APOE4/E4 participants, p-tau217 decreased with obicetrapib and increased with placebo, producing a placebo-adjusted difference of about 20%. Other biomarkers also moved in a favorable direction in APOE4/E4 participants, including GFAP and NfL [1].

This is biologically interesting. It supports the idea that lipid modulation may affect Alzheimer’s-related biomarkers, especially in people with APOE4.

But the subgroup details are important.

Of the 1,535 participants in the biomarker analysis, 1,045 were APOE3/E3, 338 were APOE3/E4, 20 were APOE2/E4, and only 29 were APOE4/E4 [1].

That means the most striking APOE4/E4 finding came from a small subgroup.

Small subgroups can reveal real biology, but they can also produce unstable estimates. A large effect in 29 people is worth following, but it requires replication before it can carry the weight of any clinical claim.

The cardiovascular enrichment also matters. In the BROADWAY biomarker subset, most APOE4 carriers had established ASCVD because the parent trial required ASCVD and/or HeFH. From the subgroup table, approximately 87% of APOE4 carriers in the biomarker substudy had ASCVD. About 81% had hypertension, and about 38% had diabetes. Among APOE4/E4 participants specifically, about 86% had ASCVD, about 83% had hypertension, and about 45% had diabetes [1].

That is not a typical APOE4 prevention cohort.

BROADWAY’s APOE4 carriers were genetically at risk, but they were also heavily enriched for vascular and metabolic risk. That combination may be central to why the signal appeared.

How to Read the BROADWAY Analysis Fairly

The BROADWAY biomarker analysis has real strengths. It came from a randomized, double-blind, placebo-controlled parent trial. The Alzheimer’s biomarker analysis was pre-specified. The overall biomarker sample was substantial. The biological rationale is plausible. And the main p-tau217 finding is worth following.

A fair reading also requires caution.

First, this was a biomarker substudy of a cardiovascular lipid trial, not a dedicated Alzheimer’s prevention trial. It did not test whether obicetrapib slows memory decline, prevents mild cognitive impairment, or prevents dementia.

Second, the substudy examined many outcomes: p-tau217 overall, APOE subgroups, APOE4/E4 participants, age subgroups, p-tau217/Aβ42:40, GFAP, NfL, p-tau181, lipid correlations, and obicetrapib blood-level correlations. These analyses are useful, but the paper states that p values were not adjusted for multiplicity [1]. That means the pre-specified main biomarker result deserves the most weight, while subgroup and secondary biomarker findings are best treated as hypothesis-generating.

Third, a subgroup pattern is not the same as a formal interaction analysis. If the treatment effect looks larger in APOE4 carriers, that does not automatically prove that APOE genotype modifies treatment response. That requires a treatment-by-APOE interaction analysis. Without that, the careful statement is that the signal appeared stronger in APOE4 carriers, not that APOE4 has been proven to modify response.

Finally, baseline p-tau217 may matter. APOE4/E4 participants had higher p-tau217 at baseline, and the paper reported that participants with higher baseline p-tau217 appeared to show greater treatment effects [1]. That raises an important question: was the stronger APOE4/E4 signal due to APOE4 biology, higher baseline p-tau217, lipid risk, ASCVD, or some combination of all these factors?

That is the key precision-prevention question.

The best target population may not be APOE4 alone. It may be APOE4 plus elevated p-tau217 plus lipid dysfunction. Or it may be elevated Alzheimer’s biomarkers plus measurable lipid dysfunction, even without APOE4.

The Statin Lesson: LDL May Define the Responder

Statins provide a useful lesson for obicetrapib.

In a 12-month randomized, placebo-controlled trial, Li and colleagues tested simvastatin in cognitively normal adults. Participants were 45 to 64 years old, statin-naive, cognitively normal, and had normal or mildly elevated cholesterol. Forty-nine people were randomized, and 46 completed the study per protocol: 25 in the simvastatin group and 21 in the placebo group. Simvastatin was titrated to 40 mg per day. The primary outcomes were CSF Aβ42, total tau, and p-tau181 measured at baseline and after 12 months [2].

Overall, simvastatin did not significantly change CSF Aβ42, total tau, or p-tau181 compared with placebo. It also did not produce significant differences in neuropsychological test changes.

But the most informative result may have been hidden in the biology of the participants. Baseline LDL modified the treatment effect on CSF p-tau181. People with higher baseline LDL appeared to have greater reductions in CSF p-tau181 on simvastatin. The authors concluded that simvastatin-related reductions in CSF p-tau181 may be modulated by LDL cholesterol and merit further study in people with hypercholesterolemia [2].

That is directly relevant to obicetrapib.

A lipid-lowering drug may show its clearest brain biomarker effect in people whose lipid biology is abnormal. If a trial includes many people without elevated LDL, ApoB, Lp(a), or other lipid risk, the average effect may look smaller than it really is in the subgroup most likely to respond.

The lesson is not that statins prevent Alzheimer’s disease.

The lesson is that the responder population matters.

For obicetrapib, the key question may not be whether it works in all people with APOE4 or all people with elevated p-tau217. The more precise question is whether it works best in people with Alzheimer’s biology plus measurable lipid dysfunction.

The EVOKE Lesson: Biomarkers Need the Right Biology and the Right Outcome

EVOKE and EVOKE+ provide a second lesson.

These were large phase 3 trials of oral semaglutide in early Alzheimer’s disease, enrolling a combined 3,808 adults with early-stage Alzheimer’s symptoms and following participants for about two years [4].

The trials were important because GLP-1 receptor agonists have strong biological rationale. They improve metabolic health, insulin resistance, inflammation, vascular risk, and weight — pathways that may matter for brain aging.

The biomarker readout was encouraging. Semaglutide reduced several CSF biomarkers compared with placebo, including p-tau217 and other Alzheimer’s-related markers.

But the clinical readout was negative: semaglutide did not significantly slow disease progression compared with placebo [4].

This matters for two reasons.

First, it shows that moving Alzheimer’s biomarkers does not guarantee cognitive benefit.

Second, EVOKE was an Alzheimer’s disease trial, but it was not primarily designed around the metabolic phenotype where GLP-1 receptor agonists are most powerful. GLP-1 drugs often work best in people with obesity, insulin resistance, diabetes, cardiometabolic disease, or related inflammatory-metabolic risk. If those features are not central to trial enrichment, the average effect may dilute the benefit in the subgroup most likely to respond.

That is another lesson for obicetrapib.

Obicetrapib is a lipid-modifying drug. Its best test may require more than Alzheimer’s biomarkers or APOE genotype. It may require identifying the people whose Alzheimer’s risk is tied to lipid or vascular dysfunction.

In other words, the question is not only whether the drug can move p-tau217.

The question is whether the trial has selected the population most likely to benefit from the way the drug works.

The Mechanism Is Still Uncertain

There are several possible explanations for the BROADWAY p-tau217 signal.

It could be a central effect, meaning obicetrapib changes processes inside the brain. But we do not yet know whether obicetrapib enters the human brain in meaningful amounts.

It could be a vascular effect. BROADWAY participants had high cardiovascular risk, and obicetrapib improved lipid markers linked to vascular disease.

It could be an HDL-mediated peripheral effect. CETP inhibition may create larger or more functional HDL particles that improve cholesterol efflux, antioxidant defense, and inflammatory signaling.

It could also involve a peripheral sink effect, where remodeled HDL particles bind or transport amyloid-related proteins in blood, changing plasma biomarker dynamics without proving that tau pathology inside the brain has changed.

These possibilities are not mutually exclusive.

The next studies need to connect plasma p-tau217 to CSF biomarkers, amyloid PET, tau PET, vascular markers, HDL function, drug exposure, and cognition.

The key question is not only whether obicetrapib lowers p-tau217.

It is how it lowers p-tau217 — and whether that mechanism leads to clinical benefit.

APOE4, Lipids, and Trial Design

APOE4 makes obicetrapib interesting because APOE is a lipid transport gene.

In the brain, APOE helps move cholesterol and phospholipids between astrocytes, neurons, microglia, and other cells. These lipids are needed for synapses, membrane repair, myelin, and normal cellular signaling. APOE4 appears less efficient than APOE3 in several aspects of lipid handling and repair.

So it is plausible that some APOE4-related brain risk is mediated through lipid biology.

But APOE4 is not the same as a lipid phenotype.

Population studies show that APOE4 carriers, on average, have higher total cholesterol and LDL cholesterol than APOE3 carriers [6]. But averages do not define individuals. The percentage of APOE4 carriers with elevated LDL, low HDL, high ApoB, high Lp(a), metabolic syndrome, or ASCVD depends on the population being studied.

That matters for obicetrapib.

A trial enriched for APOE4, p-tau217, or GFAP may identify people at Alzheimer’s risk, but it may not identify the people most likely to respond to a lipid-modifying drug.

For obicetrapib, the most informative population may be people with both Alzheimer’s biology and lipid biology: APOE4 plus elevated p-tau217, high LDL-C or ApoB, elevated Lp(a), low HDL function, ASCVD, metabolic syndrome, or combinations of these factors.

The trial design should match how the drug works.

SPINOZA is coming next

The name SPINOZA also carries a useful meaning. Baruch Spinoza was a 17th-century Dutch philosopher whose work helped shape Western ideas about reason, nature, freedom, and the unity of mind and body. That connection is fitting here: the trial asks whether changing biology outside the brain — lipid metabolism in the blood and vasculature — might influence Alzheimer’s biology inside or around the brain.

SPINOZA is a phase 2b randomized, double-blind, placebo-controlled trial testing obicetrapib in people with preclinical Alzheimer’s disease. Unlike BROADWAY, it is designed around Alzheimer’s prevention rather than cardiovascular lipid lowering [5].

The primary endpoint is change in plasma p-tau217 at week 52. Secondary endpoints include Aβ42/40, p-tau217/Aβ42, p-tau217/Aβ42/40, p-tau181, and brain-derived tau. Cognition is exploratory and measured with the Preclinical Alzheimer Cognitive Composite 5, or PACC5 [5].

SPINOZA plans to enroll up to 400 participants, including at least 100 APOE4/E4 participants, plus 125–150 APOE3/E4 and 125–150 APOE3/E3 participants [5].

That is an important design. It prospectively tests the signal in a preclinical Alzheimer’s population and intentionally includes a larger APOE4/E4 group.

The key point is not to criticize SPINOZA before it starts. The key point is that its most informative analyses may depend on careful phenotyping.

Since obicetrapib is a lipid-modifying drug, the trial could be especially informative if it reports and analyzes LDL-C, ApoB, Lp(a), HDL-C, triglycerides, metabolic syndrome, diabetes, ASCVD, statin use, PCSK9 inhibitor use, baseline p-tau217, GFAP, and drug exposure.

Not all APOE4 carriers have high LDL, low HDL, high ApoB, high Lp(a), metabolic syndrome, or ASCVD. If many participants have preclinical Alzheimer’s biology but relatively normal lipid biology, the average treatment effect could be diluted.

That does not make the trial weak. It makes subgroup and interaction analyses essential.

For a lipid drug, the most informative trial design considers both Alzheimer’s biology and lipid biology.

Conclusion: From Signal to Reproducible Science

BROADWAY produced a promising p-tau217 signal. The signal appeared strongest in APOE4 carriers, especially APOE4/E4 participants. The finding is biologically plausible and deserves prospective testing.

But it should not be overstated.

BROADWAY does not prove that obicetrapib prevents Alzheimer’s disease. It does not prove that obicetrapib slows memory decline. It does not prove that APOE4 carriers, as a broad group, benefit more. And it does not tell us whether the p-tau217 effect is central, peripheral, or both.

What BROADWAY does is point to a serious hypothesis: lipid biology may be a modifiable pathway in Alzheimer’s prevention, especially in people with APOE4, vascular risk, lipid dysfunction, and elevated Alzheimer’s biomarkers.

That is an important hypothesis.

Now it needs reproducible testing.

The next studies could select participants based on biology, test APOE interactions directly, test lipid interactions directly, measure biomarkers carefully, include cognitive outcomes, follow people long enough, and avoid turning subgroup findings into clinical claims before they are reproduced.

The main lesson is that Alzheimer’s prevention trials need to match the drug to the biology it is designed to modify.

For obicetrapib, that biology is lipid metabolism.

The right conclusion is not:

“Obicetrapib prevents Alzheimer’s disease.”

The right conclusion is:

“Obicetrapib produced a promising lipid-biology signal that now needs rigorous prospective testing in the right biological population.”

That is how reproducible prevention science can move forward.

Take-Home Messages

• Trial context matters. BROADWAY was a cardiovascular lipid trial, not a dedicated Alzheimer’s prevention trial.

• Mechanism matters. Obicetrapib’s clearest biology is peripheral lipid remodeling.

• Subgroups need discipline. The APOE4/E4 signal is promising, but the subgroup was small and needs interaction testing and replication.

• Biomarkers are not cognition. EVOKE and EVOKE+ showed that p-tau217 and other CSF biomarkers can improve without clinical benefit.

• APOE4 is not a lipid phenotype. The most informative population may be people with Alzheimer’s biology plus measurable lipid or vascular dysfunction.

• The mechanism remains open. The p-tau217 effect could be central, vascular, peripheral, HDL-mediated, or mixed.

• Precision prevention means matching the drug to the biology and the responder population. Obicetrapib acts on lipid metabolism, so the most informative trials may be those that identify people whose Alzheimer’s risk is tied to lipid or vascular dysfunction, not only APOE genotype or p-tau status.

Relevant Disclosure:

I sit on the advisory board of New Amsterdam Pharma, which is developing Obicetrapib

References

1. Davidson MH, Szarek M, Scheltens P, et al. Effect of obicetrapib, a potent cholesteryl ester transfer protein inhibitor, on p-tau217 levels in patients with cardiovascular disease. Journal of Prevention of Alzheimer’s Disease. 2025. doi:10.1016/j.tjpad.2025.100394.

2. Li G, Mayer CL, Morelli D, et al. Effect of simvastatin on CSF Alzheimer disease biomarkers in cognitively normal adults. Neurology. 2017;89:1251–1255.

3. Corcoran E, Kettlety M, Mogul U, Azah JN, Cork SC. The effects of GLP-1 receptor agonists on Alzheimer’s pathophysiology: a systematic review. Molecular and Cellular Neuroscience. 2026;137:104091.

4. Novo Nordisk. Headline results from the EVOKE and EVOKE+ phase 3 trials of oral semaglutide in early Alzheimer’s disease. Reported 2025.

5. SPINOZA Study. A Phase 2b, randomized, double-blind, placebo-controlled, parallel group study to assess the efficacy and safety of obicetrapib in participants with preclinical Alzheimer’s disease. Protocol Synopsis Version 1.0, 01 Apr 2026.

6. Bennet AM, Di Angelantonio E, Ye Z, et al. Association of apolipoprotein E genotypes with lipid levels and coronary risk. JAMA. 2007;298(11):1300–1311.

7. Alberti KGMM, Eckel RH, Grundy SM, et al. Harmonizing the metabolic syndrome: a joint interim statement of the International Diabetes Federation Task Force on Epidemiology and Prevention; National Heart, Lung, and Blood Institute; American Heart Association; World Heart Federation; International Atherosclerosis Society; and International Association for the Study of Obesity. Circulation. 2009;120(16):1640–1645.

Before discussing our recent trial, I want to thank the hundreds of research participants who made PreventE4 possible. This trial asked a lot of people: screening visits, blood draws, cognitive testing, MRI scans, study capsules, follow-up visits, and, for many, lumbar punctures twice. Prevention trials depend on people who are willing to contribute their time, often for the benefit of future generations. We are deeply grateful.

Should You Take DHA Supplements?

Many readers will come to this question first: Should I take a DHA or fish oil supplement for brain health?

PreventE4 does not give a simple yes or no answer.

It was a two-year prevention trial in cognitively healthy older adults at risk for dementia. That matters. Two years may be too short to detect whether a nutrient changes the long course of Alzheimer’s disease, especially before memory symptoms begin.

Whether someone chooses to take an omega-3 supplement is still a personal decision. We hope that the information gained from this trial will help you make that decision.

But the goal of PreventE4 was not only to answer whether people at risk of dementia should take DHA.

The deeper goal was to gain a better understanding of DHA biology and Alzheimer’s risk.

Why We Did PreventE4

DHA is an omega-3 fat that the brain needs.

It helps build brain cell membranes. It supports synapses, the connections between brain cells. It also helps regulate inflammation.

For years, studies have suggested that people with higher omega-3 levels tend to have better brain health.

From 1994 to 2008, my friend and mentor, Helena Chui, led the Aging Brain Study at USC. I joined that team later. In the Aging Brain Study, we found that cognitively healthy older adults with lower blood DHA had more amyloid on brain scans, smaller memory-related brain regions, and worse memory performance [1].

That finding was important, but it also showed the limits of observational studies.

When we see low DHA in the blood, it can mean at least two things.

It may mean a person is eating too little DHA, usually from fatty fish or other omega-3 sources.

Or it may mean the body or brain is breaking DHA down faster — what scientists call catabolism.

Both can also be true at the same time.

That distinction matters because a blood DHA level is not just a measure of intake. It may also be a clue about how the body and brain are handling DHA.

To answer this question we needed to do a DHA supplementation trial.

Why APOE4 Matters

APOE4 is the strongest common genetic risk factor for late-onset Alzheimer’s disease.

APOE is involved in moving lipids, or fats, around the brain. Since DHA is a brain lipid, we wondered whether APOE4 changes how DHA gets into the brain or how the brain uses it.

Earlier work suggested that APOE4 and Alzheimer’s disease stage may affect DHA delivery to the brain [2,3].

We conducted PET imaging with radiolabeled DHA to find more clues. Cognitively healthy APOE4 carriers showed higher DHA incorporation into gray matter, especially in the entorhinal cortex, a brain region affected early in Alzheimer’s disease [4].

One interpretation is that the APOE4 brain may be pulling in more DHA because it is using it faster, losing it faster, or trying to compensate for stress.

A later high-dose DHA pilot trial measured DHA in cerebrospinal fluid, or CSF, the fluid that surrounds the brain. DHA supplementation increased CSF DHA, but the increase was modest, and APOE4 carriers showed signals of lower omega-3 delivery than non-carriers [5].

Together, these studies led to PreventE4.

The question was simple: if we give enough DHA, can we get it into the central nervous system in people at risk, especially APOE4 carriers, before dementia begins?

Does the Form of DHA Matter?

Another common question is whether the form of DHA matters.

Some people argue that triglyceride-based DHA is the wrong form for brain delivery, and that phospholipid DHA would be superior. This is an important scientific question. Different DHA forms may behave differently.

But it is not correct to say that triglyceride DHA cannot enter phospholipid pools.

In our earlier pilot work, participants took triglyceride-based algal DHA. After supplementation, DHA increased not only in triglycerides, but also in phosphatidylcholine and cholesteryl ester lipid pools in both plasma and CSF [6]. These lipid pools appear to exchange with one another.

So the form of DHA may matter, but triglyceride DHA is not biologically isolated from phospholipid DHA. And they both reach the brain.

The Omega-3 Index

One practical way to measure omega-3 status is the omega-3 index.

This is the percentage of EPA plus DHA in red blood cell membranes. It is useful because red blood cells give a longer-term picture of omega-3 status than a single diet questionnaire.

Some brain health clinics now use the omega-3 index as part of a broader prevention workup [8], with a low or suboptimal index defined as < 6 and an optimal index>10. But these cut-offs are not validated in intervention trials.

The omega-3 index is not an Alzheimer’s test. A low number does not mean someone will develop dementia. And raising the number with supplements has not been proven to prevent cognitive decline.

It is best understood as one piece of a larger picture: diet, exercise, APOE genotype, vascular risk, inflammation, and brain biomarkers.

The Bigger Question: How Does the Brain Use DHA?

PreventE4 was designed to answer the delivery question and its implications on AD risk: once DHA reaches the brain, do we see signs of decreased AD risk?

This matters especially for APOE4.

In Alzheimer’s disease brain tissue, we see signs that polyunsaturated fatty acids are being broken down and remodeled in an inflammatory environment. This includes changes in omega-3 and omega-6 balance, lower DHA-related resolving mediators, and activation of inflammatory lipid pathways [9].

One enzyme that may be involved is cPLA2. This enzyme cuts fatty acids out of cell membranes. When it is overactive, it can release arachidonic acid and drive inflammatory lipid signaling.

In APOE4 and Alzheimer’s disease, we found that cPLA2 appears to be more active [9,14].

Put simply, the APOE4 brain may not only need more DHA delivered to it. It may also need help using DHA in the right way.

This is an important shift in thinking.

The question is not only: “How much omega-3 do you eat?”

It is also: “How does your brain metabolize omega-3s?”

What PreventE4 Was Designed to Test

I am deeply grateful to Lon Schneider for helping me design this trial. We had intense discussions on who we should recruit, what the outcomes should be, and how we interpret these findings. Wendy Mack was instrumental in planning and estimating sample sizes and analysis plans.

PreventE4 was a double-blind, placebo-controlled prevention trial in cognitively healthy older adults between ages 55 and 80 [10].

Participants had to have at least one dementia-related risk factor. These included obesity, high blood pressure, high cholesterol, low physical activity, or fewer than 12 years of education. They also had to have preserved daily function and no clinical dementia.

The trial intentionally looked for people with limited omega-3 intake. Participants were excluded if they were taking omega-3 supplements or consuming more than 200 mg per day of DHA [10].

In plain language, we were trying to find people who did not already eat much fish, had dementia risk factors, but did not yet have dementia.

That sounds straightforward.

It was not.

What It Took to Run the Trial

PreventE4 was conceived in 2016. Funding came in 2017 from NIH, two years after the application was submitted. This was followed by funding from ADDF. Recruitment began in May 2018.

The plan sounded simple: recruit cognitively healthy older adults with dementia risk factors, low omega-3 intake, and enough APOE4 carriers to test whether genotype changed DHA delivery.

In practice, it was enormously difficult.

We had to screen thousands of people to find the right participants. They had to be older adults who did not like to eat much fish, had dementia risk factors, were cognitively healthy, were willing to be followed for two years, and, in many cases, were willing to undergo a lumbar puncture.

This was especially challenging for APOE4 homozygotes, who are much less common than non-carriers or people with only one APOE4 copy.

The trial required constant outreach, prescreening, saliva collection, genotyping, phone calls, follow-up visits, and reassurance. Staff members worked tirelessly to call participants, answer questions, mail saliva kits, receive tubes, organize samples, and prepare batches for genotyping.

I remember the mechanics of it vividly.

Once a week, the genotyping workflow had to keep moving. I remember visiting the lab and finding packages dropped off by FedEx, sometimes many of them, waiting to be processed. On Fridays, I would pick up those packages and help deliver them to the genotyping team.

It was not glamorous work. It was not the part of science that people usually see. But without those tubes, labels, calls, packages, and staff members, there would be no PreventE4.

Then COVID changed everything.

Recruitment slowed. In-person visits became harder. Some participants dropped out. Safety rules changed how we could see people, collect samples, and keep the study moving. In 2021, the sample size had to be increased to account for the unexpected dropouts.

What began as a carefully powered prevention trial became a test of persistence.

Recruitment eventually ended in May 2024.

The Intervention

Participants were randomized to either 2 grams per day of DHA or an identical placebo for two years [10].

The DHA was algal-derived, which allowed a high, standardized dose.

Both the DHA and placebo groups also received a high-dose vitamin B complex. This was intentional. B-vitamin status and homocysteine metabolism can affect brain aging and may interact with omega-3 biology. By giving vitamin B supplementation to both groups, the trial reduced the chance that low B-vitamin status would confound the DHA results.

The primary outcome was measured at six months in the lumbar puncture subset. The question was whether DHA supplementation changed the ratio of DHA to arachidonic acid, or DHA/AA, in CSF [10].

CSF is not the brain itself. But it is one of the closest fluids we can safely measure in living people.

Secondary outcomes were measured over two years and focused on the brain. These included resting-state functional MRI, diffusion tensor imaging, and volumetric MRI [10]. In simpler terms, the trial asked whether DHA affected brain connectivity, white matter structure, hippocampal volume, and other imaging markers.

Exploratory outcomes included cognition, measured with the Repeatable Battery for the Assessment of Neuropsychological Status, or RBANS, along with blood biomarkers, inflammatory lipids, oxylipins, amyloid-related markers, phosphorylated tau, and microbiome sampling [10].

What We Found

The omega-3 index helped confirm that we recruited the right population and that the intervention did what it was supposed to do.

Testing showed that omega-3 levels in red blood cells climbed dramatically, from 4.9% to 11%. A 6% increase (shown in figure below)

DHA levels in CSF, the fluid that surrounds the brain, rose by an average of 17% after six months [13].

This confirmed that DHA reached its intended target.

The same increases were seen in people with an APOE4 gene, the group thought most likely to benefit from supplementation [13].

That finding matters.

At the level of the blood, people responded to DHA. The supplement was taken, absorbed, and incorporated into red blood cell membranes. At the level of CSF, DHA also increased.

So the simplest explanation is not that the dose failed to reach the body or the central nervous system.

But over two years, DHA supplementation did not improve cognition or slow hippocampal atrophy.

As you can see, there is a complete overlap between the treatment and placebo arms on cognition. No difference!

APOE4 carriers did show signs of greater vulnerability. They started slightly higher in the begining and learned less on the RBANS cognitive battery toward the end (Figure below). Something similar happened with hippocampal volumes. That finding is consistent with what we know about APOE4’s effect on cognition and hippocampal volume during preclinical (predementia) phases.

But being in the DHA treatment arm did not change that pattern.

This point is important.

Prevention trials in cognitively healthy people are hard because ceiling effects can make it difficult to detect change over only two years. But RBANS is a sensitive cognitive tool, and PreventE4 was a well-designed randomized trial. So while ceiling effects may have limited our ability to detect a small signal, we also have to be clear: we did not see a treatment signal on cognition or hippocampal volume.

Raising the omega-3 index to >10 did not affect AD biomarkers in 2 years. This is a real limitation to relying on the omega-3 index as an AD diagnostic.

That is the central result of PreventE4.

The intervention worked biologically. It changed omega-3 levels in blood and CSF. But it did not produce a measurable cognitive or hippocampal-volume benefit over two years in cognitively healthy older adults.

Why Prevention Trials Are So Hard

The first thing to remember is that PreventE4 was a prevention trial conducted in people without dementia.

That is both the strength and the challenge.

When people are cognitively normal at the start of a study, there may be very little room to improve on standard memory tests over only two years. This is called a ceiling effect. And participants do better on repeated testing. This is known as practice or learning effects. If a person is already doing well, a test may not be sensitive enough to show whether the brain is being protected.

Alzheimer’s disease also does not always progress in a straight line.

For years, changes may be slow and difficult to detect. Then, as someone approaches mild cognitive impairment or dementia, decline can accelerate.

In other words, there may be two slopes: a slow early slope and a much steeper later slope (Figure above).

A two-year study in cognitively healthy people may happen during the slow slope, before standard cognitive tests or hippocampal volume show large treatment effects. But waiting until the steep slope begins may be too late, because neurodegeneration may be harder to reverse once neurons and synapses are already being lost.

That is the central dilemma of prevention research.

The best time to intervene may be before we can easily measure clinical decline.

Still, this limitation does not erase the finding. PreventE4 used sensitive cognitive testing, MRI measures, biomarkers, and a randomized placebo-controlled design, and APOE4 carriers had a signal that they learned less and had worse hippocampal volume changes. The study showed clear biological target engagement, but it did not show that DHA changed the cognitive or hippocampal-volume trajectory over two years.

Why We Tested One High Dose

One design decision is worth explaining.

We chose to test a high dose of DHA rather than several different doses.

That was intentional.

Since one of the central questions was whether APOE4 modifies DHA delivery and response, we chose a design that gave us the best chance of detecting that interaction.

With this dose, the biological signal was convincing. DHA increased in blood, and DHA increased in CSF, but we did not see an interaction with APOE4.

The effect of APOE4 on DHA brain transport is stronger in patients with mild cognitive impairment or dementia.

So the problem was not simply an inadequate dose or failed delivery at this clinical stage.

The harder question is what CSF DHA actually tells us.

CSF is not the same as a neuron. It is not a direct biopsy of a brain cell membrane. We cannot say with certainty that a rise in CSF DHA means the same rise occurred inside neurons or synapses.

But CSF is one of the best windows we have into the living human brain. It surrounds the brain and spinal cord, and changes in CSF often reflect changes in central nervous system biology. So while CSF DHA is not a perfect measure of neuronal DHA, it is a reasonable marker of brain delivery.

What This Means for Omega-3 Supplements

PreventE4 gives us a clearer way to think about omega-3s and brain health.

The message is not that omega-3s are unimportant. DHA is essential for the brain. It is part of neuronal membranes, synapses, and inflammatory signaling.

The message is also not that every person should or should not take a supplement. That decision is personal.

The more important lesson is that omega-3 supplementation alone is not a blunt solution for Alzheimer’s prevention.

A capsule cannot fully replace the biological context in which omega-3s seem to be most helpful.

In Mediterranean-style dietary patterns, omega-3s come with a broader lifestyle: fatty fish, vegetables, legumes, nuts, olive oil, physical activity, social connection, and lower cardiometabolic risk.

Whole foods bring more than one nutrient. Salmon contains DHA and EPA, but also protein, vitamin D, vitamin B12, selenium, potassium, and other nutrients. Walnuts and seeds provide plant-based omega-3s along with fiber, minerals, and polyphenols.

That context may matter as much as the omega-3 number itself.

It is plausible that a healthy lifestyle does more than raise omega-3 levels. It may also create the biology needed to use DHA effectively.

A person who eats fatty fish, exercises, sleeps well, has good vascular health, and has lower chronic inflammation may not only have higher omega-3 levels. They may also have a brain environment that transports, retains, remodels, and uses DHA better.

That is different from taking a capsule in the setting of poor sleep, high stress, sedentary behavior, obesity, insulin resistance, and an ultraprocessed diet.

In that setting, DHA may reach the blood and even the central nervous system, but the broader biology may not be ready to use it effectively.

This is the key shift.

Omega-3s are not magic. They are part of a biological system. And a quick fix with supplements may not work.

Where We Go Next

PreventE4 did not end the omega-3 story. It sharpened the next question.

The problem may not simply be getting DHA into the body or even into CSF during preclinical AD. The problem may be whether the brain can retain, remodel, and use omega-3s in the right way.

This is why we think the next generation of research needs to focus on DHA metabolism.

How does inflammation change brain DHA use? What is the role of gut dysbiosis?

And can we change the brain environment so omega-3s work better?

This is where lifestyle becomes important.

A healthy lifestyle may not only raise omega-3 levels. It may also create the biology needed to use DHA effectively. A person who eats fatty fish, exercises, sleeps well, has good vascular health, and has lower chronic inflammation may have a brain environment that transports, retains, remodels, and uses DHA better.

That may be more meaningful than any single omega-3 index number, supplement label, or blood level.

This is also where our work on brain-penetrant cPLA2 inhibitors comes in.

cPLA2 is an enzyme that cuts polyunsaturated fatty acids out of cell membranes. When it is overactive, it can release both arachidonic acid and DHA and drive unresolved inflammation. In APOE4 and Alzheimer’s disease, cPLA2 appears to be more active. That may push the brain toward a state where polyunsaturated fatty acids are being broken down, remodeled, or consumed faster than they can be used for healthy membrane repair and anti-inflammatory signaling [9,14].

In simple terms, giving DHA may not be enough if the brain is stuck in a lipid-inflammatory state that keeps burning through these fats.

Our future direction is to target that biology directly.

We are developing brain-penetrant cPLA2 inhibitors designed to reduce harmful inflammatory lipid signaling and help restore a healthier omega-6 to omega-3 balance in the brain. The goal is not to replace omega-3 nutrition. The goal is to change the brain environment so omega-3s can be used more effectively [14].

That is still early-stage work. It is not yet a human Alzheimer’s treatment.

That is where nutrition, lipid biology, neuroinflammation, lifestyle, and drug discovery now come together.

Take-Home Messages

• PreventE4 does not give a simple yes or no answer about whether an individual should take DHA supplements.

• The trial showed that high-dose DHA reached its biological targets: omega-3 levels in red blood cells rose from 4.9% to 11%, and CSF DHA rose by about 17%.

• The same biological response was seen in APOE4 carriers.

• DHA did not improve cognition or slow hippocampal atrophy over two years.

• APOE4 carriers learned less on RBANS and had greater hippocampal atrophy, but DHA treatment did not change that pattern.

• That does not mean omega-3 biology is unimportant. It means delivery alone may not be enough.

• The next question is DHA metabolism: how the brain transports, retains, remodels, and uses omega-3s.

• A holistic lifestyle may help create the biology needed to use DHA effectively.

• Future prevention may require combining nutrition, lifestyle, biomarkers, and therapies that target inflammatory lipid metabolism, including cPLA2.

Publication

Yassine HN, Ghasem Pour S, Juarez M, Arellanes IC, Ali N, Dikeman D, Sanchez A, Park J, Kerman B, Duro MV, Asante I, Louie S, Kono N, D’Orazio L, Chui H, Mack WJ, Harrington MG, Braskie MN, Schneider LS.
CNS Target Engagement of High-dose DHA Supplementation in Older Adults at Risk for Dementia: A Randomised, Double-blind, Placebo-controlled Trial.

Acknowledgments

PreventE4 was a team effort. Hussein N. Yassine designed and supervised the trial, obtained funding, and wrote the manuscript. Sara Ghasem Pour helped with data analysis and manuscript drafting. Nada Ali assisted with manuscript drafting. Marlene Juarez helped with cognitive testing. Isabella C. Arellanes and Dante Dikeman coordinated study visits. Ashley Sanchez assisted with blood sample collection and storage. Bilal Kerman, Marlon V. Duro, and Isaac Asante assisted with biomarker measurements. Stan Louie supervised biomarker management. Jackson Park assisted with study coordination. Naoko Kono led data management. Lina D’Orazio supervised cognitive testing. Helena Chui helped with study design. Wendy J. Mack supervised study design and all data analysis. Michael G. Harrington performed lumbar punctures. Meredith N. Braskie oversaw the imaging biomarkers. Lon S. Schneider supervised study design, grant applications, and manuscript writing. Naoko Kono and Wendy J. Mack accessed and verified the underlying data. All authors read and approved the final version of the manuscript.

This study was funded in part by the National Institutes of Health/National Institute on Aging, including RF1AG076124, R01AG055770, R01AG067063, R01AG054434, R21AG056518, and P30AG066530 to Hussein N. Yassine; the Alzheimer’s Drug Discovery Foundation, GC-201711–2014197 to Hussein N. Yassine and Lon S Schneider; and donations from the Vranos and Tiny Foundations and Ms. Lynne Nauss to Hussein N. Yassine. Wendy J. Mack, Meredith N. Braskie, and Lon S. Schneider also received funding from NIH R01AG054434 and the Alzheimer’s Drug Discovery Foundation. We also thank the members of the Data and Safety Monitoring Board for their oversight and guidance.

References

1. Yassine HN, Feng Q, Azizkhanian I, et al. Association of serum docosahexaenoic acid with cerebral amyloidosis. JAMA Neurology. 2016.

2. Yassine HN, Rawat V, Mack WJ, et al. The effect of APOE genotype on the delivery of DHA to cerebrospinal fluid in Alzheimer’s disease. Alzheimer’s Research & Therapy. 2016;8:25.

3. Yassine HN, Braskie MN, Mack WJ, et al. Association of docosahexaenoic acid supplementation with Alzheimer disease stage in apolipoprotein E ε4 carriers: a review. JAMA Neurology. 2017;74(3):339–347.

4. Yassine HN, Croteau E, Rawat V, et al. DHA brain uptake and APOE4 status: a PET study with [1-¹¹C]-DHA. Alzheimer’s Research & Therapy. 2017;9:23.

5. Arellanes IC, Choe N, Solomon V, et al. Brain delivery of supplemental docosahexaenoic acid (DHA): a randomized placebo-controlled clinical trial. EBioMedicine. 2020;59:102883.

6. Bantugan MA, Xian H, Solomon V, et al. Associations of ApoE4 status and DHA supplementation on plasma and CSF lipid profiles and entorhinal cortex thickness. Journal of Lipid Research. 2023;64(6):100354. doi:10.1016/j.jlr.2023.100354.

7. Yassine HN, Samieri C, Livingston G, et al. Nutrition state of science and dementia prevention: recommendations of the Nutrition for Dementia Prevention Working Group. The Lancet Healthy Longevity. 2022;3:e501–e512.

8. Yassine HN. The omega-3 index in Alzheimer’s disease: ready for prime time? American Journal of Clinical Nutrition. 2022.

9. Ebright B, Assante I, Poblete RA, et al. Eicosanoid lipidome activation in post-mortem brain tissues of individuals with APOE4 and Alzheimer’s dementia. Alzheimer’s Research & Therapy. 2022;14:152.

10. Yassine HN, Arellanes IC, Mazmanian A, et al. Baseline findings of PreventE4: a double-blind placebo controlled clinical trial testing high dose DHA in APOE4 carriers before the onset of dementia. Journal of Prevention of Alzheimer’s Disease. 2023.

11. Yassine HN, Carrasco AS, Badie DS. Designing newer omega-3 supplementation trials for cognitive outcomes: a systematic review guided analysis. Journal of Alzheimer’s Disease. 2024.

12. Ebright B, Duro MV, Chen K, Louie S, Yassine HN. Effects of APOE4 on omega-3 brain metabolism across the lifespan. Trends in Endocrinology & Metabolism. 2024.

13. Yassine HN, Ghasem Pour S, Juarez M, et al. CNS target engagement of high-dose DHA supplementation in older adults at risk for dementia: a randomized, double-blind, placebo-controlled trial. EBioMedicine. 2026.

14. Sadybekov AV, Duro MV, Wang S, et al. Development of potent, selective cPLA2 inhibitors for targeting neuroinflammation in Alzheimer’s disease and other neurodegenerative disorders. npj Drug Discovery. 2026.

She carries a genetic risk for Alzheimer’s disease and wanted to know whether she should take rapamycin to reduce her risk.

This post is dedicated to her.

My answer today is: no.

Not because rapamycin is uninteresting. It is one of the most important drugs in aging biology. It extends lifespan in mice, affects a major growth-and-repair pathway, and may improve some aspects of immune function in older adults.

But the human evidence does not show that rapamycin works in people the way it works in mice.

We do not yet have convincing evidence that rapamycin slows human aging, improves Alzheimer’s biomarkers, prevents cognitive decline, or reduces dementia risk. The strongest human data are much narrower: vaccine response, infection-related outcomes, and tissue-specific markers such as skin aging.

That is not enough to justify taking rapamycin today for Alzheimer’s prevention. But it gets more complicated. How do we know it’s working for aging or AD?

What Is Rapamycin?

Rapamycin, also called sirolimus, is a drug originally discovered from bacteria found on Easter Island, also known as Rapa Nui.

It blocks part of a pathway called mTOR, short for mammalian target of rapamycin.

Think of mTOR as one of the cell’s main growth switches. When food and energy are plentiful, mTOR tells cells to grow, build proteins, divide, and store nutrients. When mTOR activity goes down, cells may shift more toward repair, recycling, and maintenance.

This is why rapamycin became so interesting to aging researchers.

Aging is not only about time passing. It is also about how cells handle damage, inflammation, protein buildup, and repair. By lowering mTOR activity, rapamycin may push cells away from constant growth and toward cellular maintenance.

But this does not mean that “slowing metabolism” automatically slows aging.

That is too simple.

A better way to think about rapamycin is that it changes the balance between growth mode and repair mode. That may help in some tissues, at some ages, and in some disease states. But growth signals are also needed for muscle repair, wound healing, immune defense, and recovery after exercise.

The real question is not whether mTOR is good or bad.

The real question is: how much mTOR activity, in which tissue, at what age, and at what time?

Why Rapamycin Became Famous

The strongest evidence for rapamycin comes from mice.

The landmark study was published in 2009 through the National Institute on Aging’s Interventions Testing Program, or ITP.

This was not a small mouse experiment. It was a large, multi-site study using genetically diverse mice. More than 1,900 mice were included. Rapamycin treatment began late in life, at about 20 months of age.

The result was striking: rapamycin extended lifespan in both male and female mice.

Figure 1: from Harrison et al. 2009 — t rapamycin extended lifespan in genetically heterogeneous mice when started late in life

One way the investigators reported this was by examining the age at which 90% of the mice had died. That age increased from about 1,078 to 1,179 days in males, and from about 1,094 to 1,245 days in females.

This mattered because the drug worked even when started late in life. That suggested rapamycin might affect the biology of aging itself, not just growth or development.

Later mouse studies supported the same general idea. Rapamycin became one of the most reproducible lifespan-extending drugs in mice. But those studies also showed that the effect depends on dose, sex, timing, and treatment schedule.

That is impressive.

But mouse lifespan extension is not the same as human disease prevention.

Rapamycin and Alzheimer’s Mouse Models

Rapamycin has also been tested in Alzheimer’s mouse models.

In some models, including 3xTg-AD mice, rapamycin or mTOR inhibition has been reported to reduce amyloid and tau-related pathology and improve memory-like performance.

The mechanism makes sense.

mTOR inhibition can increase autophagy, the cell’s recycling system. Autophagy helps cells clear damaged proteins and worn-out cell parts. Since Alzheimer’s disease involves protein buildup and cellular stress, improving cleanup is an attractive idea.

But Alzheimer’s mouse models are not the same as human Alzheimer’s disease.

They are simplified versions of the disease. Many are engineered to develop amyloid or tau changes quickly. Human Alzheimer’s develops over decades and is shaped by aging, blood vessels, sleep, infections, metabolism, immune history, and genetics such as APOE4.

So the mouse studies are useful for understanding possible mechanisms.

They are not proof that rapamycin prevents Alzheimer’s disease in humans.

Why Mice Are Not Small Humans

The mouse data justify human trials.

They do not justify routine human use.

Mice and humans age differently. Mice are short-lived and fast-growing. Humans grow slowly and live for decades. A drug that slows a growth pathway in a mouse may not have the same effect in a long-lived human.

Laboratory mice also live in controlled environments. Their food, movement, temperature, and exposure to germs are carefully managed.

Humans live in the real world.

We experience infections, stress, poor sleep, medications, injuries, exercise, obesity, diabetes, dental disease, and decades of immune memory.

This matters especially for the immune system.

By older age, each person has a unique immune biography. One person may have had shingles, repeated respiratory infections, autoimmune disease, insulin resistance, or chronic inflammation. Another person may not.

So two 70-year-olds may have very different immune systems.

If rapamycin helps one kind of immune aging, it may not help another. It may even hurt some people if the dose, timing, or baseline biology is wrong.

That is why biomarkers matter. And for immune aging, we are learning that they need to be tissue or organ-specific.

In mice, we can give rapamycin and measure lifespan. In humans, we need to know what biology we are trying to improve.

Is it high mTOR activity? Poor vaccine response? Chronic inflammation? Weak autophagy? Senescent cells? Metabolic dysfunction? Brain inflammation?

Right now, we do not have a simple clinical test that answers this.

How Can Rapamycin Improve Immunity?

Rapamycin can seem confusing.

At high or continuous doses, it can suppress the immune system. That is why rapamycin and related drugs have been used in transplant medicine.

But low-dose or intermittent mTOR inhibition may act differently, especially in older adults.

Older immune cells may be chronically stimulated, inflamed, and inefficient. They may be “on” all the time, but not very good at responding to a new threat.

A short course of mTOR inhibition may reduce some of that background noise. It may help immune cells clean up damaged parts and respond more effectively to a challenge, such as a vaccine.

So rapamycin may not “boost” immunity like pressing the gas pedal.

It may help some older immune cells respond more cleanly by taking them out of a stressed, inefficient, or exhausted state. I have dedicated a previous post to why immune exhaustion is a hot topic in AD.

That is the idea behind the vaccine-response studies, and why it may be relevant for the brain. But we still cannot link them.

What Do Human Studies Show?

Human studies of rapamycin and rapamycin-like drugs do exist. They are interesting, but limited.

Mannick 2014: Vaccine Response

The best-known positive human study was published by Mannick and colleagues in 2014.

This randomized trial tested RAD001, also called everolimus, a rapamycin-like drug, in older adults before influenza vaccination. Participants received short-term treatment before vaccination. The main outcome was antibody response to the flu vaccine.

The result was encouraging: short-term mTOR inhibition improved vaccine response in older adults.

This was important because rapamycin is often thought of as an immunosuppressing drug. The study suggested that dose and schedule matter. Low-dose or intermittent mTOR inhibition may improve some aspects of immune function in older people.

This is a positive study because the outcome is well defined: antibody titers. And it has clear implications. If our goal is to improve the immune response to a vaccine in older adults, rapamycin works. We need to stop and not extrapolate beyond what the study showed.

Mannick 2018: Infections

A later Mannick study, published in 2018, extended this idea, but the design was more complicated.

This was a Phase 2a randomized, placebo-controlled trial in 264 older adults. Participants received low-dose mTOR-inhibitor therapy for 6 weeks.

This study did not test rapamycin itself. It tested rapamycin-pathway drugs, including everolimus and dactolisib, alone or in combination.

The strongest signal came from a low-dose combination treatment. This group had fewer reported infections over the following year. The study also found increased activity of antiviral gene programs and improved flu-vaccine response.

Together, the 2014 and 2018 studies suggest that mTOR-pathway drugs can affect immune aging in humans in relation to an immune response to a vaccine and infections. That is the strongest positive human signal so far.

Topical Rapamycin in Skin

Another human study tested topical rapamycin for skin aging.

In this small randomized study, adults over 40 applied rapamycin to one hand and a placebo to the other. The study reported reductions in p16INK4A, a marker linked to cellular senescence, and increases in collagen VII, a protein important for skin structure.

This suggests rapamycin can affect aging-related markers in human tissue. But it was a local skin study, not a systemic aging or brain-aging study. Another example where rapamycin wins in well-defined outcomes with clear readouts.

Rapamycin and Exercise

However, a small exercise study raises an important caution. One effect on skin or a vaccine response does not mean better health.

In this 13-week randomized trial, 40 sedentary older adults received either once-weekly low-dose rapamycin or a placebo while doing a light home exercise program.

The expectation was that rapamycin might improve exercise benefits by improving aging biology. The signal went the other way.

The placebo group improved more than the rapamycin group on some physical-function measures. The clearest difference was in the 30-second sit-to-stand test, a simple measure of lower-body strength and function. Grip strength and self-rated well-being also reportedly favored the placebo. Side effects such as soreness and fatigue were more common in the rapamycin group, and one serious infection was reported.

This study was small and short. It does not prove that rapamycin always blunts exercise benefits. But it matters because exercise is one of the strongest proven interventions for healthy aging and dementia prevention.

Exercise works by creating a controlled stress. Muscle is challenged, then rebuilds stronger. That rebuilding requires mTOR. If rapamycin is active during the recovery window, it may interfere with the muscle’s ability to adapt. We saw something similar with metformin, albeit via a different mechanism.

So, To Rapamycin or Not To Rapamycin? Paraphrasing Shakespeare

In humans, the clearest positive signals are the immune response and selected tissue markers. Rapamycin-like drugs have improved vaccine response in older adults. A later study suggested fewer infections in selected treatment groups. Topical rapamycin changed some aging-related markers in skin.

These are real biological signals.

But they are not the same as showing that rapamycin slows human aging, protects the brain, or prevents Alzheimer’s disease.

That distinction is the main point.

For Alzheimer’s prevention, we would want evidence that rapamycin improves meaningful brain-related outcomes: amyloid, tau, neurodegeneration, brain inflammation, cognition, or dementia risk.

We do not have that evidence.

Instead, the field has mostly shown that rapamycin can change selected immune or tissue readouts under specific conditions. That is interesting biology. It is not yet a prevention therapy.

Where the Field Seems to Be Moving

The most active human work no longer seems to be “rapamycin for everyone to slow aging.”

It seems to be moving toward more specific questions:

Can rapamycin improve immune aging?

Can it slow ovarian aging?

Can it affect periodontal disease?

Can it change skin aging markers?

Can it be timed or dosed in a way that preserves benefit while avoiding harm?

That shift is important. It suggests that the field recognizes the main problem: aging is not one pathway, and rapamycin is not one simple anti-aging switch.

A useful rapamycin trial needs a clear biological target and a clear readout.

For Alzheimer’s prevention, we do not have that.

The Real Missing Piece: Immune-Aging Biomarkers

The biggest lesson from rapamycin is not only about rapamycin.

It is about measurement.

For cardiovascular disease, we can measure blood pressure, ApoB, LDL cholesterol, HbA1C, and other risk factors. We can treat those risk factors and track whether the treatment worked.

For immune aging, we do not have an equivalent.

We can measure antibody response after a vaccine. That is useful, but it is only one test of one part of the immune system. It does not tell us whether rapamycin improved brain aging, neuronal autophagy, senescent-cell burden, or Alzheimer’s risk.

This is the central challenge.

To use rapamycin intelligently, we would need biomarkers that tell us who has the kind of aging biology rapamycin can improve, what dose is enough, whether the drug hit the intended pathway, and whether the benefit is worth the risk.

That is hard because the human immune system is not one number. It is shaped by age, sex, infections, vaccines, metabolism, obesity, sleep, exercise, medications, genetics, chronic inflammation, and decades of immune memory.

So the future of rapamycin may depend less on enthusiasm and more on biomarkers.

Not “Who wants to try rapamycin?”

But: Who has the biology that rapamycin is likely to improve, and how will we know it worked? And in what organ? Topical for skin aging? Adjuvant to a vaccine?

Rapamycin and Alzheimer’s Prevention

Rapamycin has a plausible connection to Alzheimer’s biology. mTOR is linked to autophagy, metabolism, inflammation, immune aging, and synaptic function. These all matter for brain aging.

But plausibility is not proof.

For someone with a genetic risk for Alzheimer’s disease, the key question is not whether rapamycin is biologically interesting. It is. The question is whether the existing human studies show a prevention effect.

They do not.

Today, I would not recommend rapamycin for Alzheimer’s prevention outside a clinical trial.

The Risk Side

Rapamycin is not a supplement. It is a real drug.

It can affect immunity, cholesterol and triglycerides, blood sugar, wound healing, mouth ulcers, infection risk, and drug interactions. The risk depends on dose, schedule, age, baseline health, and other medications.

Some people argue that once-weekly low-dose rapamycin may avoid many problems seen with daily immune-suppressing doses.

That may be true.

But “may be safer” is not the same as “proven safe for long-term prevention in healthy people.”

Prevention raises the bar. A prevention drug must be very safe, because many people taking it may never develop the disease they are trying to avoid.

Areas of Uncertainty

Several questions remain unanswered.

What is the right dose?

Is weekly dosing better than daily dosing?

Should rapamycin be used only in older adults?

Does APOE genotype matter?

Does baseline inflammation matter?

Does metabolic health matter?

Does rapamycin help people who already exercise, sleep well, and have good metabolic health?

Could it blunt exercise adaptation in some settings?

Can it improve Alzheimer’s biomarkers?

Most importantly: what specific outcome? That determines whether it is working.

Until we can answer these questions, use for aging or Alzheimer’s prevention is not recommended.

What I Would Tell My Friend

I would tell my friend that rapamycin is scientifically important, but not ready for Alzheimer’s prevention.

For now, I would focus on interventions with stronger human evidence: exercise, blood pressure control, sleep, metabolic health, hearing, social connection, vascular risk reduction, and treatment of insulin resistance or diabetes when present.

The next step is better and targeted trials with better immune-aging biomarkers tied to outcomes more concrete than human aging.

Take-Home Messages

• Rapamycin is one of the most compelling anti-aging drugs in mouse studies.

• Mouse lifespan extension does not prove human benefit.

• Alzheimer’s mouse-model data are useful for understanding mechanisms, but they do not prove Alzheimer’s prevention in humans.

• Human trials exist, but the strongest positive evidence is improved vaccine response and fewer infections in selected settings.

• Low-dose rapamycin may improve immune response by reducing age-related immune dysfunction, not by simply “boosting” the immune system.

• Rapamycin may interfere with exercise adaptation if mTOR is suppressed during the muscle-rebuilding window.

• The human studies so far do not show that rapamycin works in people the way it works in mice.

• We do not yet have convincing evidence that rapamycin improves Alzheimer’s biomarkers such as amyloid, tau, neurodegeneration, or cognition.

• The biggest missing piece is a practical immune-aging biomarker that tells us who might benefit and whether the drug worked.

• Rapamycin is a real drug with real risks, especially if used long-term.

• For aging or Alzheimer’s prevention, the current evidence does not justify rapamycin use today.

• The right conclusion is not “rapamycin does not work.” The better conclusion is: what is a better indication for rapamycin? And how do we design smarter trials?

Key References

1. Harrison DE, Strong R, Sharp ZD, et al. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature. 2009.

2. Miller RA, Harrison DE, Astle CM, et al. Rapamycin-mediated lifespan increase in mice is dose and sex dependent and metabolically distinct from dietary restriction. Aging Cell. 2014.

3. Caccamo A, Majumder S, Richardson A, Strong R, Oddo S. Molecular interplay between mTOR, amyloid-beta, and tau: effects on cognitive impairments. Journal of Biological Chemistry. 2010.

4. Mannick JB, Del Giudice G, Lattanzi M, et al. mTOR inhibition improves immune function in the elderly. Science Translational Medicine. 2014.

5. Mannick JB, Morris M, Hockey HP, et al. TORC1 inhibition enhances immune function and reduces infections in the elderly. Science Translational Medicine. 2018.

6. Chung CL, Lawrence I, Hoffman M, et al. Topical rapamycin reduces markers of senescence and aging in human skin: an exploratory, prospective, randomized trial. GeroScience. 2019.

7. Kaeberlein M, Galvan V. Rapamycin and Alzheimer’s disease: time for a clinical trial? Science Translational Medicine. 2019.

If the Alzheimer’s brain is inflamed, should we reduce inflammation? First, we need to answer a more basic question:

How do we know which patient actually has harmful brain inflammation?

That is the main problem this post is about.

The field is moving toward immune-based treatments before we have reliable tools to identify the right immune state in the right patient. The recent XPro1595 trial is a useful case study. It is not just a trial of one drug. It is a test of whether Alzheimer’s research is ready for precision neuroinflammation trials.

The Main Problem: We Lack a Good Brain Inflammation Marker

In Alzheimer’s disease, we now have increasingly useful biomarkers for amyloid and tau. They are not perfect, but they tell us something fairly specific about Alzheimer’s pathology.

Inflammation is harder.

We do have good markers of general inflammation. CRP, for example, is a useful blood marker. If CRP is high, it tells us that the body may be in an inflammatory state.

But CRP does not tell us where the inflammation is coming from.

A high CRP could reflect arthritis, infection, obesity, heart disease, gum disease, poor sleep, or many other conditions outside the brain. It may matter for brain health, but it does not prove that the Alzheimer’s brain itself is inflamed.

The same problem applies to ESR and HbA1C. They may capture body-wide inflammation, metabolic stress, or vascular risk. But they do not measure brain inflammation directly.

APOE4 is more closely tied to Alzheimer’s biology and may shape immune vulnerability in the brain. But APOE4 is a genetic risk marker, not a real-time inflammation test.

This creates a major problem for trials.

If we cannot measure brain inflammation well, then we may not know who should receive an immune-targeted treatment. We also may not know whether the treatment worked biologically, even if the clinical results move in the right direction.

In drug development, that matters.

A proper Phase 2 trial should do more than look for a clinical signal. It should help define the population, the dose, the target, and the biological readout. It should show target engagement: evidence that the drug hit the pathway it was designed to hit.

There Is Inflammation in the Alzheimer’s Brain. But Is It Good or Bad?

There is strong evidence that the brain’s immune system is active in Alzheimer’s disease.

In brain tissue, immune cells cluster around amyloid plaques. Genetic studies point toward immune pathways, including TREM2, CD33, INPP5D, PLCG2, and APOE.

APOE4, the strongest common genetic risk factor for late-onset Alzheimer’s disease, is especially important here. It is not only about cholesterol. In the brain, APOE4 may affect lipid handling, amyloid clearance, blood-brain barrier function, and immune tone.

Single-cell studies also show that microglia can shift into disease-associated microglia, or DAM. DAM are often found near amyloid plaques. Early on, this may be protective: surrounding plaques, clearing debris, and containing injury.

But a protective response can become costly if it is pushed for too long.

In an APOE4 brain, where amyloid handling and immune signaling may already be stressed, DAM activation may be harder to sustain. Over time, a cleanup response may become less efficient, more inflammatory, or exhausted.

So the question is not whether inflammation exists in Alzheimer’s disease.

It does.

The better question is whether that inflammation is protective, harmful, or failing.

The Stage-Specific Model

One way to think about Alzheimer’s inflammation is by stage.

Early in the disease, inflammation may be mostly reactive and compensatory. The brain sees amyloid, damaged synapses, or cellular stress, and immune cells respond. In this setting, more immune activity may sometimes help.

This was the logic behind TREM2 agonists: enhance microglial cleanup and improve the brain’s response to amyloid. But recent TREM2 agonist trials have been disappointing, reminding us that boosting immune activity is not automatically helpful.

Later in the disease, the immune response may change. It may become chronic, senescent, or exhausted. The cells are still activated, but they may no longer be solving the problem. Instead of clearing damage, they may keep sending alarm signals and contribute to synapse loss.

This creates a different treatment goal.

Early disease may require supporting protective immune responses.

Later or inflammation-heavy disease may require inhibiting or re-aligning harmful immune signals.

This is why “anti-inflammatory treatment” is too blunt a phrase. The better goal may be stage-specific immune rebalancing.

The MINDFuL trial and the rationale for Soluble TNF

TNF, or tumor necrosis factor, is an immune signaling molecule. It is not automatically bad. The body and brain use TNF signaling for immune defense, repair, and communication between cells.

But TNF comes in different forms.

Soluble TNF can move through tissue and amplify inflammatory signaling.

Transmembrane TNF stays attached to cells and may help support normal immune balance, repair, and protective glial function.

This distinction is the rationale for XPro1595. The drug was designed to block soluble TNF while preserving transmembrane TNF. Mechanistically, XP is a dominant-negative TNF biologic. It binds to soluble TNF and forms inactive complexes, preventing soluble TNF from properly activating its receptors.

That is a reasonable biological idea. But it also raises a hard question:

Do we know which Alzheimer’s patients actually have excess soluble TNF signaling in the brain?

Right now, that answer is not clear.

XPro1595 is not really a DAM drug. It is not simply an immune exhaustion drug either. It is a soluble-TNF pathway drug. It assumes that, in some Alzheimer’s patients, inflammatory signaling has become maladaptive and that blocking soluble TNF may restore a healthier immune balance.

That assumption needs biomarkers.

What Justified Testing XPro1595?

The rationale for testing XPro1595 had three parts.

First, the biology of TNF suggested that selective soluble TNF blockade might be safer and more precise than broad TNF suppression.

Second, preclinical studies suggested that soluble TNF blockade could reduce glial activation, improve debris clearance, and restore synaptic plasticity.

Third, an earlier Phase 1b Alzheimer’s study reportedly showed dose-dependent reductions in cerebrospinal fluid biomarkers of neuroinflammation and neurodegeneration.

That last point is important because cerebrospinal fluid is closer to the brain than blood. But there is a caveat: those Phase 1b data were described as a manuscript in preparation. That means readers cannot fully judge the sample size, the biomarkers, the exposure-response relationship, or whether soluble TNF itself was measured.

So the rationale for MINDFuL was reasonable.

But it was not airtight.

What MINDFuL Tested

The MINDFuL trial was a Phase 2, randomized, double-blind, placebo-controlled study in early Alzheimer’s disease.

Participants had mild cognitive impairment or mild Alzheimer’s dementia. Treatment lasted 24 weeks. XPro1595 was given once weekly by subcutaneous injection.

The primary endpoint was EMACC, a cognitive composite designed to detect change in early Alzheimer’s disease.

That endpoint makes sense for Phase 2 signal detection. But it also has limits. EMACC is not the same as showing clear benefit on a more familiar clinical endpoint such as CDR-SB, which connects cognition and daily function.

The overall trial did not meet the primary endpoint-Figure below. Nillsh..

The study then focused on a subgroup called ADi, meaning Alzheimer’s disease with inflammation. These participants were amyloid-positive and had at least two inflammation-related markers.

The Core Critique

The core critique is not that XPro1595 had no rationale. It did.

The core critique is that the trial moved into Phase 2 without a sufficiently clear way to define the inflammatory population or prove target engagement in the brain.

The ADi subgroup was defined using CRP, ESR, HbA1C, and APOE4.

These markers may enrich for risk. They may identify people with systemic inflammation, metabolic stress, APOE4 biology, or faster expected decline.

But they do not prove harmful brain inflammation.

They also do not prove excess soluble TNF signaling in the central nervous system.

That distinction matters.

If the drug targets soluble TNF, then the ideal Phase 2 trial would show several linked steps:

The selected patients had evidence of relevant brain inflammation.

The drug reached the central nervous system.

The drug engaged soluble TNF signaling.

That engagement changed downstream biomarkers.

Those biomarker changes tracked with clinical benefit.

MINDFuL showed possible downstream biological activity, especially through p-tau217 and GFAP signals in the ADi subgroup. But these changes are not convincing for target engagement, even in the dose compliant ADi, an effect size of 0.26-0.22 is modest at best (figure below).

What do we do with these results?

Better Markers of Brain Inflammation

If this field is going to move forward, we need better biomarkers.

GFAP is one useful blood marker, but as seen above, not exactly responding to inhibition of soluble TNF. It reflects astrocyte activation and is closer to brain biology than CRP or ESR. But GFAP does not tell us which inflammatory pathway is active. It does not prove soluble TNF signaling. And worse, participants were not selected or stratified by baseline GFAP.

PET imaging may help. Some tracers, such as TSPO PET, aim to measure microglial activation. These are more brain-specific than CRP, but they still have limitations. TSPO is not exclusive to one cell type, tracer performance varies, and genetic differences can affect binding.

Experimental PET approaches may also help measure brain lipid biology.

Arachidonic acid, or AA, is involved in inflammatory lipid signaling. DHA, an omega-3 fatty acid, is important for brain membranes and inflammation resolution. AA and DHA PET are not routine clinical tools, but they point to an important idea: unresolved inflammation may involve lipid signaling, membrane turnover, and failed resolution, not just cytokines.

The field needs biomarkers that can distinguish:

body-wide inflammation

astrocyte activation

microglial activation

lipid inflammatory signaling

soluble TNF pathway activity

protective versus harmful immune states

Without this, we are partly guessing.

The Uncertainty

This is where the field needs humility.

Inflammation in Alzheimer’s may be protective early and harmful later.

Peripheral inflammation is not the same as brain inflammation.

APOE4 may increase inflammatory vulnerability, but it is not a real-time inflammation measure.

GFAP, PET imaging, and lipid PET approaches may improve the picture, but none is yet a perfect clinical tool.

Most importantly, we do not yet have a routine biomarker showing that a patient has excess soluble TNF signaling in the brain.

So XPro1595 remains scientifically interesting. But MINDFuL does not prove that we know how to identify the right patients or measure treatment efficacy with enough precision.

The Bigger Lesson

The future of Alzheimer’s inflammation treatment may not depend only on better drugs.

It may depend on better measurement.

The right patient.

The right disease stage.

The right immune pathway.

The right biomarker.

The right timing.

XPro1595 may or may not become an approved Alzheimer’s treatment. But it already teaches an important lesson:

A precision immune therapy requires precision immune biomarkers.

Without them, we risk running trials in vaguely defined populations and interpreting subgroup signals without knowing whether the drug reached the brain, hit the target, or treated the biology we intended to treat.

Take-Home Messages

• The main barrier to targeting inflammation in Alzheimer’s disease is not only treatment. It is measurement.

• We have useful markers of general inflammation, such as CRP, but we lack reliable markers of brain inflammation.

• A blood marker can tell us that the body is inflamed without proving that the brain is inflamed.

• Inflammation in Alzheimer’s may be protective early, harmful later, or ineffective when the immune response becomes exhausted.

• APOE4 may shape brain immune vulnerability, but it is not a real-time inflammation test.

• A precision immune therapy requires precision immune biomarkers.

• A strong Phase 2 trial should show more than a clinical signal. It should show brain penetration, target engagement, biomarker change, and clinical benefit.

• Without better biomarkers, Alzheimer’s neuroinflammation trials risk enrolling vaguely defined populations and producing results that are hard to interpret.

• We will likely need multiple blood-based biomarkers to detect protective or harmful inflammation markers, and senescent markers may offer clues into the latter.

Key References

1. Jaeger J, Staats KA, Barnum S, et al. XPro1595 in early Alzheimer’s disease with inflammation: results from the phase 2 MINDFuL trial. npj Dementia. 2026.

2. Heneka MT, Carson MJ, El Khoury J, et al. Neuroinflammation in Alzheimer’s disease. Lancet Neurology. 2015.

3. Hansen DV, Hanson JE, Sheng M. Microglia in Alzheimer’s disease. Journal of Cell Biology. 2018.

4. Keren-Shaul H, Spinrad A, Weiner A, et al. A unique microglia type associated with restricting development of Alzheimer’s disease. Cell. 2017.

5. Bellenguez C, Küçükali F, Jansen IE, et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias. Nature Genetics. 2022.

6. Vitek MP, Brown CM, Colton CA. APOE genotype-specific differences in the innate immune response. Neurobiology of Aging. 2009.

7. Ising C, Venegas C, Zhang S, et al. NLRP3 inflammasome activation drives tau pathology. Nature. 2019.

8. Reviews on TSPO PET, GFAP, and brain lipid metabolism imaging may be useful additions for readers who want a deeper dive into neuroinflammation biomarkers.

A husband and wife came to see me in clinic.

She was 63 years old. She had two copies of APOE4, the strongest common genetic risk factor for late-onset Alzheimer’s disease. Over the previous three years, her function had declined significantly. She had been diagnosed with Alzheimer’s dementia.

Her husband had seen information about ALZ-801, an investigational oral Alzheimer’s drug being developed for people with two copies of APOE4. The message was powerful: his wife had the risk gene copy, she had Alzheimer’s, and this drug seemed designed for someone like her.

He asked me a direct question: How can we get this drug to her as soon as possible?

That question is why I am writing this.

Alzheimer’s disease creates urgency. APOE4 creates a sense of precision. How can we help her?

This post is about two related stories: taurine, a naturally occurring molecule now being studied for ApoE4 biology, and ALZ-801, an investigational oral drug built around homotaurine/tramiprosate chemistry.

Both are scientifically interesting. Neither should be treated as proven.

Why ApoE4 matters

ApoE is a protein involved in fat transport and brain health. Most people carry one of three common forms: ApoE2, ApoE3, or ApoE4. ApoE4 is a major AD risk factor

One idea is that ApoE4 may be more likely than ApoE3 to misfold or clump together. That is where taurine enters the story.

A recent paper reported that taurine can reduce ApoE4 aggregation in laboratory systems, change ApoE4 protein behavior, and shift some gene patterns in ApoE4 brain organoids toward patterns seen with ApoE3.

That is interesting. But it does not mean taurine treats Alzheimer’s disease.

Why a promising lab result is not the same as a treatment

A lab result can be exciting because it gives scientists a lead. It says: this pathway might matter. But a lead is not a medicine.

To become a medicine, a molecule has to answer many questions.

Does it work in a clear dose-response pattern? Scientists ask whether more drug produces more effect, and whether the concentration needed is realistic. For many drugs, this is measured with an IC50, the amount needed to block a process by 50 percent.

But protein clumping is messy biology. It depends on time, temperature, protein concentration, salt, assay conditions, and whether the molecule binds one specific target or weakly touches many surfaces.

That matters for taurine. In the ApoE4 paper, taurine appears to interact weakly and in multiple places. That can still matter biologically, but it is not the same as a strong, clean, highly selective drug effect.

Does it reach the brain? A molecule can work in a test tube and fail in the body. The brain is protected by the blood-brain barrier, which limits what gets from the blood into brain tissue. For an Alzheimer’s drug, reaching the brain is essential.

Does the lab dose match what people get from food or supplements? This is where translation often breaks down. In the taurine paper, the aggregation assay used 20 millimolar taurine to suppress ApoE4 aggregation, and 30 millimolar taurine to test whether pre-formed aggregates could be reduced. Those are test-tube concentrations, not dietary exposures.

A 20 millimolar taurine solution contains about 2,500 milligrams of taurine per liter. One liter of a 20 millimolar taurine solution contains roughly the taurine found in about 3.7 to 31 kilograms of beef; a 30 millimolar solution corresponds to about 5.5 to 47 kilograms. This is an imperfect comparison because a test tube is not a human body, but that is the point: eating steak does not recreate a millimolar concentration to break the apoE aggregation shown in the test tube. At least after a beefy meal!

The point is not that dietary taurine is irrelevant. The point is that foods and supplements must not be treated as if they automatically reproduce the concentrations, timing, and exposure used in laboratory experiments.

Does it have off-target effects? A molecule that weakly interacts with many surfaces may affect many pathways. Sometimes that is useful. Sometimes it creates unwanted effects. Either way, it has to be measured.

PK means pharmacokinetics: what the body does to the drug. Is it absorbed? How long does it last? Does it enter the brain? How is it cleared?

PD means pharmacodynamics: what the drug does to the body. Does it hit the intended biology? Does it move a meaningful biomarker? Does that translate into better memory or function?

A molecule can look promising and still fail because the PK is poor, the PD is weak, or the effect does not matter clinically.

This does not make the biology unimportant. It means the biology must be tested carefully before it becomes a claim about treatment.

ALZ-801: applying the drug-development ladder

ALZ-801 is different from taurine because it is not simply a supplement story. It is a drug-development story.

ALZ-801, also called valiltramiprosate, is a prodrug of tramiprosate. Tramiprosate is also known as homotaurine or 3-APS. Homotaurine is chemically related to taurine, but ALZ-801 is not taurine. It was designed to deliver tramiprosate more predictably in the body.

Tramiprosate was originally developed as an Alzheimer’s drug called Alzhemed. The idea was to interfere with amyloid-beta oligomers, small soluble amyloid clusters that may damage synapses before large plaques form.

The broad tramiprosate Phase 3 program did not show convincing benefit in the overall Alzheimer’s population. Later, homotaurine was sold over the counter in Canada as a memory supplement called Vivimind.

So how did this older molecule return as ALZ-801? The public story has four pieces: lab studies suggesting anti-amyloid aggregation activity; subgroup analyses suggesting a possible APOE4/4 signal; a prodrug designed for better absorption and tolerability; and a Phase 3 trial in people with early Alzheimer’s disease and two copies of APOE4.

That is a reasonable path to a Phase 3 trial. It is not a reason to assume the drug works before the randomized trial answers the question.

What ALZ-801 still had to prove

Is there a clean IC50? Not in the simple way people may imagine. ALZ-801 is not like a classic drug that blocks one enzyme at one active site. Its proposed effect is to reduce amyloid oligomer formation, which is a complex clumping process. The public development story is built around aggregation assays, stoichiometry, PK modeling, and projected brain exposure, not a single easy-to-interpret human IC50.

How was the dose chosen? Public papers describe a PK bridging strategy. The Phase 3 dose, 265 mg twice daily, was selected to produce tramiprosate exposure similar to 150 mg twice daily tramiprosate, the dose later emphasized from older APOE4/4 subgroup analyses. In other words, the dose was not chosen from a clean prospective human brain target-engagement trial; it was bridged from the older tramiprosate program and from projected exposure needed to inhibit amyloid oligomer formation.

Does it cross the blood-brain barrier? The public evidence is supportive but still indirect. The development literature describes tramiprosate and 3-SPA as brain-penetrant and refers to CSF/metabolite data, animal brain-penetration work, and projected human brain exposure. I did not find a public FDA review package that independently lays out a definitive human BBB and target-engagement analysis for ALZ-801.

What about off-target effects? ALZ-801 is usually discussed as an amyloid-oligomer drug, but homotaurine-like molecules may have other biological effects, including possible effects on GABA receptors. That does not mean the drug is unsafe or ineffective. It means the mechanism should not be oversimplified.

What is the PD readout? Cognition is not the PD readout. Cognition is the clinical outcome. A cleaner PD readout would show that the drug changed Alzheimer’s biology in humans, such as a prespecified change in p-tau217, p-tau181, amyloid measures, GFAP, NfL, or another biomarker connected to disease activity. A convincing story would line up exposure, brain penetration, target engagement, biomarker movement, and clinical benefit.

The data on ALZ-801 are still incomplete.

Timing matters: preventing amyloid clumps is not the same as removing them

Amyloid biology unfolds over years. Amyloid-beta can misfold, form small oligomers, seed further aggregation, and eventually contribute to plaques. By the time someone has memory symptoms, especially someone with two copies of APOE4, amyloid plaques may already be widespread. The amyloid plaques appear decades before one has a diagnosis of mild cognitive impairment.

That matters for ALZ-801. If a drug mainly prevents new amyloid oligomers from forming, when is the best time to use it? Before plaques form? Around the time amyloid is seeding? After plaques are already present?

Does ALZ-801 dissolve existing plaques? Does it reduce amyloid PET signal? Does it prevent plaques from shedding toxic oligomers? Does it slow downstream tau spread? Or does it need to be used earlier than the stage when most patients are diagnosed?

These questions remain open. This does not mean ALZ-801 cannot work. It means the timing of treatment may be critical and that there are many unaswered questions still.

The problem with confidence before placebo-controlled evidence

The path to ALZ-801 included lab studies, improved PK, brain-penetration arguments, APOE4 subgroup signals, and open-label Phase 2 biomarker studies. That may be enough to justify a Phase 3 trial. It is not enough to assume success.

Open-label Phase 2 studies can help with safety, dosing, tolerability, and biomarker hypotheses. But they cannot prove clinical efficacy in Alzheimer’s disease. Without a placebo group, it is hard to know whether apparent cognitive stabilization reflects the drug, patient selection, natural history, missing data, expectations, or analysis choices.

The company’s 2023 press release claimed 31% reduction in plasma p-tau181 at 24 months, 28% preservation of hippocampal volume compared with matched ADNI external controls, and improvement on cognitive tests at 6 months with sustained stabilization above baseline for 24 months. It also stated that treated patients had a 24% improvement on RAVLT Total Score compared with matched ADNI subjects

But there was no placebo arm. Matching to ADNI is not the same as having a placebo arm. FDA standards for drug approval still require evidence that can distinguish a real drug effect from placebo effect, spontaneous change, or biased observation.

But it can become a loophole in the public narrative. When open-label data are described as clinical improvement or treatment benefit, readers may hear proof where the study design only supports a hypothesis. That is especially risky in Alzheimer’s disease, where patients and families are actively searching for options.

What did APOLLOE4 report?

APOLLOE4 tested valiltramiprosate in people with early Alzheimer’s disease who had two copies of APOE4. It was randomized, double-blind, and placebo-controlled. That was the right test.

The overall clinical result was negative. The trial did not meet its primary cognitive endpoint in the overall study population. That is the anchor result.

Fig: In the overall APOLLOE4 Phase 3 population, ALZ-801 did not show a statistically significant clinical benefit over placebo on the primary cognitive endpoint, ADAS-Cog13, or on key functional measures. The ALZ-801 and placebo curves separate modestly on some outcomes, but the trial did not meet its primary endpoint, so these results do not establish clinical efficacy.

The trial did report slowing of hippocampal atrophy on MRI and nominally significant clinical findings in a prespecified mild cognitive impairment subgroup. These findings are worth studying. They may justify a new trial focused on earlier-stage patients.

But they do not erase the main result: the overall Phase 3 clinical endpoint was not met. Secondary endpoints and subgroup findings can fail when tested directly in a new trial. The responsible next step is to test the MCI hypothesis prospectively, not to promote the drug publicly as if an MCI indication has already been proven.

What about biomarkers? The peer-reviewed Phase 3 paper reports that plasma biomarkers were collected, including Aβ42, Aβ40, p-tau181, p-tau217, GFAP, and NfL. But it also says that serial assessments of core Alzheimer’s plasma biomarkers were still in progress and would be reported separately.

Alzheon later announced p-tau217 results, including decreases beginning at 26 weeks, larger effects in the MCI subgroup, and correlations with cognition, function, and hippocampal volume. Those results should be taken seriously, but they are not the same as a complete, independently interpreted, peer-reviewed biomarker package with all methods, subgroup definitions, missing-data handling, and statistical corrections clearly shown.

Being conservative is not being dismissive

There is an urgent need for safe, accessible treatments for APOE4 patients. The approved anti-amyloid antibodies are difficult treatments. They require infusions, MRI monitoring, infrastructure, and careful safety discussions. APOE4 carriers, especially APOE4 homozygotes, face particular safety concerns.

An oral drug with good safety, brain penetration, and real disease-modifying benefit would be a major advance. So the goal is not to dismiss ALZ-801. The goal is to leave room for it to succeed without treating success as already proven.

If future trials in earlier APOE4/4 patients show consistent clinical benefit, supported by clear biomarker movement and acceptable safety, that would matter.

But until then, the fairest summary is this: ALZ-801 has a rational mechanism, improved PK compared with older tramiprosate, a plausible but not fully independently verified brain-exposure argument, and company-reported biomarker signals. It also missed its primary clinical endpoint in the overall APOLLOE4 Phase 3 population. The MCI signal is a reason to study the drug further, not a reason to market it to the public as proven for MCI.

What if a patient wants to take the risk?

This is the hardest question in clinic. Some APOE4 carriers and families may say: if the drug appears safe and the disease is devastating, why not try it?

That instinct is understandable. But risk is not only side effects. There is also the risk of false hope, cost, distraction from proven care, delay in planning, and loss of trust if the promise does not hold. A drug with few obvious side effects can still fail to help, and an unproven treatment can still change decisions.

The right answer is not to block hope. It is to keep hope inside a system that can tell whether the drug truly works. For now, that means carefully designed trials, transparent communication, and avoiding public claims that go beyond the evidence.

The danger of direct-to-consumer promises

The husband in my clinic was responding to a message that seemed to match his wife’s biology. That is why direct-to-consumer communication about investigational Alzheimer’s drugs must be handled carefully.

Alzheimer’s disease is devastating. APOE4 status can make a drug feel personally targeted. Families may hear developed for APOE4/4 and ask why they cannot access it now. A company may communicate through emails, websites, newsletters, webinars, advocacy groups, trial-recruitment materials, or advertisements. Some of that communication may be legal. Some of it may be framed as education.

But legal is not the same as scientifically responsible.

When a company contacts consumers with language that implies benefit before benefit has been proven, or highlights a subgroup such as MCI after the overall Phase 3 trial missed its primary endpoint, it risks turning scientific uncertainty into marketing. The audience includes patients, spouses, adult children, and caregivers making decisions under pressure.

False or overstated promises can shape decisions. Families may pursue access, interpret subgroup findings as proof, delay planning, or organize expectations around a treatment whose benefit is still uncertain.

They also damage trust. If patients are led to believe that a drug is likely to help, and then Phase 3 results disappoint, many will remember that science raised their hopes and then changed the message.

Science can survive negative trials. Negative trials are part of progress. What science cannot afford is a cycle in which early biology becomes marketing, subgroup signals become promises, and families are asked to believe before the evidence is ready.

Take-home messages

• ApoE4 biology matters, and therapies tailored to APOE4 patients are urgently needed.

• Taurine is interesting, but not proven. The taurine-ApoE4 paper is a mechanistic study, not evidence that taurine prevents or treats Alzheimer’s disease.

• Food molecules are not automatically therapies. Red meat contains taurine, wine contains resveratrol, and green tea contains EGCG, but eating or drinking these foods does not reproduce the high, controlled concentrations used in many preclinical aggregation experiments.

• ALZ-801 is a serious drug-development story, but still unproven. Its dose selection and brain-exposure argument are based on public PK and bridging data, not a clean public IC50-to-human-brain target-engagement chain.

• APOLLOE4 did not meet its primary cognitive endpoint overall. MCI subgroup and secondary findings are worth studying, but they should not be promoted to the public as a proven MCI indication without a direct confirmatory trial.

• Conservative interpretation is not dismissal. An oral, safe, effective therapy for APOE4 patients would be a major advance. The evidence should be strong enough that patients can trust the claim.

References

1. Legrand A, Cerna KA, Marques SM, et al. Taurine inhibits apolipoprotein E4 aggregation. Biomedicine & Pharmacotherapy. 2026.

2. Hey JA, et al. Clinical Pharmacokinetics and Safety of ALZ-801, a Novel Prodrug of Tramiprosate in Development for the Treatment of Alzheimer’s Disease. Clinical Pharmacokinetics. 2018.

3. Hey JA, Kocis P, Hort J, et al. Discovery and Identification of an Endogenous Metabolite of Tramiprosate and Its Prodrug ALZ-801 that Inhibits Beta Amyloid Oligomer Formation in the Human Brain. CNS Drugs. 2018.

4. Abushakra S, et al. Clinical Pharmacokinetics of Oral ALZ-801/Valiltramiprosate and its Active Agents in APOE4/4 Subjects with Early Alzheimer’s Disease. Clinical Pharmacokinetics. 2025.

5. Abushakra S, et al. Clinical Efficacy, Safety and Imaging Effects of Oral Valiltramiprosate in APOE4/4 Homozygous Individuals with Early Alzheimer’s Disease: The APOLLOE4 Phase III Randomized Controlled Trial. CNS Drugs. 2025.

6. Alzforum. Alzhemed / tramiprosate therapeutic profile.

7. Alzforum. ALZ-801 therapeutic profile.

8. FDA / eCFR. 21 CFR 314.126: Adequate and well-controlled studies.

9. FDA. Fast Track designation information and FDA Global Substance Registration System note that UNII/substance records do not imply regulatory review or approval.

10. FDA. Unproven Alzheimer’s Disease Products.

11. Cleveland Clinic Journal of Medicine. Table: Taurine content of meats, seafood, and dairy products.

12. Hey JA, et al. Effects of Oral ALZ-801/Valiltramiprosate on Plasma Biomarkers, Brain Hippocampal Volume, and Cognition: Results of 2-Year Single-Arm, Open-Label, Phase 2 Trial in APOE4 Carriers with Early Alzheimer’s Disease. Drugs. 2024.

13. Alzheon. Topline Results from Pivotal APOLLOE4 Phase 3 Trial of Oral Valiltramiprosate/ALZ-801. 2025.

It is also a good moment to think about rest, recovery, and the quiet biological work the brain does when the day finally ends.

Consider a 56-year-old man -- let’s call him David.

David works in the media industry, where deadlines, breaking news, public pressure, and long screen-heavy days can make work unusually stressful.

He recently learned that he carries two copies of APOE4, the strongest common genetic risk factor for late-onset Alzheimer’s disease.

He occasionally loses a word mid-sentence.

He eats well, exercises daily, and keeps a healthy weight.

But one pattern stands out: he wakes around 4 a.m., lies there exhausted, and cannot fall back asleep.

Most evenings, he watches television late.

He feels anxious at night.

A few times a week, he has caffeine in the afternoon or wine with dinner.

This post is written for him -- and for anyone wondering whether sleep is just rest, or something more active and protective.

I am grateful to my colleague and friend Dr. Miranda Lim -- neuroscientist, sleep researcher, and expert in sleep and neurodegeneration -- for her generous feedback on this piece.

1. The Brain’s Nightly Cleaning Crew

Sleep is active brain maintenance.

It is not simply the absence of being awake.

During sleep, fluid moves through the spaces around brain cells and helps wash away metabolic waste.

This system is called the glymphatic system -- glial because it depends on support cells in the brain, and lymphatic because it behaves like a drainage system.

Among the waste products it helps clear are amyloid-beta and tau, two proteins central to Alzheimer’s disease.

In a landmark mouse study, the brain cleared waste far more efficiently during sleep than during wakefulness [1].

Other studies have shown that amyloid and tau follow sleep-wake rhythms, rising during wakefulness and falling during sleep [2,3].

One bad night does not cause Alzheimer’s.

But repeated sleep disruption, year after year, may remove one layer of protection from a brain that is already aging.

The relationship runs in both directions.

Poor sleep may worsen amyloid and tau biology.

Early amyloid and tau changes may also damage the brain circuits that regulate sleep, making sleep more fragmented over time [4].

2. APOE4 and Sleep: Why David’s Night Matters

Why might sleep matter more in APOE4 carriers?

One way to think about it is this: APOE4 may make amyloid handling less efficient and the brain’s immune response more reactive.

If the nightly clean-up system is also weakened by insomnia, sleep apnea, alcohol, irregular timing, or anxiety, the same amount of sleep disruption may carry a higher cost.

Human studies suggest that sleep and APOE4 interact.

In older adults, better sleep consolidation appeared to weaken the relationship between APOE4 and Alzheimer ’s-related pathology [5].

Newer work also suggests that sleep disturbance may be more strongly linked to amyloid burden in APOE4 carriers than in non-carriers [6].

Animal work supports the same idea: APOE4 and sleep disruption can act together to accelerate amyloid and tau-related changes [7].

The practical point is leverage.

David cannot change his genes.

But he can change the nightly environment his genes operate in.

For an APOE4 carrier, treating insomnia, sleep apnea, irregular sleep timing, alcohol-related sleep fragmentation, and nighttime anxiety is not cosmetic wellness.

It is risk management.

3. Not All Sleep Is Equal: Understanding the Cycles

Adults usually need seven to nine hours of sleep.

But quality matters as much as quantity.

Sleep moves through stages in cycles of roughly 90 to 120 minutes.

Most people cycle through these stages three to five times per night.

A typical night is not flat.

It looks like a staircase: deeper sleep early in the night, then longer stretches of dream-rich REM sleep toward morning.

Figure: Sleep cycles across a typical night. Deep sleep is more common early in the night. REM sleep lengthens toward morning. Adapted from Shah et al., Portable Monitoring, 2013.

The light green zone at the beginning is the hypnagogic state. This is the drifting period when thoughts loosen, images appear, and the brain crosses from wakefulness into sleep. Its mirror image is the hypnopompic state, the groggy transition that happens while waking. Both are normal borderlands between consciousness and sleep.

NREM Stage 1, or N1, is the lightest stage of sleep — the brief doorway between wakefulness and sleep, when thoughts loosen, awareness fades, and the hypnagogic state begins.

NREM Stage 2 is the workhorse of the night. It often makes up about half of total sleep and is marked by sleep spindles, brief bursts of brain activity involved in memory processing.

NREM Stage 3 is deep sleep, also called slow-wave sleep (SWS). This stage is linked to physical restoration, immune signaling, memory stabilization, and possibly glymphatic clearance. The purple segments in the figure refer to lucid dreamless sleep. This is a rare state, reported especially among experienced meditators, in which a person may have clear awareness during deep sleep without the usual story, images, or dream content.

Deep sleep declines with age. A long-term study in JAMA Neurology found that lower slow-wave sleep was associated with higher dementia risk, although the study was observational and cannot prove cause and effect [8].

REM sleep is the stage of vivid dreams. The brain is highly active, but the body is temporarily paralyzed, so we do not act out dreams. Some people experience lucid dreaming during REM: they realize they are dreaming while the dream is still happening, and sometimes can influence the dream. REM helps process emotion, fear, anxiety, and complex memories. Alcohol, sleep apnea, and some medications can fragment or suppress REM.

The dashed arrows mark brief returns toward wakefulness, often called WASO -- wake after sleep onset. Tiny awakenings happen to everyone. The problem is not waking for a moment. The problem is waking and staying awake for 20 minutes or more.

The goal is not to force one stage at the expense of another. The brain needs the full natural sequence: light sleep, Stage 2, deep sleep, REM, and brief awakenings that quickly return to sleep.

David’s 4 a.m. awakening may begin as a normal sleep-cycle boundary. It also occurs in the part of the night when REM sleep is usually longer and more frequent. If anxiety fragments this REM-rich period, the night may feel as if it ends too early.

4. Start With a Sleep Audit

Before buying devices or supplements, start with a simple sleep diary.

Track for at least two to three weeks, including weekends.

Each morning, write down bedtime, wake time, estimated total sleep, number of awakenings, and how refreshed you feel.

If you wake during the night, estimate how long it takes to fall back asleep.

Each evening, note caffeine timing, alcohol, exercise, screen exposure, and anxiety level.

Anxiety deserves its own line in the diary.

It is one of the most common and underrecognized sleep disruptors.

For David, the diary might reveal a pattern: caffeine after 2 p.m., television until 11 p.m., wine with dinner, and then a 4 a.m. awakening.

None of these is dramatic alone.

Together, they can fragment sleep.

The goal is not perfection.

The goal is pattern recognition.

5. Six Habits With the Best Evidence

Keep a steady schedule.

Wake up at the same time every day, including weekends.

A stable wake time is one of the strongest signals to the body clock.

Protect the last hour.

Dim lights after dinner.

Avoid emotionally charged television, work email, and bright screens close to bed.

Blue-wavelength light suppresses melatonin, the hormone that tells the brain darkness has arrived.

Keep the bedroom cool.

A room around 65 to 68 degrees F, or 18 to 20 degrees C, helps the body drop its core temperature, which supports sleep onset and deep sleep.

Cut caffeine early.

Caffeine can last five to seven hours, and often longer in older adults.

What was harmless at 35 may be disruptive at 56.

Do not use alcohol as a sleep aid.

Alcohol may help you fall asleep faster, but it fragments the second half of the night and suppresses deep sleep and REM.

Move your body daily.

Regular exercise improves sleep pressure, mood, insulin sensitivity, and cardiovascular health.

Avoid vigorous exercise close to bedtime if it leaves you wired.

6. Treat Anxiety and Insomnia Together: Why CBT-I Matters

Many people think insomnia is a nighttime problem.

Often, it is also a daytime anxiety loop.

The brain starts to learn that the bed is a place for effort, monitoring, frustration, and worry.

That learned association is called conditioned arousal.

CBT-I -- Cognitive Behavioral Therapy for Insomnia -- is designed to break that loop.

It is not simply sleep hygiene.

CBT-I combines stimulus control, sleep scheduling, cognitive therapy, relaxation skills, and careful adjustment of time in bed.

It helps the brain relearn a simple rule: bed equals sleep, not struggle.

The American College of Physicians recommends CBT-I as the initial treatment for chronic insomnia in adults [9].

AASM guidance also supports behavioral-psychological treatment and, when medication is needed, pairing it with CBT-I rather than using medication alone [10].

For David, CBT-I is especially relevant because anxiety may be shortening or fragmenting the REM-rich final third of his night.

The goal is not to convince him to stop worrying by willpower.

The goal is to change the sleep system so his body no longer practices worry at 4 a.m.

CBT-I can be done with a trained therapist, through some sleep clinics, or through validated digital CBT-I programs when access is limited.

It can feel uncomfortable at first because it asks people to change habits that feel protective.

But unlike sedatives, it teaches a durable skill.

7. When to Refer to a Sleep Specialist

Not every sleep problem needs a sleep lab.

But some do.

Refer for evaluation if there is loud snoring, witnessed pauses in breathing, gasping, choking, morning headaches, dry mouth, high blood pressure, or daytime sleepiness despite enough time in bed.

These are red flags for obstructive sleep apnea.

Sleep apnea repeatedly drops oxygen and forces the brain to wake just enough to breathe.

That can destroy deep sleep and REM without the person remembering the awakenings.

Referral is also appropriate when insomnia lasts more than three months, causes daytime impairment, or does not improve with basic behavioral steps.

Other reasons include restless legs, violent dream enactment, narcolepsy symptoms, shift-work disorder, complex medication questions, or persistent early-morning awakening with depression or severe anxiety.

For older adults and APOE4 carriers, the threshold can be lower.

A treatable sleep disorder is not something to watch passively.

8. Sleep Medications: Useful, but Not the Same as Natural Sleep

There are times when sleep medication is reasonable.

Jet lag, a hospital stay, acute grief, or a short period of severe insomnia may justify short-term help.

The problem is assuming that a pill that causes unconsciousness is the same as healthy sleep.

Many common sleep aids -- benzodiazepines, Z-drugs such as zolpidem, and over-the-counter antihistamines such as diphenhydramine -- sedate the brain.

They may reduce the memory of waking up.

But they can also alter sleep architecture, suppress REM or deep sleep, increase falls, and create dependence or rebound insomnia.

This matters even more in older adults.

Antihistamine sleep aids have anticholinergic effects, which are generally a poor fit for brain health when used chronically.

Newer orexin receptor antagonists, such as suvorexant and lemborexant, work differently.

They reduce the brain’s wakefulness signal rather than broadly sedating the brain.

They may preserve sleep architecture more faithfully, but they still require medical supervision and their long-term effects are still uncertain.

Melatonin belongs in a separate category.

It is not a sedative.

It is a darkness signal that can help shift the body clock for jet lag or delayed sleep timing.

Low doses, often 0.5 to 1 mg, are usually enough for circadian purposes.

More is not necessarily better.

Do not take melatonin before driving.

9. Rings, Watches, and New Sleep Technologies

Sleep devices can be helpful when they are used for trends, not truth.

A ring or watch can show bedtime regularity, total sleep opportunity, heart rate patterns, movement, and whether weekends are drifting later.

For David, a device might show that his sleep is shorter after wine or caffeine, or that his weekend schedule shifts by two hours.

But consumer devices are not sleep labs.

They usually estimate sleep from movement, heart rate, temperature, and algorithms.

They do not directly measure brain waves the way polysomnography does.

Recent validation studies suggest many devices are fairly good at estimating sleep versus wake, but less reliable at distinguishing deep sleep from REM or light sleep [11].

This is why a low deep sleep score should not create panic.

And a beautiful sleep score should not override symptoms like snoring, gasping, or daytime sleepiness.

The best use is pattern recognition.

Use the device to ask: what happens when I keep a steady wake time, stop alcohol, treat anxiety, or get morning light?

The worst use is sleep perfectionism.

Some people become more anxious because they chase a score, a problem sometimes called orthosomnia.

If tracking makes sleep feel like a test, stop tracking for a while and return to a diary.

On the research frontier, more advanced tools are emerging.

Wearable EEG headbands can measure brain activity more directly and may eventually guide gentle stimulation during naturally occurring deep sleep transitions [12].

Other experimental tools are trying to measure glymphatic function in humans during sleep [13,14].

They are not yet a substitute for the basics: regular timing, darkness at night, morning light, exercise, apnea treatment, and anxiety care.

10. What We Still Do Not Know

Sleep and Alzheimer’s influence each other in both directions.

Poor sleep may increase amyloid and tau.

But early Alzheimer’s changes can also damage the brain circuits that regulate sleep.

That makes cause and effect hard to untangle.

We do not yet know the exact amount of deep sleep needed for brain clearance.

We do not know whether every APOE4 carrier benefits equally from sleep intervention.

We do not know whether fixing sleep after symptoms begin can slow disease progression.

So the honest answer is this: sleep is not a cure for Alzheimer’s.

But it is one of the most biologically plausible, low-cost, and broadly beneficial targets we have.

Take-Home Messages

· Sleep is active brain maintenance. The glymphatic system helps clear amyloid-beta, tau, and other waste during sleep.

· APOE4 raises the stakes, not destiny. Genes are not fate, but APOE4 may make good sleep more important.

· All sleep stages matter. Deep sleep may support clearance, REM supports emotional and memory processing, and the full cycle matters more than any single stage.

· Brief awakenings are normal. Lying awake for 20 minutes or more is the pattern to address.

· Treat anxiety as part of treating sleep. CBT-I directly targets the 4 a.m. worry loop.

· Refer when red flags appear. Snoring, gasping, witnessed pauses, daytime sleepiness, restless legs, unusual nighttime behaviors, or chronic insomnia deserve evaluation.

· Use devices wisely. Rings and watches can reveal patterns, but they are not diagnostic tests and should not become another source of anxiety.

· Be careful with sedatives. Unconsciousness is not the same as natural sleep.

For David, the prescription is not fear.

It is a plan: protect the night, treat the disruptors, and give the brain the conditions it needs to do its quiet work.

References

1. Xie L, Kang H, Xu Q, et al. Sleep drives metabolite clearance from the adult brain. Science. 2013;342(6156):373-377.

2. Kang JE, Lim MM, Bateman RJ, et al. Amyloid-beta dynamics are regulated by orexin and the sleep-wake cycle. Science. 2009;326(5955):1005-1007.

3. Holth JK, Fritschi SK, Wang C, et al. The sleep-wake cycle regulates brain interstitial fluid tau in mice and CSF tau in humans. Science. 2019;363(6429):880-884.

4. Vanderheyden WM, Lim MM, Musiek ES, Gerstner JR. Alzheimer’s disease and sleep-wake disturbances: amyloid, astrocytes, and animal models. Journal of Neuroscience. 2018;38(12):2901-2910.

5. Lim ASP, Yu L, Kowgier M, Schneider JA, Buchman AS, Bennett DA. Sleep modifies the relation of APOE to risk of Alzheimer disease and neurofibrillary tangle pathology. JAMA Neurology. 2013;70(12):1544-1551.

6. Liang Q, et al. Moderating effect of APOE epsilon4 on the association of sleep disturbance with regional amyloid-beta burden in cognitively normal older adults. Frontiers in Aging Neuroscience. 2025.

7. Wang C, Nambiar A, Strickland MR, et al. APOE-epsilon4 synergizes with sleep disruption to accelerate amyloid-beta deposition and plaque-associated tau seeding and spreading. Journal of Clinical Investigation. 2025.

8. Himali JJ, Baril A-A, Cavuoto MG, et al. Association between slow-wave sleep loss and incident dementia. JAMA Neurology. 2023;80(11):1326-1333.

9. Qaseem A, Kansagara D, Forciea MA, Cooke M, Denberg TD. Management of chronic insomnia disorder in adults: a clinical practice guideline from the American College of Physicians. Annals of Internal Medicine. 2016;165(2):125-133.

10. Buysse DJ, et al. Combination treatment for chronic insomnia disorder in adults: an American Academy of Sleep Medicine clinical practice guideline. Journal of Clinical Sleep Medicine. 2026.

11. Robbins R, Weaver MD, Sullivan JP, et al. Accuracy of three commercial wearable devices for sleep tracking in healthy adults. Sensors. 2024;24(20):6532.

12. Hathaway E, Morgan K, Carson M, et al. Transcranial electrical stimulation targeting limbic cortex increases the duration of human deep sleep. Sleep Medicine. 2021;81:350-357.

13. Dagum P, Giovangrandi L, Levendovszky SR, et al. A wireless device for continuous measurement of brain parenchymal resistance tracks glymphatic function in humans. Nature Biomedical Engineering. 2025;9:1656-1676.

14. Dagum P, Elbert DL, Giovangrandi L, et al. The glymphatic system clears amyloid beta and tau from brain to plasma in humans. Nature Communications. 2026;17(1):715.