40 Hz sound and Alzheimer's: the evidence ladder

A light flickering forty times a second, with a tone clicking at the same rate, is the most scientifically serious idea in the whole of sound-and-brain research, and the most badly reported. In mouse models of Alzheimer’s disease the results are striking and have been replicated for a decade. In humans, as of today, one randomised sham-controlled trial has published its results, it enrolled 76 people, it was not powered to test whether the treatment works, and it found no separation between the active and sham groups on its primary measure. The first properly powered trial finished collecting data in August 2026 and has not reported.

So the honest grade depends on which rung you are looking at. Much of the coverage moves from one rung to the next mid-sentence, without telling the reader it has. Here is each rung, and what it does and does not show.

The ladder

Rung	Study	What it showed	Our grade
1. Mice, invasive	Iaccarino 2016, Nature	Driving interneurons at 40 Hz optogenetically cut amyloid-β and transformed microglia	Not graded: mouse, and the method cannot be used in people
2. Mice, non-invasive	Martorell 2019, Cell	Seven days of 40 Hz tones improved memory and reduced amyloid in 5XFAD mice; light plus sound did more than either	Not graded: mouse
3. Mice, mechanism	Murdock 2024, Nature	40 Hz stimulation drove fluid clearance through the brain; blocking that clearance abolished the amyloid removal	Not graded: mouse
4. Aged primates	Wang 2026, PNAS	Nine rhesus monkeys aged 26 to 31 exposed to 40 Hz sound; changes in cerebrospinal fluid amyloid and tau	Not graded: animal
5. Humans, feasibility	Chan 2022, PLOS ONE	40 Hz stimulation safely produces measurable gamma entrainment in people, confirmed on intracranial electrodes in two	Feasibility, not efficacy
6. Humans, randomised, sham-controlled	OVERTURE (Hajós 2024), Frontiers in Neurology	76 patients, 6 months. Safe and well tolerated. No separation on the primary endpoint. Three secondary measures reached nominal significance	The only controlled human efficacy data there is
7. Humans, open label	Chan 2025	Five people used a device daily for two years; three declined less than registry-matched patients	weak
8. All human evidence, independently reviewed	Hyun 2026, GeroScience	Eight studies. Structural signals that failed a sensitivity check; no consistent cognitive benefit	mixed
9. Humans, pivotal	HOPE (NCT05637801)	673 participants, quadruple-masked, 12 months. Completed 17 August 2026	No results posted

Rungs 1 to 3: the mouse work, which is strong

The idea started with an observation rather than a therapy. Gamma oscillations, the brain’s fast 20 to 50 Hz rhythms, are reduced in Alzheimer’s disease. In 2016 Hannah Iaccarino and colleagues in Li-Huei Tsai’s lab at MIT showed that gamma oscillations fell in a mouse model of the disease before plaques formed or memory declined, and then that driving the relevant interneurons at 40 Hz, using optogenetics, reduced amyloid-β in the brain and changed the shape and gene expression of microglia, the brain’s immune cells. Forty hertz did this; other frequencies did not (doi:10.1038/nature20587).

Optogenetics requires genetic modification and an implanted fibre, so the next question was whether an ordinary sound could do the same job. In 2019 Anthony Martorell and colleagues, same lab, reported that seven days of 40 Hz auditory tones improved spatial and recognition memory in 5XFAD mice and reduced amyloid in the auditory cortex and hippocampus, that it also reduced phosphorylated tau in a separate tauopathy model, and that combining sound with light did more than either alone (doi:10.1016/j.cell.2019.02.014). That paper is why the technology exists.

In 2024 Mitchell Murdock and colleagues, again the same lab, supplied a plausible mechanism: 40 Hz stimulation increased the flow of cerebrospinal fluid into the cortex and interstitial fluid out of it, with changes in aquaporin-4 along astrocyte endfeet and dilated meningeal lymphatic vessels. The persuasive part is the negative control. When they inhibited that glymphatic clearance, the amyloid removal disappeared (doi:10.1038/s41586-024-07132-6).

Three things about these three papers. They are published in Nature, Cell and Nature. They are mechanistically coherent in a way that very little sound research is. And all three carry the same senior author, which means the mouse work is one lab’s sustained programme rather than independent groups converging. None of that is a criticism. It does mean the replication that matters has to happen in people.

Rung 4: nine monkeys

Rodent brains are not human brains, so in January 2026 Wenchao Wang and colleagues in Xintian Hu’s group took what they describe as a rare opportunity: nine rhesus monkeys aged 26 to 31, which is old for the species, exposed to 40 Hz auditory stimulation. They tracked amyloid and tau in cerebrospinal fluid, on the reasoning that CSF concentrations track pathology in the brain tissue (doi:10.1073/pnas.2529565123).

This study is the source of most of the “listening to this sound could treat Alzheimer’s” headlines of 2026, including Newsweek’s, which reports increases in CSF proteins of around 206 and 201 percent after a week. Nine animals, a non-human species, a biomarker rather than a symptom. It raises the plausibility that a mouse finding carries into a primate brain. It is not a treatment result in a person, and nobody involved says it is.

Rungs 5 and 6: what has actually been tested in people

This is where the reporting falls apart, so it is worth being slow.

Diane Chan and colleagues ran the feasibility work: a phase 1 study in 25 cognitively normal volunteers, 16 patients with mild Alzheimer’s dementia, and two patients with epilepsy who happened to have intracranial electrodes in place, which allowed direct confirmation that 40 Hz light and sound really does entrain gamma activity inside the human brain. That part is settled. The same paper reports a small single-blind randomised placebo-controlled phase 2A pilot in mild Alzheimer’s (doi:10.1371/journal.pone.0278412).

Then OVERTURE. This is the one trial that matters for the question “does it work”, and it belongs in every article about 40 Hz. Cognito Therapeutics ran a randomised, double-blind, sham-controlled six-month trial of its device in 76 participants with mild-to-moderate Alzheimer’s, randomised 2:1, self-administered for an hour a day at home with a care partner (NCT03556280, published as Hajós and colleagues 2024, doi:10.3389/fneur.2024.1343588).

What it found, in the paper’s own words: adverse events were similar between groups, no amyloid-related imaging abnormalities in anyone, adherence 85 to 90 percent in both arms. And: “There was no statistical separation between active and sham arm participants in primary outcome measure of MADCOMS or secondary outcome measure of CDR-SB or ADAS-Cog14. However, some secondary outcome measures including ADCS-ADL, MMSE, and MRI whole brain volume demonstrated reduced progression in active compared to sham treated participants, that achieved nominal significance.”

The paper also states plainly that “the trial was not statistically powered to evaluate potential efficacy outcomes”. Its primary outcome measures were safety and tolerability. On those it succeeded: the device is safe, people use it, nothing alarming showed up on MRI. That is a real and necessary result.

The three measures that moved were secondary, and “nominal significance” means significant before any correction for having looked at several outcomes. A later post-hoc analysis by Kern and colleagues reframed the same data as time saved, reporting 4.83 months of progression saved on daily function, 4.59 on MMSE and 4.09 on whole-brain atrophy over the six randomised months (doi:10.1002/trc2.70118). Cognito’s own clinical studies page presents the same three measures as a “77% Reduction in decline of daily function”, a “76% Reduction in decline of cognitive function” and a “69% Reduction in MRI brain volume loss”.

Those percentages are not invented, and they are not what the trial was built to measure. They are three secondary outcomes with uncorrected p-values from an underpowered feasibility study whose primary endpoint showed nothing, converted into the most favourable available framing. A reader who meets the 77 percent before they meet the primary endpoint has been handed the conclusion in the wrong order.

Rung 8: the first independent look

In October 2026 Myeonghae Hyun, Yeseung Park, Seo Yoon Park and Ki Woong Kim published the first systematic review of gamma sensory stimulation in older adults, with a registered protocol and PRISMA reporting, searching four databases from 2016 to 2025. We graded it mixed this week.

Eight human studies exist. Across them, EEG confirms target engagement: the brain does follow the stimulus, with the strength varying by settings and by person. Controlled structural studies suggest possible slowing of hippocampal atrophy, ventricular enlargement and white-matter decline. Functional connectivity points the same way. The authors pooled hippocampal and ventricular volume change and then reported that those pooled estimates did not survive conservative sensitivity analysis. The fluid biomarker evidence reduces to exploratory immune findings from one uncontrolled study. And cognitive findings “did not establish a consistent benefit”.

Their summary is that human evidence for gamma entrainment is “promising but preliminary and hypothesis-generating”. That is also ours, and it is an independent group with no device to sell saying it.

Rung 9: HOPE, and how to judge it

The HOPE trial is the one to wait for. It randomised 673 participants across about 70 US sites to active or sham 40 Hz light and sound for an hour a day for roughly twelve months, and the registry describes the masking as quadruple: participants, study partners, investigators and outcome assessors all blinded. Primary completion was 22 June 2026 and the study completed on 17 August 2026. As of 26 October 2026 the registry shows no results posted and no results submission, and the record has not been updated since 9 September.

Two things to hold on to when it does report.

The primary endpoint is a composite called iADMARS, which integrates the MMSE and the ADCS-ADL. Those are two of the three secondary measures that reached nominal significance in OVERTURE. Building a pivotal trial’s primary endpoint from the signals a small predecessor produced is ordinary, defensible drug-development practice, and it is also the reason the result will be decisive either way: this is the hypothesis OVERTURE generated, now tested properly, with the blinding and the numbers to settle it.

Judge the primary endpoint, not the secondaries. If iADMARS separates from sham, that is the first real evidence that 40 Hz stimulation helps people with Alzheimer’s disease, and it will deserve to be reported as such. If it does not, expect the same shape of coverage that followed OVERTURE: a subgroup, a secondary measure, a time-saved reframing. We will grade whichever happens against the same rubric we used here.

What this means if you are considering a device

We do not give medical advice, and Alzheimer’s disease is the last place to start. Two factual points stand on their own.

The device used in the human trials is a prescription investigational device delivering light and sound at a verified 40 Hz with EEG confirmation that the brain is following it. A 40 Hz tone on a streaming service is not that: it has no light component, no entrainment verification, and the human evidence such as it is comes from combined audiovisual stimulation delivered for an hour a day under supervision. The mouse work found that light plus sound did more than either alone.

And the honest state of the evidence, today, is that the best-designed human trial to report so far did not separate from its sham on the measure it was judged by, and the first independent review found no consistent cognitive benefit across everything published. The safety record is good. The efficacy question is open, and the answer may already exist in a database somewhere, unpublished.

Sources

• Iaccarino and colleagues, “Gamma frequency entrainment attenuates amyloid load and modifies microglia”, Nature 540, 2016. doi:10.1038/nature20587
• Martorell and colleagues, “Multi-sensory gamma stimulation ameliorates Alzheimer’s-associated pathology and improves cognition”, Cell 177, 2019. doi:10.1016/j.cell.2019.02.014
• Murdock and colleagues, “Multisensory gamma stimulation promotes glymphatic clearance of amyloid”, Nature, 2024. doi:10.1038/s41586-024-07132-6
• Wang and colleagues, “Long-term effects of forty-hertz auditory stimulation as a treatment of Alzheimer’s disease: insights from an aged monkey model study”, PNAS, January 2026. doi:10.1073/pnas.2529565123
• Chan and colleagues, “Gamma frequency sensory stimulation in mild probable Alzheimer’s dementia patients: results of feasibility and pilot studies”, PLOS ONE, 2022. doi:10.1371/journal.pone.0278412
• Hajós and colleagues, “Safety, tolerability, and efficacy estimate of evoked gamma oscillation in mild to moderate Alzheimer’s disease”, Frontiers in Neurology 15, 2024 (OVERTURE, NCT03556280). doi:10.3389/fneur.2024.1343588
• Kern and colleagues, “Time saved in activities of daily living and whole-brain volume: post hoc analysis of a randomized feasibility trial of gamma oscillation treatment”, Alzheimer’s & Dementia: TRCI 11, 2025. doi:10.1002/trc2.70118
• Chan and colleagues, two-year open-label extension, Alzheimer’s & Dementia, 2025. Our summary, doi:10.1002/alz.70792
• Hyun, Park, Park and Kim, “Sensory gamma entrainment in older adults with or without cognitive impairment: a systematic review and exploratory meta-analysis”, GeroScience, 2026. Our summary, doi:10.1007/s11357-026-02564-6
• HOPE study record, ClinicalTrials.gov NCT05637801, checked 26 October 2026.
• Cognito Therapeutics, clinical studies page, read 26 October 2026.