What Blood Proteomics Says About Menopause and Brain Aging

A study published in Nature Medicine on 22 September 2026 measured blood proteins across the menopause transition and linked them to later cognitive decline¹. The finding worth understanding is narrower and more interesting than the headline. Sixteen proteins moved with menopause stage, and they tracked hormone levels better than they tracked age.

This matters for anyone measuring their own plasma proteome, because it is a clean example of a signal that chronological age hides.

What was measured

The discovery cohort was small and unusually well characterized. Eighty women aged 43 to 58 were staged by STRAW+10 criteria into premenopausal, perimenopausal and postmenopausal groups. Thirty-six age-matched men were included for comparison. Serum went onto a NULISAseq central nervous system panel covering 118 proteins.

Staging matters more than sample size here. Most menopause research sorts women by age or by self-report, which blurs the transition badly. STRAW+10 uses bleeding patterns and hormone levels, so the groups are separated by where a woman is in the transition rather than by how old she is.

Sixteen proteins were elevated in postmenopausal relative to premenopausal women after adjusting for age. They fall into four groups:

• Inflammatory: CCL13, IL-12p70, CXCL1 and CCL26, and also CCL2, CCL11, TEK and CD63
• Synaptic and neuronal: CNTN2, ACHE, SNAP25 and BDNF
• Metabolic: IGFBP7, IGF1R
• Alzheimer’s biology: BACE1, p-tau231

The authors ran principal components analysis on those sixteen and used the first component as a single menopause proteomic score. That component explained 28.7% of the variance among them.

Why the score is not just age

This is the part worth dwelling on. In these women, the menopause score correlated with age at $r = 0.36$, which looks like an age effect. After adjusting for menopause stage the association disappeared and reversed sign, to $\beta = -0.14$ with a confidence interval spanning zero.

The unadjusted effect sizes make the comparison concrete. Relative to premenopausal women, perimenopausal status carried a 1.36 standard deviation higher score and postmenopausal status a 3.25 standard deviation higher score. Each additional year of age carried 0.24 standard deviations.

So moving through the transition is worth roughly thirteen years of chronological age on this measure. In the men, age did not correlate with the same composite at all.

The score also tracked estradiol, follicle-stimulating hormone and progesterone. It did not track testosterone, sex-hormone binding globulin or DHEAS. Several of the brain-relevant proteins were most strongly tied to FSH, including BACE1, ACHE and CNTN2.

What replicated, and how well

The discovery cohort has eighty people in it. The validation does most of the work.

The authors took 2,814 age-matched pre- or perimenopausal and postmenopausal women aged 45 to 60 from the UK Biobank, measured on the Olink platform rather than NULISAseq. In that sample 1,146 proteins were elevated and 154 were reduced in postmenopausal women. The largest hits were markers of ovarian hormone signalling, PAEP and FSHB.

Thirteen of the original sixteen proteins were present on both platforms. Nine replicated clearly. Effect size concordance across the two platforms was $r = 0.39$, which is moderate rather than strong. Different assays measuring nominally the same protein often disagree this much, and that is a limitation of cross-platform proteomics generally rather than of this study.

The cognitive and dementia results

Four independent cohorts of older women, 11,925 in total, were used to ask whether the score predicts anything later in life.

In the Alzheimer’s Disease Neuroimaging Initiative, 673 women followed for about five years, a higher baseline score predicted steeper cognitive decline ($\beta = -0.30$). A smaller cohort at UCSF showed the same direction in 100 women. A cross-sectional sample of 93 women showed lower global cognitive performance with higher scores.

The dementia incidence result needs reading carefully. Among 11,059 UK Biobank women followed a mean of 15.7 years, a higher menopause score was associated with Alzheimer’s dementia at a hazard ratio of 1.15, with a confidence interval of 1.01 to 1.31 and 219 events. The lower bound sits just above one.

The same analysis found no significant association with frontotemporal dementia, vascular dementia, or all-cause dementia. All-cause dementia had the most events, 387, and returned a hazard ratio of 1.08 that did not reach significance.

A hazard ratio of 1.15 on one dementia subtype, with a confidence interval that nearly touches one, is a real but small signal. We would describe this as consistent with the hypothesis rather than as evidence that the score predicts who gets dementia.

The finding that complicates the story

Men had higher menopause proteomic scores than women overall, after adjusting for age.

That sounds like a contradiction and is not. The score is a composite of inflammatory, synaptic and metabolic proteins that rise for many reasons. What the study shows is that within women the composite moves sharply across the menopause transition. It does not show that the composite is specific to menopause, and an absolute value on it means little without knowing who is being compared to whom.

Anyone reading a proteomic panel should hold that distinction. A score built to detect a within-person or within-group change is not automatically a marker you can interpret on its own.

Symptoms

In 89 older women with a mean age of 69, a history of hot flashes was associated with a higher menopause proteomic score, as were sleep problems and total symptom count. Mood swings, night sweats, depression and sexual dysfunction were not significantly associated.

CCL2 was the only individual protein associated with vasomotor symptoms in both midlife and older women. That is the kind of narrow, repeated result that is more useful than a broad one.

What this means if you are measuring your own proteome

Three things follow for someone with access to their own plasma proteomics.

The transition is visible in blood, and it is not visible as age. If you are a woman in your forties or fifties measuring proteins, a panel interpreted against age-matched norms will miss a shift this size. Hormone measurements alongside the proteomics are what make the pattern legible.

Single timepoints are weak here. Every meaningful result in this study came from comparing groups or following people over time. One panel gives you a number with no reference point, and the score’s own behaviour in men shows why an absolute value is hard to read.

The proteins involved are mostly ordinary inflammatory and metabolic markers. Most are not exotic. CCL2, IL-12p70, IGF1R and BDNF appear on commercial panels. The signal is in how they move together. It is not in any one of them.

If a pattern in your own data raises a question about your health, take the data and the question to a clinician. This study describes associations in groups and does not support any decision about treatment.

Questions people also ask

Does menopause cause Alzheimer’s disease? This study does not show that and does not claim it. The authors note that all women with ovaries go through menopause while most do not develop dementia. The study itself is observational. They frame the transition as a window for observing early brain aging rather than as a cause.

What is a menopause proteomic score? It is the first principal component of sixteen blood proteins that were elevated in postmenopausal relative to premenopausal women in this study. Higher values mean a more postmenopausal-looking protein profile. It is a research measure and is not available as a clinical test.

Is p-tau231 a menopause marker? In this cohort p-tau231 was the only canonical Alzheimer’s blood biomarker that differed significantly by menopause stage. Phosphorylated tau 181, amyloid beta 40, amyloid beta 42, GFAP and neurofilament light did not. The authors state that the clinical meaning of small p-tau231 elevations in midlife is unclear.

Can I get this panel? Not as a clinical test. The discovery panel was NULISAseq, a research platform, and the validation used Olink. Many of the individual proteins are available on research proteomics panels, and the score itself has not been validated for individual use.

Did hormone therapy change the results? The authors ran the analysis stratified by hormone therapy history as a sensitivity check, and their findings held. The study was not designed to test whether hormone therapy alters the proteomic signature, and it does not support any conclusion about taking it.

Footnotes

1. Madeline Wood Alexander, Jennifer S. Rabin, Michelle Caunca, et al. Blood proteomics of menopause map to brain aging and dementia risk. Nature Medicine, 2026. https://doi.org/10.1038/s41591-026-04648-4 ↩