The Last Biomarker Standing: A Nerve-Damage Protein Predicts Who Lives Past 100 — and Who Stays Sharp

RapAdmin · June 29, 2026, 3:08am

In nearly 500 Japanese centenarians, a blood marker of raw neuronal injury (neurofilament light chain, NfL) — not the famous Alzheimer’s proteins amyloid and tau — was the strongest predictor of both worse cognition and earlier death. At the extreme edge of the human lifespan, generic brain-and-body wear appears to matter more than classic Alzheimer’s pathology.

For three decades, the story of brain aging has been dominated by two villains: amyloid-beta and tau. Clear them out, the thinking went, and you protect the aging mind. But the very oldest humans keep breaking that script — many centenarians carry heavy amyloid and tau burdens yet stay cognitively intact, while others decline without them.

A new study from Keio University’s Center for Supercentenarian Medical Research, published in JAMA Network Open, leans into that paradox. Drawing on two of the world’s longest-running centenarian cohorts, the team measured three blood biomarkers in 495 people aged 100 to 116 and tracked who died over a follow-up extending up to 17 years (95.5% died during the study). They asked a deceptively simple question: among amyloid (the Aβ42/40 ratio), phosphorylated tau (p-tau181), and neurofilament light chain, which one actually tracks cognition and survival when you’re already past 100?

The answer was consistent and a little surprising. NfL — a structural protein released into the blood when axons are damaged, regardless of the cause — was the marker that mattered. Higher NfL went with lower cognitive scores and, more strikingly, with a meaningfully higher risk of dying. Each standard-deviation rise in NfL was associated with roughly a 36% higher mortality hazard, the largest effect of any biomarker tested. The amyloid ratio showed a weaker link to cognition, and after full statistical adjustment, neither amyloid nor p-tau181 predicted death at all.

The “big idea” is a shift in framing. NfL is not specific to Alzheimer’s; it rises with vascular injury, inflammation, kidney decline, and general neurodegeneration. Its dominance here suggests that at the limits of human life, what kills and dims centenarians is less a single disease and more the cumulative, systemic erosion of the nervous system — what the authors tie to falling kidney function, anemia, and low albumin. NfL appears to be a readout of that whole-organism decline.

A crucial caveat: studying only people who already reached 100 introduces “collider” bias — you’ve conditioned on survival itself — so these associations cannot be cleanly generalized to younger adults, and they are correlations, not proof of cause. Still, as a candidate “summary statistic” of late-life biological frailty, NfL is now a serious contender.

Actionable Insights

The honest takeaway: this study tells you how to read the dashboard, not how to fix the engine. What the paper offers is a useful signal you can measure.

NfL works as a tracking marker. Of the three brain-related blood proteins tested, one called neurofilament light chain (NfL) — a protein that leaks into the blood when nerve cells are damaged — was the strongest predictor of who would die sooner. The study found that for every one standard-deviation increase in NfL (a standard deviation is just a standard “step size” that describes how spread out the values are in the group), the risk of dying rose by about 36%. When the researchers re-ran the analysis using only people with complete data, that risk climbed to roughly 70–80% per step. To put that in context, a 36% bump per step is a fairly strong signal for a single blood test — on par with well-known heart-health markers the same research group had flagged in earlier work.

The link to thinking ability is real but modest. Higher NfL also tracked with lower scores on a standard 30-point cognitive test (the Mini-Mental State Exam). But the effect was small: each one-step rise in NfL corresponded to losing only about one point on that 30-point scale. Statistically dependable, but too small to notice in any one person. The amyloid marker showed a similarly small effect.

How to think about it practically. NfL can be ordered commercially today. The key thing to understand is that it’s a general alarm, not an Alzheimer’s-specific one — it goes up with all kinds of nerve, blood-vessel, and inflammation-related stress. It also rises when kidney function drops and when someone is anemic or poorly nourished. So a high reading should prompt you to check kidney function, blood counts, and nutrition just as much as brain health — a single elevated value doesn’t automatically point to the brain.

Missing Data / What Would Move Confidence

Longitudinal NfL trajectories (not single timepoint); neuroimaging or autopsy correlation within this cohort; GFAP (a glial marker that often complements NfL); and external replication in non-Japanese centenarians. Until then, NfL here is a promising prognostic integrator, not a validated clinical tool — which the authors appropriately concede.

Source:

Open Access Paper: Biomarkers, Cognitive Function, and Mortality in Centenarians
Institution: Center for Supercentenarian Medical Research, Keio University School of Medicine (with Eisai-Keio Innovation Laboratory; Tokyo University of Agriculture and Technology)
Country: Japan
Journal: JAMA Network Open, Published Online: May 7, 2026
Impact Evaluation: JAMA Network Open carries a Journal Impact Factor of 11.7 in the newest release (June 2026, based on 2025 citation data; the prior 2024 JCR value was 9.7), therefore this is a High impact journal

RapAdmin · June 29, 2026, 3:13am

How to get the NfL test:

The NfL blood test went from research-only to widely orderable in the last few years. Here’s the practical landscape in the US, roughly cheapest to most expensive. Note these are the underlying reference labs (Labcorp, Quest, Mayo) plus the consumer-facing storefronts that resell their draws — the storefront is usually how you order without a doctor.

Provider / storefront	Underlying lab	Approx. cash price	Doctor order needed?
Ulta Lab Tests, Walk-In Lab, True Health Labs, Marek Health (DTC resellers)	Labcorp or Quest	~$150–$300	No (they provide the requisition)
iHLTH	Quest/Labcorp CLIA network	$350 flat, results 2–4 days	No
Labcorp (direct, or Labcorp OnDemand)	Labcorp (Simoa-based LDT)	~$200–$400 list; varies	Yes for direct; OnDemand may not
Quest Diagnostics	Quest LDT	~$200–$400	Usually yes
Mayo Clinic Laboratories (NFLP)	Mayo	clinician-ordered, billed through provider	Yes
Apollo Health “BrainScan,” other panels	bundles NfL + p-tau217 + GFAP	~$300–$600 for the panel	No

A few practical points that matter more for your use case than the price:

The single firm number I can confirm today is iHLTH at a flat $350 with no referral. The DTC marketplace prices (Ulta, Walk-In Lab, etc.) fluctuate and frequently run promotions, so treat the $150–$300 band as indicative rather than locked. If you want to anchor on the actual Simoa assay used in most of the centenarian/longevity literature, Labcorp and Quest both run Simoa/Quanterix-style platforms, and the DTC resellers route to one of those two — so ordering through a reseller gets you the same assay at a lower sticker price than walking into the lab directly.

Three interpretation caveats that bear directly on reading your own result, and that connect back to the paper:

NfL is strongly assay- and lab-dependent — absolute values are not interchangeable across Labcorp, Quest, Mayo, or any research cohort. If you plan to track it serially (the only really informative use for a healthy person), stay on the same lab and assay every time.

It’s heavily confounded by age, kidney function, BMI, and biotin. The paper itself showed NfL rising with age and falling with eGFR, albumin, and hemoglobin, and Labcorp explicitly flags that high-dose biotin interferes — stop biotin ≥72 hours before the draw, and interpret any elevation alongside an eGFR. A “high” NfL in someone with reduced kidney function may be partly renal, not neural.

Reference ranges are age-banded and still being standardized. A single value in a healthy person is hard to act on; the signal is in the trajectory or in a value far outside the age band.

If your interest is the broader brain-aging panel rather than NfL alone, the bundled p-tau217 + GFAP + NfL offerings give a more complete amyloid/inflammation/injury picture — though, consistent with the centenarian paper, NfL is the component most tied to general aging and mortality rather than amyloid-specific disease.

One note worth stating plainly: as a screening test in an asymptomatic person, NfL has no established action threshold — there’s no validated “if elevated, do X” pathway yet, which the study authors themselves flag. It’s most defensible as a longitudinal tracking marker or in the context of actual symptoms or known risk.

RapAdmin · June 29, 2026, 3:22am

What Measures and Levels Matter:

There’s an important honest answer up front: there is no single universal “safe” NfL number. Two things make absolute cutoffs unreliable — NfL climbs steadily with normal aging, and different labs/assays (Simoa/Quanterix, Roche, Siemens Atellica, Ella) report meaningfully different absolute values that can’t be compared head-to-head. So the field reads NfL as a percentile or Z-score within your own age band, not as a fixed threshold. Here’s what the numbers actually look like.

Normal (age-expected) upper limits in healthy people

These are blood (serum/plasma) upper reference limits — roughly the 95th–97.5th percentile of healthy people in each age group, from large normative studies using sensitive assays:

Age band	“Upper normal” (≈95th–97.5th %ile)
18–50	~10 pg/mL Oxford Academic
51–60	~15 pg/mL Oxford Academic
61–70	~20 pg/mL Oxford Academic
70+	~35 pg/mL Oxford Academic

A more granular US dataset that excluded people with kidney disease, stroke, prior heart attack, and obesity gives slightly higher decade-by-decade ceilings: 20s: ≤8.4, 30s: ≤11.4, 40s: ≤15.4, 50s: ≤20.8, 60s: ≤28.0, 70s: ≤37.9, 80+: ≤51.2 pg/mL. ResearchGate

The underlying trend is steady: the median rises from about 5.5 pg/mL under age 25 to about 42.5 pg/mL above age 80, increasing roughly 3% per year of age. So a value of 30 pg/mL is alarming in a 35-year-old and completely ordinary in an 82-year-old. Springer ResearchGate

How “elevated” and “high-risk” are read

Within the age band’s normal range (below ~95th percentile): considered age-appropriate.
Above the age-specific 95th percentile: flagged as elevated — worth investigating, but on its own it’s a yellow light, not a diagnosis.
Well above the band (roughly 2–3× the age-expected median, or a high Z-score): this is where risk associations get strong. In multiple sclerosis prognostic work, values above the 80th, 95th, and 99th percentiles each predicted progressively worse outcomes — risk is continuous, not a single cliff.

For scale on the high end: in active neurological disease (ALS, frontotemporal dementia, acute injury), values commonly run into the dozens to several hundred pg/mL, far above any age band.

What this looks like at the extreme — your centenarian paper

The Shikimoto study is a useful illustration of how detached these numbers become at the limits of life. Its centenarians averaged 114.6 pg/mL (SD 108.4), ranging from 12.5 up to 1,239.5 pg/mL — values that would be wildly abnormal in a 50-year-old are simply the baseline at 104. That’s exactly why the authors used Z-scores rather than raw cutoffs: the mortality signal (HR 1.36 per 1-SD step) is about being high relative to your centenarian peers, not about crossing an absolute line.

Three caveats that change interpretation

Use your reporting lab’s own age-banded reference range. A Labcorp Simoa value and a Siemens Atellica value are not interchangeable; one NHANES dataset using the Atellica platform reported a population mean near 16–18 pg/mL regardless of age, which differs from the Simoa-based numbers above. Match the cutoff to the assay. PubMed Central
A single high value isn’t specific to the brain. NfL rises with reduced kidney function, recent stroke, heart attack, and other systemic stress — the same pattern the centenarian paper showed (NfL inversely tracking eGFR, hemoglobin, albumin). An elevated reading should prompt a look at kidney function before assuming a neurological cause.
There’s no validated action threshold for an asymptomatic person. Even a value above the 95th percentile doesn’t map to a defined treatment pathway in a healthy individual. For longevity tracking, the trajectory over repeated draws on the same assay is far more informative than any one number against a cutoff.

Yoo · June 29, 2026, 12:53pm

Super important biomarker.

The nervous system aging seems to be a major bottleneck in life/healthpan and not something we have decent tools to manage.

For example, resistance training has great effect on the muscular system but motor neuron loss is still not prevented.

This marker appears general (detects all kind of nervous system damage), stable within age groups and shows a monotonous increase over time. What is describes makes mechanistic sense.