Another and more recently article on “protein signature” in blood that could predict healthy aging:
"In a new study, published Friday (Oct. 9) in the journal Immunity, scientists found that individuals with more memory cells that make granzyme K than cells that make granzyme B tend to have healthier aging.
In the future, the balance of cells making granzyme K and granzyme B in a person’s blood could be used as a test to determine “whether you’re on the right trajectory of aging or not,” study co-author Maxim Artyomov, an immunologist at Washington University in St. Louis, told Live Science."
AI generated summary:
Your T-cell mix may track how you’re aging, but the test behind the claim hasn’t been run
Bottom line
As people age, a group of memory immune cells (CD8 effector memory T cells) tends to tilt one of two ways. Some people end up dominated by cells that make granzyme K, and they tend to have fewer chronic conditions. Others end up dominated by cells that make granzyme B, and that tilt goes with CMV infection, autoimmune disease and comorbidity. In 48,620 UK Biobank participants, a blood-protein estimate of the granzyme B side predicted modestly higher death rates and more diabetes, kidney failure, liver disease and sepsis over about 17 years. There’s nothing to act on yet: no clinical test exists and no intervention has been tried. The long-term link rests on an indirect estimate of half the ratio, adjusted only for age and sex.
The study in plain English
Part 1: an immune atlas (one snapshot in time). The authors pooled single-cell sequencing of blood from 2,609 “ostensibly healthy” adults aged 18–97, across 8 cohorts and about 12 million cells. Participants were mostly of European and Asian ancestry. The most reproducible aging signal was thymic decline: the loss of new and naive T cells. Among older people, the cells split toward the granzyme K or granzyme B side.
The ratio itself barely moves with age. What drives it down is:
- CMV infection, the strongest driver
- lupus
- comorbidity, though that comparison rests on only 23 older people with chronic conditions in the new King’s College London cohort
Part 2: UK Biobank (followed over time, but indirect). UK Biobank has no single-cell data. So the authors trained a model on 329 people to estimate the share of granzyme B cells from blood proteins. In testing on people it hadn’t been trained on, that estimate explained only about a quarter of the real variation. They then applied it to 48,620 people.
Results. Among 60–70-year-olds, about 27 in 100 people in the top quarter of the score died over roughly 16 years, versus about 19 in 100 in the bottom quarter. That’s about 8 extra deaths per 100, read off the survival curves. The age- and sex-adjusted HR was 1.43. For scale, being male in the same model carried an HR of 1.66. The score also predicted:
- type 2 diabetes (HR 1.74)
- gout (1.80)
- kidney failure (1.60)
- liver disease (1.74)
- sepsis (1.50)
There was no signal for dementia, Alzheimer’s or Parkinson’s.
Why it matters
Should you care? The biology is genuinely interesting. Immune aging looks like two diverging paths rather than one slope, and the paper finds a measurable axis with a likely driver (CMV). It’s a sharper version of the old Swedish “immune risk profile” (an inverted CD4:CD8 ratio plus CMV). As a predictor, though, nobody has shown it adds anything. The authors didn’t test it against hs-CRP, kidney function, HbA1c, or the existing UK Biobank proteomic aging clock, which already predicts death and 18 diseases (Argentieri 2024).
Actionability: none.
-
No test. Standard clinical flow cytometry can’t tell these cells apart. Measuring them takes single-cell sequencing or specialized staining.
-
No intervention. Nothing has been shown to shift the ratio, and nobody has shown that shifting it would change outcomes.
-
No handle on the main driver. There’s no CMV vaccine, and treating latent CMV in healthy people isn’t indicated.
-
Nothing to add to a biomarker panel. The protein score tracks CRP and IL-6. You can already measure hs-CRP.
Don’t read “more granzyme K is good” literally. In tissue, granzyme K–producing CD8 cells drive inflammation: granzyme K activates the entire complement cascade and is implicated in rheumatoid arthritis (Donado, Nature 2025). In mice, these same authors earlier tagged granzyme K cells as the cells behind age-related inflammation. The “good” signal applies only to the ratio in circulating blood. Anyone selling a granzyme K product off this paper is selling nonsense.
Caveats and blind spots
-
The headline claim isn’t what was tested over time. UK Biobank never measured the ratio. It measured a protein proxy for the granzyme B side only. The top proxy proteins (GNLY, KLRD1, GZMH) are also abundant in NK cells, so the score may be a general “cytotoxic and inflammatory load” reading rather than a T-cell measurement.
-
The adjustment is thin. The models adjust only for age and sex: not smoking, BMI, alcohol, deprivation, kidney function or existing disease. The outcomes that light up (diabetes, gout, liver and kidney disease) are classic metabolic outcomes. Much of the signal could simply mean that sicker, more inflamed people die sooner.
-
Reverse causation. The score was higher in people who already had long-standing illness. The mortality analysis didn’t exclude existing disease. Artyomov himself told Live Science the ratio “might be just a consequence, not a cause.”
-
CMV wasn’t adjusted for in the mortality model. The CMV–mortality link is itself contested. Some UK cohorts found one (Savva 2013); a pooled analysis of about 10,000 older adults found none (Chen, JID 2021).
-
Other gaps.
- Nobody was sampled twice, so we don’t know whether a person’s ratio is stable.
- There’s no medication data.
- African and Hispanic ancestry are underrepresented.
- UK Biobank skews healthy and mostly white British.
-
Conflict of interest. Five authors, including Artyomov, are named on a provisional patent application “involving matters related to this study.” That’s useful context for his “future test” quote. Funding came from the nonprofit Aging Biology Foundation and NIH, with no industry sponsor.
Strengths: It’s peer reviewed in Immunity and large. Findings counted only if they replicated in at least 5 of the 8 cohorts. Data and code are open, and the null results were reported.
Verdict: 3/5 (interesting but preliminary)
The atlas alone would earn a 4: it’s large, replicated across cohorts and transparent. The clinical claim alone is closer to a 2. It rests on a weak proxy for half the ratio, adjusted only for age and sex, never compared against existing markers, and promoted by authors with a pending patent. Overall it’s real biology pointing in a direction, with nothing to act on yet.
Sources: