Scientists Identify Universal Genetic Markers of Aging and Mortality Across Mammalian Species

In May 2026, a team led by Alexander Tyshkovskiy and Vadim Gladyshev at Harvard Medical School published what may be the most comprehensive molecular atlas of mammalian aging to date. Their study, published in Nature, integrated more than 11,000 transcriptomes — snapshots of which genes are active in a given cell at a given time — across more than 25 tissue types in four mammalian species: mice, rats, macaques, and humans [1]. The result: a set of gene expression signatures that the authors describe as "universal transcriptomic hallmarks" of aging and mortality, conserved across species, organs, and cell types.

The claim is ambitious. If correct, it means the molecular machinery of aging is not merely similar across mammals but fundamentally shared — and measurable. The implications stretch from basic biology to insurance law, from pharmaceutical development to the ethics of predicting when someone will die.

What the Study Actually Found

The research team conducted RNA sequencing (RNA-seq) on tissues from mice subjected to 20 different compound treatments in the National Institute on Aging's Interventions Testing Program (ITP), then integrated that data with over 4,000 additional rodent tissue samples representing responses to genetic, pharmacological, and dietary interventions with established survival data [2]. They identified aging-associated gene expression changes that were conserved not just across tissues within a single species but across species boundaries.

Two genes stood out as particularly robust markers: CDKN1A (which encodes the cell cycle inhibitor p21, a key mediator of cellular senescence) and LGALS3 (encoding galectin-3, a protein involved in inflammation and fibrosis). Protein levels of both were independently associated with mortality and multimorbidity in the UK Biobank, a prospective cohort of approximately 500,000 participants [1].

Through network analysis, the team identified and annotated 26 co-regulated gene modules associated with aging and longevity across tissues. These modules mapped onto distinct functional pathways: inflammatory response, mitochondrial function, lipid metabolism, and extracellular matrix organization [1]. The researchers then built "module-specific clocks" — predictive models that capture aging-associated changes within each functional component separately, rather than collapsing aging into a single number.

"The same genes are associated with aging in, for example, liver and heart in rats and humans," Tyshkovskiy told Scientific American [3]. The team also released a public tool called TACO (Transcriptomic Age Calculator Online) to allow other researchers to test whether candidate interventions shift transcriptomic age [3].

How This Compares to Epigenetic Clocks

The field of aging biomarkers has been dominated for over a decade by epigenetic clocks — algorithms that estimate biological age from DNA methylation patterns at specific genomic sites. Steve Horvath's original 2013 clock, and subsequent iterations like GrimAge and PhenoAge, have become the standard tools in both research and a growing commercial market for biological age testing.

The transcriptomic approach differs in a critical way. As David Sinclair of Harvard noted regarding the new study, the researchers "developed transcriptomic clocks that don't just estimate age; they measure progressive loss of cellular function" [3]. Epigenetic clocks, by contrast, identify statistical associations between methylation patterns and chronological age but offer limited insight into why aging occurs.

Source: OpenAlex

Data as of Jan 1, 2026CSV

Research on epigenetic clock aging has grown steadily, with over 26,700 papers published since 2011 and a peak of 4,203 in 2025 [4]. But a 2020 study in eLife found that correlations between different biological aging indicators — including epigenetic and transcriptomic clocks — were small, with all correlation coefficients below 0.2 [5]. This suggests that different clock types capture largely non-overlapping biological variance. Rather than rendering epigenetic clocks obsolete, the transcriptomic hallmarks appear to measure a complementary dimension of aging biology.

A transcriptomic clock called BiT Age, developed independently, demonstrated median absolute errors of 5.55 years in human prefrontal cortex samples, approaching what its developers described as the theoretical accuracy limit for transcriptome-based age prediction [6]. The Tyshkovskiy study extends this work by linking transcriptomic signatures not just to chronological age but to mortality risk and intervention response.

Source: OpenAlex

Data as of Jan 1, 2026CSV

The broader field of aging biomarker transcriptomics has seen explosive growth, with nearly 119,000 papers published and a peak of 23,842 in 2025 [4] — roughly five times the volume of epigenetic clock research, reflecting the increasing recognition that gene expression data captures functional aging processes that methylation alone may miss.

The Longevity Outlier Problem

The study's "universal" label faces its stiffest test from species that break the normal rules of mammalian aging. The naked mole rat (Heterocephalus glaber) lives more than 35 years — over ten times the lifespan of a comparably sized house mouse — and shows negligible senescence, meaning its mortality rate does not increase with age in the way predicted by standard aging models [7]. Bowhead whales (Balaena mysticetus) can exceed 200 years, with unique mutations in DNA repair genes including ERCC1, ERCC3, and histone deacetylases HDAC1 and HDAC2 [8].

The Tyshkovskiy study analyzed four species: mouse, rat, macaque, and human. Naked mole rats, bowhead whales, and Brandt's bats — the organisms most likely to stress-test a "universal" aging framework — were not included. This is a significant limitation. The four species studied, while spanning roughly 90 million years of evolutionary divergence, are all relatively conventional agers. Whether the 26 co-regulated modules hold in species with radically different aging trajectories remains an open question.

Comparative transcriptomic work across 26 mammalian species has found that genes negatively correlated with maximum lifespan are primarily involved in energy metabolism and inflammation, while genes positively correlated with longevity show enrichment in DNA repair, microtubule organization, and RNA transport [9]. Some of these pathways overlap with the Tyshkovskiy modules, but the degree of concordance has not been formally tested.

The naked mole rat's longevity appears to involve mechanisms — such as unique modifications to the HAS2 gene that produce high-molecular-mass hyaluronan, conferring cancer resistance [7] — that are species-specific rather than shared across mammals. Convergent amino acid changes in longevity-associated genes have been found between distantly related species like naked mole rats and killer whales [8], suggesting that evolution has found multiple, independent solutions to the problem of aging.

Cause, Correlation, and the Limits of Observational Genomics

The study is observational. It identifies gene expression patterns that change with age and correlate with mortality, but correlation is not causation. Do these transcriptomic signatures drive aging and death, or are they downstream consequences — molecular smoke rather than fire?

The researchers addressed this partly through the ITP intervention data: compounds that extended lifespan in mice also shifted transcriptomic signatures in the predicted direction. The study also showed that rejuvenation induced by heterochronic parabiosis (surgically joining the circulatory systems of young and old mice), early embryogenesis, and cellular reprogramming reversed aging-associated transcriptomic changes [2]. These experimental manipulations provide stronger evidence for a causal role than simple age-correlation studies, though they fall short of the gold standard of genetic knockout experiments targeting individual module genes.

Gladyshev himself acknowledged a key constraint: "Currently in humans, we don't have a single intervention that extends lifespan" [3]. Without the ability to experimentally manipulate human lifespan and observe whether transcriptomic signatures change accordingly, the causal chain remains partially unresolved.

The Skeptics: Is "Universal" Aging a Category Error?

Not all biogerontologists accept the premise that aging has a universal molecular mechanism. A 2016 paper in Aging argued that "the process of aging is not universal and its mechanisms have not been widely conserved among species," concluding that because the same forces can disrupt organismal functions in different ways depending on the organism, "no universal mechanism of aging can exist" [10].

A 2026 paper in Biogerontology reinforced this position, arguing that aging is "not a disease" and challenging the assumption of a uniform pathological pattern across species [11]. The evolutionary argument runs as follows: aging is not directly selected for by evolution but arises as a byproduct of declining selection pressure after reproductive age. Because different species face different ecological pressures and have different life histories, the specific molecular failures that accumulate with age should differ between lineages.

The counterargument from the Tyshkovskiy team is empirical: the data show conservation. But critics would note that finding shared changes across four mammalian species — all of which are placental mammals with relatively similar body plans — may reflect shared physiology rather than a universal aging program. Testing the framework against marsupials, monotremes, or the full diversity of bats and cetaceans would provide a more rigorous test.

The question of phylogenetic non-independence also applies. Mouse and rat are closely related rodents; human and macaque are both primates. Finding that two rodents and two primates share aging signatures does not establish the same level of universality as finding conservation across, say, a bat, an elephant, a whale, and a shrew. The statistical thresholds for designating a marker "universal" have not been publicly evaluated with phylogenetic correction methods that account for shared evolutionary history.

Who Funds Aging Research — and What Gets Studied

The landscape of aging research funding shapes which questions get asked. The U.S. National Institute on Aging (NIA) allocated approximately $4.4 billion in FY2025, making it one of the largest public funders of geroscience [12]. But NIA funding is spread across Alzheimer's disease, basic biology of aging, and behavioral research, with only a fraction directed at cross-species transcriptomics.

Source: Public reporting and press releases

Data as of May 1, 2026CSV

Private capital has flooded into the field. Altos Labs raised $3 billion in 2022 — the largest initial funding round for any biotech startup — to pursue life extension through partial epigenetic reprogramming using Yamanaka factors [13]. The Hevolution Foundation, backed by Saudi Arabia's Public Investment Fund, has committed $1 billion annually to bridge the gap between preclinical aging research and clinical trials [12]. Calico, Alphabet's longevity subsidiary, has invested an estimated $1.5 billion.

These funding sources shape research priorities. Altos Labs focuses on cellular reprogramming, not transcriptomic biomarkers. Hevolution funds broad geroscience but emphasizes translational work — moving findings toward clinical application. The Tyshkovskiy study, conducted at Harvard Medical School and Brigham and Women's Hospital, relied primarily on academic infrastructure and publicly available datasets like the ITP.

What gets left out is telling. Species like naked mole rats and bowhead whales — the most informative for testing universality claims — require specialized collection infrastructure and produce smaller sample sizes. The four species in this study (mouse, rat, macaque, human) are the ones most readily available in biomedical research, not necessarily the ones best suited to answering the evolutionary question.

From Markers to Medicine: Translational Timeline

The gap between identifying transcriptomic signatures and producing a validated clinical tool is substantial. The TACO calculator is a research tool, not a diagnostic. Moving to a human clinical application would require:

1. Analytical validation: Demonstrating that the assay reliably measures the intended transcripts across platforms and laboratories.
2. Clinical validation: Showing that transcriptomic age predicts clinically meaningful outcomes (mortality, disease onset, functional decline) in diverse human populations.
3. Clinical utility: Proving that acting on the test result improves patient outcomes — the highest bar and the one most aging biomarkers have not cleared.

The UK Biobank validation of CDKN1A and LGALS3 protein levels provides early clinical validation evidence, but the full transcriptomic clock has not been tested prospectively in a clinical trial. Realistic timelines for FDA clearance of an aging diagnostic range from 5 to 15 years, depending on the regulatory pathway chosen [14].

On the commercial side, Insilico Medicine spun off Deep Longevity in 2020 to commercialize AI-based aging clocks, and holds patents on deep transcriptomic markers of biological aging (US Patent 10,325,673) [15]. Whether the Tyshkovskiy findings will generate new patent filings or licensing agreements has not been publicly disclosed.

Legal Gaps: When Mortality Prediction Meets Insurance

A validated mortality-prediction tool built on transcriptomic markers would collide directly with existing gaps in genetic discrimination law. In the United States, the Genetic Information Nondiscrimination Act (GINA) of 2008 prohibits discrimination based on genetic information in health insurance and employment — but explicitly does not cover life insurance, disability insurance, or long-term care insurance [16]. Insurers in these markets can legally use genetic and biomarker information to set premiums or deny coverage.

Source: World Bank Open Data

Data as of Dec 31, 2024CSV

Life expectancy varies substantially across countries — from 84.4 years in Switzerland to 54.6 years in Nigeria [17] — and populations with lower baseline life expectancy could face disproportionate consequences from mortality-prediction tools that flag them as higher-risk. Biomarkers detectable up to 20 years before symptoms appear create a window in which individuals can be penalized for conditions they have not yet developed [18].

Fewer than half of U.S. states have laws extending genetic discrimination protections beyond GINA's scope [16]. California's CalGINA (2011) is the most comprehensive, covering emergency medical services, housing, and education in addition to insurance.

In the European Union, the GDPR classifies genetic data as "special category data" under Article 9, prohibiting processing without explicit consent [19]. However, insurance-specific genetic discrimination laws vary by country: Austria, Belgium, Denmark, France, Ireland, Poland, and Portugal prohibit insurers from using genetic information, while Germany, the Netherlands, and Switzerland allow its use for policies above certain coverage thresholds [20].

In Asia, protections are even more fragmented. Japan's Act on the Protection of Personal Information covers genetic data broadly, but lacks specific prohibitions on insurance discrimination based on aging biomarkers. China's Personal Information Protection Law (2021) classifies genetic data as sensitive but provides limited enforcement mechanisms in the insurance context.

The informed consent question is equally unresolved. If a transcriptomic aging test reveals a biological age significantly higher than chronological age, does the patient have a right not to know? Current U.S. informed consent frameworks for genetic testing were designed for disease-specific results, not for broad mortality-risk assessments that implicate dozens of biological pathways simultaneously.

What Comes Next

The Tyshkovskiy study establishes a framework, not a finished product. Its core contribution — the 26 co-regulated modules and the demonstration that transcriptomic aging signatures are conserved across four mammalian species — provides a new scaffold for aging research. The TACO tool gives the field a standardized way to test whether interventions shift molecular aging signatures.

But the "universal" claim requires further testing against the full diversity of mammalian aging strategies. The study's four species, while carefully chosen, represent a narrow slice of mammalian evolution. The genes CDKN1A and LGALS3 may be genuinely universal markers, or they may be features of a shared placental-mammal aging program that breaks down in species that have evolved fundamentally different solutions to the problem of growing old.

The translational path from transcriptomic signature to clinical diagnostic is long and uncertain. The regulatory path from clinical diagnostic to insurable risk factor is shorter — and less well guarded.

Data and charts in this article draw on World Bank, OpenAlex, and publicly reported funding figures. All statistics are cited to primary sources.