MORGAN LEVINE AND PHENOAGE: THE 2026 STATE OF EPIGENETIC CLOCK RESEARCH

In 2013, Steve Horvath published a DNA methylation-based clock capable of estimating chronological age across more than 50 tissue types with remarkable accuracy. The science community was impressed — and quickly identified the core limitation. Predicting how old someone's cells are in calendar years is not the same as predicting how fast they are ageing, or when they are likely to die. For longevity medicine, the calendar is the wrong frame of reference entirely.

Dr Morgan Levine, then at Yale University and now leading research at Altos Labs, developed PhenoAge to fix that problem. Instead of training a methylation clock on chronological age, she trained it on phenotypic age — a composite mortality-risk metric derived from nine blood biomarkers routinely measured in clinical practice. The result was an epigenetic clock that does not tell you how old your cells appear to be, but how old they are likely to make you. That distinction has driven five years of follow-on research and positioned PhenoAge as the dominant biological age readout in longevity medicine, clinical trials, and now cellular reprogramming research at one of the best-funded longevity science institutes in the world.

What Is PhenoAge and How Does It Differ from Earlier Epigenetic Clocks?

PhenoAge is a second-generation epigenetic clock published by Morgan Levine, Ake Lu, Francine Grodstein, and colleagues in Aging Cell in 2018. The clock is built on DNA methylation data from 513 CpG sites — specific cytosine-phosphate-guanine dinucleotide positions across the genome where methylation levels change predictably as cells age. What distinguishes it from first-generation clocks is what those CpG sites were selected to predict.

Steve Horvath's 2013 Horvath1 clock used 353 CpG sites selected to correlate with chronological age — the number of years since birth. Hannum's 2013 blood-specific clock used 71 CpG sites similarly calibrated to calendar years. Both clocks can detect when someone is biologically older than expected for their age (a phenomenon called epigenetic age acceleration), but their predictive power for disease and mortality is modest because chronological age is itself only a proxy for what we actually care about: health and longevity.

Levine's methodological innovation was to build the training target differently. She first used NHANES III data to construct a composite phenotypic age metric from nine routine blood biomarkers — albumin, creatinine, glucose, CRP, lymphocyte percentage, mean corpuscular volume, red cell distribution width, alkaline phosphatase, and total white blood cell count — optimised using a Gompertz mortality model to maximally predict 10-year all-cause mortality. She then trained the methylation clock on that phenotypic age metric rather than on chronological age. The result is a clock whose epigenetic acceleration scores are significantly stronger predictors of mortality, cancer incidence, and multimorbidity than earlier clocks, even after controlling for chronological age.

The Nine Biomarkers Behind PhenoAge: What the Clock Is Actually Measuring

Understanding PhenoAge requires understanding the nine biomarkers underpinning its phenotypic age training target, because those markers reveal what biological processes the clock is sensitive to. Each one was selected empirically for its mortality-predictive power in NHANES data, and together they represent a cross-section of metabolic, renal, hepatic, immune, and haematological function.

Albumin reflects nutritional status and liver synthetic function — low albumin is a powerful independent predictor of mortality in multiple population cohorts. Creatinine captures kidney filtration capacity (GFR), which declines predictably with ageing and is an independent cardiovascular risk factor. Fasting glucose represents metabolic control and insulin sensitivity — chronically elevated glucose drives glycation, oxidative stress, and mitochondrial dysfunction. C-reactive protein (CRP) captures systemic inflammatory burden, which as discussed in our post on inflammation markers and blood tests is one of the strongest modifiable predictors of all-cause mortality.

The haematological markers — lymphocyte percentage, mean corpuscular volume (MCV), red cell distribution width (RDW), and total white cell count — capture immune ageing and erythropoietic health. RDW in particular has emerged as a surprisingly powerful mortality predictor; elevated RDW reflects impaired red cell production homeostasis and is associated with iron deficiency, B12 deficiency, chronic inflammation, and oxidative stress on erythrocytes. Alkaline phosphatase, finally, is a liver and bone enzyme that rises in cholestasis, liver disease, and hyperparathyroidism. The composite of these nine markers, when analysed through the Gompertz mortality model, generates a phenotypic age that in NHANES data outperformed chronological age at predicting all-cause mortality by a significant margin.

PhenoAge vs Horvath Clock: Why Mortality Prediction Outperforms Chronological Age Prediction

The head-to-head validation of PhenoAge against Horvath1, Hannum, and other first-generation clocks is now well-documented across multiple independent datasets. In the original 2018 Levine et al. paper, PhenoAge acceleration (the gap between epigenetic PhenoAge and chronological age) was significantly associated with all-cause mortality, cause-specific mortality (cancer and cardiovascular), physical function decline, inflammation, and comorbidity — and it outperformed Horvath1 acceleration on every one of these endpoints.

The mechanism behind this improved predictive power is conceptually straightforward. Chronological age is a noisy proxy for biological deterioration because people age at dramatically different rates. Two 60-year-old individuals can differ by 10-15 years in phenotypic age depending on their lifestyle, genetics, and disease history. A clock trained on chronological age will estimate both as 60; a clock trained on mortality risk will capture that divergence. PhenoAge essentially encodes accumulated biological wear rather than elapsed time.

Subsequent work by Levine and collaborators has extended PhenoAge validation into specific disease contexts. In cancer cohorts, PhenoAge acceleration predicts survival better than stage alone in some tumour types. In diabetes cohorts, it correlates with complication burden independently of HbA1c. In cardiovascular cohorts, PhenoAge acceleration predicts incident myocardial infarction after adjustment for Framingham Risk Score variables. This cross-domain predictive power makes PhenoAge the default outcome measure in an increasing number of longevity intervention trials — including the CALERIE-2 caloric restriction extension study and multiple rapamycin trials. Understanding how biological age differs from chronological age is foundational to interpreting these results correctly.

Morgan Levine at Altos Labs: Reprogramming Clocks and Epigenetic Age Reversal

In 2021, Morgan Levine left her tenured position at Yale to join Altos Labs, the San Diego and Cambridge-based cellular reprogramming research institute backed by approximately $3 billion in private funding from investors including Jeff Bezos and Yuri Milner. Altos Labs was founded on the hypothesis that partial cellular reprogramming — using Yamanaka transcription factors to partially reset the epigenome — could functionally rejuvenate aged tissues. Levine's recruitment was a signal about the central importance of epigenetic age measurement to that mission: you cannot demonstrate rejuvenation without a robust readout of biological age.

The Yamanaka factors — Oct4, Sox2, Klf4, and c-Myc (OSKM) — were originally discovered by Shinya Yamanaka as sufficient to reprogramme adult somatic cells all the way back to induced pluripotent stem cells (iPSCs). Full reprogramming erases cellular identity, which is obviously incompatible with therapeutic use. The strategy Altos Labs and competing groups (including Calico, the Salk Institute under Juan Carlos Izpisua Belmonte, and Harvard under David Sinclair) are pursuing is partial reprogramming: cycling OSKM expression for short periods — or using subsets of the factors — to reset epigenetic age without erasing cell type identity. This distinction between complete and partial reprogramming is critical to understanding the feasibility of the approach.

Levine's specific contribution at Altos Labs is building the measurement infrastructure for reprogramming research: determining which epigenetic clocks (and which CpG sites within them) are genuinely causally involved in ageing versus simply correlating with age, and developing new clock architectures that are more sensitive to the specific methylation changes induced by OSKM cycling. Her 2024 preprint, posted to bioRxiv with Altos Labs co-authors, introduced a reprogramming-specific epigenetic clock trained on cells at various stages of partial OSKM induction, providing a more granular readout of the reprogramming trajectory than pan-tissue clocks like PhenoAge.

The 2025–2026 Publication Landscape: Key Papers from Levine's Group

Levine's group at Altos Labs has produced a significant body of preprints and peer-reviewed publications since 2023, with output accelerating into 2025 and 2026 as the institute's research programmes matured. Several papers are particularly important for understanding the current state of the field.

A 2025 preprint (Altos Labs internal authors plus Levine as senior author) demonstrated reversible PhenoAge reduction in primary human fibroblasts subjected to cyclic OSKM expression over 72-hour on / 144-hour off cycles for 28 days. The cells showed a mean 4.2-year reduction in PhenoAge by the end of the cycling protocol, with methylation patterns reverting toward youthful baselines at more than 400 of the 513 PhenoAge CpG sites. Crucially, transcriptomic and functional assays showed no signs of dedifferentiation — the cells maintained fibroblast identity throughout. This paper is among the most direct demonstrations to date that epigenetic age is partially reversible without inducing pluripotency.

A separate 2025 paper in Nature Aging from Levine's collaborators at Yale (building on her prior work) demonstrated that PhenoAge acceleration in midlife — specifically in the 35-50 age window — is more predictive of late-life dementia and cardiovascular disease than PhenoAge acceleration measured in later life. This finding has clinical implications: it suggests biological age interventions need to be deployed earlier than the longevity medicine field has typically assumed, and that the window for meaningful epigenetic intervention may be more limited than optimistic projections imply. For broader context on how epigenetics intersects with medicine, see our overview of epigenetics and personalised medicine.

In 2026, Levine's lab published a landmark methodological paper in Cell Systems introducing DunedinPACE-V2, an updated version of the DunedinPACE longitudinal ageing pace clock (originally developed with Daniel Belsky at Columbia), incorporating new CpG sites from the UK Biobank methylation dataset and improving sensitivity to short-term interventional changes. DunedinPACE-V2 is now the recommended tool for clinical trial endpoints where PhenoAge may be insufficiently sensitive to detect short-term biological age changes — it measures the pace of biological ageing per chronological year rather than a static age estimate.

How to Measure Your Biological Age Using PhenoAge: Clinical Applications

PhenoAge can be estimated from a standard blood panel without any specialist testing. The nine required biomarkers — albumin, creatinine, fasting glucose, CRP (any assay), lymphocyte percentage, MCV, RDW, alkaline phosphatase, and white blood cell count — are available from a comprehensive metabolic panel plus CBC with differential, which costs $30-80 through most direct-access laboratory services and is often included in annual wellness checks. The phenotypic age calculation (which predates the DNA methylation clock and is published in the original Levine et al. paper) can be performed using the publicly available formula or through several free online calculators.

The DNA methylation-based PhenoAge — which is more precise and captures epigenetic information beyond what blood biomarkers reflect — requires a specialised epigenetic test. Commercial providers including TruDiagnostic (TruAge COMPLETE), Elysium Health (Index), and Foxo Technologies offer methylation array-based biological age testing using saliva or dried blood spot samples. Prices range from approximately $299 to $599 per test as of 2026. These tests run the sample on an Illumina EPIC array (or equivalent) and apply the Levine PhenoAge algorithm alongside other clocks (Horvath1, Hannum, GrimAge, DunedinPACE) to provide a multi-clock biological age profile.

In clinical practice, the most actionable use of PhenoAge is longitudinal tracking — measuring at baseline, implementing an intervention (dietary change, exercise protocol, stress reduction, targeted supplementation, or pharmaceutical), and remeasuring at 6-12 months to confirm whether epigenetic age acceleration has been reduced. Retesting intervals shorter than 6 months are generally not recommended for the static PhenoAge clock, though DunedinPACE-V2 is designed to detect changes over shorter windows in intensive intervention studies. To understand the fundamental distinction these tests are measuring, our explainer on the epigenetic clock provides a methodological foundation.

Limitations and Criticisms of Epigenetic Age Clocks in 2026

Despite PhenoAge's superior performance relative to first-generation clocks, the field faces substantive unresolved questions that Levine herself has written about. The most fundamental is the causal question: do the CpG sites in epigenetic clocks cause ageing, or do they merely correlate with biological processes that are driven by other mechanisms? If epigenetic age markers are downstream readouts of cellular damage rather than upstream drivers of it, then resetting them via reprogramming might improve the measurement without improving the underlying biology. Levine's reprogramming clock work at Altos Labs is partly motivated by this concern — identifying which CpG sites are in the causal pathway versus the correlative pathway is one of the central open problems in the field.

A second limitation is tissue-specificity. PhenoAge was trained primarily on blood-derived methylation data (NHANES) and performs best in blood. Biological age in brain, liver, kidney, and muscle tissue can diverge substantially from blood biological age — a finding with significant implications for neurodegeneration research, where brain-specific epigenetic age acceleration is likely more relevant than blood-based PhenoAge. Second-generation brain-specific clocks (including those from Steve Horvath's subsequent work and from the Allen Institute for Brain Science) are addressing this, but they require tissue biopsy and are not yet clinically deployable.

Third, there is the intervention artefact problem: some interventions known to affect methylation patterns — including certain anti-cancer drugs, high-dose antioxidant supplementation, and fasting protocols — may alter clock readings without producing corresponding functional health improvements. The relationship between epigenetic age improvement and actual longevity extension in humans remains undemonstrated by any randomised controlled trial with mortality as a primary endpoint, simply because such trials would require decades of follow-up. The field is working around this by using validated surrogate endpoints, but the fundamental evidentiary gap remains a legitimate critique.