“Biological age” is often presented as if one habit controls one number. Sleep better, eat better, exercise more — and the clock supposedly moves in the right direction. Real ageing biology is much harder to compress into a single switch.
A 2026 FinnTwin12 study followed 847 participants from early adolescence into young adulthood and screened 186 unique exposures spanning lifestyle, socioeconomic conditions and the physical environment. The researchers then asked how much of later epigenetic-age variation those measured exposures could predict — and what remained unexplained.
The result is not a list of anti-ageing hacks. It is a map of complexity: the measured exposome explained a meaningful share of variation, but most variation still lay outside those measured factors, with twin modelling pointing to both genetics and person-specific environmental influences.
Quick answer
How much of epigenetic ageing did the measured exposome explain?
In this FinnTwin12 analysis, the 186-exposure set explained 25.7% of variance in PCGrimAge and 30.8% in DunedinPACE at age 22. That is a meaningful signal, but it also means most variation was not explained by the measured exposome. Twin modelling suggested that both genetics and environmental influences unique to each twin contributed to what remained.
Key takeaways
- The final peer-reviewed analysis included 847 FinnTwin12 participants measured at ages 12, 14, 17 and 22.
- Researchers screened 186 unique lifestyle, socioeconomic and environmental exposures; this final exposure count differs from the earlier preprint.
- Blood DNA methylation at age 22 was used to calculate PCGrimAge and DunedinPACE.
- The measured exposome explained 25.7% of PCGrimAge variance and 30.8% of DunedinPACE variance.
- Smoking, alcohol use and youth unemployment were among predictors associated with higher epigenetic-age signals, but some environmental predictors showed counterintuitive directions that should not be interpreted causally.
- Twin modelling suggested that much of the remaining “missing exposome” reflected environmental influences not shared by twin siblings, alongside a substantial genetic component.
What does “exposome” mean?
The exposome is an attempt to study the full landscape of exposures that accumulate across life rather than isolating one behaviour at a time. Depending on the study, that can include smoking, alcohol, sleep, activity, socioeconomic conditions, neighbourhood features, pollution, green space and many other contextual variables.
The idea matters because real life does not happen one variable at a time. People who differ in one behaviour often differ in several others, and those factors can interact, cluster and change across development.
A longitudinal twin cohort gives the question more structure
FinnTwin12 is a Finnish longitudinal twin cohort. In this analysis, 847 participants contributed data across ages 12, 14, 17 and 22. The design allowed researchers to connect exposures accumulated through adolescence with epigenetic-age measures derived from blood DNA methylation at age 22.
The twin structure adds something an ordinary cohort cannot: it helps separate population variation into genetic, shared-environmental and non-shared environmental components. That does not make the analysis experimental, but it makes the unexplained remainder more informative.
The final study tested 186 unique exposures
The peer-reviewed version of record screened 186 unique exposures across lifestyle, socioeconomic and environmental domains. This is important because an earlier preprint described a larger raw exposure count. TwinPare uses the final published number.
The researchers combined exposome-wide association analysis with multiple data-driven machine-learning approaches designed to detect non-linear patterns and interactions that conventional one-variable-at-a-time models can miss.
The measured exposome explained part — not most — of the signal
For PCGrimAge, the exposure set explained 25.7% of variance. For DunedinPACE, it explained 30.8%. Those are meaningful proportions for complex human biomarkers, but they are nowhere near a complete explanation.
This is one reason “biological age calculator” language can become misleading. Even in a rich longitudinal dataset with many exposures, most variation in these two epigenetic-age outcomes remained outside the measured exposure set.
Some familiar factors appeared — but the model also produced surprises
Smoking, alcohol use and youth unemployment were among predictors associated with higher epigenetic-age signals. Other environmental features also entered the models.
Some directions were counterintuitive. For example, variables related to green space and aerial black carbon did not behave like a simple “more healthy exposure equals younger age” story. That is exactly why machine-learning predictor lists should not be converted into causal advice. Correlation structure, non-linearity, residual confounding and model specification can all influence which variables emerge and in which direction.
The missing exposome was not just “more lifestyle”
After accounting for the measured exposures, the researchers used twin modelling to study what remained. The unexplained variation — what they called the “missing exposome” — was attributed primarily to environmental influences not shared by twin siblings, distinct from a substantial genetic component in the model.
“Non-shared environment” is broader than personal choice. It can include different experiences, illnesses, social relationships, micro-environments, chance events, measurement error and exposures the study did not capture. It should not be translated into “lifestyle explains the rest”.
Why identical lives do not follow from identical genes
Twins can share family, age and much of their inherited biology while still accumulate different experiences over time. Those differences may become more visible as people move through school, work, relationships, neighbourhoods and independent routines.
The study therefore fits one of TwinPare Research’s central ideas: twin similarity is scientifically valuable, but twin divergence is equally informative. Shared DNA narrows some explanations; it does not make two life histories identical.
Why two epigenetic clocks can tell different stories
PCGrimAge and DunedinPACE are both DNA-methylation-based ageing measures, but they were built for different purposes. One is more closely linked to ageing-related mortality and physiological risk signatures, while the other estimates a pace-of-ageing signal.
A factor that predicts one clock strongly may not predict another in the same way. That is a feature of the field, not a technical nuisance. “Biological age” is not one universally agreed measurement hiding behind different brand names.
How this connects to the diet co-twin study
The previous article in this Healthy Ageing series asked a narrower question: does the association between healthier eating and favourable epigenetic-age measures survive a genetically identical co-twin comparison? In that Danish study, the individual-level association weakened markedly inside monozygotic pairs.
FinnTwin12 asks a broader question across many exposures and an earlier life stage. The two studies do not replicate one another. Read together, they reinforce a more useful principle: apparent ageing signals need stronger designs before they become causal stories.
What Really Changes Your Biological Age?
The honest answer is unlikely to be one behaviour, one gene or one score. The best evidence points toward layered influences that accumulate across biology, development, environment and lived experience.
For TwinPare Research, the next step is therefore not to promise an anti-ageing formula. It is to keep asking which associations replicate, which survive twin controls, which clocks agree, which factors are genuinely modifiable and which biomarker changes are linked to outcomes people actually care about.
Source notes
The sources have been verified and editorially reviewed for this article. The limitations below show which level of conclusion the sources support.
- [opperbeck-finntwin12-exposome-epigenetic-ageing-2026] Exploring the exposome and unexplained variance in epigenetic ageing – Insights from a longitudinal twin study in adolescence and early adulthood Annika Opperbeck; Zhiyang Wang; Ilkka Rautiainen; Aino Heikkinen; Miina Ollikainen; Sylvain Sebert; Elina Sillanpää. Environment International, 2026. Evidence type: Longitudinal FinnTwin12 analysis of 847 participants measured at ages 12, 14, 17 and 22, screening 186 unique lifestyle, socioeconomic and environmental exposures against PCGrimAge and DunedinPACE at age 22 with exposome-wide, machine-learning and twin-modelling approaches Limitation: The final version of record reports 186 unique exposures, not the larger exposure count described in the earlier preprint. The exposure set explained 25.7% of PCGrimAge variance and 30.8% of DunedinPACE variance. Predictors came from multiple domains and some directions were counterintuitive, so individual associations should not be read as causal advice. Twin modelling attributed much of the remaining unexplained variation to environmental influences not shared by twin siblings while also retaining a substantial genetic component. PubMed DOI
- [nygaard-diet-epigenetic-ageing-twins-2026] Dietary pattern and epigenetic aging: a longitudinal twin study M. Nygaard; H. R. Elliott; G. Davey Smith; et al.. GeroScience, 2026. Evidence type: Longitudinal monozygotic co-twin study of diet quality and five DNA-methylation ageing measures in 132 complete twin pairs Limitation: This companion article asks a narrower diet-specific question in a different adult cohort. Cross-sectional individual-level associations weakened substantially within genetically identical twin pairs, so it should not be treated as a replication of FinnTwin12. Publisher DOI
Editorial source review
This section shows how the article's key factual claims are linked to the source.
Phrasings that require caution
- Do not say the 186 measured exposures caused the observed epigenetic-age differences.
- Do not say the exposome explains biological ageing in general; the study analysed two specific DNA-methylation measures at age 22.
- Do not convert 25.7% or 30.8% explained variance into a personal prediction.
- Do not describe non-shared environmental variance as lifestyle alone.
- Do not interpret counterintuitive predictor directions as health recommendations.
- Do not call PCGrimAge or DunedinPACE a definitive whole-body biological age.
- Do not imply machine-learning variable importance establishes intervention targets.
- Do not treat this study as a replication of the Danish diet co-twin study.
- Do not imply TwinPare currently calculates epigenetic age or the exposome.
| ID | Claim | Source support | Caution |
|---|