You’re caught up.

Home/Issues/Issue 09/Study

Cross sectional analysis of two population cohorts with independent replication Online publication date confirmed against the PubMed publication date field and the PMC record.

Ten scores for a healthy diet almost never picked the same people, and still landed on the same biology

AI narration, generated on first listen
Journal
Nature Communications 17(1)
Authors
Tavares JF, Liu D, Breteler MMB
Institution
the German Centre for Neurodegenerative Diseases
Published
29 August 2026
Source
PMID 42668319 · DOI 10.1038/s41467-026-77064-4
Design
Cross sectional analysis of two independent population based cohorts. Ten diet quality scores were derived from food frequency questionnaires, then tested against epigenetic age acceleration measures and against epigenome wide DNA methylation, with the discovery findings carried into a replication cohort and meta-analysed.
Sample
6,470 adults from the Rhineland Study, 57 percent women, mean age 56.3 plus or minus 13.6 years, aged 30 to 95. Replication in 1,034 adults from EPIC-Potsdam.

What an epigenetic clock is

An epigenetic clock estimates how old someone's body looks biologically by reading chemical tags on the DNA rather than counting birthdays. The versions used here are expressed either as the gap in years between that estimate and actual age, so a positive number means looking older than you are, or as a pace of ageing scaled so that one is the population average rate per calendar year.

Drawn from background physiology, not from this paper.

Why they ran it

The authors write that the extent to which adherence to various healthy dietary patterns overlaps, and whether they affect health outcomes through similar or distinct molecular mechanisms, has not been explicitly examined, and that whether diet induced methylation changes converge on distinct biological pathways remains unclear.

Drawn from the paper's introduction.

Ten diet quality scores were calculated for every participant in the Rhineland Study, then recalculated in EPIC-Potsdam as an independent replication. The scores agreed far less than their shared purpose would suggest. Of 6,470 adults, 18 sat in the top quarter of adherence on all ten scores at once, and 77 managed it on at least nine. Meanwhile 1,212 people made the top quarter on at least one score, so the label of good eater attaches to roughly one person in five or to almost nobody depending entirely on which instrument is picked up.

Correlations between the scores ran from 0.61 at the closest, between the unhealthy plant based index and the dietary inflammatory index, down to 0.18 between the Nordic diet score and EAT-Lancet.

Against the biological ageing measures the picture inverted. All ten scores were significantly associated with the pace of ageing measure after false discovery rate correction, in the expected direction, though the sizes differed. DASH moved the pace of ageing by minus 0.10 SD per standard deviation of the score (95 percent CI minus 0.12 to minus 0.08) and the Nordic score by minus 0.07 SD (95 percent CI minus 0.09 to minus 0.05), while the dietary inflammatory index ran the other way at plus 0.07 SD (95 percent CI 0.05 to 0.08). EAT-Lancet was the exception on the clock measures, showing no association with either age acceleration measure and producing no epigenome wide significant site in either cohort.

Across nine patterns, 240 methylation sites reached epigenome wide significance, from 54 for AHEI-2010 down to 3 for the unhealthy plant based index, and 39 of those replicated. The specific sites and genes barely overlapped between scores. The pathways they belonged to did.

The numbers

People in the top 25 percent on all ten scores at once18 of 6,470
People in the top 25 percent on at least nine of ten77
People in the top 25 percent on at least one score1,212
DASH, pace of ageing, per 1 SD (4.4 points)minus 0.10 SD (95% CI minus 0.12 to minus 0.08)
Nordic, pace of ageing, per 1 SD (3.4 points)minus 0.07 SD (95% CI minus 0.09 to minus 0.05)
Dietary inflammatory index, pace of ageing, per 1 SD (1.7 units)plus 0.07 SD (95% CI 0.05 to 0.08)
Scores associated with pace of ageing after FDR correctionall 10
EAT-Lancet score against either age acceleration measureno association, and no epigenome wide significant site in either cohort
Epigenome wide significant sites across nine patterns240, from 54 (AHEI-2010) to 3 (unhealthy plant based index); 39 replicated
Correlation between scores, highest and lowest0.61 (unhealthy plant based vs inflammatory index); 0.18 (Nordic vs EAT-Lancet)

Why this might happen

Proposed by the authors This is the explanation the authors offer in their discussion. This study did not test it.

The authors describe a convergent pathway model. They found minimal overlap in the specific sites or genes associated with different diet quality scores, yet substantial overlap in the biological pathways those sites belonged to, so that different healthy diets may influence distinct sets of epigenetic markers while converging on similar processes involving cell signaling, metabolism and neurogenesis.

They note pattern specific arms within that convergence, with the DASH diet showing changes in heart function related pathways and the MIND diet in central nervous system and cognition related pathways.

They leave the transcriptional link open. The majority of identified sites did not show expression associations in blood, which they write does not preclude biological relevance, since blood methylation may serve as a biomarker of systemic dietary exposures whose functional consequences occur in tissues such as adipose, liver or the gastrointestinal tract.

Drawn from Discussion, PMC13526023.

What this does not show

  • This does not show that diet changes biological ageing. Every measurement was taken at one moment in observational cohorts, so the direction of the arrow is not established and no one was assigned to eat anything.
  • This does not show the effects are large. The associations are reported in standard deviation units and are small, with the strongest around a tenth of a standard deviation in the pace of ageing measure.
  • This does not generalise beyond the populations studied. The authors state their findings come from a predominantly German population of European ancestry in both cohorts, and that generalizability to other ancestries or dietary habits remains to be investigated.
  • This does not establish that the shared pathways are real biology rather than shared measurement. Diet was captured by food frequency questionnaire, which the authors note is subject to measurement error, and one cohort's dietary data were collected in the 1990s.

Where this leaves us

The ten competing definitions of a good diet identify almost entirely different people as good eaters, which means the choice of score is doing a lot of the work in any study that uses one, and yet the molecular signatures they leave behind land in overlapping territory.

Adults of European ancestry in two German cohorts, aged 30 to 95 in the discovery sample.

It undercuts reading any single diet score as the measure of diet quality, and it argues that results attached to one score should not be quoted as if they were about healthy eating in general.

Caveats worth holding

  • Both cohorts are German and predominantly of European ancestry.
  • Diet was measured by food frequency questionnaire; EPIC-Potsdam dietary data were collected in the 1990s.
  • The majority of identified methylation sites showed no expression association in blood.
  • Cross sectional design in both cohorts, so no temporal ordering.

Newsletter

Each issue by email, when the newsletter launches. Leaving your address puts you on the list, nothing is sent yet.