Nutritional associations with decelerated epigenetic aging: vegan diet in a Dutch population

Abstract

Background

The 2021 Aging Report of the European Union projected significant increases in healthcare costs, with some member states expecting up to a 60% rise over the next 50 years, primarily due to an aging population and related diseases. Interventions targeting aging have been proposed to reduce this burden by extending healthspan. Recent evidence suggests that vegan diets may slow the aging process.

Results

Using data from the Netherlands Twin Register (n = 22,124), dietary habits of meat eaters (n = 21,614), pescetarians (n = 294), vegetarians (n = 194), and vegans (n = 22) were examined, which were collected in a 2014–2016 survey period. Health parameters such as BMI, waist circumference, and insulin sensitivity showed improved health with more plant-based diets, though results were not adjusted for confounders. Epigenetic age was assessed using the Hannum, Horvath, PhenoAge, GrimAge, and Dunedin Pace of Aging clocks for 3049 participants with DNA methylation (DNAme) data from a 2004–2011 sample collection period and were compared to calendar age from the time of blood draw. Although discordant twin pairs with and without vegan diets (n = 3) were too few for statistical significance, their results suggested a potential benefit of veganism. In the subpopulation with complete data on DNA methylation, dietary habits, and covariates (n = 1198), veganism was significantly associated with lower epigenetic aging scores on the Hannum and Horvath clocks, even after adjusting for confounders (age, sex, smoking, education, physical activity, BMI, and alcohol use). Analysis of individual covariates in the model found that higher education, physical activity, being female, and non-smoking were associated with reduced epigenetic age, while higher BMI was linked to increased epigenetic age; however, since these parameters were used with the primary purpose of accounting for confounders, caution should be used in interpreting these results. Finally, dietary analysis showed that abstaining from pork was associated with lower biological ages (Dunedin Pace of Aging), whereas abstaining from poultry was linked to higher biological ages (PhenoAge, Dunedin Pace of Aging).

Conclusions

Although with a small sample size and a large time gap between nutritional survey collection and blood collection for participant’s epigenetic ages, these findings suggest that dietary choices, particularly plant-based diets, may influence epigenetic aging. The results highlight the potential of veganism to reduce epigenetic age and underscore the importance of further research to clarify the relationship between diet and aging. Larger cohorts and clinical trials would be necessary to gain more certainty on our initial findings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-025-01934-9.

Background

As populations around the world continue to age, there exists an increasing urgency for strategies that promote healthy aging. Increase of life expectancy over the last century, while a remarkable achievement, has introduced new societal challenges related to age-related health issues, healthcare costs, and quality of life in older age. With the elderly population expanding, especially in high-income countries, there is a critical need to identify lifestyle factors that can enhance longevity and minimize age-associated health declines. Societies worldwide are seeking effective ways to support healthy aging, not only to ease healthcare burdens but also to ensure older adults maintain their vitality, independence and overall well-being. Indeed, given the global aging population, strategies to promote healthy aging are increasingly relevant [1].

Among various lifestyle factors impacting aging, diet plays a profound role. Research consistently links dietary habits, patterns, or food components with health outcomes across the lifespan, influencing mortality and health in older age [2–6]. Plant-based diets, including veganism, have gained significant attention for their potential health benefits. Studies indicate that plant-based diets may reduce inflammation, lower the risk of cardiovascular diseases, and promote overall health [7–9]. Research also shows that vegan diets, which exclude any animal products, beneficially impact aging at the molecular level [10]. As the search for dietary approaches to promote longevity intensifies, understanding the specific impacts of veganism on aging can provide valuable insights.

The biological age (BioAge) of a person refers to the functional and physiological state of a person’s cells, tissue, and organs [11]. While chronological age is fixed, BioAge varies between individuals in relation to lifestyle factors, including diet [12]. DNA methylation (DNAme) biomarkers of aging—also known as biological aging ‘clocks’—measure age-related changes in DNA methylation patterns and have become a powerful approach for assessing BioAge. Several well-known clocks include the ‘Hannum’ and Horvath (also known as ‘Intrinsic Epigenetic Age Acceleration’ or ‘IEAA’) clocks, which were among the first to link methylation patterns to age [13, 14]; as well as ‘PhenoAge’, designed to predict morbidity and mortality by capturing health-related changes in the body [15], ‘GrimAge’, which incorporates DNA methylation markers for aging-related plasma proteins to better predict life expectancy and disease risk [16], and the Dunedin Pace of Aging clock (‘DunedinPACE’), a measure that allows for insight into how quickly an individual is aging biologically [17].

In the Netherlands, the Netherlands Twin Register (NTR) presents a unique opportunity to study the effects of veganism on biological aging within a well-characterized population [18–22]. The NTR includes extensive data on lifestyle, health, genetics, DNA methylation, and environmental factors among Dutch twins and their family members. This database allows researchers to disentangle the effects of genetic and environmental influences on aging, making it ideal for studying the complex interplay between diet and aging processes. In this study, we investigate the association between dietary intake—especially veganism—and biological aging in this Dutch population.

Methods

Cohort characteristics

The Netherlands twins Register (NTR) was founded in 1987 and collects data about Dutch twins and their families on a variety of parameters including genetics, lifestyle, health status, DNA methylation profiles, and dietary intake [18–21]. This study used whole blood Illumina 450k array DNA methylation data and survey data from various waves of survey collection in adult twins and their families, including survey 11, which had a special focus on nutrition [18, 22]. Consumption of eggs, dairy, poultry, pork, shellfish, fish, beef, or alcohol was surveyed with a binary question where 1 = yes and 2 = no. Education was assessed on a scale from 1–4 from lower to higher education. Smoking was assessed at blood draw on a 3-point scale where 1 = currently, 2 = previously, and 3 = never. Physical activity level was assessed based on the recall of voluntary regular exercise behavior in leisure time [23]. Education level was taken from the same survey as dietary information, while physical activity level was an average from separate assessments. Sex was assessed as 1 = male and 2 = female; if missing entries existed, these were omitted from the analysis. Three health parameters, measured as part of the NTR biobank project [24], were used in this study. These included BMI, waist circumference, and insulin sensitivity. BMI was calculated, as kg/m² where kg is a person's weight in kilograms and m² is height in meters squared. Waist circumference was assessed using a tape measurer between the top of the hip and bottom of the ribs. Insulin sensitivity was estimated using fasted insulin and glucose levels using the Homeostasis Model Assessment (HOMA) [25] provided by the Radcliffe Department of Medicine.

Data from our study were collected across various time frames corresponding to study waves in the NTR. A 2014–2016 survey period provided the information on dietary intake and information on potential confounders including smoking status, education, and alcohol intake. A 2004–2011 study wave involved the biobanking and provided DNA methylation data for biological age estimations and additional information on BMI, waist circumference, and estimated insulin sensitivity. Finally, the exercise level was assessed as an average value from measures collected in three different study waves spanning 2000–2017.

Dietary qualifications

Dietary survey data were collected in a 2014–2016 survey period. Survey responders were classified as either meat eaters (not actively denying the consumption of either cow, pork, or chicken n = 21,614), pescetarians (denying the consumption of cow, pork and chicken, and affirming the consumption of either fish or shellfish, n = 294), vegetarians (denying the consumption of cow, pork, chicken, fish, and shellfish, and affirming the consumption of either dairy or eggs, n = 194), and vegans (denying the consumption of each component; cow, pork, chicken, fish, shellfish, dairy, and eggs, n = 22).

DNA methylation

Samples for DNA methylation were collected in 2004–2011 sample collection period, and age at blood draw was used as the ‘chronological age’ to compare to ‘biological age’ measures from these collections. DNA methylation was assessed with the Infinium HumanMethylation450 BeadChip Kit (Illumina, San Diego, CA, USA) by the Human Genotyping facility (HugeF) of ErasmusMC, the Netherlands (http://www.glimdna.org/) as part of the Biobank-based Integrative Omics Study (BIOS) consortium [26]. DNA methylation measurements have been described previously [19, 26] Genomic DNA (500ng) from whole blood underwent bisulfite treatment with the Zymo EZ DNA Methylation kit (Zymo Research Corp, Irvine, CA, USA), and 4 µl of bisulfite-converted DNA was measured on the Illumina 450k array following the manufacturer’s protocol. A number of sample- and probe-level quality checks and sample identity checks were performed, as described in detail previously [19]. In short, sample-level QC was conducted with the assistance of MethylAid [27]. Probes were set to missing in a sample if they had an intensity value of exactly zero, a detection p > 0.01, or a bead count of < 3. After these steps, probes that failed based on the above criteria in > 5% of the samples were excluded from all samples. (Only probes with a success rate ≥ 0.95 were retained.) The methylation data were normalized with functional normalization [28].

DNA methylation biomarker of biological aging

DNA methylation age acceleration measures of Hannum, Horvath, PhenoAge, and GrimAge were computed through the Horvath epigenetic age calculator (https://dnamage.genetics.ucla.edu/) and included the following outputs:

• Hannum age acceleration—“AgeAccelerationResidualHannum” (AARH),

• Intrinsic epigenetic age acceleration; the residual resulting from regressing the DNAm age estimate from Horvath on chronological age and blood cell count estimates.—“IEAA”,

• PhenoAge acceleration—“AgeAccelPheno” (AAPheno),

• GrimAge acceleration—“AgeAccelGrim” (AAGrim)

DunedinPACE was calculated based on code accessible on GitHub (https://github.com/danbelsky/DunedinPACE).

Statistics and data visualization

Analyses were done with R [29] version 3.5.1. Statistics were calculated either using a Kruskal–Wallis test, or where relevant, a paired Student’s t test for comparing twin pairs. A generalized estimation equation (GEE) fit of the data, incorporating family relatedness, was used to assess the association of dietary regimens, food components, and/or lifestyle/demographic factors with epigenetic aging measures using the gee() function in R package ‘gee’ version 4.13-26 and setting ID to FamilyNumber (a variable unique to each family in the NTR) [30]. For all epigenetic clocks, deltaAge (i.e., the residual of biological age estimation minus chronological age at time o blood draw) was used as the outcome variable. We assumed an"exchangeable"correlation structure (corstr ="exchangeable") to model the within-family correlation in the repeated measurements. The GEE model used a Gaussian error distribution, appropriate for the continuous outcome variables (deltaAges). For each analysis, the specific epigenetic aging clock’s deltaAge (i.e., From Hannum, Horvath, PhenoAge, GrimAge, or the DunedinPace of aging) was modeled separately as the dependent variable. Covariates included year of birth (yob), sex, smoking status, education, physical activity, alcohol intake, and BMI, to be considered with either dietary pattern or food components. For the dietary patterns, all diets were included in a single model and thus mutually adjusted for one another, alongside the standard covariates. Similarly, for the dietary component analysis, all dietary components were included in a single model and thus mutually adjusted for one another, alongside the standard covariates. See Supplemental Fig. 1 and Supplemental Table 1 for specific information on what parameters were included in each model and number of individuals, see Tables 1 and 2 for demographic characteristics, and see Supplemental Tables 2 and 3 for outcomes of the models. Multicollinearity of the parameters included in the GEE model was assessed using the Generalized Variance Inflation Factor (GVIF) in an equivalent linear model. All GVIF^(1/(2*Degrees-of-freedom) values were < 2, including for the dietary group variable, indicating that multicollinearity was not of large concern. Visualization of data was performed using R packages ggplot2 [31] version 3.2.1, ggpubr [32] version 0.2.5, UpSetR [33] version 1.4.0, or circlize [34] version 0.4.16, and using colors provided within RColorBrewer [35] version 1.1-2.

Fig. 1.

Study design and biological age calculations in the Netherlands Twins Register. A Study design, comparing DNAme BioAges, health metrics, and food preference nutritional survey data to identify associations between dietary preferences and health. B Top left panel: conceptualization of comparing chronological age (X-axis) with BioAge (Y-axis) and how individuals can be considered either biologically older or younger. Other panels: comparison of chronological age (X-axis) to BioAge measures including GrimAge, PhenoAge, IEAA, Hannum, and the DunedinPACE. Age is measured in years at the time of blood draw. Note: Dunedin PACE, in which a value of 1 is equal to a pace of aging of 1 year per year, was multiplied by participant age to visualize results on a similar scale as the other age predictors. C Cross-comparison of the deltaAges (DNAmAge minus calendar age; positive values are commonly referred to as epigenetic age acceleration and negative values are commonly referred to as epigenetic age deceleration) for each BioAge estimation method in the population. Scatterplot points color scale: light blue = higher density of points, dark blue = lower density of points. Correlation and significance: Pearson’s. Asterix: ***p < 0.001. n = number of individuals

Table 1.

Dietary preferences; study population used in Fig. 3C

	Meat eater	Pescetarian	Vegetarian	Vegan
Total (n)	1146	19	13	2
Monozygotic (nt)	662	12	9	1
Non-drinkers (%)	16.2% (n = 186)	21.1% (n = 4)	38.5% (n = 5)	100% (n = 2)
BMI (mean)	24.2076009	21.88304448	21.19014753	22.89118026
BMI (SD)	3.979335555	3.338623047	3.459284868	2.601510003
Physical activity level (mean)	647.3608275	794.5789474	673.3076923	1220.25
Physical activity level (SD)	730.1576991	1235.795206	1266.612136	482.6003782
High education (%)	55.7% (n = 638)	73.7% (n = 14)	69.2% (n = 9)	0% (n = 0)
Never smoker (%)	60.3% (n = 691)	73.7% (n = 14)	76.9% (n = 10)	100% (n = 2)
Female (%)	69.3% (n = 794)	94.7% (n = 18)	92.3% (n = 12)	100% (n = 2)
Birth year (mean)	1967.43	1968.74	1972.31	1975

1198 individuals possessed data entries for all variables (age, sex, smoking status, education, physical activity level, BMI, and alcohol consumption) in addition to being classified into a dietary group and having DNAme biological age scores

BMI = body mass index. Average = ‘avg’. Standard deviation = ‘sd’. Number of individuals = ‘n’. Number of monozygotic twin pairs = ‘n^t’

Table 2.

Food components; study population used in Fig. 4A

Total (n)	1180
Monozygotic twin (nt)	684
No alcohol	16.7% (n = 197)
BMI (avg)	24.13
BMI (SD)	3.98
Physical activity (avg)	650.99
Physical activity (SD)	747.33
Highest education	56% (n = 661)
Never smoking	29.2% (n = 344)
Female	70% (n = 826)
Year of birth (avg)	1967.52
Year of birth (SD)	12.39
Beef consumers	95.9% (n = 1132)
Fish consumers	91.9% (n = 1085)
Shellfish consumers	56.8% (n = 670)
Pork consumers	83.4% (n = 984)
Poultry consumers	91.6% (n = 1081)
Dairy consumers	96.4% (n = 1137)
Eggs consumers	98.6% (n = 1163)

1180 individuals possessed data entries for all variables (age, sex, smoking status, education, physical activity level, BMI, and alcohol consumption) in addition to food survey information on dietary components and having DNAme biological age scores

BMI = body mass index. Average = ‘avg’. Standard deviation = ‘sd’. Number of individuals = ‘n’. Number of monozygotic twin pairs = ‘n^t’

Results

Part 1: Biological age estimations in The Netherlands Twins Register

In order to explore the relationship between biological aging and diet, especially veganism, we turned to the cohort of the Netherlands Twins Register (NTR), which contained dietary preference information based on 2014–2016 survey data from 22,124 individuals. Furthermore, from 2004 to 2011 DNA methylation profiles of blood had been performed for 3,049 individuals paired with this survey data, which could be used to calculate biological ages of individuals [18–21, 36]. Age at blood draw for DNAme was used to assess ‘deltaAges’ of individuals; the difference in estimated biological age compared to calendar age. This would allow the cross-referencing of dietary trends against biological aging differences to identify diets or nutritional patterns associated with decelerated aging (Fig. 1A, and see Supplemental Fig. 1 for extended overview of analysis in this study).

Various approaches to calculating biological age using DNA methylation profiles have been developed. The NTR database contained biological age estimates for the clocks developed by Hannum [14], Horvath (IEAA) [13], PhenoAge [15], GrimAge [16], and Dunedin Pace of Aging (DunedinPACE) [17] studies (Fig. 1B). Cross-comparison of these biological aging scores indicated some degree of correlation, though likewise independent associations as well (Fig. 1C), and we therefore proceeded to assess the relationship of dietary preferences to biological aging using all BioAge measures independently.

Part 2: Common dietary types and health associations

In order to assess the association between a vegan diet and epigenetic age, we first partitioned the questionnaire respondents as being meat eaters (n = 21,614), pescetarians (n = 294), vegetarians (n = 194), or vegans (n = 22) (Fig. 2A). Notably, classification into these dietary groups required an active denial of the consumption of particular food components (Fig. 2A); therefore, these percentages of the population—97.7% meat eater, 1.33% pescetarian, 0.88%, vegetarian, and 0.09% vegan—are likely conservative estimates and provide a high probability that correspondents did in fact follow the diet described.

Fig. 2.

Dietary preferences in a Dutch population. A Overview of responses from the food and nutrition survey in the Netherlands Twins Register. A black dot indicates the participant actively denies consuming the indicated food. Allows partitioning participants as meat eaters (default, n = 21,614), pescetarian (yellow, n = 294), vegetarian (gray, n = 194), or vegan (red, n = 22). B Comparison of BMI in the designated diet groups (blue, meat eater, n = 8863; yellow, pescetarian n = 74; gray, vegetarian n = 37; red, vegan n = 3). C Comparison of waist circumference in the designated diet groups (Blue, meat eater, n = 9269; yellow, pescetarian n = 76; gray, vegetarian n = 37; red, vegan n = 3). D Comparison of insulin sensitivity (HOMA 2S) in the designated diet groups (Blue, meat eater, n = 8445; yellow, pescetarian n = 69; gray, vegetarian n = 31; red, vegan n = 3). B–D Means are shown above boxplots. Statistical significance was calculated using a Kruskal–Wallis test. Comparisons were not corrected for any additional covariates (to maximize the n per comparison). n = number of individuals

As a first analysis of these dietary regimens, we assessed parameters associated with better health in general. We cross-correlated the survey response dataset with health data available for each respondent on three specific parameters well known to be associated with disease incidence: BMI, waist circumference, and insulin sensitivity. Here, we found a trend for decreasing BMI for pescetarian (n = 74, p = 6.6e−5), vegetarian (n = 37, p = 0.00094), and a stepwise trend suggesting similar benefits for vegan (n = 3, p = 0.17) diets compared to meat eaters (n = 8863) (Fig. 2b). Furthermore, the same stepwise trend was observed with waist circumference comparing pescetarian (n = 75, p = 1.9e−5), vegetarian (n = 37, p = 0.00025), and vegan (n = 3, p = 0.015) diets to meat eaters (n = 9269) (Fig. 2c). Assessing insulin sensitivity provided a similar step-wise trend comparing pescetarian (n = 69, p = 0.00029), vegetarian (n = 31, p = 0.0076), and vegan (n = 3, p = 0.26) diets to meat eaters (n = 8445) (Fig. 2d).

Of consideration, while this stepwise association tended to allocate non-meat eaters with the greater health benefit, these associations were not significant for vegans, likely owing to the low number of vegans present in the overlapping datasets between the survey and health parameters. Nonetheless, the trends for pescetarian and vegetarian advantages over meat eaters were significant (p < 0.05), and the stepwise trend suggests similar benefits for vegans.

Part 3: Epigenetic aging in relation to common dietary practices

We next proceeded to assess the epigenetic ages of the participants according to each dietary regimen. Here, we found a general agreement between the epigenetic aging trends and the health trends. For example, GrimAge showed a stepwise benefit for each diet, with pescetarians having a median deltaAge for GrimAge of − 1.4 years (n = 26); vegetarians having − 1.45 years (n = 21), and vegans having − 2.38 years (n = 5) (Fig. 3A). However, none of these differences were significant as compared with meat eaters who possessed a median deltaAge GrimAge of 0.027 years (n = 2996) (Fig. 3A). Although one clock, the Hannum clock, showed a statistically significant association with a vegan diet—whereby vegans possessed a median deltaAge of − 2.4 years compared to meat eaters (p = 0.038)—we chose to present GrimAge in Fig. 3 as a representative example of a non-significant but suggestive trend (p = 0.08), to avoid overstating the findings, as no other clock reached statistical significance. Full results for all clocks are provided in the accompanying results table (Supplemental Table 1). The various non-significant differences in most of the biological age predictors were likely due to the low sample sizes of each diet with overlapping DNAme bioAge data available. Furthermore, to allow for a larger population size in this analysis, no corrections were made for covariates present in the population data. Following this, we performed an exploratory analysis in vegan diet-discordant twin pairs. After removing non-discordant twins, triplets, and twins without DNAme data, six individuals remained from three twin pairs discordant for a vegan diet, of which two pairs were monozygotic and one pair was dizygotic. Here we found a similar direction of effect for age deceleration, though again not significant (p = 0.322) (Fig. 3B). Furthermore, here too, comparison was not adjusted for additional covariates, though comparisons were within twin sets.

Fig. 3.

Association of Biological Aging with Dietary Preferences. A GrimAge deltaAges for participants following each dietary regimen, meat eaters (n = 2996), pescetarians (n = 26), vegetarians (n = 21), and vegans (n = 5). Means are shown above boxplots. Statistical significance calculated using a Kruskal–Wallis test. B Subset of diet-discordant twins for vegan and non-vegan twin pairs (n = 3 pairs, 6 total). P value is calculated using a paired Student’s t test. C Generalized estimation equation (GEE) fit of the data which takes into account family relatedness built on n = 1198 individuals with DNAme data to estimate biological ages (x-axis) based on dietary preference and additional factors (y-axis). Color designates being associated with higher (red) or lower (blue) biological age, while size and shape denote p value significance (square p < 0.05, circle p > 0.05). n = number of individuals. Covariates are shown for transparency in model adjustment but are not presented for causal inference; interpretations should focus on the primary exposure (dietary group)

To gain better resolution on the influence of diet on biological aging, we applied a generalized estimation equation (GEE) model, which accounts for family relatedness and adjusts for relevant covariates. Specifically, we included sex, year of birth, smoking status, educational attainment, physical activity level, body mass index (BMI), and alcohol consumption as covariates in the model. A total of 1,198 individuals had complete data for all these variables, though our specific diets had low sample sizes (Table 1). Nonetheless, this analysis revealed that several lifestyle and demographic factors—including sex, smoking, education, physical activity, BMI, and alcohol intake—were independently associated with deltaAge values from various biological aging clocks. Furthermore, a vegan diet was significantly associated with decelerated aging as measured by the Hannum and Horvath (IEAA) clocks (Fig. 3C).

Interestingly, these lifestyle and demographic variables showed various directions of effect on biological age that is of interest on their own. For example, BMI, scored on a continuous scale from low to high, was associated with higher (red) biological age the higher the BMI. Physical activity, scored on a continuous scale from less to more, was associated with lower (blue) biological age the more physical activity was present. Education was scored on a 4-point scale from low to high, and having higher education was associated with lower (blue) biological age. Smoking was scored on a 3-point scale with ‘3’ being ‘never’ and was associated with lower biological age (blue) if smoking was never done. Sex was scored as male or female and being female was associated with lower (blue) biological age. While of interest in follow-up studies, we should note that covariates are shown for transparency in model adjustment but are not presented for causal inference; interpretations should focus on the primary exposure (dietary group).

Part 4: Associations between dietary components and epigenetic age

To further investigate the influence of individual dietary components on biological aging, we constructed GEE models assessing the impact of specific food items—including eggs, dairy, poultry, pork, shellfish, fish, and beef—while adjusting for the same covariates used previously: chronological age, sex, smoking status, education level, physical activity level, BMI, and alcohol consumption. This approach allowed us to analyze the full survey cohort with available DNA methylation data and information on confounders (n = 1180; Table 2), rather than limiting analyses to predefined dietary groups.

Here we found that nutritional components, as with diets in general, possessed significant associations with biological age across various biological aging measures, though they did not possess as strong an influence as factors such as BMI and smoking status (Fig. 4A). For example, pork consumption—answered as either ‘yes’ or ‘no’—was associated with lower biological age (blue) if answered as ‘no’, an observation occurring across all biological aging scores though only reaching significance in the Dunedin Pace of Aging metric (Fig. 4A). Meanwhile, poultry consumption possessed an opposite trend (abstinence from consumption, i.e., answering ‘no’ to its consumption, was associated with higher biological age). The other food components including eggs, dairy, shellfish, fish, and beef possessed more mixed results, either being associated with lower or higher biological ages, often non-significantly, depending on the biological aging score (Fig. 4A).

Fig. 4.

Associations of food components with biological age measures. A Generalized estimation equation (GEE) fit of the data built on n = 1180 individuals for estimation of biological ages (x-axis) and food component survey data and additional factors (y-axis). Color designates being associated with higher (red) or lower (blue) biological age, while size and shape denote p value (square p < 0.05, circle p > 0.05). All food components were included in a single model and thus mutually adjusted for one another. See Supplemental Table 3. Covariates are shown for transparency in model adjustment but are not presented for causal inference; interpretations should focus on the primary exposure (component). B–C Chord diagrams depicting the relationship between either B specific factors including BMI, physical activity level, education, smoking status, sex and age, or C food components, and the biological aging scores. Thickness of chord and directionality color is directly related to significance and model coefficient of panel A. n = number of individuals

To visualize the associations of our measured variables to biological aging scores better, we plotted these as chord diagrams, either between the socio-demographic factors (Fig. 4B) or the nutritional components (Fig. 4C) and biological age measures. This illustrated the clear influence smoking has with age acceleration (Fig. 4B) or the more mixed results obtained from food groups (Fig. 4C). Evaluating the relationship between the diets and aging in this manner also illustrated the striking observation that veganism was consistently associated with lower biological age, for all aging measures, and represented the most profound dietary pattern associated with lower biological age (Fig. 5) (supplemental Table 2). Overall, these analyses reveal a complex relationship between dietary components, diet, and associations with either acceleration or deceleration of biological age.

Fig. 5.

Comparison of dietary patterns with biological aging measures. Chord diagrams depicting the relationship between dietary preferences (left side of chord diagram), and the biological aging scores (right side of chord diagram). Thickness of chord and directionality color is directly related to significance and model coefficient of Fig. 3C (see supplemental Table 2)

Discussion

Our study investigated the relationship between a vegan diet and biological aging within the Netherlands Twin Register, aiming to determine whether vegan dietary patterns are associated with decelerated aging. However, the primary goal of assessing the impact of veganism on biological aging was limited by the low number of twin pairs discordant for a vegan diet. Among the biological aging clocks analyzed, only the Hannum clock indicated a potential significant association between a vegan diet and decelerated aging, based on data from just three discordant twin pairs (Supplemental Table 1, row 4). While this finding is intriguing, the small sample size and effect with only one biological aging measure underscores the need for caution in interpretation and calls for replication in larger cohorts. Subsequently, we assessed diets at the population level (n = 1198), where we found veganism significantly correlated with lower biological aging scores in the Hannum and Horvath clocks, even after adjusting for confounders (age, sex, smoking, education, physical activity, BMI, alcohol).

In addition to our findings on vegan diets associated with decelerated aging, our study provided other notable insights. First, lifestyle factors such as smoking, physical activity, and demographic factors including education and sex were observed to show stronger associations with epigenetic age compared to dietary patterns. Second, specific food components—especially pork consumption—emerged as potentially being associated with accelerated aging. Finally, the various biological age measures—such as those derived from different epigenetic clocks—appear to capture distinct aspects of the aging process, highlighting the complexity and multidimensionality of biological aging. These findings emphasize the interplay between diet, lifestyle, and broader biological mechanisms.

It should be noted that to account for the non-independence of twin and family data in our analyses, we applied generalized estimation equations (GEE) with family number as the clustering unit. This approach corrects for within-family relatedness and shared environmental factors in a robust manner by assuming an exchangeable correlation structure—modeling equal correlation among all individuals within each family. However, while effective for capturing broad familial clustering, this method does not explicitly model within-family associations in monozygotic (MZ) or dizygotic (DZ) twin pairs. As such, our estimates reflect population-level associations rather than within-pair contrasts. Nonetheless, we performed twin-pair controlled comparisons, though the number of discordant twin pairs (n = 3 pairs) was likely too low for meaningful insights. Future studies may benefit from applying co-twin control designs (e.g., as proposed by Gonggrijp et al., 2023) [37] when larger samples of discordant twin pairs become available, to more rigorously disentangle genetic and environmental contributions to diet–aging relationships.

Given that data collection occurred across multiple waves, with DNA methylation data gathered between 2004 and 2011 and dietary survey data collected between 2014 and 2016, we opted to use year of birth as a standardized proxy for participant age. This decision allowed for more consistent age adjustment across datasets. Interestingly, year of birth remained significantly associated with several biological aging measures even after adjustment, suggesting either residual confounding, partial age dependence of certain epigenetic clocks, or cohort effects. These associations do not imply that older individuals had younger biological ages, but they do underscore the nuanced interpretation required when adjusting for chronological age in cross-cohort data.

We note that although we did not adjust directly for total caloric intake, BMI was included as a covariate and likely captures much of the variation in long-term energy balance. This is particularly relevant given the lower caloric density of many vegan diets and the established link between calorie restriction and reduced biological aging [5]. In our sample, vegans exhibited lower average BMIs, raising the possibility that at least part of the observed association between veganism and slower aging reflects differences in energy balance. A more detailed investigation of caloric intake—as well as macronutrient composition and meal timing—would be valuable in future work to clarify the role of dietary energy in modulating biological age.

A key challenge in our adjusted analyses was the small number of vegans and vegetarians with complete covariate data. While the broader dataset included 22 vegans and 194 vegetarians, only a fraction had usable overlapping data for essential health and lifestyle variables. Specifically, only five vegans had BMI and waist circumference data, three twin pairs were discordant for veganism, and just two individuals had complete data for all modeled confounders—including age, sex, smoking status, education, physical activity, BMI, and alcohol intake. We chose not to impute missing values, as doing so would have involved interpolation for a substantial portion of the vegan subgroup—out primary study focus—and could introduce bias. Consequently, adjusted associations for vegans and vegetarians should be interpreted with caution. This limitation is further compounded by sex imbalance: all vegans in our dataset were female, and only one vegetarian with complete data was female. As a result, we were unable to disentangle the effects of sex from diet in adjusted models, and this restricts interpretation of subgroup findings.

Another important limitation is the potential for reverse causation. It remains possible that some individuals adopted plant-based diets in response to emerging health conditions such as metabolic syndrome or cardiovascular disease, rather than diet being the upstream influence on biological aging, and these may concordantly cause weight loss and lower BMI, for example. While we adjusted for several health-related variables, the observational and cross-sectional nature of our analysis precludes determination of causality. Longitudinal studies that track dietary patterns and biological age trajectories over time, or randomized dietary interventions, are needed to address this issue more definitively.

Additionally, dietary classifications in this study relied on self-reported dietary recall, which is susceptible to both recall errors and reporting biases. While our classification strategy emphasized the exclusion of specific food components to ensure high specificity in dietary grouping, this rigorous definition likely contributed to small group sizes—particularly for strict vegans. Although this approach enhances internal validity, it comes at the cost of statistical power. The limited sample size of vegan-discordant twin pairs, along with substantial missing values in lifestyle covariates, reduced our ability to detect significant associations and constrained our capacity to explore effect modification by sex or other covariates. These design challenges highlight the need for larger and more comprehensively characterized cohorts.

The time lag between dietary assessment and biological sample collection further introduces potential misclassification. Because DNA methylation data were obtained several years prior (i.e., 2004–2011) to the dietary survey data (i.e., 2014–2016), we cannot confirm that participants maintained the same dietary patterns at the time their biological samples were collected. This temporal disconnect may have attenuated the strength of observed associations. Future studies would benefit from concurrent collection of dietary and biological data, as well as follow-up measurements to assess the stability of dietary habits and introduce a longitudinal component to analyses of biological aging.

Although we conducted a discordant twin analysis focusing on veganism—as per the primary a prior goals of the study—this analysis was based on a small number of pairs and did not yield statistically significant results. We chose not to extend the discordant analysis to other dietary groups (e.g., pescetarians or flexitarians), despite their higher sample sizes, to avoid deviating from the study’s central hypothesis or introducing post hoc bias. However, future work could expand this approach, particularly as more twin datasets with detailed dietary exposure become available. Likewise, we did not conduct formal co-twin control analyses for the individual food components, as the intended nature of these additional analyses was solely exploratory. Notably, prior work within the same cohort has explored heritability for dietary behaviors including vegetarianism and pescetarianism [22], emphasizing that both genetic and shared environmental factors influence diet.

Our study highlights the potential influence of food components and dietary patterns on biological aging and presents a framework for future investigations into diet and aging. Our work is in line with recent studies that have found dietary patterns to associate with epigenetic age. For example, one study that analyzed data from 1995 participants using the “Dietary Approaches to Stop Hypertension” (DASH) score and three epigenetic age acceleration markers (Dunedin Pace of Aging, GrimAge acceleration, and PhenoAge acceleration) found associations between diet and epigenetic aging, where higher diet quality associated with lower deltaAges for all three biological aging scores [38]. Notably, the DASH diet is a high-quality food plan that emphasizes plant-based proteins from nuts and legumes, along with fruits, vegetables, low-fat dairy products, and limits red meat, sweets, and sugary drinks, which is partially in line with the pescetarian, vegetarian, and vegan diets outlined in our study. Recently, another study specifically assessed veganism in relation to epigenetic aging in a clinical trial with twins [10]. In the study, 22 pairs of twins were given vegan or non-vegan diets for eight weeks and found that the vegan cohort, following the plant-based eating pattern and excluding animal products, showed significant reductions in epigenetic age acceleration (Dunedin Pace of Aging, GrimAge acceleration, and PhenoAge acceleration) [9, 10]. This is in line with another recent study, that used the Dunedin Pace of Aging clock in a population of 48 females following a nutrient-dense plant-rich diet (Nutritarian) and 49 females on a standard American Diet (SAD), which found that the Nutritarian group exhibited slower epigenetic age acceleration compared to the SAD diet [39]. Our findings are also in line with the recent findings of Cribb et al. (2025) [40] reported that diets high in fiber were associated with reduced epigenetic age acceleration, whereas high-protein intake was linked to faster aging, in a population of 5310 Australians. Similarly, another recent study by Ravi et al. (2025) [41] performed in 826 twins from Finland observed that plant-rich diets correlated with slower biological aging across several epigenetic clocks. Together, these studies provide converging evidence that dietary composition may play a meaningful role in modulating biological age.

Taken together, while our results on vegan diets suffered from a small sample size, nonetheless, veganism significantly correlated with lower biological aging in two aging clocks (Hannum and Horvath), even after adjusting for confounders (age, sex, smoking, education, physical activity, BMI, alcohol), findings that align with those from other studies, reinforcing the significant role of lifestyle, dietary patterns, and specific food components in influencing biological aging. These findings provide a foundation for future research, which has the potential to uncover actionable insights into how dietary interventions can promote healthy aging and yield substantial benefits for individuals and society as a whole.

Author contributions

G.S., G.E.J. and L.L. conceived and designed the study. L.L., J.v.D. and E.d.G. curated the datasets and provided input on study design. G.E.J. performed the data analyses. G.E.J and G.S. wrote the manuscript with contribution from all authors.

Funding

Data collection in the Netherlands Twin Register was funded by the Netherlands Organization for Scientific Research (NWO), Grant Numbers 31160008, 904-61-193 and 985-10-002. The Netherlands Organization for Scientific Research (NWO): Biobanking and Biomolecular Research Infrastructure (BBMRI-NL, NWO 184.033.111) and the BBRMI-NL-financed BIOS Consortium (NWO 184.021.007), Genotype/Phenotype Database for Behaviour Genetic and Genetic Epidemiological Studies (ZonMw Middelgroot 911-09-032); Netherlands Twin Registry Repository: researching the interplay between genome and environment (NWO-Groot 480-15-001/674).

Data availability

The datasets generated and analyzed during the current study are part of the Netherlands Twin Register repository and are available at the Netherlands Twin Register upon reasonable request, and if proposed data usage is compatible with the original informed consent and IRB permissions. No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Written informed consent was obtained from all participants. The study was approved by the Central Ethics Committee on Research Involving Human Subjects of the VU University Medical Centre, Amsterdam, an Institutional Review Board certified by the U.S. Office of Human Research Protections (IRB Number IRB00002991 under Federal-wide Assurance-FWA00017598; IRB/institute codes, NTR 03-180).

Competing interests

The authors declare no competing interests.