Study populationsCooperative health research in the region of Augsburg (KORA)

KORA FF4 (2013–2014) is the second follow-up of KORA S4 cohort. Its baseline was conducted between 1999–2000 with participants of German nationality from the region of Augsburg, Germany, aged between 25 and 74 years old. From the 4261 participants enrolled in KORA S4, 2279 also participated in the second follow-up study. The KORA cohort ethical approval was granted by the ethics committee of the Bavarian Medical Association (REC reference numbers FF4: #06068) and it was carried out in accordance with the principles of the Declaration of Helsinki. Details about the KORA FF4 study protocol and recruitment were published elsewhere [36].

From the 2279 participants in the KORA FF4 study, individuals without DNAm or dietary data, pregnant women, and participants with severe blood disorders were excluded, resulting in a total number of 1354 participants.

Leiden longevity study (LLS)

The Leiden Longevity Study (LLS) [37] was established in 2002, with the aim of investigating the genetic component of exceptional survival and its interaction with environmental factors. From 2002 to 2006, long-lived, Dutch, Caucasian siblings (n = 944) were recruited with their offspring (n = 1671) and their offspring’s partners (n = 744).

Family eligibility required at least two long-lived living siblings who met a stringent, sex-specific age criterion (aged at least 89 years for males and 91 years for females). At the time of the study’s initiation, less than 0.5% of the Dutch population fulfilled this requirement as an individual, and sibships with multiple eligible members were estimated to represent less than 0.1% [38].

In the LLS population, offspring and other first-degree relatives are enriched for familial influences on longevity [37]. Their partners serve as controls, having comparable age, socio-economic status, location, lifestyle and environmental factors but without this genetic advantage. Recruited, living subjects completed a pedigree, questionnaires, and a non-fasted venous blood sample was drawn for isolation of DNA, RNA, serum, and plasma.

Dietary polyunsaturated fatty acid intakeCooperative health research in the region of Augsburg (KORA)

The participants were requested to answer at least two 24-h food lists (24HFL) and a food frequency questionnaire (FFQ) [35, 39]. Through a two-step model, the daily food intakes were estimated. In step one, a logistic linear mixed model was applied to estimate the probability of food item consumption for each participant based on the 24HFL data. The models were adjusted for age, sex, BMI, physical activity, smoking, education, and additionally for the frequency of food consumption (assessed by the FFQ). Since the 24HFL does not assess the amounts of food consumed, these were estimated in step two using data from the second Bavarian food consumption survey (BVS II), a cross-sectional study to assess the dietary habits of the Bavarian population. The amount distribution was not symmetrical, so the quantities were transformed using Box-Cox transformation and modeled using linear mixed-effect models adjusted for age, sex, BMI, smoking, physical activity and educational level. The estimated quantities were then transformed back to the original scale, obtaining the amount consumed per individual per item. By multiplying the probability of consumption and the estimated amount, the individual's usual intake of each food item was estimated. The nutrients provided by each food item were taken from the German Food Composition Table Bundeslebensmittelschlüssel (Version 3.0.2) [35].

The usual intakes (in mg per day) of the n-3 PUFAs alpha-linolenic acid (ALA, (C18:3, n-3)), stearidonic acid (SDA, (C18:4, n-3)), eicosapentaenoic acid (EPA, (C20:5, n-3)), docosapentaenoic acid (DPA, (C22:5, n-3)) and docosahexaenoic acid (DHA, (C22:6, n-3)), and the n-6 PUFAs linoleic acid (LA, (C18:2, n-6)), eicosadienoic acid (EDA, (C20:2, n-6)) and arachidonic acid (ARA, (C20:4, n-6)) were estimated for each participant and used as exposure variables in the EWAS models.

Leiden longevity study (LLS)

In the LLS study, a self-administered FFQ was used to assess dietary intake. The daily intake of the four fatty acids ALA, EPA, DHA and LA was estimated. The FFQ was designed for the Dutch population and based on the Vet Express [40] and extended with vegetables, fruit and foods to 104 items for estimating the intake of specific PUFA’s and other nutrients. The participants were asked to report the food intake during the previous month [41].

White blood cells percentage

The percentage of monocytes, basophils, eosinophils, neutrophils and lymphocytes in whole blood from each participant from the KORA study analysis was determined through differential blood count from the participant’s blood sample using the Coulter LH 750 device from Beckman Coulter and the Sysmex XN device. Three participants with malignant neoplasm of lymphatic and hematopoietic tissue (ICD9 codes: 200–208) were excluded from our analysis.

In the LLS, the percentage of white blood cell (WBC) types (neutrophils, lymphocytes, monocytes, eosinophils, and basophils) was measured with a blood Differential test in fasted blood samples. These cell types were included in all analyses.

DNA methylation dataCooperative health research in the region of Augsburg (KORA)

Genomic DNA extracted from whole blood from 1928 individuals from KORA FF4 was bisulfite converted using the EZ-96 DNA Methylation Kit (Zymo Research, Orange, CA, USA) in two batches (N = 488, N = 1440). Subsequent methylation analysis was performed on an Illumina (San Diego, CA, USA) iScan platform using the Infinium MethylationEPIC BeadChip according to standard protocols provided by Illumina.

Raw DNA methylation data were extracted with Illumina Genome Studio (version 2011.1), methylation module (v1.9.0), and processed using R (v3.0.1) following the CPACOR pipeline of Lehne et al. [42] including exclusion of 65 SNP probes and background correction using minfi [43]. Probes were set to missing if the detection p value ≥ 0.01 or the number of beads < 3. Samples were excluded if the detection rate was ≤ 0.95. Quantile normalization was performed on intensity values separated by color channel, probe type and M/U subtypes. The resulting methylated and unmethylated signals were used to calculate β values, a measure of percent methylation between 0 and 1.

Leiden longevity study (LLS)

DNA methylation data of whole blood samples were generated from 821 unrelated LLS participants by the Human Genotyping facility (HuGe-F, Erasmus MC, Rotterdam, The Netherlands) within the Biobank-Based Integrative Omics Studies (BIOS) consortium. DNA of LLS was analyzed using the Illumina 450 k BeadChip array. Genomic DNA (500 ng) was isolated and bisulfite converted using the Zymo EZ-96 DNA methylation kit (Zymo Research Corp, Irvine, CA, USA). 4 μl was then hybridized on the Infinium HumanMethylation450 BeadChip array (Illumina Inc, San Diego, CA, USA) according to the manufacturer’s protocol.

IDAT files were generated by the Illumina iScan BeadChip scanner and data quality was assessed in R using sample dependent and sample independent quality metrics reported by the Bioconductor package MethylAid (van Iterson et al., 2014) with default settings. Unreliable or outlying values were removed, including those indistinguishable from background noise (detection p value > 0.01), based on a low number of beads (n < 3), or with zero values for signal intensity. Following background correction and probe-type normalization, the data were checked for outlying samples using plots of the first two principal components (PCs) and any samples or probes with less than 95% success rate were removed. However, there were no outliers found in these checks indicating high quality data. The resulting methylated and unmethylated signals were used to calculate β values which range from 0 (completely unmethylated state) to 1 (completely methylated).

Potential confounding variables

The covariates selected for this study were age (years), sex (male/female), BMI (kg/m2), smoking (current/former/never smoker), WBC%, physical activity (active/ non-active), energy intake (kcal/day), current estrogen therapy (yes/no) and PUFA supplement intake (yes/no).

Physical activity was assessed on a four-level graded scale based on the amount of regular leisure time exercise per week during summer and winter in KORA. Based on this assessment, participants were categorized into active and non-active [44].

In LLS, a base model was run adjusting for age (years), sex (Male/female), BMI (kg/m2), smoking status (current/former/never), blood cell type proportions (monocytes, basophils, eosinophils, neutrophils, lymphocytes), and plate number.

Statistical analysis

In KORA, the association analysis between DNAm beta values and each PUFA was carried out in R, using linear regression model, with the methylation beta value as the dependent variable and the PUFA intake as the independent variable. To adjust for potential confounding, surrogate variables were calculated for each PUFA using the sva R package [45]. The model was adjusted for age, sex, BMI, smoking, WBC% and surrogate variables (Model 1). WBC% (monocytes, basophils, eosinophils, neutrophils and lymphocytes) were calculated from blood subfractions. We removed probes found on sex chromosomes and those that contain common genetic variants. We also removed probes that were ambiguously mapped as well as probes that were removed from the Illumina arrays [46]. The total number of probes was 697,732 and therefore we used a Bonferroni threshold of 7.17 × 10–8 (0.05/697,732) to determine significance.

In LLS, the association analysis between DNAm beta values and each PUFA was carried out in R package limma, with the methylation beta value as the dependent variable and PUFA intake as the independent variable. The base model was adjusted for age, sex, BMI, smoking, WBC% and plate number (Model 1) all as fixed effects.

We ran a second model (model 2) similar to the main one (model 1) but further adjusted for physical activity, energy intake, estrogen therapy and, if applicable, PUFA supplement intake, added to the model as fixed effects. The PUFA supplement intake was used as confounder only for the models analyzing ALA, DHA and EPA as independent variable. In LLS, only total energy intake (kcal/day) was additionally adjusted for in the extended model. We performed quality control for our analysis and the Manhattan and QQ plots can be found in the Supplementary Materials.

Meta-analysis

For the meta-analysis, we performed inverse variance fixed effects meta-analysis using METAL version 2011–03-25 [47]. For multiple testing correction, we used Bonferroni correction. A p value < 1.23 × 10–7 (0.05/406,132 DMPs) was used as the significance threshold.