The lipidomic correlates of epigenetic aging across the adult lifespan: A population‐based study

Abstract Lipid signaling is involved in longevity regulation, but which specific lipid molecular species affect human biological aging remains largely unknown. We investigated the relation between complex lipids and DNA methylation‐based metrics of biological aging among 4181 participants (mean age 55.1 years (range 30.0–95.0)) from the Rhineland Study, an ongoing population‐based cohort study in Bonn, Germany. The absolute concentration of 14 lipid classes, covering 964 molecular species and 267 fatty acid composites, was measured by Metabolon Complex Lipid Panel. DNA methylation‐based metrics of biological aging (AgeAccelPheno and AgeAccelGrim) were calculated based on published algorithms. Epigenome‐wide association analyses (EWAS) of biological aging‐associated lipids and pathway analysis were performed to gain biological insights into the mechanisms underlying the effects of lipidomics on biological aging. We found that higher levels of molecular species belonging to neutral lipids, phosphatidylethanolamines, phosphatidylinositols, and dihydroceramides were associated with faster biological aging, whereas higher levels of lysophosphatidylcholine, hexosylceramide, and lactosylceramide species were associated with slower biological aging. Ceramide, phosphatidylcholine, and lysophosphatidylethanolamine species with odd‐numbered fatty acid tail lengths were associated with slower biological aging, whereas those with even‐numbered chain lengths were associated with faster biological aging. EWAS combined with functional pathway analysis revealed several complex lipids associated with biological aging as important regulators of known longevity and aging‐related pathways.

enable the development of more tailored and effective preventive and therapeutic strategies aimed at maximizing health span.
Dynamic DNA methylation regulates gene expression and is responsive to environmental and lifestyle changes. With increasing age, the methylation status of numerous DNA cytosine-phosphateguanine (CpG) sites differentially changes across the genome (Bell et al., 2012(Bell et al., , 2019Florath et al., 2014;Horvath, 2013;Horvath & Raj, 2018;Jones et al., 2015). Hence, DNA methylation patterns have been used to estimate biological age through so-called epigenetic clocks, also known as epigenetic aging estimators. The most recent of these include DNAm Phenotypic Age (PhenoAge), trained on mortality-related clinical biomarkers (Levine et al., 2018), and DNAm GrimAge, developed using plasma proteins that are associated with age-related conditions (Lu et al., 2019). Both PhenoAge and GrimAge more closely capture the high inter-individual variability in the underlying biological aging processes and are strong predictors of mortality (Beynon et al., 2022;Hillary et al., 2021;Lemke et al., 2022;Li et al., 2021;Protsenko et al., 2021;Roberts et al., 2021;Verschoor et al., 2021). Recent studies have shown that GrimAge outperforms other biological age estimators in the prediction of age-related diseases and all-cause mortality and in reflecting multisystem dysfunction McCrory et al., 2021;Protsenko et al., 2021).
The determinants of biological aging remain largely unknown, yet lipid metabolism has been suggested to be involved (Hahn et al., 2017;Mutlu et al., 2021). Key pathways that have been implicated in the aging process, including the insulin-like growth factor-Akt-mTOR pathway, the nuclear factor kB (NF-kB) pathway, and the AMP-activated protein kinase (AMPK) pathway, are also crucial regulators of lipid metabolism (Jesko et al., 2019;van der Spoel et al., 2015;Weir et al., 2017). Various circulating lipid species have been linked to age-related phenotypes, including cardiovascular diseases (Tabassum et al., 2019), insulin resistance (Lemaitre et al., 2018), obesity (Yin et al., 2020), chronic kidney disease (Afshinnia et al., 2021), and Alzheimer's disease (Huynh et al., 2020). Further evidence for the crucial role of lipid metabolism in aging stems from studies of nonagenarians and centenarians, as well as their offspring, which found favorable lipid profiles in healthy agers (Collino et al., 2013;Gonzalez-Covarrubias et al., 2013;Montoliu et al., 2014;Pradas et al., 2019;Vaarhorst et al., 2011). Notably, in the Leiden Longevity Study, it was found that the offspring of nonagenarians had higher levels of phosphocholine (PC) and sphingomyelin (SM) species and lower levels of phosphoethanolamine (PE) (38:6) and long-chain triacylglycerols (TAGs), independent of total triglyceride levels (Gonzalez-Covarrubias et al., 2013;Vaarhorst et al., 2011).
Similarly, Pradas et al. found that higher levels of alkyl-PC with shorter chain lengths and double bonds and/or lower levels of alkenyl-PE with longer chain lengths and double bonds were associated with human longevity (Pradas et al., 2019).
A recent large-scale epigenome-wide association study (EWAS) (N = 16,265) found that hundreds of CpGs were associated with HDL, LDL, and TAGs in either trans-ethnic or ethnic-specific metaanalyses (Jhun et al., 2021). EWAS combined with Mendelian randomization analyses indicated that inter-individual variations in lipid levels are likely causally related to changes in DNA methylation (Dekkers et al., 2016;Jhun et al., 2021). Furthermore, many of the lipid-related CpGs have also been linked to age-related phenotypes, including metabolic syndrome, type 2 diabetes, and coronary artery disease (Gomez-Alonso et al., 2021;Hedman et al., 2017;Xie et al., 2019). Taken together, current evidence suggests that variations in lipid levels may exert at least part of their (pathological) effects through epigenomic remodeling.
Despite the intriguing connection between lipid metabolism and aging, it is still unknown whether and how inter-individual differences in lipid profiles contribute to different rates of biological aging in the general population. The heterogeneous chemical structure of lipids poses challenges for their accurate quantification, and until now only a few lipid species have been investigated in the context of human aging and age-related health outcomes. Yet the vast diversity of lipid functions is reflected by the wide variation in the structure and composition of lipid molecules, which ultimately determine their specific effects (Harayama & Riezman, 2018). Recently, highthroughput, in-depth molecular characterization of many lipid species has become available through Metabolon's complex lipid assay platform (Wu et al., 2022). This lipidomics platform provides absolute quantitation of 14 lipid classes across phospholipids, sphingolipids, and neutral lipids, as well as the complete fatty acid composition of each lipid class, covering more than 900 absolute concentrations of their constituent molecular species. These recent technological developments make it possible to investigate the contribution of complex lipids to biological aging at a population level. Furthermore, we performed epigenome-wide association analyses of biological aging-associated lipid molecular species and pathway analysis to gain biological insights into the mechanisms underlying the effects of lipidomics on biological aging.

| Estimations of AgeAccelPheno and AgeAccelGrim
The characteristics of the study population are presented in Table 1.
AgeAccelPheno and AgeAccelGrim were significantly higher in men than in women. The concentrations of the measured plasma lipid classes are consistent with the quantitative plasma lipidomic analyses of a similar study (Eichelmann et al., 2022).

| Associations of lipid class and molecular species with AgeAccelPheno and AgeAccelGrim
Age, sex, and batch-adjusted partial correlations showed that all lipid classes except for MAG, LPE, and LPC were moderately to highly correlated (Pearson's r > 0.3). The strongest correlations for LDL and cholesterol were with CE, PC, PI, and sphingolipids, whereases HDL was only weakly correlated with almost all main lipid classes. As expected, clinically measured total triglycerides were highly correlated with DAG and TAG, and moderately correlated with phospholipids and sphingolipids. BMI was only weakly correlated with the main lipid classes ( Figure S1).  Note: The missingness for each variable is less than 5%.
*Comparison between women and men, adjusted for age.
LCER classes, AgeAccelPheno and AgeAccelGrim decreased with 0.25-1.00 year per each SD concentration increase in each molecular species. More molecular species were associated with AgeAccelGrim, and with larger effect sizes, than with AgeAccelPheno ( Figure 1).
After adjustment for HDL and/or LDL levels, the associations of many molecular species belonging to the neutral lipids and phospholipids classes-especially TAG, DAG, PC, and PEwith AgeAccelPheno and AgeAccelGrim became non-significant, Conversely, adjustment for HDL and/or LDL levels did not materially affect the results for sphingolipids. These findings are in line with the known function of lipoproteins as key regulators of mainly neutral lipids and phospholipids metabolism, whereas they have a smaller influence on sphingolipids metabolism (Borodzicz et al., 2015). The results remained almost identical after adjustment for BMI ( Figure 1).
We found that DAG, MAG, and TAG levels tended to be stable across the different age groups and PI levels tended to decrease over the years. Lipid levels in CE, PC, and sphingolipids tended to decrease from the age of 60, with significantly lower levels of these lipids in participants from the 80-89 age group. Unfortunately, only seven participants were older than 90 years, which rendered it impossible to obtain precise estimates for this age group as reflected in the vary wide confidence intervals ( Figure S2).

| Associations of the total number of carbons and double bonds with AgeAccelPheno and AgeAccelGrim
Within the neutral lipid category, higher levels of TAG molecular species with even numbers of carbons were associated with higher AgeAccelPheno ( Figure 2a) and AgeAccelGrim (Figure 2b) Figure S3).

| Associations of fatty acid composition across lipid classes with AgeAccelPheno and AgeAccelGrim
The number of acyl chain carbons in a lipid's fatty acid tail may define specific biological effects (Harayama & Riezman, 2018). We analyzed were associated with lower AgeAccelPheno and AgeAccelGrim. In addition, shorter fatty acid tails were related to larger effect sizes.
Importantly, the direction of the effects also depended on the lipid class (Figures 1 and 3).

| Associations of saturation of fatty acid tails with AgeAccelPheno and AgeAccelGrim
Differences in the content and fraction of mono-and polyunsaturated lipids determine membrane peroxidation, which has been linked to longevity (Gonzalez-Covarrubias et al., 2013;Pradas et al., 2019). For MUFA lipids, the effect on AgeAccelPheno and AgeAccelGrim was stronger with shorter (even numbered) chain lengths of the fatty acid tail. In contrast, for PUFA lipids with the same chain length, fewer double bonds were related to a stronger effect on AgeAccelPheno and AgeAccelGrim ( Figure 4).

| Sex interaction and sex-stratified analyses
We found that lipid species, which were differently associated with AgeAccelPheno and AgeAccelGrim in men and women, predominately belonged to TAG and some phospholipid species ( Figure S4 Figure S6).

F I G U R E 2
Associations of the total number of carbons and double bonds in lipid species with AgeAccelPheno and AgeAccelGrim. Individual lipid species are depicted by filled circles and arranged by lipid class according to the total number of carbon atoms (x-axes) and the number of double bonds (y-axes). The circle color indicates the magnitude and direction (positive or negative) of the effect size, and the circle size corresponds to the significance level. Lipids with the same number of carbon atoms and double bonds are pulled apart vertically to increase their visibility. FDR, false discovery rate; LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; PC, phosphatidylcholine; PE(O), phosphatidylethanolamine ether; PE(P), phosphatidylethanolamine plasmalogen; PE, phosphatidylethanolamine; PI, phosphatidylinositol; TAG, triacylglycerol.

| Uncovering biological pathways involved in the association of lipidome with AgeAccelPheno and AgeAccelGrim
Epigenome-wide association analyses were performed for 525 lipids, which were identified as being associated with either AgeAccelPheno or AgeAccelGrim (Figure 1). This approach resulted in the identification of lipid-associated CpGs that were subsequently used as uniquely defined proxies for the respective lipids, enabling KEGG pathway analysis to delineate the underlying biological pathways modulated by AgeAccel-associated lipids. A total of 65 pathways were identified, including many known longevity-related pathways such as the mTOR signaling pathway, AMPK signaling pathway, MAPK signaling pathway, and growth hormone synthesis, secretion, and action pathway. Moreover, pathways involved in age-related diseases, including type 2 diabetes mellitus, insulin resistance and secretion, cortisol synthesis, and secretion, and longterm depression were among those related to lipid-associated CpGs.

| DISCUSS ION
We investigated 14 complex lipid classes, covering 964 molecular species and 267 fatty acid composites, with biological aging. We in silico pathway analysis revealed that lipids that were associated with biological aging estimators were mainly involved in known longevity and aging-related pathways, revealing their role as potential determinants of biological aging across the lifespan in the general population.
Our observation that higher levels of odd-numbered fatty acid tail lengths (15:0 and 17:0) were associated with slower biological aging, whereas even-numbered fatty acid tail lengths were associated with faster biological aging, fits with findings from previous studies. The EPIC-InterAct study (n = 27,296) found that higher levels of odd-chain saturated fatty acids (15:0; 17:0) were associated with a reduced risk of type 2 diabetes, whereas the risk was increased for people with higher levels of even-chain saturated fatty acids (Forouhi et al., 2014). Likewise, the EPIC-Norfolk study (n = 7354) found that higher levels of even-chain saturated fatty acids were associated with a higher risk of incident coronary heart disease (CHD), which carbohydrates and alcohol are converted to fatty acids in the liver or adipose tissue. Biochemical experiments also demonstrated toxic effects of 16:0, 18:0, and 24:0, including activation of inflammatory cytokines and lipotoxicity to pancreatic β cells (Jalili & Hekmatdoost, 2021;Zhou et al., 2020). This suggests that the lipid composition of the diet might have an impact on the rate of biological aging.
Very little work has explicitly assessed the value of LPC species as potential human blood-derived biomarkers of human aging.
Circulating LPCs are generated by phospholipases A2 from the PC.
Previous studies have shown that LRs and their components regulate nutrient-sensing pathways, calcium homeostasis, synaptic and neurotrophin signaling (Zhang et al., 2022). Indeed, our EWAS combined with functional pathways analyses revealed that biological aging-related lipids converged on pathways related to lipid signaling pathways (i.e., sphingolipids, phosphatidylinositol and phospholipase D signaling pathway), nutrient-sensing pathways (i.e., AMPK, MAPK, insulin, cAMP, and mTOR signaling pathway), and neuroplasticity pathways (i.e., cholinergic/glutamatergic/GABAergic synapse, axon guidance, calcium and neurotrophin signaling pathway). It is noteworthy that nutrient-sensing alterations reportedly affect the lifespan (de Lucia et al., 2020). It is thus possible that plasma complex lipids fluctuations may disrupt LRs in lipids and nutrient-sensing signaling, thereby contributing to biological aging (de Lucia et al., 2020;Zhang et al., 2022). In addition, LRs have been detected at synapses and are essential for synapse development, Aβ production, and cholesterol efflux. Complex lipid changes and subsequent loss of these microdomains may therefore possibly impact pre-and postsynaptic function, ultimately leading to (neuro)pathological aging (Egawa et al., 2016;Mesa-Herrera et al., 2019).

HDL and LDL levels are well-established biomarkers for various
CVDs and are widely used in assessing CVD risk in the clinic (Singh et al., 2020;Vallejo-Vaz et al., 2017). We demonstrate that the effects of neutral lipids and phospholipids on biological aging largely depend  (Boren et al., 2022). In contrast, sphingolipid metabolism is more independent of lipoproteins (Borodzicz et al., 2015). Our findings provide a detailed overview of the differential effects of lipoproteins across a wide range of lipid classes and species and underscore the importance of accounting for their potentially confounding effects in lipidomics analyses.
The main strength of our study is that we were able to delineate

| CON CLUS ION
In conclusion, diverse complex lipid species are associated with different rates of biological aging, with lipid class as well as fatty acid chain length and saturation as key determinants of their influence on biological aging. These findings emphasize the importance of investigating in-depth lipidomics in aging research beyond the standard clinical lipid panel. Since lower levels of LPC species were predominantly associated with slower biological aging and have been linked to age-related biological mechanisms (e.g., oxidative stress and mitochondrial dysfunction), they represent promising candidate human blood-derived biomarkers of human aging. Finally, investigating the sources of different lipids which have disparate association patterns with biological aging may increase our understanding of the underlying biological mechanisms.

| Study population
This study was based on the Rhineland Study, an ongoing singlecenter, population-based cohort study among people aged 30 years and above in Bonn, Germany. All individuals living in two pre-defined recruitment areas are invited to participate in the study. The only exclusion criterion is an insufficient command of the German language to provide informed consent. One of the Rhineland Study's primary objectives is to identify determinants and markers of healthy aging, applying a deep-phenotyping approach. At baseline, participants complete an 8-h in-depth multi-domain phenotypic assessment, and various types of biomaterials are collected. Approval to undertake the study was obtained from the ethics committee of the University of Bonn, Medical Faculty. We obtained written informed consent from all participants in accordance with the Declaration of Helsinki.
For the current analysis, we used baseline data of the first 4471 participants of the Rhineland Study for whom complex lipids data were available. We excluded participants without methylation data (n = 290). The final analytical sample consisted of 4181 participants.

| DNA methylation quantification
Genomic DNA was extracted from buffy coat fractions of anti- probe-level quality control were performed using the "minfi" package (Fortin et al., 2017) in R. Probes with a missing rate > 1% (at a detection p-value > 0.01) were excluded. Samples with sex mismatch or a missing rate at >1% across all probes were also excluded following previously published recommendation guidelines for analyzing methylation data (Wu & Kuan, 2018).

| Estimation of biological age
Biological age was estimated as PhenoAge and GrimAge, which were calculated based on the algorithms developed by Levine et al. (2018) and Lu et al. (2019), using 513 and 1030 CpG sites, respectively.
Discrepancies between an individual's chronological and estimated biological age (PhenoAge/GrimAge), referred to as AgeAccelPheno and AgeAccelGrim, were defined as the residual (in years) that results from regressing PhenoAge/GrimAge on chronological age.
AgeAccelPheno and AgeAccelGrim represent the residual variation in estimated biological age independent of chronological age, indicating whether individuals have biologically aged faster or slower in comparison to their chronological age (Fox et al., 2023). Specifically, an increase in AgeAccelPheno and/or AgeAccelGrim with changing lipid levels is referred to as accelerated biological aging, whereas a decrease in AgeAccelPheno and/or AgeAccelGrim with changing lipid levels is referred to as slower biological aging (Jain et al., 2022;McCrory et al., 2021). In total, 1050 molecular species and 278 fatty acid compositions covering these 14 classes were measured. Individual lipid species that contained more than 90% missing values across all the participants were not included (86 molecular species and 11 fatty acid compositions), leaving a total of 964 molecular species and 267 fatty acid compositions for the analyses (Table S2)

| Demographic and health variables
We included age and sex as demographic factors.
Differences between women and men were compared using linear regression for continuous variables, and logistic regression for categorical variables adjusting for age. Age and sex-adjusted partial correlations were used to assess the correlations among standard clinical lipid measures (i.e., LDL, HDL, total cholesterol, TAG, LDL/ HDL ratio), BMI, and the 14 main lipid classes and the partialcorrelation matrix was generated using "corrplot" R package. All lipid variables were z-transformed to have a mean of 0 and a standard deviation of 1 before further analyses to enable a better comparison of the effect sizes across different lipid classes. We used complete data without the imputation of missing values for all analyses.

| Association analysis
We used multiple linear regression analyses to quantify the association between each lipid class, molecular species, fatty acid composition (independent variables), and AgeAccelPheno/AgeAccelGrim (dependent variables). The base model was adjusted for sex, batch information of lipids and methylation, and smoking status. As HDL and LDL transport lipids in the circulation which could confound the association between complex lipids and biological age estimators, we further adjusted for HDL and LDL, both separately and jointly. To assess whether the associations of lipid species with AgeAccelPheno/ AgeAccelGrim were independent of BMI, we additionally adjusted for BMI. We used the false discovery rate (FDR) method to account for multiple comparisons, considering FDR <0.05 as statistically significant.
The overall patterns between all lipid classes/species and AgeAccelPheno/AgeAccelGrim were shown as forest plots. Patterns across all species of specific lipid classes were also shown as forest plots. To assess whether the effect (strength and direction) of the associations depended on the total number of carbons and double bonds in the lipids, the effect estimates of the lipid species were plotted as circles with their position in the 2-dimensional lipid class graphs determined by the total carbon number (x-axis) and double bonds (y-axis). To assess whether the effect (strength and direction) of the associations depended on the number of carbons in one specific fatty acid tail, heat maps were created for the effect estimates of the lipid species at FDR <0.05 level. To examine whether the associations differed by the degree of saturation in each lipid molecular species, beta estimates across lipid classes were shown as forest plots. The plots were generated using the "ggplot2," "ggpubr," "ggrepel," and "gridExtra" R packages.
We also assessed the association between chronological age and the levels of the 14 lipid classes in seven different age groups (i.e., 30-39 years old, 40-49 years old, 50-59 years old, 60-69 years old, 70-79 years old, 80-89 years old, and ≥90 years old), adjusting for sex.

| Sex interaction and sex-stratified analysis
To examine sex differences between lipid class and lipid molecular species and AgeAccelPheno/AgeAccelGrim, we assessed the interaction effects between sex and each lipid species on AgeAccelPheno/ AgeAccelGrim. In case statically significant sex-lipid interactions were identified, additional sex-stratified analyses were performed.

| Pathway analysis
Epigenome-wide association analyses were performed to examine the association between AgeAccelPheno/AgeAccelGrim-associated lipid species (independent variable) and DNA methylation level (outcome) using multiple linear regression while adjusting for age, sex, smoking status, batch information of lipids and methylation, first 10 genetic principal components (PCs) which account for residual population stratification. The associated CpGs were used as proxies for each lipid to interrogate the Kyoto Encyclopedia of Genes and Genomes (KEGG) database for pathway analysis using g:Profiler (https://biit.cs.ut.ee/gprof iler/). A bar plot to visualize the KEGG pathway results was generated using "ggplot2" R package.

AUTH O R CO NTR I B UTI O N S
Dan Liu was involved in conceptualization, methodology, formal analysis, writing-original draft preparation, and visualization; N.
Ahmad Aziz was involved in conceptualization, methodology, supervision, and writing-reviewing and editing; Elvire N. Landstra was involved in methodology and writing-reviewing and editing; Monique M.B. Breteler was involved in conceptualization, methodology, resources, writing-reviewing and editing, supervision, data curation, and funding acquisition.

ACK N OWLED G M ENTS
The work was partly supported by the German Research Foundation

CO N FLI C T O F I NTE R E S T S TATE M E NT
None of the authors has any conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The Rhineland Study's dataset is not publicly available because of data protection regulations. Access to data can be provided to scientists in accordance with the Rhineland Study's Data Use and Access Policy. Requests for further information or to access the Rhineland Study's dataset should be directed to rs-duac@dzne.de. The corresponding author takes responsibility for the integrity of the data.