Cell-free DNA methylation patterns in aging and their association with inflamm-aging

Abstract

Aim: Liquid biopsies analyzing cell-free DNA (cfDNA) methylation in plasma offer a noninvasive diagnostic for diseases, with the potential of aging biomarkers underexplored. Methods: Utilizing enzymatic methyl-seq (EM-seq), this study assessed cfDNA methylation patterns in aging with blood from 35 healthy individuals. Results: It found aging signatures, including higher cfDNA levels and variations in fragment sizes, plus approximately 2000 age-related differentially methylated CpG sites. A biological age predictive model based on 48 CpG sites showed a strong correlation with chronological age, verified by two datasets. Age-specific epigenetic shifts linked to inflammation were revealed through differentially methylated regions profiling and Olink proteomics. Conclusion: These findings suggest cfDNA methylation as a potential aging biomarker and might exacerbate immunoinflammatory reactivity in older individuals.

Plain language summary

Our bodies undergo many changes as we age, some of which might affect our health. To better understand these changes, scientists study something called ‘cell-free DNA' (cfDNA) in our blood. This cfDNA can give us clues about our health and the risk of diseases like cancer or heart conditions.

In our research, we analyzed cfDNA from the blood of 35 people to identify patterns associated with aging. We discovered that approximately 2000 specific spots in our DNA change in a way that's linked to aging. These changes might help us figure out someone's biological age – essentially, how old their body seems based on various health factors, which can differ from their actual age.

We also found that these DNA changes could indicate how aging might make the body's defense system – which fights off diseases – react more intensely. Understanding this could be crucial for managing health as we get older.

Our study suggests that cfDNA could be a useful marker for aging, offering a new approach to understanding and possibly managing the health effects associated with growing older.

TWEETABLE ABSTRACT

A new study reveals cfDNA methylation in blood as a promising noninvasive marker for aging. Analysis of 35 individuals highlights key aging signatures and approximately 2000 age-related changes in DNA methylation, linking aging with inflammation. #AgingResearch #LiquidBiopsy.

Plain language summary

Summary points.

• First application of EM-seq for aging in healthy individuals, achieving high-quality data from limited cell-free DNA (cfDNA).

• Older individuals showed increased cfDNA concentrations, suggesting higher cell turnover or death rates.

• Elevated cfDNA from neutrophils and colon epithelial cells observed in older compared with younger groups.

• Analysis identified around two thousand DMCs across age groups, emphasizing nucleosome structure genes.

• A 48-CpG-sites cfDNA methylation model strongly correlated chronological and biological age.

• Significant methylation changes, including hyper- and hypo-methylation, were detected in older individuals.

• Gene promoter analysis linked increased inflammation pathways with age, supported by proteomics data.

• The study highlights cfDNA methylation's potential as an aging predictive tool, especially for inflammation-related changes.

1. Introduction

Aging is a natural and intricate phenomenon characterized by a gradual and irreversible decline in physiological functions across multiple organs [1]. This decline in physiological functions makes aging a significant risk factor for various diseases, such as diabetes, Alzheimer's disease (AD), Parkinson's disease (PD), cardiovascular diseases, chronic obstructive pulmonary disease, osteoporosis and osteoarthritis [2]. Individual-level aging measurements, particularly the study of biological markers of aging, can capture interindividual differences in disease onset, functional decline and mortality [3,4]. These measurements play a pivotal role in distinguishing chronological age from biological aging. As the aging population increases, individual-level biological markers, such as epigenetic biomarkers, genomic biomarkers, transcriptional biomarkers and clinical biomarkers, become crucial for evaluating interventions aimed at promoting healthy aging.

Plasma cell-free DNA (cfDNA) is derived from DNA released after cellular damage and breakdown and undergoes enzymatic cleavage into fragments of approximately 167 nucleotides in size [5]. The dynamic processes of cfDNA generation and clearance render it an ideal molecular marker for various diseases, including cancer screening, prenatal diagnoses and non-tumor-related conditions, such as autoimmune disorders [6]. A previous study found that the tissue origin of cfDNA can be traced through tissue- and organ-specific cfDNA information, such as nucleosome-coating patterns, organ-specific methylation site information and the association between adjacent methylation sites, which provides a basis for aging and disease assessment of organs and tissues through cfDNA [7]. A study on cfDNA in different populations found that, compared with young people, the nucleosome distribution of cfDNA in the elderly was more irregular, which once again suggested that cfDNA could be used as a biomarker of aging [8].

DNA methylation, which involves the addition of a methyl group to the fifth carbon of cytosine (5-methylcytosine [5mC]), is a typical epigenetic alteration. Correspondingly, the methylation patterns observed in cfDNA are consistent with the cells or tissues of origin [7]. Although DNA methylation has been widely recognized as an epigenetic clock, little is known about the specific changes in cfDNA methylation during the aging process [9,10]. While cfDNA methylation has been used as a surrogate source of tumor DNA for cancer diagnosis and prognosis [11], its potential role as a biomarker for aging remains unexplored.

Inflamm-aging, characterized by chronic inflammation during aging, is partly attributed to senescent cells that secrete inflammatory compounds [12]. A meta-analysis revealed that elevated levels of both total and unmethylated cfDNA are linked to systemic inflammation and increased frailty in nonagenarians [13]. CfDNA plays a role in innate immunity, mainly by TLR-9 in specific immune cells [14,15]. However, the extent of methylation in cfDNA is cell-specific and decreases with age, enhancing its immunostimulatory potential in elderly individuals [10]. This is further intensified in conditions such as systemic lupus erythematosus (SLE), where cfDNA with reduced 5mC levels is more inflammatory [16]. Furthermore, the interaction of cfDNA with proteins such as RAGE, which take up cfDNA into endosomes and whose levels increase with age, may further promote TLR-9 activation [17,18]. Despite these insights, the relationship between cfDNA methylation and inflamm-aging is still not well understood.

Although there are existing reports of cfDNA as a biomarker for biological aging [8], research on aging-related changes in cfDNA methylation has been limited. This study aims to explore the cfDNA methylation profiles during aging and to evaluate the potential of cfDNA methylation as a biomarker for aging. To understand the relationship between cfDNA methylation and age-associated inflammation, we analyzed 92 inflammatory markers using Olink Proteomics to assess inflammation levels.

2. Materials & methods

2.1. Volunteer recruitment

We recruited 35 healthy volunteers from the Chinese population, across various age groups. These included 11 young adults 22–31 years old (5 males and 6 females), 10 middle-aged adults 57–64 years old (5 males and 5 females), and 14 older adults 75–88 years old (7 males and 7 females). All participants had a body mass index (BMI) <30. All participating donors were healthy. The study excluded individuals with a history of conditions, such as dementia, PD, AD, cancer, infectious diseases and severe inflammatory diseases, including hepatitis B or C, Crohn's disease and systemic lupus erythematosus (Figure 1 & Supplementary Table S1).

Figure 1.

Overview of the selection criteria and study design. The histogram depicts the age distribution across young (n = 11), middle-aged (n = 10), and older participants (n = 14).

The study was performed in accordance with the Helsinki Declaration and approved by the Ethics Committee of Beijing Hospital (reference number: 2019BJYYEC-054-02). Informed consent was obtained from all participants, and no identifying information was included in the text.

2.2. cfDNA preparation & EM-seq library production

cfDNA was extracted from 2 to 3 ml of plasma samples using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany). Quantification of genomic DNA was carried out using a Qubit 3.0 Fluorometer (Thermo Fisher Scientific, USA). The extracted cfDNA was then subjected to adapter ligation using a NEBNext^® Enzymatic Methyl-seq Kit (New England Biolabs, MA, USA) [19]. Post-ligation, 5-methylcytosine and 5-hydroxymethylcytosine in cfDNA were oxidized using TET enzymes, preventing them from undergoing deamination by subsequent APOBEC enzymes. Conversely, unmodified cytosines are deaminated by APOBEC enzymes, converting them into uracil. Subsequently, the enzymatically converted cfDNA was subjected to polymerase chain reaction (PCR) amplification and purification to obtain the final EM-Seq library. The prepared library was then sequenced on an Illumina NovaSeq 6000 system (Illumina) in 150-bp paired-end sequencing mode.

2.3. A single-base analysis of cfDNA methylation sites

Sequencing on the Illumina NovaSeq 6000 system was followed by an image analysis, base calling and quality control to generate the raw reads. Quality control was initially performed using the Q30 threshold. Adapters and low-quality reads were removed using the Cutadapt software program (version 1.9.3), resulting in high-quality reads. These reads were then aligned to the reference genome using the Bismark software program (version 0.16.3). The Bismark Methylation Extractor component was used to determine the genome-wide methylation status of each cytosine (C) site. EM-seq facilitates the examination of C methylation at a single-base resolution. We used the methylation_extractor command in Bismark to quantify the methylation of each C across the entire genome. The outcomes were presented in a bedGraph format file, along with a coverage file that included counts for each C site.

2.4. Identification of cfDNA-contributing tissues & cells

We implemented the method developed by Dor et al. (2018) to identify the cellular origin of cfDNA [7]. This algorithm utilizes non-negative least-squares regression and references DNA methylation signatures, covering 25 major human cell types that are frequently involved in various diseases [7]. The deconvolution algorithm approximated plasma cfDNA methylation profiles as a linear combination of cell type-specific methylation profiles from the reference atlas. This approach has proven effective in identifying the cellular origins of cfDNA, even with a limited dataset comprising approximately 4000 CpG sites. In our analysis, only CpG sites with at least fivefold coverage in each sample were included in the deconvolution process, which was conducted using the R software program (version 3.6.3). To ascertain the concentration of cfDNA from particular tissues or cell types, we calculated the relative estimated proportions of cfDNA and multiplied these by the total cfDNA concentration found in the same plasma sample. Our final analysis focused on 13 tissue and cell types that demonstrated measurable cfDNA concentrations in the majority of participants.

2.5. Identification of differentially methylated CpG sites

Our analysis of differential methylation across groups involved calculating the average methylation rate at each CpG site within the groups. The log2 fold change (log2(FC)) between groups was computed using the formula log2(FC) = log2((avgA + 1)/(avgB + 1)), where ‘avgA’ and ‘avgB’ represent the average methylation rates in groups A and B, respectively. A t-test was employed to assess the significance of the differences in methylation rates. To address multiple testing issues and decrease the false discovery rate (FDR), the p-values were adjusted using the Benjamini & Hochberg method. CpG sites that had an adjusted p-value (FDR) <0.05 and |log2(FC)| >1 were identified as differentially methylated, indicating significant loci of methylation variation between the groups.

2.6. A functional epigenetic module analysis of cfDNA differentially methylated CpG sites

We conducted a functional epigenetic module (FEM) analysis, a method that combines gene interaction networks with differential methylation information [20]. This approach helps identify the interaction modules of genes that show changes in methylation levels, revealing gene sets associated with sample phenotypic data. The CHAMP package in R, version 3.6.3, was used for the FEM analysis [21].

During this process, each gene was represented as a node in the protein-protein interaction (PPI) network. The relationships between these nodes were defined according to the established connections of the network. The color coding of each node was determined based on the T-values of differential methylation, providing a visual representation of methylation changes. Nodes with T-values <-1.5 are colored in a gradient from yellow to white, indicating hypomethylation, while nodes with T-values >1.5 are shown in a gradient from light blue to blue, indicating hypermethylation [20]. Genes with T-values between these ranges are shown in gray. This color spectrum effectively highlights the extent and direction of methylation alterations across genes, aiding in pinpointing hub genes that are central to the methylation changes observed in our aging study cohort.

2.7. Mapping EM-seq methylation data to Illumina bead-Array platforms

For aligning cfDNA EM-seq sequencing data with the Infinium HumanMethylation450 BeadChip assay format, we adopted a mapping method akin to the one developed by Christensen in 2016 [22]. This method maps bisulfite sequencing methylation data to the CpG sites measured by the widely used Illumina bead-array platforms. The process facilitated the integration of our EM-seq data with existing datasets and methodologies that are predominantly based on the Illumina methylation bead array, enabling a broader comparison and validation of our findings within the context of the extensive body of research already conducted using the Infinium HumanMethylation450 platform.

2.8. Development & validation of the cfDNA methylation age prediction model

To construct the cfDNA methylation age prediction model, we employed the lasso regression method within the Glmnet package, which is renowned for its efficiency in handling high-dimensional data typical of methylation datasets. Model optimization was achieved using leave-one-out cross-validation (LOOCV), which fine-tuned the λ parameter to minimize the mean absolute error (MAE) and maximize the correlation coefficient (r) between the predicted and actual ages. This approach is critical for selecting the simplest yet most effective model, minimizing overfitting while maintaining predictive accuracy.

Subsequent validation was conducted against external datasets from the Gene Expression Omnibus (GEO) database [23,24]. This crucial step evaluated the model's generalizability and performance beyond our initial dataset. By applying the model to independent datasets with varied age ranges and demographics from GEO, we assessed its robustness and reliability for predicting biological age based on cfDNA methylation patterns. This external validation ensured the reliability of the model for broader applications in diverse research and clinical contexts, confirming its suitability for use in various population groups and settings.

2.9. Identification & annotation of differentially methylated regions

We used the Bioconductor package DSS (Dispersion Shrinkage for Sequencing data) to identify statistically significant methylated regions in our study [25]. This involved constructing negative control datasets for each sample and focusing on regions with a p-value < 0.05 and a methylation difference >0.1, as per the software program's default settings. Following the identification of differentially methylated regions (DMRs), annotations were performed to understand their potential functional implications. This process categorizes DMRs based on their relative positions to known genes, which is crucial for inferring their roles in gene regulation. By associating DMRs with specific genomic features, such as gene bodies, promoters or intergenic areas, we were able to provide insights into the biological significance of the observed methylation changes.

2.10. The KEGG pathway & gene ontology (GO) analyses of genes linked to DMRs

While single-methylated CpGs occasionally correlate with gene expression regulation and can influence disease susceptibility [26], the majority of DMRs detected in studies are several hundred to a few thousand bases in length [27]. This range aligns with the typical dimensions of gene regulatory regions, suggesting that DMRs may play a role in mediating cell type-specific transcriptional repression [28,29]. The KEGG pathway analysis involves mapping molecular datasets from genomics, transcriptomics, proteomics, and metabolomics onto KEGG pathway maps to interpret the biological functions of these molecules. In this study, we performed a pathway analysis on genes associated with DMRs in the promoter regions (TSS-2000 to TSS+2000), aiming to annotate and hypothesize the pathways in which these DMRs might be involved [27]. We set a threshold p-value of < 0.05 for significant enrichment. For the KEGG pathway analysis, multiple testing correction has been applied, with the FDR-adjusted p-values presented in Supplementary Table S4.

The GO project (http://www.geneontology.org) developed a structured, controlled vocabulary for annotating genes, gene products and sequences. We conducted a GO functional analysis of genes associated with DMRs in the promoter regions (TSS-2000 to TSS+2000) to annotate and speculate the potential roles of these methylation regions [27]. GO terms with a p-value < 0.05 were considered statistically significant, providing insights into the molecular functions, biological processes and cellular components associated with the DMRs identified in our study. This analysis helps further our understanding of the biological implications of methylation changes in aging and inflammation. In the GO analysis, we have also implemented multiple testing correction, and the FDR-adjusted p-values are documented in Supplementary Table S5.

2.11. Protein analyses

In our study, we analyzed the relative concentrations of 92 different proteins using the Proseek Multiplex Inflammation I Probe kit 96 × 96 from Olink Bioscience (Uppsala, Sweden) [30]. The concentrations of these proteins are reported as arbitrary units, representing normalized protein expression on a log2 scale. This quantification method was consistent with previously established protocols [30]. The Olink Multiplex platform allowed for the simultaneous measurement of multiple inflammation-related proteins, providing a comprehensive profile of the inflammatory status of our study participants.

2.12. Statistical analyses

In our investigation, the Kruskal–Wallis test was used to evaluate differences among the three groups, effectively addressing the small sample size and non-normal data distribution inherent to our study. This non-parametric approach was complemented by post-hoc pairwise comparisons, executed via the Wilcoxon rank-sum test, to further elucidate the distinctions among the groups. To maintain the integrity of our statistical evaluation, we implemented the Benjamini–Hochberg procedure, efficiently correcting for multiple testing.

All statistical analyses and data visualizations were performed using the R software program, version 3.6.3.

3. Results

3.1. Study design & cohort characteristics on age-related variations in health participants

Blood samples were obtained from healthy young adults (22–31 years old, n = 11), middle-aged adults (57–64 years old, n = 10), and older adults (75–88 years old, n = 14) in the Chinese population, all with a BMI <30. The Kruskal–Wallis test showed no significant differences in the BMI among the three groups, with a p-value of 0.272 determined (Figure 2A). There was a balanced sex distribution across all age groups (Figure 2A). Plasma samples were analyzed for cfDNA methylation using EM-seq, and 92 inflammation-related biomarkers were assessed using the proximity extension assay (PEA) provided by Olink proteomics (Figure 1).

Figure 2.

Clinical characteristics across three different age groups. (A) Participant demographics and BMI variations by age group. (Left) Distribution of participants' ages. (Center) Breakdown of sex within each age group. (Right) BMI distribution among young, middle-aged, and older groups, with each group's BMI presented as the mean ± standard deviation (SD). The p-value, obtained from the Kruskal-Wallis non-parametric test, assesses BMI differences across groups. (B) Blood differential counts were obtained using the SYSMEX XN-20 analyzer. The Mann–Whitney U-test evaluated differences in blood cell counts between groups. Boxplot notations: lower and upper quartiles correspond to the 25th and 75th percentiles, respectively; the horizontal bar indicates the median; whiskers extend to data points within 1.5× the IQR from the quartile boundaries.

Complete blood counts and the differential analysis of whole blood revealed a significant decline in red blood cell counts with age, consistent with previous studies (Figure 2B & Supplementary Table S1) [31]. Young individuals exhibited lower hemoglobin levels than those in the old group (p = 0.039). Monocyte, neutrophil, lymphocyte and platelet counts did not differ significantly across the three age groups (Figure 2B & Supplementary Table S1). cfDNA is sourced from a range of organs and tissues, mainly from hematopoietic cells in healthy individuals [5,7,32]. No significant age-related variations were observed in leukocyte counts.

3.2. Elevated cfDNA concentrations alongside shifts in fragment size & source profiles reveal aging signatures

The analysis indicated a marked increase in cfDNA concentration with advancing age, peaking in the old group (Figure 3A). The detection of a positive significant correlation between the participant age and cfDNA concentration substantiates this finding (r = 0.486; p = 0.003; Figure 3B). Across all age groups, a cfDNA fragment analysis revealed notable enrichment of fragments of approximately 175 bp (Figure 3C). In contrast, the young group exhibited fewer long cfDNA fragments than did the old group (Figure 3C).

Figure 3.

An analysis of age-related variations in cfDNA concentration, fragment size and origin. The concentrations of cell-free DNA (cfDNA) across all participants are presented as the mean ± standard deviation, based on log-transformed values. (A) cfDNA concentration comparisons among young, middle-aged and old groups, statistically significant differences were observed. Pairwise comparisons demonstrated that cfDNA concentrations were significantly higher in the old group than in the young and middle-aged groups. (B) The correlation between age and cfDNA concentration is represented by the Pearson correlation coefficient and its p-value. (C) Distribution of plasma cfDNA fragment sizes among the three age groups. (D) The stacked bar graph depicts the proportion of cfDNA from various tissues and cell types, differentiated by age group. Different colors designate individual cells or tissues, with the segment sizes reflecting their proportionate shares of the overall cfDNA concentration. (E) This heatmap illustrates the distribution of cfDNA concentrations from 13 cell and tissue types across different age groups. CfDNA sequences were mapped to a reference panel of methylation patterns specific to 25 cell types, identifying their origins with a focus on those detectable in most participants. The heatmap layout aligns columns with individual participants and rows with distinct cell or tissue types. (F) Cell-type-specific cfDNA concentrations are depicted as the mean ± standard deviation of log-transformed values, highlighting significant increases in the older group for certain cell types. Statistical significance is denoted by asterisks: * for p < 0.05 and ** for p < 0.01.

Plasma samples from all participants were subjected to enzymatic conversion of cfDNA followed by whole-genome sequencing. After EM-seq, cfDNA sequencing data were cross-referenced with a panel of 25 cell-specific DNA methylation signatures to identify the tissue origin of cfDNA [7]. The analysis was limited to 13 cells or tissues with quantifiable cfDNA in the majority of participants. These cell and tissue types included NK cells, T cells, B cells, hepatocytes, erythrocyte progenitors, neutrophils, monocytes, pancreatic cells, uterine cervical cells, breast cells, colon epithelial cells and cells from the head and neck/larynx (Figure 3D). Using deconvolution algorithms, we identified age-related variations in the cellular and tissue sources of cfDNA (Figure 3E). Specifically, the analysis showed that different cells and tissues contribute to cfDNA to varying degrees, depending on the age (Figure 3F). Specifically, it was observed that cfDNA concentrations in colon epithelial cells were significantly greater in the old group compared with the young group (p = 0.010). In neutrophils, the comparison revealed significantly higher cfDNA concentrations in the old group versus the young group (p = 0.003), and also when comparing the old group to the middle-aged group (p = 0.011), suggesting their roles in the age-associated rise in cfDNA levels. Conversely, cfDNA concentrations in NK cells markedly declined when moving from the middle-aged to the old group (p < 0.0001). The elevated cfDNA level in neutrophils is likely due to increased lysis or turnover with aging, as indicated by changes in their recruitment and function [33]. In colon epithelial cells, elevated cfDNA levels may stem from rapid turnover and increased cell death in the gut epithelium, contributing to conditions such as inflammatory bowel disease [34]. Similarly, B cells, T cells, pancreatic, breast and head and neck/larynx cells also demonstrated a trend of increased cfDNA concentrations in the old group, although these differences were not statistically significant. These findings reveals that higher cfDNA concentrations in older individuals, especially from neutrophils and colon epithelial cells, reflect age-related changes in cell turnover and tissue-specific dynamics

3.3. Age-group variations in cfDNA methylation cites & the FEM analysis of hub genes

We performed whole-genome methylation sequencing using EM-seq on cfDNA samples to identify differentially methylated CpG sites (DMCs) in various age groups. On average, each cfDNA sample yielded 190,091,677 clean reads, with approximately 56,011,403 CpGs, attaining a sequencing depth of over 30-fold. This depth ensured high confidence in determining the methylation status (Supplementary Table S2). A differential methylation analysis among the young, middle-aged and old groups revealed distinct patterns. In the middle-aged and young groups (Figure 4A), we identified 604 hypermethylated and 1251 hypomethylated DMCs. Similarly, in the old versus young group comparison (Figure 4B), 965 hypermethylated and 832 hypomethylated DMCs were observed. In the old vs middle-aged comparison (Supplementary Figure S1A), 1343 hypermethylated and 725 hypomethylated DMCs were identified. The DMCs in all three comparison groups were predominantly within the intergenic and intronic regions (Figure 4D, F & Supplementary Figure S1C). These DMCs were predominantly enriched in exon and repeat regions as opposed to promter regions and CpG islands (Figure 4C, E & Supplementary Figure S1B). Interestingly, it was found that over 50% of the age-related DMCs were located within repeat regions (Figure 4D, f & Supplementary Figure S1C). This finding suggests a significant association with aging. These repeat regions encompass transposable elements, such as long interspersed nuclear elements (LINEs), a category of interspersed repeats [35]. Further evidence from previous studies suggests that the presence of LINE-1 retroelements in blood's cfDNA is indicative of human aging, with LINE-1 sequences constituting a significant portion of cfDNA [8,36]. These findings imply a potential association between cfDNA patterns and the aging process.

Figure 4.

Age-related alterations in cfDNA methylation and hub gene identification. (A) The volcano plot displays differentially methylated cytosines (DMCs) between the middle-aged and young groups. Red dots denote hypermethylated sites, while blue dots indicate hypomethylated sites, all with significant adjusted p-values (all adjusted p < 0.05). (B) The volcano plot indicates DMCs between the old and young groups, highlighting hypermethylated sites in red and hypomethylated sites in blue, all statistically significant (all adjusted p < 0.05). (C) Enrichment scores of DMCs across various genomic elements in the middle-aged versus young group comparison. This includes genomic elements, such as CpG islands, promoters, exons, introns, intergenic areas and repeat regions. (D) Percentage distribution of DMCs among different genomic elements and regions in the middle-aged versus young group comparison. (E) Enrichment scores of DMCs in various genomic elements in the old versus young group comparison. (F) Percentage distribution of DMCs across various genomic elements and regions in the old versus young group comparison. (G) Network visualization from the FEM analysis contrasting the middle-aged and young groups. Yellow nodes represent genes with hypomethylation, blue nodes represent genes with hypermethylation and central hub genes pivotal to the network are highlighted in red. (H) A similar FEM network for the old versus young groups employs the same color coding to denote methylation status. Blue nodes signify genes with hypermethylation, while yellow nodes mark genes with hypomethylation, and red nodes identify central hub genes critical to age-related methylation changes.

Our study used an FEM analysis to assess age-related changes in cfDNA DMCs. This bioinformatics method identifies clusters of genomic regions co-regulated through similar epigenetic patterns, such as DNA methylation, thereby aiding in the identification of hub genes and elucidation of the complex relationship between cfDNA methylation and aging [20]. The FEM analysis identified several hub genes associated with age-related methylation changes. For instance, in the comparison between the middle-aged and young groups, genes such as MORF4L1, H4C5,NCL, and RPL5 were significantly hypomethylated, whereas H1-0 showed notable hypermethylation (Figure 4G & Supplementary Figure S1D–F) [37–41]. In the comparison between old and young groups, H1-2 showed marked hypermethylation, while HNRNPA0 and H2AC21 were significantly hypomethylated (Figure 4H & Supplementary Figure S1G) [42,43]. The comparison between the old and middle-aged groups highlighted genes such as CALML4 and SRC for hypermethylation (Supplementary Figure S1H & I) [44,45]. The FEM analysis of cfDNA DMCs revealed key insights into aging-related epigenetic changes, highlighting genes essential for chromatin remodeling, the ribosomal function and cellular signaling.

3.4. Construction of a cfDNA methylation-based age prediction model

A significant portion of current cfDNA methylation research still relies on data from methylation chips, with most established methylation clocks based on data from the Illumina Infinium HumanMethylation450 (450 K array) [46,47]. To enhance the clinical and research utility of our biological age prediction model, we aligned cfDNA methylation data from EM-seq with CpG sites measured using the widely used Illumina methylation bead-array platforms [22]. First, we performed unsupervised hierarchical clustering on 450 K array data to reveal age-related methylation patterns. Eight modules were identified, of which Cluster 1 demonstrated a linear decrease, and Cluster 8 showed a linear increase in methylation with advancing age (Figure 5A). Clusters 2–4 initially experienced a decline in methylation, followed by an increase. Conversely, Clusters 5 to 7 presented a reverse pattern, with values initially increasing before decreasing (Figure 5A). A Pearson correlation analysis of approximately 450,000 methylation sites identified 22,927 age-correlated sites (p < 0.05), uniformly distributed across all 23 pairs of chromosomes (Figure 5B).

Figure 5.

Constructing an epigenetic age prediction model using cfDNA methylation and validating it across independent datasets. (A) A cluster analysis on cfDNA methylation trajectories, performed using EM-seq data adapted to simulate the 450k array format, groups CpG sites into eight clusters. These clusters categorize CpG sites based on methylation change similarities across three age groups: Young, middle-aged and Old. Thicker lines represent average methylation trajectories for each cluster. (B) A genome-wide distribution of linearly age-correlated methylation sites is observed across all 23 pairs of chromosomes. (C) The construction of a cfDNA methylation-based age prediction model employs LASSO regression (n = 35) and incorporates 48 CpG sites. This model demonstrates strong predictive accuracy, with a high Pearson correlation coefficient between chronological and biologically predicted ages, and is supported by low MAE and RMSE values. (D) A principal component analysis (PCA) effectively differentiated the young, middle-aged, and old groups using methylation profiles from the 48 CpG sites. (E) Validation of the model's age prediction accuracy using the GSE186888 dataset (n = 23) shows consistent Pearson correlation coefficients and low MAE and RMSE values between actual and predicted ages. (F) Further validation with the GSE208758 dataset (n = 23) substantiates the model's predictive reliability, as reflected in the correlation coefficients and the error metrics, MAE and RMSE.

By employing LASSO regression, we developed a cfDNA methylation age prediction model. This model included 48 CpG sites (Supplementary Table S3). The model comprised 12 CpG sites that were positively correlated and 36 CpG sites that were negatively correlated with chronological age (Supplementary Figures S2 & 3). There was a high correlation (r = 0.994) between chronological age and predicted biological age, with a mean absolute error (MAE) of 5.12 years (Figure 5D), which effectively differentiated biological age and suggested distinct methylation profiles as biomarkers of aging using a principal component analysis (PCA) (Figure 5C). In validation tests against an independent healthy dataset (GSE186888; n = 23), the model exhibited a strong correlation (r = 0.973) between chronological age and biological age, with an MAE of 6.72 years (Figure 5E). When applied to an independent epilepsy patient dataset (GSE208758; n = 23), the correlation between chronological age and biological age decreased slightly (r = 0.855), with an MAE of 4.94 years (Figure 5F). Further validation of the independent healthy dataset GSE186888 and epilepsy patient dataset GSE208758 confirmed the accuracy of the model in correlating chronological age with biological age, albeit with slightly reduced precision in the epilepsy patient dataset. In addition, when we applied the most commonly used epigenetic clock to our cfDNA dataset, early clocks, such as Horvath and Hannum's clock, showed limited predictive power [9,10]. In contrast, recent PC clocks, especially PCGrimAge, have demonstrated excellent results despite involving a large number of CpG sites (Supplementary Figure S4) [48].

We herein report the first cfDNA methylation clock validated across independent datasets, demonstrating an accurate biological age prediction.

3.5. Profiling cfDNA DMRs reveals age-specific epigenetic patterns & inflammatory roles

The DSS software program was used to identify regions with statistically significant DMRs (p < 0.05; methylation differences >0.1), revealing 6429 hypermethylated and 7499 hypomethylated DMRs in middle-aged versus young participants (Figure 6A). Comparisons of old versus young participants displayed 7938 hypermethylated and 5601 hypomethylated DMRs (Figure 6B), while old versus middle-aged participants showed 10,133 hypermethylated and 5782 hypomethylated areas (Supplementary Figure S5A). Notably, the study found 960 genes consistently hypermethylated and 694 genes consistently hypomethylated in both the middle-aged and old groups compared with the young group, indicating age-related methylation (Figure 6C). Further investigation into the genomic locations of these DMRs showed significant changes in areas such as CpG islands, promoters, exons, introns and intergenic zones, especially in exons and promoters, which was consistent across the three age groups (Figure 6D, E & Supplementary Figure S5B). For hypermethylated and hypomethylated DMRs across each comparison group, exons are the primary location of coverage (Supplementary Figure S6A–F). In these exonic regions, hypermethylated DMRs show greater coverage than hypomethylated DMRs across all age groups. Exons are also the predominant genomic region for DMRs consistently hypermethylated or hypomethylated in the middle-aged and old groups. Notably, consistently hypermethylated DMR coverage in exons is almost double that of consistently hypomethylated DMRs (Supplementary Figure S6G–H). These observations highlight methylation's critical role in aging and its potential influence on gene regulation.

Figure 6.

Identification of age-associated cfDNA differentially methylated regions (DMRs) and the functional analysis of promoter-associated DMRs. (A) A volcano plot illustrating DMRs comparing middle-aged to young individuals. Red dots represent hypermethylated regions and blue dots indicate hypomethylated regions, all statistically significant (p < 0.05). (B) A volcano plot depicting DMRs comparing older adults to young individuals, with hypermethylated regions in red and hypomethylated regions in blue, each statistically significant (p < 0.05). (C) A Venn diagram displaying the overlap of genes associated with hypermethylated and hypomethylated DMRs between the middle-aged versus young and old versus young group comparisons. (D) Bar chart representing the distribution of DMRs across various genomic elements and regions in the middle-aged versus young group comparison. (E) Bar chart showing the distribution of DMRs across different genomic elements and regions in the old versus young group comparison. (F) Bubble chart illustrating a KEGG pathway enrichment analysis for promoter-associated DMRs in comparisons of middle-aged versus young and old versus young groups, highlighting significant enrichment in inflammation-related pathways (p < 0.05). (G) Bubble chart presenting a GO terms enrichment analysis for genes associated with DMRs in comparisons of middle-aged versus young and old versus young groups, emphasizing pathways related to cell migration and nervous system development (p < 0.05).

KEGG pathway enrichment and GO function assessments of genes linked to promoter DMRs (TSS-2000 to TSS+2000) were performed to identify significant pathways (p < 0.05). The KEGG pathway analysis revealed notable upregulation in inflammation-related pathways, including spinocerebellar ataxia and leukocyte transendothelial migration, in middle-aged versus young and old versus young group comparisons, with the old versus young group comparison also showing heightened platelet activation and chemokine signaling pathways (Figure 6F & Supplementary Table S4). Age-specific regulatory trends were observed, such as purine and sphingolipid metabolism, with some pathways, such as pancreatic secretion and Hippo signaling, being downregulated in these comparisons.

In addition, a GO term enrichment analysis indicated a significant increase in immune- and inflammation-related pathways, including cell migration and adhesion, particularly in the old versus young group, whereas pathways linked to neuron and muscle development declined (Figure 6G & Supplementary Table S5). Together with the results from KEGG, it shows the complex impact of cfDNA methylation changes on cellular functions in inflamm-aging. This suggests a need for further investigation into their association with immunosenescence, accelerated biological aging, and disease etiology, especially considering cfDNA's inherent pro-inflammatory characteristics.

3.6. Exploring age-related associations with inflammation & cfDNA methylation using Olink proteomics

Given the unclear relationship between cfDNA methylation and inflamm-aging, we initiated an in-depth investigation using Olink proteomics to analyze the correlation between 92 inflammatory factors and age, in the context of the role of cfDNA in aging-related inflammation. This revealed that 37 factors, such as CDCP1, were positively correlated with age, with CDCP1 (r = 0.885, p = 1.72E-11), OPG (r = 0.819, p = 9.97E-09), and CXCL9 (r = 0.789, p = 7.95E-08) among the most significant (Figure 7A & Supplementary Table S6). In contrast, TRANCE showed a negative correlation (r = -0.689, p = 1.30E-05).

Figure 7.

Relationship between immune inflammation factors and cfDNA methylation across age groups. (A) Bubble plot depicting Pearson correlation coefficients between age and 38 selected inflammatory factors from a total of 92 analyzed in 35 participants. The size of each bubble indicates statistical significance, with larger bubbles representing lower p-values (all p < 0.05). The color gradient from blue to red illustrates the range of correlation coefficients, with blue indicating negative and red indicating positive correlations (all |r| >0.3). (B) Box plots showing the distribution of expression for the top inflammatory factors positively correlated with age across three age groups: young, middle-aged, and old. The Kruskal-Wallis test was used to assess differences among groups due to the small sample size and non-normal distribution of the biomedical data. Post hoc pairwise comparisons were conducted using Wilcoxon's rank sum test, with the Benjamini-Hochberg procedure for multiple testing corrections. Significance levels are marked as ns (not significant), *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001. (C) Scatter plots illustrating the Pearson correlation analysis between the log-transformed concentrations of cfDNA (log10 cfDNA in ng/ml) and 6 selected inflammatory factors, each plot displaying the corresponding r and p values. (D) Heatmap visualizing the correlation between 48 CpG sites identified in the cfDNA methylation age prediction model and 38 age-related inflammatory factors. The color scale transitions from red to blue to represent positive to negative correlation values, respectively. Significance levels are marked as *p < 0.05 and **p < 0.01.

Our study analyzed differences among the three age groups for nine factors that showed the strongest correlations with aging and found that the CDCP1, OPG and CXCL9 levels notably increased in the old group (p < 0.05, p < 0.0001) (Figure 7B). In addition, markers such as CCL11, HGF, CD40, LAP TGF-beta 1, FGF-5, and CCL25 displayed significant intergroup variations (Figure 7B). These trends, particularly the increase in chemokines with age, are consistent with the cfDNA DMR KEGG and GO pathways, suggesting a potential link between cfDNA methylation changes and inflamm-aging processes.

In the Pearson correlation analysis of log-transformed cfDNA concentrations with 92 inflammatory factors, several markers showed significant positive correlations, including CD8A (r = 0.513), IL-20 (r = 0.397), IL-10RA(r = 0.392), CDCP1 (r = 0.362), HGF (r = 0.359), and TGF-β(r = 0.350) (Figure 7C). A heatmap analysis of 48 CpG sites from the cfDNA methylation age prediction model and 38 age-related inflammatory factors revealed complex correlations: some CpG sites, such as cg15728779, showed strong positive correlations with age-increasing factors (Figure 7D), while others, such as cg02235659, had negative correlations, suggesting potential age-related protective trends (Figure 7D). These patterns suggest a significant connection between cfDNA methylation and inflamm-aging, suggesting that cfDNA methylation changes might reflect or influence aging-related immune and inflammatory processes.

4. Discussion

The methylation characteristics of cfDNA from body fluids have become increasingly interesting [6]. The major findings of the present study are as follows: the first use of EM-seq-based methylation profiling on healthy individuals for aging produced high-quality data, despite limited cfDNA quantities; increased cfDNA concentrations were noted in older individuals, suggesting higher cell turnover or cell death rates with age, while younger individuals show fewer long cfDNA fragments; higher concentrations of cfDNA from neutrophils and colon epithelial cells were observed in the old group than in the other age groups; in each age group comparison, approximately 2000 DMCs were identified, with the FEM analysis identifying key hub genes in nucleosome structure; a 48-CpG-sites-based cfDNA methylation model showed a strong correlation between chronological and predicted biological age; and in the old group, significant methylation changes, including hypermethylation and hypomethylation in DMRs, were observed. A gene promoter analysis revealed a marked increase in inflammation-related pathways, consistent with the proteomics results, indicating a substantial link between cfDNA methylation and inflammation. Collectively, these results suggest the utility of cfDNA methylation as a predictive tool for aging.

Our 48 CpG site-based cfDNA methylation model showed a strong correlation between chronological and biological age (r = 0.994), effectively predicting the biological age. This represents the first instance of a cfDNA methylation clock, which marked a significant advancement in this field. Notably, within these 48 CpG sites, specific sites, such as cg22171829 in the PDK4 gene, cg00481951 in the SST gene, and cg19283806 associated with CCDC102B and TMX3 genes, have parallels in Horvath and Hannum's established methylation clocks [9,10], suggesting comparability between our cfDNA methylation clock and previous multi-tissue clocks. Validated across healthy and epilepsy-specific datasets, the model's robustness and superior predictive ability establish it as a promising tool for biological age prediction, potentially outperforming the current epigenetic clocks [9,10,48–50]. These findings not only validate the model, but also open up new avenues for the use of cfDNA methylation in aging research and biomarker discovery.

Recent studies have shown that the difference between DNA methylation age and chronological age, referred to as Δage, reflects biological aging [51]. Δage, known as age acceleration, has prompted further research beyond just chronological comparisons. A study employing a Loess model to adjust Δage (cAgeDiff) has provided more in-depth insights [52]. The potential of methylation age as a predictive marker for disease and mortality is noteworthy. Its correlation with biological markers and its capability to predict aging-related diseases such as cancer and cardiovascular disease underscore its importance [53–55]. Our study presents a cfDNA methylation clock as a novel alternative for assessing age acceleration, offering a fresh perspective in the evaluation of DNA methylation clocks. Despite existing challenges, this underscores the necessity for more comprehensive and longitudinal research to accurately determine the predictive capability of cfDNA methylation for health outcomes and longevity. Future investigations are required to expand sample sizes and include a wider range of disease populations to validate the effectiveness of the cfDNA methylation clock in detecting age acceleration.

Furthermore, the changes in inflammatory markers during aging, especially the increase in chemokines, are consistent with the KEGG and GO pathway analyses of cfDNA DMRs. The positive correlations observed between certain inflammatory factors and cfDNA concentrations suggest an association with methylation changes and the alterations in immune function and inflammatory processes related to aging [18]. One possible mechanism involves cfDNA from degenerated adipocytes, which is known to partially activate macrophages via the TLR9 pathway, leading to inflammation of adipose tissue [15]. MCP-1, also known as CCL2, belongs to the CC chemokine family and plays a key role in inflammation by attracting and activating various inflammatory cells and factors [56]. An example is the observed increase in MCP-1 with aging, potentially linked to increased cfDNA release from cell death, which in turn enhances MCP-1 expression in macrophages through the TLR9 pathway [14,17]. These findings provide a comprehensive view of the age-related epigenetic landscape through cfDNA DMRs, enriching our understanding of inflamm-aging and suggesting new research directions.

Our study has made significant advancements in deciphering the complex relationship between age-related changes in cfDNA methylation and the dynamics of inflammation, thereby enhancing our understanding of immune aging and inflamm-aging. Another key achievement of our research was the development of a cfDNA methylation-based model for predicting biological age, which holds great promise in the field of gerontology. Building upon these findings, one of our ongoing studies has collected blood transcriptomic data from participants. We intend to correlate this data with cfDNA methylation patterns in future research, aiming to determine if age-related alterations are consistent when age is analyzed as a continuous variable. This method is expected to enhance our comprehension of the aging process and its biological underpinnings, potentially revealing more about the mechanisms behind age-related changes.

5. Limitations of the study

The study faces several limitations, including a small sample size of 35 participants, which may limit the generalizability of our findings and introduce concerns regarding potential overfitting with the LASSO model. Additionally, our study's design excluded certain age groups (32–56, 65–75 years of age), a decision primarily influenced by budget constraints. We specifically chose distinct age groups to better observe pronounced age-related cfDNA methylation changes, focusing on three age groups with significant differences to highlight these variations. Future research aims to address these limitations by expanding the sample size and including a more diverse range of age groups, thereby enhancing the study's robustness. The cross-sectional nature of our study could also limit insights into the dynamic changes in cfDNA methylation over time. The complex influence of lifestyle and environmental factors on cfDNA methylation may further obscure the specific impacts of aging. Moreover, the clinical applicability of our cfDNA methylation-based age prediction model requires further validation. Additionally, the observed relationship between cfDNA methylation and inflamm-aging necessitates more experimental research to establish a definitive causal link.

6. Conclusion

This study uses EM-seq to explore cfDNA methylation profiling in healthy individuals for aging and introduces the first cfDNA epigenetic clock. Gene promoter analysis of cfDNA DMRs revealed a marked increase in inflammation-related pathways, consistent with the proteomics results, indicating a substantial link between cfDNA methylation and inflammation. Future research involving larger and more varied cohorts is expected to further clarify the role of cfDNA methylation in aging, and the clinical applications of this biomarker in aging and age-related diseases.

Funding Statement

This study was supported by the National Natural Science Foundation of China (No.82170856 and No.82304565) and the Chinese Academy of Medical Sciences (CAMS) Innovation Fund for Medical Sciences (No. 2021-1-I2M-050).

Supplemental material

Supplemental data for this article can be accessed at https://doi.org/10.1080/17501911.2024.2340958

Author contributions

J-P Cai designed the study. S-J Lia, L-T Zenga, Y-M Danga, Y-Q Maa and L-Q Zhanga collected the samples and obtained clinical data. S-J Lia, X Gao, Z-H Wanga and J Lia drafted the manuscript. S-J Lia performed bioinformatics analyses of the data. Q-Y Wanga, Y-M Zhanga, H-L Liuc, R-M Qia and J-P Cai reviewed and revised the manuscript. All authors have read and approved the final manuscript.

Competing interests disclosure

The authors declare no conflicts of interest. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Writing disclosure

No writing assistance was utilized in the production of this manuscript.

Ethical conduct of research

The study was performed in accordance with the Helsinki Declaration and approved by the Ethics Committee of Beijing Hospital (reference number: 2019BJYYEC-054-02). Informed consent was obtained from all participants, and no identifying information was included in the text.

Data sharing statement

Raw data, processed data and accompanying metadata have been deposited to the Gene Expression Omnibus (GEO) database under the accession code GSE259312. The datasets generated and/or analyzed during the study are available from the corresponding author upon reasonable request.