Changes in nuclear and mitochondrial DNA methylation in cow blood associated with age and disease

ABSTRACT

DNA methylation is among the most promising biomarkers for age prediction, enabling the development of epigenetic clocks that correlate methylation profiles with chronological age. In this study, we investigated the relationship between ageing and disease susceptibility, focusing on both nuclear and mitochondrial DNA methylation in dairy cows. Genome-wide DNA methylation profiling was performed using enzymatic methyl-seq, covering 53 million CpG sites. The dataset included 96 cows with different phenotypes, sampled cross-sectionally and ranging from 2 to 9 years of age. We applied elastic net regression to identify the most predictive CpG sites for age estimation, achieving a mean absolute error of 111 days with a strong correlation to chronological age r = 0.97. Beyond chronological age prediction, we assessed the impact of disease status on epigenetic ageing. Our results revealed accelerated epigenetic ageing in cows susceptible to diseases, suggesting a link between health-related stress and disrupted DNA methylation dynamics. We further identified age-associated promoter methylation changes, particularly in MAB21L1, which may play a role in molecular ageing mechanisms. Additionally, we observed a decline in mitochondrial DNA methylation with age, notably in genes encoding Cytochrome c oxidase (COX), indicating a possible connection between mitochondrial dysfunction and epigenetic regulation. An inverse correlation between D-loop methylation and mtDNA copy number was also observed. This study demonstrates the potential of epigenetic models for biological age prediction in livestock, while recognizing that their accuracy may vary among species with different lifespans.

Introduction

Epigenetics is influenced by environmental factors, which can modulate gene expression and play a crucial role in disease susceptibility and resilience over time. Among epigenetic mechanisms, DNA methylation has emerged in recent years as a promising biomarker of biological ageing [1].

Environmental exposures can trigger or exacerbate diseases, with their impact depending on the intensity and duration of exposure, thereby accelerating biological ageing. In livestock, particularly cattle, the average lifespan under natural conditions typically ranges from 15 to 20 years. However, intensive farming practices for dairy and meat production often reduce this lifespan dramatically, with many animals living only 4 to 6 years [2]. This reduction in longevity is frequently accompanied by the emergence of undesirable phenotypes and an increase in susceptibility to diseases.

Age comes with increased DNA damage, which induces epigenetic alterations during repair, thereby compromising genomic integrity. These changes in gene expression may represent a key hallmark of ageing [3,4]. More recently, Yang et al. (2023) [5] demonstrated that epigenetic alterations driven by DNA repair mechanisms contribute significantly to the ageing of organs and tissues in mammals. Nevertheless, it remains unclear whether all these epigenetic modifications are a direct consequence of ageing or whether they could be a cause [6].

DNA methylation provides a reliable method to estimate biological age by leveraging mathematical combinations of methylation levels at specific CpG sites. These predictions can then be compared to an individual’s chronological age [7–9]. Several other previous studies have investigated DNA methylation changes over time in various tissues [10,11].

The epigenetic clock is widely used in humans to predict biological age, which may differ between tissues, unlike chronological age, which increases uniformly (every 365 days). Studies have demonstrated the accuracy of a pan-tissue epigenetic clock based on the percentage of methylation of specific CpG sites in estimating chronological age in humans [8,12]. Blood, being easily accessible, serves as a valuable tissue for such studies due to its cellular diversity and systemic interaction with all organs [13,14].

In several mammalian species, DNA methylation levels have been shown to correlate with age, typically displaying a deceleration in modification rates with increasing age [15]. In livestock production systems, where birth dates are often not accurately recorded, having a robust biomarker to determine biological age is important [16]. Incorporating the biological age of females into later life selection decisions could identify the ones that are at risk of developing health issues or reduced reproductive performance [17]. As such, the epigenetic clock holds great promise as an objective biomarker for predicting productivity and health potential in animals.

Although numerous studies have reported age-related changes in blood gene methylation, no study to date has specifically investigated the relationship between nuclear and mitochondrial DNA methylation, age, and various disease phenotypes in cattle.

In this study, we investigated a novel temporal DNA methylation pattern in Holstein dairy cows to explore the relationship between ageing and disease susceptibility based on methylation profiles from both nuclear and mitochondrial genomes. We refer to this pattern as ‘ChronoMeth.’ Using blood samples from 96 cows ranging in age from 1.8 to 9.2 years, we identified the most predictive CpG sites for estimating age. Additionally, we examined how health status influences epigenetic ageing, with the goal of early identification of at-risk animals to improve replacement selection strategies. Finally, we show that mitochondrial DNA methylation is also modulated by age.

Materials and methods

Animal selection and sample collection

Blood samples were collected from Holstein dairy cows (Bos taurus; females) at the University of British Columbia’s experimental farm in Vancouver between November 2020 and February 2021. A year later, we selected 48 cows that had been culled during the year due to complications related to infertility (BFR, n = 6), mastitis (BMS, n = 6), lameness (BFL, n = 12), metabolic disorders (BMT, n = 6), or low production (n = 18), which were not identified at the time of sampling. This culling reason was recorded in farm management logs and represented the primary cause for culling. These cows formed the ‘diseased group.’ For comparison, we selected a second group of 48 cows that remained in the herd without any health issues. This second group represented the ‘control group.’ The ages of the cows in both groups ranged from 1.8 to 9.2 years (Table S1). Samples were allocated to groups retrospectively according to culling records

Enzymatic methyl sequencing

One mL of blood from each cow was sent to the Génome Québec sequencing platform for DNA extraction, enzymatic methyl-seq (EM-seq) library preparation [18], and whole-genome sequencing. Libraries were pooled at equimolar concentrations and sequenced using Illumina NovaSeq technology, generating approximately 450 million 150-bp paired-end reads per library.

During library preparation, there is a DNA conversion step that involves three primary enzymes: TET2 and T4-BGT detect and transform 5mC and 5hmC into a product that the APOBEC3A enzyme cannot deaminate. Subsequently, APOBEC3A deaminates unmodified cytosines by converting them to uracils. As a result, unmethylated cytosines are transformed into uracils and then into thymines during polymerase chain reaction (PCR) amplification, while methylated cytosines remain as cytosines. To verify the extent of DNA conversion of each sample, two internal controls were added before the conversion step: unmethylated lambda (Table S2) and CpG methylated pUC19. By estimating the methylation levels of lambda and pUC19, we assess the quality of DNA conversion.

Methyl calling

Reads were filtered, trimmed, and deduplicated using fastp version 0.23.1 with default parameters [19]. Read conversion, alignment to the Bos taurus reference genome (ARS-UCD1.2.104), and methyl calling were conducted using Bismark [20]. This process was implemented in the DRAGEN Methylation Pipeline, which was run on an Illumina DRAGEN server using default parameters. In brief, reads from the EM-seq library were converted into both a C-to-T and G-to-A version and then aligned to a correspondingly converted reference genome. This approach ensured that sites from both DNA strands were captured. Methyl calling was achieved by comparing each sequenced read with the unconverted consensus bovine reference genome (Bos_taurus.ARS-UCD1.2).

Methylation calls from Bismark were filtered to retain sites covered by at least six reads for nuclear DNA analysis and at least five reads for mitochondrial DNA analysis.

Annotation

Differentially methylated sites or regions were linked to adjacent genes or the closest genes using the annotatePeak function in the default mode of the R package ChIPseeker (version 1.38.1). The annotation considered regions spanning 3 kb upstream of the TSS, designating them as promoters. Depending on the location relative to the closest gene, each annotated site or region was categorized as ‘Promoter,’ ‘5’ UTR,” ‘3’ UTR,” ‘Exonic,’ ‘Intronic,’ ‘Downstream,’ or ‘Distal intergenic.’ For these analyses, the reference genome used was Bos_taurus.ARS-UCD1.2 from Ensembl (version 108) with supplementary genomic annotations provided by the org.Bt.eg.db package (version 3.18.0).

Functional enrichment analysis

Gene clusters, based on their functional similarity, were examined using the DAVID tool, focusing on four resource categories: BP, CC, MF, and KEGG pathways. This tool uses the hypergeometric test algorithm. This analysis was conducted using Bos_taurus.ARS-UCD1.3 annotation. Functional categories showing the most significant enrichment, defined as those with a false discovery rate (FDR) below 0.05, were retained for further interpretation. Functional enrichment analysis was performed on CpGs in promoters.

Genes potentially expressed in blood

Genes were examined based on blood gene expression data from Holstein lactating cattle downloaded from the supplementary data of Gao et al., (2022) [21]. This study utilized single-cell transcriptomics of peripheral blood mononuclear cells (PBMCs) to profile seven major cell types, including CD4 T cells, CD8 T cells, B cells, monocytes, NK, innate lymphoid cells (ILC), and dendritic cells (DC).

Statistical analyses and visualization methods

All statistical analyses were performed using R software (version 4.3.1). Genomic annotations were based on the Bos_taurus.ARS-UCD1.2 (version 108) and the TxDb.BT9 database.

We used the MethyLasso tool (version 1.0) [22] as a segmentation-based approach to analyse DNA methylation patterns and identify the most significant differentially methylated regions (DMRs), defined by a q-value < 0.05 and a minimum methylation difference of 10%. This analysis complemented our CpG-based differential methylation analyses.

Selection of markers

We applied the Elastic Net approach by setting the alpha parameter to 0.5. Internal cross-validation using the cv.glmnet function was performed to automatically select the optimal regularization parameter (lambda). A leave-one-out cross-validation (LOOCV) strategy was implemented using the glmnet package (version 4.1–8), where each cow was iteratively used as the test set while the remaining individuals served as the training set.

The training dataset selected from a correlation analysis between DNA methylation levels and chronological age in 48 control cows. CpG sites were filtered based on a correlation coefficient > 0.5, a q-value < 0.01 and the exclusion of all sites containing missing values.

To prevent overfitting and assess the model’s generalizability,

Differential methylation analysis: mitochondrial genome

Differential methylation analysis was performed at both the cytosine (site-specific) and regional levels using the MethylKit package (version 1.28.0). Methylation data obtained from Bismark alignments were filtered to retain CpG, CHH, and CHG sites covered by at least five reads in a minimum of ten samples per group.

For each site, a logistic regression model was used to compute p-values, which were subsequently adjusted to q-values using the SLIM method [23]. To account for potential differences in bisulphite conversion efficiency, the methylation level of lambda phage DNA in the CpG context was included as a covariate.

Methylation differences were calculated as the difference between the mean methylation level in young animal controls (n = 20; 2.9 to 5.8 years) and that in adult controls (n = 20; 6.5 to 9.1 years). Only sites with q-values below 0.05, after false discovery rate (FDR) correction using the SLIM method, were considered significant for downstream analyses.

Mitochondrial DNA copy number

DNA was extracted from 300 μL of blood using the DNeasy Blood & Tissue Kit (Cat# 69,504, Qiagen) according to the manufacturer’s instructions. Each ddPCR reaction (final volume of 20 μL) contained 1× EvaGreen ddPCR Supermix (Bio-Rad), 100 nM of the specific primers (listed in table S3), and 10 ng of DNA. Reactions were partitioned into droplets using a Q × 100 droplet generator (Bio-Rad), and the emulsions were transferred to a 96-well plate. PCR amplification was performed under the following cycling conditions: 95°C for 5 min; 50 cycles of 95°C, 60°C, and 72°C for 30 s each; and a final hold at 12°C. Droplet fluorescence was measured using a Q × 100 droplet reader (Bio-Rad), and thresholds for positive droplets were automatically defined by the software. Results were expressed as the concentration of input template molecules per μL. To quantify mtDNA copy number, we targeted a sequence within the non-coding D-loop region. In parallel, we measured MX1, a bovine single-copy nuclear gene, to estimate the total number of cells in each reaction. The final mtDNA copy number was normalized to reflect copies per cell equivalent by dividing the total number of D-loop copies by the total number of input cells, estimated from the nuclear single-copy gene MX1. Thus, mtDNA copy number = (D-loop copies)/(MX1 copies), representing the number of mitochondrial genomes per cell.

Results

Age distribution

The age distribution varied among the cow groups that developed disease one year after sampling. Cows with lameness ranged from 1.8 to 9.2 years, those with fertility issues from 3.3 to 4.6 years, low-production cows from 1.8 to 5.6 years, mastitis-affected cows from 2.8 to 7.8 years, and cows with metabolic disorders from 3.0 to 5.0 years. In contrast, control cows exhibited a broader age range, from 2.9 to 9.1 years (Figure 1(A)).

Figure 1.

Age distribution of cows across six phenotypic groups. A: age of cows: control (GD, purple), lameness (BFL, red), fertility issues (BFR, brown), metabolic disorders (BMT, blue), low production (BLP, green), and mastitis (BMS, turquoise). B: correlation plot illustrating the relationship between mean DNA methylation levels and chronological age in control cows, showing a negative Pearson correlation coefficient (r = −0.43).

To assess and estimate epigenetic age, we focused on the control group and examined the relationship between DNA methylation levels and chronological age. Based on average methylation levels across approximately 53 million CpG sites, we observed a moderate negative Pearson correlation (r = −0.43, p = 0.0021) with chronological age (Figure 1(B)), suggesting a potential association between epigenetic modifications and ageing processes in dairy cows, although substantial inter-individual variation indicates that methylation alone does not fully explain age-related differences.

Functional enrichment analysis

To investigate the relationship between DNA methylation and age, we conducted a correlation analysis using data from 48 control cows. CpG sites were filtered based on a correlation coefficient > 0.5, a q-value < 0.01, and no more than 10 missing values (NA) per CpG site across the 48 cows. This stringent filtering yielded 52,855 age-correlated CpG positions. From these, we focused specifically on CpGs located within promoter regions, identifying 2125 unique promoter-associated gene IDs for Gene Ontology (GO) enrichment analysis and Venn diagram comparisons (Figure 2(A)).

Figure 2.

CpGs correlated with age. A: DAVID functional enrichment analysis of genes targeted by age-correlated CpG sites (r > 0.5) located within promoter regions. Enriched terms are categorized into biological process (bp), molecular function (MF), cellular Component (cc), and KEGG pathways.B: venn diagram illustrating the overlap of promoter-associated genes targeted by age-correlated CpG sites (r > 0.5) and those expressed in bovine blood, based on the Gao et al. (2022) dataset [21].

DAVID functional enrichment analysis was performed across four categories: Biological Process (BP), Cellular Component (CC), Molecular Function (MF), and KEGG pathways. The analysis revealed several significantly enriched terms, which were organized into six main functional categories:

• Regulation of transcription and interaction with RNA polymerase II (4 terms)

• Cellular localization and associated structures (6 terms)

• Protein interaction and signalling (5 terms)

• Cell differentiation and development (2 terms)

• Nervous system and cellular communication (2 terms)

• Pathologies and disease-related processes (1 term)

To refine the biological relevance of these findings, we cross-referenced the identified genes with the Gao et al. (2022) [21] dataset of genes expressed in bovine blood. Among the 2125 promoter-associated genes, 472 gene symbols were potentially expressed in blood and targeted by at least one age-correlated CpG site (Figure 2(B)).

Data preparation for the prediction model

To build a model based on the previously analysed methylation data, all CpG sites with missing values were excluded to ensure data quality and consistency, resulting in a final dataset comprising 1890 CpG positions.

Principal Component Analysis (PCA) was performed to explore the distribution of cows based on their methylation profiles (Figure 3). The first principal component (PC1) accounted for 20% of the total variance and allowed for the separation of the cows into four age-based groups, each containing 12 individuals:

Figure 3.

Distribution of 1890 age-correlated CpG sites. principal Component analysis (PCA) based on 1890 age-correlated CpG sites (r > 0.5) with no missing values. Control cows are grouped into four age categories: young (red; 1067–1848 days), middle-aged (green; 1893–2259 days), older (blue; 2283–2668 days), and oldest (purple; 2680–3338 days). The PCA reveals age-related clustering based on DNA methylation patterns.

• Young: 1067 – 1848 days

• Middle-aged: 1893 – 2259 days

• Older: 2283 – 2668 days

• Oldest: 2680 – 3338 days

A total of 121 CpG positions were located in promoter regions, targeting 113 unique gene IDs. Among these, several promoters were targeted by multiple CpG sites, including:

• PIG (Phosphatidylinositol glycan anchor biosynthesis class U and T), targeted by 3 CpG sites

• NEU3 (Neuraminidase 3), targeted by 2 CpG sites

• SOX7 (SRY-box transcription factor 7), targeted by 2 CpG sites

• MAB21L1 (Mab-21 like 1), targeted by 3 CpG sites

These genes are involved in key biological processes such as transcriptional regulation, cellular signalling, and phosphatidylinositol glycan anchor biosynthesis functions that are potentially crucial to cellular maintenance and ageing.

Epigenetic age prediction model

Epigenetic clocks provide a robust and reliable method for estimating an animal’s age based on its DNA methylation profile. In this study, we developed an epigenetic age prediction model using elastic net regression. The training dataset comprised 1890 CpG sites, selected from a correlation analysis between DNA methylation levels and chronological age in 48 control cows.

The final model achieved high predictive accuracy, with a Mean Absolute Error (MAE) of 111 days (3 months and 21 days) and a correlation coefficient of 0.97 (p < 2.2e-16) between predicted and actual age (Figure 4). These results underscore the strong predictive performance of our epigenetic clock for estimating bovine biological age.

Figure 4.

Correlation plot between chronological age and predicted age. based on the elastic net regression model, using leave-one-out cross-validation (LOOCV). The model achieved a Pearson correlation coefficient (r) of 0.97 (p < 2.2e-16) and a mean absolute error (MAE) of 110.9 days, demonstrating high accuracy in epigenetic age prediction.

Age acceleration and disease prediction

We applied the trained epigenetic age prediction model originally developed and validated using control cows to a cohort of cows affected by various health conditions, including mastitis (n = 6), low production (n = 18), lameness (n = 12), fertility issues (n = 6), and metabolic disorders (n = 6). For diseased cows, missing methylation values were imputed using the missForest R package, which employs Random Forests to perform accurate and non-parametric imputation while preserving nonlinear relationships between CpG sites.

Overall, the model showed a mean absolute error (MAE) of 896 days (~2.5 years) and a correlation of 0.16 (p = 0.27) in diseased cows, indicating a weak association between predicted epigenetic age and chronological age in this population (Figure 5(A)).

Figure 5.

Age acceleration. A: prediction of chronological age in diseased cows using the elastic net regression model. The scatter plot shows actual versus predicted age, with control cows in green and diseased cows in red. B: plot of actual age versus mean absolute error (MAE) in age prediction across phenotypes: control cows (GD, green), lameness (BFL, blue), fertility issues (BFR, purple), low production (BLP, red), mastitis (BMS, brown), and metabolic disorders (BMT, yellow). Mean epigenetic age acceleration (in days) among diseased phenotypes was as follows: low production = 1142 days (~3.1 years), metabolic disorders = 872 days (~2.4 years), fertility issues = 742 days (~2.0 years), lameness = 560 days (~1.5 years), and mastitis = 477 days (~1.3 years). C: principal Component analysis (PCA) based on the 62 CpG markers used for age prediction. The plot displays the distribution of control and diseased cows with no clear clustering by phenotype. D: boxplot of epigenetic age acceleration across health statuses, highlighting differences between control (green) and diseased (red) cows. E: relationship between actual age and the mean methylation levels of predictive markers, stratified by phenotype.

Age acceleration, defined as the difference between epigenetic and chronological age, varied across phenotypes. Cows with low production exhibited the greatest age acceleration, while mastitis-affected cows showed the least (Figure 5(B)).

The PCA of control and diseased cows based on the 62 CpG markers used for age prediction revealed no clear clustering by phenotype, suggesting that these markers are primarily associated with age rather than disease status (Figure 5(C)).

Among the 48 diseased cows, seven animals had an MAE below one year. Most cows exhibited negative MAE values (Figure 5(D)), meaning their predicted epigenetic age was more advanced than their actual chronological age. However, three individuals presented positive MAE values, suggesting a younger epigenetic age than their actual age (low production: +106 days, mastitis: +505 days, lameness: +989 days).

Interestingly, methylation trends at the 62 CpG sites varied by phenotype over time (Figure 5(E)). Increasing methylation with age was observed in control cows (r = 0.87, p = 1.29e-15), fertility issues (r = 0.32, p = 0.54), and low production (r = 0.33, p = 0.19). In contrast, decreasing methylation with age was observed in lameness (r = −0.22, p = 0.50), mastitis (r = −0.34, p = 0.52), and metabolic disorders (r = −0.50, p = 0.31).

These findings suggest potential phenotype-specific epigenetic ageing patterns, with some health conditions associated with accelerated or altered ageing trajectories in dairy cows. One must remember that these changes are seen in the white blood cells, which have mainly immune functions. Decreased methylation could mean more expression in this case.

Gene evolution and differentially methylated regions (DMRs)

To identify genes with age-associated epigenetic changes, we analysed promoter methylation levels in control cows. Genes were selected based on a Spearman correlation coefficient |rho| > 0.6 and a Benjamini-Hochberg (BH) adjusted q-value < 0.05.

Among the most strongly correlated genes, MAB21L1 showed the highest positive correlation with age (rho = 0.80, q-value < 2.2e-16), indicating a marked increase in promoter methylation over time. Other positively correlated genes included NKX1-1 and TRIM71. In total, 12 genes exhibited increased promoter methylation with age, while 7 genes showed decreased promoter methylation (Figure 6(A), Table 1).

Figure 6.

Genes with age-associated epigenetic changes. A: graph showing the relationship between age and average promoter DNA methylation levels across healthy cows, highlighting trends in age-associated methylation changes. B: genomic distribution of differentially methylated regions (DMRs) identified between young cows (2.9–5.8 years) and adult cows (6.5–9.1 years), classified by genomic feature. C: density scatter plot showing overall DNA methylation levels in young and adult cows, based on all significant differentially methylated regions (DMRs). Each dot represents the average methylation level of a DMR, while the blurred blue clouds indicate regions of high point density, reflecting the overall distribution of DMRs across both age groups. Darker blue areas above the diagonal represent a higher concentration of DMRs with increased methylation in adult cows compared to young ones, whereas areas below the diagonal reflect hypomethylation with age.

Table 1.

List of genes targeted by CpG sites located within promoter regions. Defined as 0 to 3000 bp upstream of the transcription start site (TSS), showing a significant correlation with age in control cows. The table includes gene names, correlation coefficients, q-values,and average methylation trends.

To further explore age-related methylation changes, we performed a Differentially Methylated Region (DMR) analysis using the MethyLasso tool. The control cow population was divided into two age groups:

• Young group (n = 20): 2.9 to 5.8 years

• Adult group (n = 20): 6.5 to 9.1 years

A total of 197 DMRs were identified, ranging in size from 24 bp to 1063 bp and containing between 10 and 145 CpGs per region. All DMRs met the criteria of a minimum methylation difference of 10% and a q-value < 0.05. Notably, 73.6% of the DMRs were located in genic regions (Figure 6(B)).

Overall, older cows exhibited higher average methylation levels (51.16%) compared to younger cows (44.76%) (Figure 6(C)).

We specifically focused on the 10.15% of DMRs located in promoter regions. Notably, the MAB21L1 gene contained two promoter-associated DMRs:

1. First DMR: 260 bp region with 54 CpGs, showing a 14.68% increase in methylation in older cows

2. Second DMR: 348 bp region with 63 CpGs, showing a 14.55% increase in methylation in older cows

These results further support the involvement of MAB21L1 in age-related epigenetic regulation, with promoter methylation levels increasing significantly with age.

Other genes, such as NUDT10, RBM33, and INSIG1, showed a substantial decrease in promoter methylation with age, each being associated with a high number of CpGs in their DMRs (21, 20, and 15 CpGs, respectively), suggesting they may also play roles in the ageing process (Table 2).

Table 2.

List of genes targeted by differentially methylated regions (DMRs) located within promoter regions, defined as 0 to 3000 base pairs upstream of the transcription start site (TSS). The table includes genomic coordinates, DMR length, number of CpG sites, direction of methylation change (hypo-/hypermethylation), and average methylation difference between young and adult cows.

MAB21L1	348	63	−14.55	1.48E-02	−2158	mab-21 like 1
MAB21L1	260	54	−14.68	6.64E-04	−1887	mab-21 like 1
NSD1	135	41	−14.67	2.11E-02	−2664	nuclear receptor binding SET domain protein 1
INKA2	155	32	−14.67	4.94E-03	−1517	inka box actin regulator 2
NUDT10	91	21	11.48	1.33E-02	−108	nudix (nucleoside diphosphate linked moiety X)-type motif 10
GABRQ	70	20	−10.21	1.92E-02	−330	gamma-aminobutyric acid type A receptor subunit theta
HOXB9	189	20	−12.32	3.31E-02	−1392	homeobox B9
RBM33	488	20	16.84	8.36E-04	−2042	RNA binding motif protein 33
INSG1	29	15	12.05	2.99E-02	−986	insulin induced gene 1
SULT1C4	42	12	−11.54	5.74E-03	−248	sulfotransferase family, cytosolic, 1C, member 4
SYTL2	78	12	−13.81	5.74E-03	−943	synaptotagmin like 2
TBCE	62	12	−13.64	1.70E-02	−1360	tubulin folding cofactor E
IPO7	90	10	−11.22	4.65E-02	−1968	importin 7
TMSB4	104	10	25.42	4.65E-02	−966	thymosin beta 4, X-linked

Age and mitochondrial DNA methylation

We next analysed the mitochondrial genome to determine whether distinct mtDNA methylation signatures are associated with ageing for the Young group (n = 20): 2.9 to 5.8 years and the Old group (n = 20): 6.5 to 9.1 years. Methylation was assessed across 6434 cytosine sites (716 CpG, 903 CHG, and 4818 CHH), including both CpG and non-CpG contexts. A total of 2869 sites were differentially methylated between young and older cows (q < 0.05; fold change > 1.5), with the majority exhibiting higher methylation in younger animals, suggesting an age-associated decline in mtDNA methylation (Figure 7(A)).

Figure 7.

Mitochondrial DNA methylation and age prediction in dairy cows. (a) distribution of methylation levels across different mitochondrial genes. Blue represents young animals, and red represents adults. (b) relationship between chronological age and the mean DNA methylation levels at mitochondrial cytosine sites significantly correlated with age (|correlation| ≥ 0.5, q-value < 0.05). The trend shows a decrease in mitochondrial methylation with increasing age in control cows. (c) correlation plot between chronological age and predicted age based on the mitochondrial epigenetic clock. The model was built using elastic net regression and evaluated via leave-one-out cross-validation (LOOCV). It achieved a Spearman correlation coefficient of 0.82 (p = 1.909e-08) and a mean absolute error (MAE) of 302.9 days, indicating high accuracy in mitochondrial age prediction (d) number of mtDNA copies per cell and per group. (e) methylation levels per group for DMCs in the D-loop region.

Strand-specific analysis revealed a greater number of differentially methylated cytosines (mtDMCs) on the light strand compared to the heavy strand (2,031 vs. 929), consistent with previously reported strand-biased methylation patterns in mtDNA [24,25]. These results suggest that mitochondrial gene methylation is dynamically regulated with age in dairy cows.

To identify age-associated methylation sites, we applied a stringent filter (|rho| ≥ 0.5; q < 0.05) using data from 40 healthy control cows. We identified 61 age-associated cytosines: 21 CpG, 3 CHG, and 37 CHH sites, which collectively showed a strong negative correlation with age (Spearman’s ρ = −0.77, p = 1.07 × 10^{− 7}; Figure 7(B)). These cytosines were distributed across the mitochondrial genome, including eight protein-coding genes (ATP6, COX1, COX2, COX3, CYTB, ND1, ND4, ND6), four tRNAs (tRNA-Arg, tRNA-His, tRNA-Ile, tRNA-Met), and both mitochondrial rRNAs.

To further explore the potential of mitochondrial methylation as an ageing biomarker, we developed an mtDNA-based epigenetic clock using elastic net regression. From 916 cytosines significantly correlated with age (q < 0.05), we constructed a predictive model using leave-one-out cross-validation (LOOCV) to reduce overfitting and estimate individual age predictions. The final model retained 10 methylation markers (8 CHH, 1 CHG, 1 CpG) and achieved strong performance, with a mean absolute error (MAE) of 303 days (~10 months) and a high correlation between predicted and chronological age (r = 0.82, p = 1.91 × 10^{− 8}; Figure 7(C)).

The retained markers were primarily located in genes encoding subunits of Cytochrome c oxidase (COX1, 2, and 3), with additional sites in the displacement loop (d-loop), tRNA-Phe, and ND5.

We quantified mtDNA copy number per cell by normalizing the total D-loop copies to the nuclear M×1 gene copies obtained via ddPCR. This analysis aimed to explore the relationship between mitochondrial DNA methylation in the D-loop region and mtDNA copy number across age groups. Comparison between age groups revealed that older cows had significantly higher mtDNA copy numbers (mean ± SD: 70.1 ± 35.4) than younger cows (41.7 ± 19.0; p = 0.009) (Figure 7(D)). In parallel, the mitochondrial D-loop region showed that methylation levels (43 DMCs, q < 0.05) were significantly higher in younger cows compared to older ones (p = 1.8E-05) (Figure 7(E)).

Discussion

This study provides novel insights into age-associated DNA methylation patterns across different cattle phenotypes, investigating the interplay between nuclear and mitochondrial methylation, chronological age, and health status. To our knowledge, this is the first comprehensive analysis integrating these elements in dairy cows. We analysed whole-genome methylation profiles from mature Holstein females across five disease phenotypes, covering approximately 53 million CpG sites.

The use of enzymatic methyl-seq (EM-seq) provided excellent genome-wide CpG coverage and preserved DNA integrity, allowing for high-confidence methylation calls, even in GC-rich regions [18,26]. This level of accuracy enabled us to reliably exclude CpG sites with missing data and still generate a robust model for age estimation and correlation analysis.

Compared to previous studies, our model achieved a correlation of r = 0.97 and a mean absolute error (MAE) of 3 months and 21 days, outperforming earlier bovine studies, which reported MAEs of 8 months or more [2,27,28].

At the global level, methylation levels decreased with age, consistent with age-related demethylation observed in mammals [29]. However, specific predictive CpG markers selected by our model showed an increase in methylation with age. Additionally, differentially methylated regions (DMRs) were more methylated in older cows compared to younger ones.

Functional enrichment analysis of age-correlated CpG sites revealed significant enrichment for genes involved in transcriptional regulation, cellular localization, and protein interactions, suggesting that gene expression and cell signalling may influence ageing. Interestingly, 38% of the identified genes are expressed in blood, indicating a possible epigenetic drift for genes involved in the immune system. This observation may differ from the findings of Gonzalez et al. (2025) [30], who reported that maternal age had no significant effect on immune function in adult offspring. Our results showing a premature ageing associated with disease are also original in this species.

When comparing control and disease-susceptible cows, we observed distinct methylation trajectories. In control, fertility-issue, and low-production groups, methylation levels of markers tended to increase with age, whereas in lameness, mastitis, and metabolic disorder groups, methylation levels of markers decreased over time. This is consistent with the nature of these phenotypes. Fertility issues may reflect aspects of normal ageing and are potentially less directly linked to the blood cells, although they are influenced by multiple interacting factors [31,32] and tissues. Low production is not a disease in itself, but may be associated with previous traits as calves [33] or other physiological stresses. In contrast, conditions such as lameness, mastitis, and metabolic disorders often involve inflammation and immune activation. The inverted methylation profile observed in these groups could reflect abnormal biological ageing, particularly affected by cell populations predominant in blood.

The low-production group exhibited the highest epigenetic age acceleration, in line with previous reports showing that high milk yield can accelerate ageing in dairy cows [2]. However, drawing a direct connection between those findings and our own remains challenging, as the cows in our study were not necessarily selected based on milk production. Calves exposed to pathologic or other suboptimal management conditions rarely fulfil their genetic potential [34] as milk producers, and what we see here may be related to the early-life context.

Notably, in diseased cows, epigenetic and chronological age converged as animals aged, suggesting that disease may disrupt typical ageing trajectories at earlier stages but aligns more closely over time. Crucially, the ‘age’ epigenetic markers selected from control cows did not apply well to diseased animals, suggesting that pathological conditions may disrupt standard age-related methylation patterns. However, due to the weak correlation observed between predicted epigenetic age and chronological age in diseased cows, it remains difficult to draw definitive conclusions.

This finding underscores a key consideration in the development of epigenetic clocks: health status must be taken into account when applying such models to diverse populations.

Among the most significant findings, MAB21L1 emerged as a key gene of interest. Its promoter methylation increased significantly with age, and two DMRs were identified within its promoter region. MAB21L1 is known to regulate neuronal development and has been implicated in neurodevelopmental and neurodegenerative disorders [35,36]. Notably, MAB21L1 May also regulate p53 activity, a central modulator of ageing and genomic stability [35,37].

Mab21L1 has been shown to promote the survival of lens epithelial cells by upregulating αB-crystallin and suppressing the ATR/CHK1/p53 stress response pathway, thereby playing a protective role against cataract formation [35]. Our finding of age-associated hypermethylation in the Mab21L1 promoter region supports the hypothesis that epigenetic silencing of this gene over time may contribute to the onset or progression of age-related cataracts. However, the functional interpretation remains speculative, as most of the available evidence for MAB21L1 originates from human and neurodevelopmental studies. The physiological role of this gene in bovine systems remains to be established, and future functional validation in relevant tissues will be required to determine whether similar age-related regulatory mechanisms occur in cattle.

This study also provides the first evidence of mitochondrial DNA (mtDNA) methylation profiles in bovine blood cells. As previously observed in other tissues and species, mtDNA shows consistently lower methylation levels than nuclear DNA [38,39]. Despite these lower baseline levels, mtDNA methylation has already been implicated in biological ageing, senescence, and various disease processes, as summarized by [40].

In the present study, global mtDNA methylation was significantly lower in older animals compared to young. To investigate whether these patterns could serve as a proxy for biological age, we adapted the epigenetic clock framework to mtDNA. A predictive model incorporating 10 cytosine sites demonstrated a strong correlation between predicted and chronological age in healthy cows.

Notably, the cytosines most strongly associated with age were predominantly located within regions encoding subunits of Cytochrome c oxidase (COX), a central enzyme of the mitochondrial respiratory chain. Specifically, COX1, 2, and 3, which form the catalytic core of complex IV [41], play essential roles in ATP production and the regulation of oxidative stress [42]. These results are consistent with prior studies reporting that increased methylation within COX subunit genes has been linked to ageing and cardiovascular disease in humans [43,44].

Another study using senescent mesenchymal stem cells demonstrated hypomethylation of CpG sites within the COX1 gene compared to non-senescent controls [45]. Similarly, hypermethylation of the COX2 gene has been associated with reduced COX2 protein expression in cells from older individuals [46].

In dairy production systems, cows experience substantial physiological stress during lactation, as milk yield often exceeds their natural physiological capacity. This increased energy demand heightens oxygen consumption and ROS production [2,47]. Because mitochondrial function is closely linked to metabolic health, changes in the activity of Cytochrome c oxidase could potentially accelerate ageing through disrupted gene regulation and cellular stress responses. Therefore, monitoring mitochondrial function particularly within COX subunits can serve as an early indicator of physiological stress, fertility decline, or subclinical disease risk associated with cellular ageing.

In humans, it is well established that both the quality and quantity of mitochondrial DNA (mtDNA) are influenced by age [48]. Mitochondrial DNA copy number (mtDNAcn) in peripheral blood cells generally declines with age, and higher mtDNAcn is associated with better health outcomes [49,50]. In other studies, significant associations have been reported between mtDNAcn and D-loop methylation [51,52].

Our results revealed an inverse relationship between D-loop methylation levels and mtDNAcn in bovine blood cells, with younger cows exhibiting higher D-loop methylation and lower mtDNA copy numbers compared to older cows. The D-loop region is the primary regulatory control site of the mitochondrial genome. Previous studies have demonstrated that hypomethylation of the D-loop is associated with an increase in mtDNA copy number [53].

In our study, the lower D-loop methylation observed in older cows may reflect a compensatory mechanism to counteract age-related mitochondrial dysfunction, oxidative stress, or increased metabolic demands associated with intense lactation, thereby resulting in higher mtDNA content per cell. According to Castellani et al., 2020 [54], this pattern suggests that changes in mtDNAcn can influence nuclear DNA methylation and may impact health and disease. These findings further indicate that epigenetic regulation of the D-loop may act as a molecular switch modulating mitochondrial biogenesis in response to physiological ageing in cattle.

This mitochondrial dysregulation aligns with findings in humans, where COX gene methylation, mtDNA copy number, and oxidative stress are interrelated markers of biological ageing and disease susceptibility [55,56]. Environmental stressors such as particulate matter exposure have also been shown to modulate mitochondrial and inflammatory pathways [57], suggesting that both internal metabolic and external environmental pressures can shape mitochondrial epigenetic regulation in mammals.

This study was conducted on cows that developed diseases 2 years after sampling, meaning they were healthy and showed no visible phenotypes. This raises a fundamental question: were the observed DNA methylation differences already in place because the animals were biologically predisposed to disease, or did these methylation differences actively contribute to the later onset of disease? It is well established that ageing increases vulnerability to disease [58]. Therefore, these cows may have exhibited a biologically advanced ageing profile relative to their chronological age, which could have predisposed them to future health issues. In this context, ageing appears to be a potential causal factor in disease development. However, longitudinal studies would be needed to confirm these hypotheses and determine whether the observed epigenetic alterations act as causal factors or are associated with disease development.

What remains unclear, however, is why these cows exhibited such distinct methylation patterns despite living under the same environmental and farming conditions. This suggests the involvement of additional factors, possibly genetic, early-life epigenetic programming, or subtle environmental variations contributing to individual differences in ageing trajectories and disease susceptibility.

This study has some limitations that should be considered when interpreting the results. The relatively small sample size within some disease categories limits statistical power and the generalization of the phenotype-specific effects. In addition, the weak age correlation observed in diseased cows suggests that the current marker set, trained on healthy individuals, may not be fully applicable to diseased populations. Moreover, as the analyses were based exclusively on blood samples, the observed methylation patterns may not fully capture epigenetic changes occurring in disease-relevant tissues. For instance, Wang et al. (2023) [59] demonstrated that in subclinical mastitis, the mammary gland exhibits local methylation alterations reflecting tissue-specific immune responses, which, while responsive to infection, may not be predictive when assessed from peripheral blood. Therefore, validation across multiple organs will be necessary to determine the systemic versus tissue-specific nature of these findings. In addition, the cross-sectional design prevents causal inference, making it difficult to distinguish whether methylation differences represent predisposing factors or associated with the disease. Longitudinal and multi-cohort studies will be required to assess the stability of these associations across different breeds, environments, and management systems.

Despite the overall correlation between methylation and age, substantial inter-individual variability was observed, suggesting that additional factors such as genetic background may contribute to these differences. This variability highlights the complexity of epigenetic ageing and underscores the need for multi-factorial models.

The retrospective classification of animals into ‘diseased’ and ‘control’ groups based on culling records may have introduced selection bias, as culling decisions are influenced by multiple interacting factors such as farm management, economics, or subclinical conditions. Furthermore, although the missForest imputation method effectively handled the low proportion of missing data in this study, future research should assess the robustness of the age-prediction model under different imputation strategies and varying levels of data completeness. Finally, the inter-marker variability and the fact that the age-prediction model was trained on healthy individuals suggest limited transferability to other populations and highlight the need for validation in independent cohorts.

Conclusion

Overall, these findings pave the way for improved precision livestock farming strategies by integrating epigenetic monitoring. Such an approach could help preserve animal health, longevity, and productivity, while anticipating and mitigating the negative consequences of intensive farming. Taken together, our findings support the development of a novel integrative framework, which we term ChronoMeth, to address the complexity of age-related epigenetic changes. By combining cytosine methylation profiles from both nuclear and mitochondrial genomes, ChronoMeth offers a multidimensional view of ageing and disease susceptibility. In contrast to conventional epigenetic clocks based solely on nuclear DNA, this dual-compartment approach enables refined age prediction and provides new insights into the systemic nature of ageing. As such, ChronoMeth represents a promising tool for studying the epigenetic architecture underlying complex traits and biological ageing.

Funding Statement

This research was supported by a grant from Genome Canada and Genome Quebec.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data supporting the findings of this study are openly available in the NCBI Gene Expression Omnibus (GEO). This study includes a total of 96 samples. Among them, 60 samples were previously published under accession number GSE290131 and are also available in the Sequence Read Archive (SRA) under BioProject PRJNA1223015 [60]. In addition, three mastitis samples were retrieved from the study by Bouzeraa et al. (2024) [61] under accession number GSE256195, resulting in a total of 63 reused samples for comparative analysis.

The current submission includes 33 newly generated samples, which are available under accession number GSE296484.

Arrive compliance statement

Reporting for animal research in this study adheres to the ARRIVE 2.0 guidelines.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15592294.2025.2598087

Ethical approval and consent

Ethical approval to conduct the study was provided by the Animal Protection and Ethics Committee of Agriculture and Agri-Food Canada (approval number [A20-0196]). The study adhered to the guidelines set by the Canadian Council on Animal Care.