Age-associated DNA methylation changes in naive CD4⁺ T cells suggest an evolving autoimmune epigenotype in aging T cells

Abstract

Aim:

We sought to define age-associated DNA methylation changes in naive CD4⁺ T cells.

Materials & methods:

Naive CD4⁺ T cells were collected from 74 healthy individuals (age 19–66 years), and age-related DNA methylation changes were characterized.

Results:

We identified 11,431 age-associated CpG sites, 57% of which were hypermethylated with age. Hypermethylated sites were enriched in CpG islands and repressive transcription factor binding sites, while hypomethylated sites showed T cell specific enrichment in active enhancers marked by H3K27ac and H3K4me1. Our data emphasize cancer-related DNA methylation changes with age, and also reveal age-associated hypomethylation in immune-related pathways, such as T cell receptor signaling, FCγR-mediated phagocytosis, apoptosis and the mammalian target of rapamycin signaling pathway. The MAPK signaling pathway was hypermethylated with age, consistent with a defective MAPK signaling in aging T cells.

Conclusion:

Age-associated DNA methylation changes may alter regulatory mechanisms and signaling pathways that predispose to autoimmunity.

Understanding human aging only within genomic constrains is impossible, as genetics appears to explain only a small proportion of the observed variation in lifespan and health [1–3]. DNA methylation is arguably a promising complement to the genetic component of aging [4,5]. A growing number of human diseases and normal processes are linked to epigenetic alterations, emphasizing the importance of DNA methylation changes. Methylation changes have been identified as a hallmark of cancer, immunologic and neurologic disorders, as well as aging [6–10]. For example, it has been demonstrated that, using ‘epigenetic clock’, tumor tissues appear to age 40% faster than the corresponding normal tissue [11,12].

The connection between age-associated methylation sites and immunologic functions has been noted, but remains relatively less explored than age–cancer epigenetic similarity. The majority of age-associated DNA methylation studies use whole blood, or unfractionated peripheral blood mononuclear cells (PBMCs) [11,13–16]. Although age-associated DNA methylation changes have been suggested as a multitissue phenomenon, and several studies indeed show high similarity among each other, concerns remain that cell type and composition may significantly affect interpretation of the age-associated DNA methylation changes [10–13,16–22]. Thus, studying purified immune cell subsets may be the most feasible way to investigate age-associated changes and the corresponding immune function alterations.

This study focuses on age-associated changes in DNA methylation of naive CD4⁺ T cells, a lymphocyte population that differentiates into effector CD4⁺ T cells after activation by complexes of MHC and antigen [23]. Defects in T cell DNA methylation have been previously described in lupus, an autoimmune disease characterized by autoantibody production against a number of nuclear antigens [24,25]. The prevalence of these autoantibodies increases with age in healthy individuals [25], and T cells from aging individuals have been shown to demethylate and overexpress some of the same genes demethylated and overexpressed in T cells from lupus patients [26]. Furthermore, naive CD4⁺ T cells have previously been shown to have distinct hypomethylation and epigenetic poising of interferon-regulated genes in lupus [27]. In this study, we focused on characterizing age-sensitive DNA methylation sites in human naive CD4⁺ T cells, and explore the potential relationship between these age-dependent DNA methylation changes and autoimmunity.

Methods

Study demographics

Seventy four healthy female individuals were included in this study. A subset of these individuals was recruited for a prior study comparing the DNA methylation changes in naive CD4⁺ T cells of healthy European–Americans and African–Americans (GEO accession: GSE79237) [28]. The age of the participants ranged from 19 to 66 years, with a mean age of 40.8 and a median age of 40. The racial distribution of study participants was as follows: 47 European–Americans, 22 African–Americans, three Asians, one Hispanic and one Indian/Arab. Healthy participants were recruited from the Oklahoma Medical Research Foundation, University of Michigan Health System and Henry Ford Health System. The institutional review boards of the participating institutions approved this study and all participants provided written, informed consent prior to enrollment.

Naive CD4⁺ T cell isolation & DNA extraction

Naive CD4⁺ T cells were isolated from peripheral blood as previously described [27,28]. Briefly, PBMCs were isolated from peripheral blood using Ficoll-Paque^™ PLUS (GE Healthcare, PA, USA). Naive CD4⁺ T cells were isolated from PBMCs by indirect labeling using the naive CD4⁺ T cell Isolation Kit II, human (Miltenyi Biotec, CA, USA). Other cells were bound by biotinylated antibodies and using antibiotin-labeled magnetic beads, bound to a column allowing untouched naive CD4⁺ T cells to flow through. DNA was isolated from untouched naive CD4⁺ T cells using the DNeasy Blood and Tissue Kit (Qiagen, MD, USA) according to manufacturer's instructions. Concentration and 260/280 absorbance ratio were measured using a NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific, MA, USA).

Naive CD4⁺ T cell DNA methylation profiling

Five hundred nanogram of naive CD4⁺ T-cell DNA was bisulfite-converted using the EZ DNA Methylation^™ Kit (Zymo Research, CA, USA) according to manufacturer's instructions. The bisulfite converted DNA was hybridized to Illumina Infinium HumanMethylation450 BeadChip arrays (Illumina, CA, USA) according to manufacturer's instructions. The BeadChip arrays were scanned with the iScan reader (Illumina). All samples passed Illumina array quality control measures.

Preprocessing of methylation data

The DNA methylation data were imported into a MethyLumiSet object using R software with the methylumi v. 2.16.0 package. Probes confounded with array batch (using BeadChip ID number) were removed (n = 1164). Nonspecific (n = 29,155), polymorphic (n = 62,344) and chromosome Y (n = 294) probes were also removed based on best practice recommendations [29]. Background correction and quantile normalization was performed using the dasen method from the wateRmelon v. 1.10.0 package. The batch effect of the Infinium I and II chemistries was adjusted using the BIMQ method [20]. Several visualization strategies provided by the wateRmelon and lumi v. 2.22.1 packages were utilized to ensure the quality of background correction/normalization. The batch effect was confirmed by principal component analysis (PCA) and removed using the ComBat function in the sva v. 3.18.0 R package. Three individuals (2 European–Americans and 1 African–American; mean age 38.6 years) were removed from further analysis at this stage, as ComBat cannot adjust for batch effect in a batch consisting of one sample. The background corrected, normalized and batch effect removed dataset was used for further analysis.

Regression analysis

To evaluate the association of methylation differences with age, a beta regression model was calculated using the betareg v. 3.0.5 R package, along with linear regression and Pearson correlation coefficient approaches. The beta regression model has been shown to be particularly well suited to test associations based on the distribution of methylation β values [30,31]. The model included race, BeadChip ID and sample chip placement as covariates. A Benjamini–Hochberg adjusted p-value threshold of 0.05 was selected as the threshold of statistical significance prior to performing the regression analysis. All subsequent genomic and epigenomic enrichment analyses, epigenomic similarity analysis, and functional enrichment analyses, as described below, were performed using age-dependent DNA methylation changes identified using this regression analysis.

Selection of CpG sites showing substantial age differences

To detect CpG sites showing large change in DNA methylation during aging, the β values were transformed to M values using the equation . The median M values between individuals in the higher 75th (n = 18; 53–66 years) and lower 25th (n = 17; 19–32 years) percentile of the age range were compared. CpG sites with |ΔM| >1 were selected [32]. The rank sums of the M values of the two groups were further compared using Wilcoxon test.

Genomic & epigenomic enrichment analysis

Positional and epigenomic enrichment analyses were performed as described previously [33]. Briefly, the enrichment analysis evaluates whether a set of age-associated CpG sites colocalizes with genome annotation datasets in a statistically significant manner, utilizing genomic coordinates of the CpG sites and genomic annotations in the hg19/GRCh37 human genome coordinate system. All CpG sites included on the Illumina Infinium 450K array were used as a ‘background’. Genomic coordinates of chromosome bands and transcription factor/regulator binding sites obtained by ChIP-seq from ENCODE (wgEncodeRegTfbsClusteredV3 data table) were obtained from the UCSC genome browser database [34]. Coordinates of gene/transcript types [35] were obtained from annotables R package. Genomic Evolutionary Rate Profiling (GERP) elements [36], CpG islands [37] and Functional Annotation of the Mammalian Genome enhancers [38], as well as chromatin states, 15-mark model, experimentally obtained histone modifications and gapped peaks from the Roadmap Epigenomics project [39] were obtained from the accompanying web sites. The two-tailed chi-square test was used to calculate enriched and depleted associations. While enriched associations imply significant concentration of CpG sites in the tested regions, depleted associations indicate that age-associated CpG sites are devoid of tested regions compared with background. All reported p-values were corrected using Benjamini–Hochberg procedure.

Epigenomic similarity analysis

To compare epigenomic signatures of age-associated CpG sites from different studies, we performed epigenomic similarity meta-analysis as described previously [33]. Briefly, age-associated CpG sites were tested for enrichment in multiple epigenomic annotations, and the corresponding epigenomic enrichment profiles of –log₁₀- transformed p-values were compared using Pearson correlation coefficient and PCA. Experimentally obtained histone modification marks detected with gapped peak algorithm were used. These included H3K4me1, H3K4me3, H3K4me3, H3K9me3, H3K27me3, H3K27ac and H3K36me3. This analysis allows comparing epigenomic enrichments of age-associated CpG sites in the context of reference genome annotations.

Gene set functional enrichment analysis

Genes annotated with CpG sites hypermethylated or hypomethylated with age were tested for enrichment using ‘molecular function’, ‘biological process’ and ‘cellular component’ gene ontology annotations, and Kyoto Encyclopedia of Genes and Genomes canonical pathways [40]. Enrichment p-values were calculated using chi-square test in the GOStats v2.36.0 R package. The p-values were corrected for multiple testing using the Benjamini–Hochberg procedure.

Results

Naive CD4⁺ T-cell-specific age-associated methylation changes

DNA methylation profiles generated from naive CD4⁺ T cells of 71 healthy women (age range: 19–66 years) were included in our analysis. Figure 1 shows our analysis pipeline to characterize age-associated DNA methylation changes. PCA of the DNA methylation profiles identified the first principle component to be correlated with batch ID (PC1 accounting for 10.48% variability, p = 3.10 × 10^-27; Supplementary Figure 1), suggesting technical batch effect. Although other parameters, such as race and sample chip placement, did not significantly contribute to variability of the data (not shown), all three variables including bath ID, sample chip placement and race, were included in the regression model as covariates in all analyses.

Figure 1. . Analysis pipeline used to identify and characterize age-associated DNA methylation changes in naive CD4⁺ T cells.

Age-associated CpG sites are predominantly hypermethylated in naive CD4⁺ T cells

Our approach was two sided in that we aimed to identify CpG sites significantly hypermethylated or hypomethylated with age. A total of 11,431 CpG sites were identified as having methylation values with a significant association with age (Supplementary Table 1). Fifty six of these CpG sites were previously identified as a part of the 353 CpG sites defining an ‘age–epigenetic clock’ signature [12], which was statistically significant (p = 1.09 × 10^-24, Fisher's exact test). Furthermore, 300 CpG sites identified in our study were a part of the 6366 CpG sites signature of increased variability with age, as identified in the recent study by Slieker et al. [41] (p = 2.06 × 10^-15, Fisher's exact test).

A total of 6533 (57%) CpG sites were positively correlated (hypermethylated) with age compared with 4898 (43%) negatively correlated (hypomethylated) sites (Supplementary Table 1). Table 1 lists the top 20 most robust CpG sites associated with age (as measured by r² value) observed in this study. In order to identify CpG sites showing the largest methylation differences at the extremes of age range, we compared M values between two groups of individuals in the lower 25th and upper 75th percentiles of the total age range, 56 CpG sites had a ΔM >1 and Benjamini–Hochberg corrected Wilcoxon p < 0.05 (Supplementary Table 2).

Table 1. . Top 20 CpG sites with DNA methylation levels that correlate with age in naive CD4⁺ T cells using beta regression analysis.

CpG site	Age coefficient	Regression p-value	r2	Pearson correlation	Pearson correlation p-value	Gene symbol
cg16867657	0.01470	2.42 × 10-25	0.65402	0.80879	0.00 × 1000	ELOVL2
cg18898125	0.00870	8.75 × 10-18	0.57144	0.76013	1.51 × 10-14	NEFM
cg04212239	0.01685	9.29 × 10-15	0.54095	0.74173	1.37 × 10-13	SMC4
cg19283806	-0.01833	1.64 × 10-14	0.53900	-0.72976	5.23 × 10-13	CCDC102B
cg00474657	-0.01392	7.36 × 10-14	0.52064	-0.72892	5.73 × 10-13	SPNS1
cg01697487	-0.02314	7.36 × 10-14	0.52550	-0.72819	6.20 × 10-13	MYCBPAP
cg15353603	-0.02453	7.36 × 10-14	0.52392	-0.73051	4.82 × 10-13	DPP4
cg24724428	0.01238	7.36 × 10-14	0.52689	0.72268	1.12 × 10-12	ELOVL2
cg01409343	-0.02258	7.76 × 10-14	0.51800	-0.72437	9.35 × 10-13	TMEM49
cg05762671	-0.02160	1.56 × 10-13	0.51405	-0.71969	1.53 × 10-12	KCTD19
cg02202664	-0.02398	5.05 × 10-13	0.51321	-0.71588	2.27 × 10-12	TIAM2
cg10701110	-0.01065	5.21 × 10-13	0.50237	-0.71352	2.89 × 10-12	ARHGAP30, USF1
cg07642566	0.01136	5.96 × 10-13	0.50347	0.70857	4.75 × 10-12	C19orf30, MIR7–3
cg14377791	-0.02071	7.60 × 10-13	0.51007	-0.70851	4.78 × 10-12	LOC400696
cg17851795	-0.01366	9.67 × 10-13	0.50167	-0.71006	4.10 × 10-12	GPSM3, PBX2
cg13029847	0.01120	1.07 × 10-12	0.50039	0.70446	7.13 × 10-12	SEZ6
cg23003085	-0.01710	1.21 × 10-12	0.50237	-0.70558	6.39 × 10-12	PRDX5, TRMT112
cg03917666	-0.02410	1.79 × 10-12	0.49840	-0.70761	5.23 × 10-12
cg21915449	-0.01651	4.46 × 10-12	0.48747	-0.70049	1.05 × 10-11	HAVCR1
cg15953461	-0.01524	6.91 × 10-12	0.48626	-0.69800	1.33 × 10-11

CpG site: Unique CpG locus identifier; Age coefficient: Age correlation coefficient, 1 year of age change corresponds to coefficient value change in methylation; Regression p-value: significance of the regression coefficient; r²: Percent of variance explained by the regression fit; Pearson correlation: Pearson correlation coefficient between age and methylation level; Pearson correlation p-value: Significance of the correlation.

CpG sites associated with age differ in their genomic location & regulatory context

We sought to investigate previously reported observations that CpG sites hypermethylated or hypomethylated with age are located in distinct genomic compartments and enriched in different epigenomic elements [13,14,42–44]. To identify chromosomal regions with unusually high concentration of age-associated CpG sites, we performed enrichment analysis of CpG sites on chromosomes and chromosomal bands (Figure 2). CpG sites hypermethylated with age were enriched on chromosome 2q36.2 (p = 1.55 × 10^-30), 2q24.1 (p = 8.69 × 10^-19), 2q31.1 (p = 5.19 × 10^-14), 6q16.3 (p = 2.51 × 10^-12) and several other bands. They were also depleted on chromosome 6p21.33 (p = 3.40 × 10^-09), 1p36.32 (p = 4.62 × 10^-04) and others (Supplementary Table 3). On the contrary, CpG sites hypomethylated with age were enriched on chromosome 6p21.32 (p = 1.52 × 10^-14), 6p21.33 (p = 2.81 × 10^-10), 17q23.1 (p = 1.52 × 10^-13), 3p21.2 (p = 4.81 × 10^-13), and depleted on chromosome 12q24.33 (p = 9.66 × 10^-04) (Supplementary Table 3). Expectedly, some enrichments were mutually exclusive, for example, CpG sites hypermethylated with age were strongly depleted in the 6p21.33 chromosomal band (p = 3.40 × 10^-09), while CpG sites hypomethylated with age were enriched in this band (p = 2.81 × 10^-10). These results confirm our hypothesis that age-associated hypermethylated and hypomethylated events may involve different genomic compartments.

Figure 2. . Genomic colocalization of naive CD4⁺ age-associated CpG methylation sites in chromosomal bands.

Significant enrichments (red bars) and depletions (green bars) for CpG sites hypermethylated (A & B) and hypomethylated (C & D) with age are shown.

Y-axis represents -log10-transformed p-values obtained using chi-squared test.

The differential localization of CpG sites hypermethylated or hypomethylated with age has been observed in the meta-analysis of multiple independent studies [17]. To compare genomic location and the corresponding epigenomic signatures of the age-associated CpG sites detected in our study with other studies, we performed epigenomic similarity analysis of our hypermethylated or hypomethylated age-associated sites and those identified in six other studies [13,14,19,42,43,45]. Expectedly, CpG sites hypermethylated or hypomethylated with age are grouped into distinct regulatory clusters (Figure 3). PCA further confirmed this observation, with the age-correlation accounting for 86.06% variability explained by the first principle component (p = 6.78 × 10^-05) (Supplementary Figure 2). Our list of CpG sites hypomethylated with age expectedly showed high regulatory similarity with those identified in CD4⁺ T cells by Reynolds et al. [42] (Pearson coefficient = 0.82, p < 1.00 × 10^-16), indicating that the location and regulatory impact of age-associated CpG sites identified in our study in naive CD4⁺ T cells are most similar to those previously identified in total CD4⁺ T cells (Figure 3).

Figure 3. . Epigenomic similarities among sets of CpG sites hypermethylated and hypomethylated with age identified in previous studies, compared with one another and to this study.

Set-specific epigenomic enrichment profiles were obtained using Roadmap Epigenomics histone modification data, and pair-wise correlated using Pearson correlation coefficient. The resulting correlation matrix was clustered using Euclidean distance/ward.D clustering metrics. A color gradient from blue to red indicates Pearson correlation coefficient values, with blue being -1 and red being 1. Groups of hypermethylated and hypomethylated sets of sites are marked with purple and green color of the dendrogram and the side bars, respectively. The following studies were compared (Study name, number of hypermethylated sites, number of hypomethylated sites, tissue types used): Alisch et al. (2012), 479 hypermethylated sites, 1599 hypomethylated sites, peripheral blood cells; Marttila et al. (2015), 3925 hypermethylated sites, 4615 hypomethylated sites, peripheral blood mononuclear cells; Weidner et al. (2014), 58 hypermethylated sites, 44 hypomethylated sites, whole blood; Reynolds et al. (2014), 2049 hypermethylated sites, 546 hypomethylated sites, CD4⁺ T cells; Heyn et al. (2012), 1219 hypermethylated sites, 1986 hypomethylated sites, cord blood of newborns and CD4⁺ T cells of centenarians; and Florath et al. (2012), 119 hypermethylated sites, 43 hypomethylated sites, whole blood. Data taken from [13,14,19,42,43,45].

Hypermethylated, but not hypomethylated age-associated CpG sites are enriched in CpG islands & conserved regions

Regulatory enrichment analysis identified CpG sites hypermethylated with age as enriched in known CpG islands (p = 6.15 × 10^-04) and evolutionary conserved GERP elements (p = 2.38 × 10^-36). CpG sites hypomethylated with age were significantly depleted in CpG islands (p = 1.89 × 10^-287) and GERP elements (p = 5.16 × 10^-13).

Cell type specific enrichment of CpG sites hypomethylated with age

Enrichment analysis of age-associated CpG sites for overlap with known classes of genes and transcripts [35], identified hypermethylated CpG sites to be enriched in protein-coding genes (p = 6.26 × 10^-04). CpG sites hypomethylated with age were enriched in cell type specific genes, ‘T-cell J receptor gene’ (p = 8.50 × 10^-23) and ‘all T-cell receptor genes combined’ (p = 6.12 × 10^-08). Both CpG sites hypermethylated and hypomethylated with age were depleted in ‘pseudogene’ transcript types (p = 8.13 × 10^-11 and p = 1.83 × 10^-12, respectively). These results suggest that CpG sites hypermethylated with age may regulate noncell type specific expression of protein-coding genes, while CpG sites hypomethylated with age regulate T cell specific gene expression activity.

Hypermethylated CpGs are associated with PRC2, while hypomethylated CpGs are enriched in enhancers

The different types of enhancer regions enriched in CpG sites hypermethylated or hypomethylated with age were also observed in the histone mark enrichment analysis focused on previously obtained histone mark ChIP-seq data in primary naive CD4⁺ T cells [39]. CpG sites hypermethylated with age were enriched in H3K27me3 (p = 1.06 × 10^-305), a mark of polycomb repressive complex II (PRC2) binding, and H3K4me1 (p = 1.06 × 10^-305), a classical activating mark, that together with H3K27me3, marks bivalent promoters [46]. These bivalent promoters, repressed in the absence of differentiation signals, are considered to regulate temporal expression of developmental genes. CpG sites hypomethylated with age in naive CD4⁺ T cells were enriched in H3K4me1 (p = 2.44 × 10^-228) and H3K27ac (p = 6.96 × 10^-207), a mark of active enhancers (Supplementary Table 3).

EZH2 & SUZ12 binding sites are enriched in CpGs hypermethylated with age

We investigated whether age-associated CpG sites are enriched in binding sites of transcription factors and chromatin remodeling complexes. Binding sites of two core proteins of PRC2 complex, EZH2 (p = 1.61 × 10^-304) and SUZ12 (p = 5.73 × 10^-181) were over-represented in hypermethylated CpG sites, but the opposite trend was observed in CpG sites hypomethylated with age (under-represented EZH2 (p = 4.37 × 10^-51) and SUZ12 (p = 6.13 × 10^-20). Intriguingly, enrichment analysis of the recently released EED binding sites, another core protein of PRC2 complex, identified EED as being enriched in both hypermethylated and hypomethylated CpG sites (p < 3.00 × 10^-39, Supplementary Table 3).

Among transcription factor binding sites, CpG sites hypomethylated with age were also enriched in BATF (p = 4.1 × 10^-159), a negative regulator of AP-1 mediated signaling [47], and IKZF1 (p = 1.09 × 10^-101) and BCL11A (p = 1.81 × 10^-67), transcription factors important in lymphocyte development [48–50]. Also enriched was IRF4 (p = 1.37 × 10^-45), a lymphocyte-specific transcription factor that negatively regulates Toll-like receptor signaling. CpG sites hypermethylated or hypomethylated with age were depleted in PHF8 and KDM5B transcription factor binding sites (PHF8 p = 5.54 × 10^-84 and 1.47 × 10^-25, KDM5B p = 1.06 × 10^-52 and 1.03 × 10^-16 for hypermethylated and hypomethylated age-associated CpG sites, respectively). Both transcription factors are members of the Jumonji family of proteins that play a role in chromatin remodeling and histone demethylation [51,52] (Figure 4 & Supplementary Table 3).

Figure 4. . Genomic colocalization of naive CD4⁺ T cell age-associated CpG methylation sites in transcription factor binding sites.

Significant enrichments (red bars) and depletions (green bars) for CpG sites hypermethylated (A & B) and hypomethylated (C & D) with age are shown.

Y-axis represents -log10-transformed p-values obtained using chi-squared test.

Functional enrichments of genes annotated with age-associated CpG sites

A total of 3183 and 3022 unique genes were annotated with CpG sites hypermethylated and hypomethylated with age, respectively (Table 2 & Supplementary Table 4). Genes associated with CpG sites hypermethylated with age were enriched in ‘Hedgehog signaling pathway’ (p 1.10 × 10^-05) (Table 3), downregulation of which has been associated with age-related diseases [53]. Other pathways included ‘pathways in cancer’ (p = 1.10 × 10^-05), ‘focal adhesion’ (p = 8.35 × 10^-04), ‘Wnt signaling pathway’ (p = 8.35 × 10^-04) and ‘MAPK signaling pathway’ (p = 4.35 × 10^-03) (Table 3 & Supplementary Table 5). Genes associated with CpG sites hypomethylated with age were enriched in two metabolism-related pathways, ‘insulin signaling pathway’ (p = 4.94 × 10^-04) and ‘mammalian target of rapamycin (mTOR) signaling pathway’ (p = 2.95 × 10^-03), (Table 4). Further reflecting immune links of age-associated methylation profiles, ‘Fc gamma R-mediated phagocytosis’ (p = 6.37 × 10^-04) and ‘T cell receptor signaling pathway’ (p = 8.41 × 10^-04) were also enriched in genes associated with hypomethylated CpGs. Cancer-related pathways, such as ‘small cell lung cancer’ (p = 8.41 × 10^-04), ‘acute myeloid leukemia’ (p = 8.41 × 10^-04) and ‘apoptosis’ (p = 8.41 × 10^-04) were also among the top most significant hypomethylated pathways (Table 4 & Supplementary Table 5).

Table 2. . Top 20 genes with the largest number of CpG sites hypermethylated or hypomethylated with age in naive CD4⁺ T cells.

Genes annotated with CpG sites hypermethylated with age
Gene symbol	Number of CpG probes	Description
PRRT1	22	Proline-rich transmembrane protein 1
MAD1L1	21	MAD1 mitotic arrest deficient-like 1 (yeast)
MSI2	14	Musashi RNA-binding protein 2
KCNQ1	13	Potassium channel, voltage gated KQT-like subfamily Q, member 1
SOX2OT	12	SOX2 overlapping transcript
PITX2	11	Paired-like homeodomain 2
REC8	11	REC8 meiotic recombination protein
BLCAP	10	Bladder cancer associated protein
FAM124B	10	Family with sequence similarity 124B
GNA12	10	Guanine nucleotide binding protein (G protein) alpha 12
JARID2	10	Jumonji, AT rich interactive domain 2
NNAT	10	Neuronatin
NR4A2	10	Nuclear receptor subfamily 4, group A, member 2
GALNT2	9	Polypeptide N-acetylgalactosaminyltransferase 2
SYNGAP1	9	Synaptic Ras GTPase activating protein 1
ZIC1	9	Zic family member 1
C10orf26	8	WW domain binding protein 1-like
FOXG1	8	Forkhead box G1
GIPR	8	Gastric inhibitory polypeptide receptor
H2AFY	8	H2A histone family, member Y
Genes annotated with CpG sites hypomethylated with age
Gene symbol	Number of CpG probes	Description
IFT140	15	Intraflagellar transport 140
PRDM16	15	PR domain containing 16
KIAA1949	12	Protein phosphatase 1 regulatory subunit 18
RASA3	12	RAS p21 protein activator 3
RPTOR	12	Regulatory associated protein of MTOR, complex 1
DAXX	11	Death-domain associated protein
ZC3H12D	9	Zinc finger CCCH-type containing 12D
B3GALT4	8	UDP-Gal:betaGlcNAc beta 1,3-galactosyltransferase, polypeptide 4
EHMT2	8	Euchromatic histone-lysine N-methyltransferase 2
KALRN	8	Kalirin, RhoGEF kinase
TAP1	8	Transporter 1, ATP-binding cassette, sub-family B (MDR/TAP)
TMEM49	8	Vacuole membrane protein 1
PLEC1	7	Plectin
SKI	7	SKI proto-oncogene
ABHD14A	6	Abhydrolase domain containing 14A
ABHD14B	6	Abhydrolase domain containing 14B
CISH	6	Cytokine inducible SH2-containing protein
CUTA	6	cutA divalent cation tolerance homolog (Escherichia coli)
DGKZ	6	Diacylglycerol kinase, ζ
HCCA2	6	MOB kinase activator 2

Table 3. . Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of genes annotated with CpG sites hypermethylated with age in naive CD4⁺ T cells.

Kyoto Encyclopedia of Genes and Genomes pathway	FDR-adjusted p-value	Gene symbol
Hedgehog signaling pathway	1.10 × 10-05	BMP4, ZIC2, CSNK1E, WNT10A, WNT5B, GLI3, WNT6, BMP2, BMP7, BMP8B, BTRC, CSNK1A1, FBXW11, GAS1, GLI1, LRP2, PRKACA, PRKACG, WNT1, WNT2B, WNT3, WNT5A, WNT7A, WNT9A
Pathways in cancer	1.10 × 10-05	SPI1, CDK6, FGF17, BMP4, MECOM, PDGFB, AXIN2, CTNNA2, IGF1R, TRAF3, WNT10A, WNT5B, ZBTB16, AKT3, BCL2, BRCA2, CDKN1A, CSF3R, EPAS1, FGF2, GLI3, GSTP1, JUP, LAMC1, RAC1, RARA, RXRG, SMAD3, WNT6, ABL1, APPL1, AXIN1, BCL2L1, BCR, BIRC3, BMP2, CDKN2A, EGLN3, FGF10, FGF19, FGF9, FN1, FOS, FZD6, GLI1, GRB2, HDAC1, ITGA6, JAK1, KIT, KRAS, LAMA1, LAMC2, MAP2K1, MITF, MLH1, MMP9, NKX3–1, NTRK1, PAX8, PIAS1, PIK3R3, PTEN, RALB, RALGDS, SMAD2, STAT1, STAT3, STAT5A, TCF7L1, TFG, TGFBR1, TGFBR2, TPM3, TPR, VHL, WNT1, WNT2B, WNT3, WNT5A, WNT7A, WNT9A
Adherens junction	1.27 × 10-04	TJP1, CTNND1, CTNNA2, IGF1R, LMO7, RAC1, SMAD3, ACTN1, ACTN2, ACTN4, BAIAP2, CSNK2A1, FYN, IQGAP1, NLK, PTPRJ, PVRL1, SMAD2, SNAI1, SORBS1, SRC, TCF7L1, TGFBR1, TGFBR2, WAS, WASL
Maturity onset diabetes of the young	8.35 × 10-04	NR5A2, NKX6–1, NEUROD1, PAX6, HES1, MAFA, ONECUT1, GCK, HHEX, HNF1B, NEUROG3, NKX2–2
Focal adhesion	8.35 × 10-04	PIP5K1C, PDGFB, IGF1R, PARVA, AKT3, BCL2, CCND3, FLNC, LAMC1, PAK6, PARVG, PPP1CC, RAC1, THBS2, VTN, ACTN1, ACTN2, ACTN4, BCAR1, BIRC3, CAPN2, CAV1, COL11A2, COL5A2, COL6A3, COL6A6, DOCK1, FLT1, FLT4, FN1, FYN, GRB2, ITGA10, ITGA11, ITGA6, ITGA8, ITGB5, LAMA1, LAMC2, MAP2K1, PAK2, PDPK1, PIK3R3, PPP1CA, PTEN, SRC, THBS1, THBS4, TNXB, VAV3
Wnt signaling pathway	8.35 × 10-04	LRP5, AXIN2, CSNK1E, DKK4, NKD1, PPP2CA, SFRP2, SOX17, WNT10A, WNT5B, CAMK2D, CCND3, CTNNBIP1, NFATC1, NFATC2, RAC1, SMAD3, WNT6, AXIN1, BTRC, CSNK1A1, CSNK2A1, DKK2, FBXW11, FRAT1, FZD6, NLK, PPP2R5A, PPP3R1, PRKACA, PRKACG, SMAD2, TBL1XR1, TCF7L1, WNT1, WNT2B, WNT3, WNT5A, WNT7A, WNT9A
Regulation of actin cytoskeleton	3.28 × 10-03	GNA12, ARHGEF7, FGF17, PIP5K1C, PDGFB, CHRM2, EZR, FGF2, GSN, ITGAE, ITGAM, MSN, PAK6, PPP1CC, RAC1, ACTN1, ACTN2, ACTN4, BAIAP2, BCAR1, CD14, CHRM4, CHRM5, CSK, DIAPH3, DOCK1, FGF10, FGF19, FGF9, FN1, INSRR, IQGAP1, ITGA10, ITGA11, ITGA6, ITGA8, ITGB5, KRAS, MAP2K1, MYH10, PAK2, PFN3, PIK3R3, PPP1CA, SLC9A1, SSH1, TIAM1, VAV3, WAS, WASL
Arrhythmogenic right ventricular cardiomyopathy	4.17 × 10-03	CACNG8, CTNNA2, DSP, JUP, ACTN1, ACTN2, ACTN4, CACNA2D1, CACNA2D2, CACNA2D4, CACNB2, CACNG2, CACNG3, CACNG4, ITGA10, ITGA11, ITGA6, ITGA8, ITGB5, LMNA, SGCD, TCF7L1
Neuroactive ligand-receptor interaction	4.17 × 10-03	GIPR, GRM2, CNR2, GRIK5, HTR6, P2RX1, GALR2, GRIK2, NR3C1, PTGDR, TSHR, ADORA3, CHRM2, CHRNB4, EDNRB, F2RL1, GLRA1, GRIN2D, GRM5, HTR1E, PRLHR, SSTR3, ADORA2B, ADRA1A, ADRB1, ADRB3, AVPR1A, CALCRL, CHRM4, CHRM5, CHRNA5, CNR1, DRD2, GABRA1, GABRA2, GABRB3, GABRG2, GABRQ, GABRR2, GLRA3, GRIA2, GRIN1, GRM6, GRM8, HTR2B, HTR4, HTR5A, LEP, LEPR, MCHR2, MLNR, NPY1R, NPY5R, NTSR2, OPRL1, P2RY1, SSTR2, TACR3, TBXA2R, VIPR2
MAPK signaling pathway	4.35 × 10-03	GNA12, CACNA1I, RPS6KA2, FGF17, MAP3K8, CACNG8, MECOM, PDGFB, MAP4K4, AKT3, CACNA1G, DUSP14, DUSP6, FGF2, FLNC, NFATC2, RAC1, RASGRP2, CACNA1B, CACNA2D1, CACNA2D2, CACNA2D4, CACNB2, CACNG2, CACNG3, CACNG4, CD14, DUSP4, DUSP8, ECSIT, FGF10, FGF19, FGF9, FOS, GRB2, HSPA2, KRAS, MAP2K1, MAP2K4, MAP2K5, MAP2K6, MAP3K3, MAP3K5, MAP3K6, MAPK14, MAPK8IP1, MAPKAPK2, MAPKAPK3, NLK, NTRK1, PAK2, PPP3R1, PRKACA, PRKACG, RASGRP4, RPS6KA4, TGFBR1, TGFBR2, ZAK*

FDR: False discovery rate.

Table 4. . Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of genes annotated with CpG sites hypomethylated with age in naive CD4⁺ T cells.

Kyoto Encyclopedia of Genes and Genomes pathway	FDR-adjusted p-value	Gene symbol
Insulin signaling pathway	4.94 × 10-04	RPTOR, BAD, PDE3B, PIK3CD, MAPK10, SREBF1, AKT1, FLOT1, G6PC, PIK3R1, PIK3R2, PRKAG2, PYGM, AKT3, CBL, CRKL, EIF4E, EIF4E2, EIF4EBP1, FASN, FLOT2, GYS1, INPP5D, INS, IRS1, MKNK2, PHKG2, PIK3CA, PIK3R5, PPP1CA, PPP1R3B, PRKAR2B, PRKCI, PTPN1, PTPRF, RAPGEF1, RPS6KB1, RPS6KB2, SOCS1, SOCS2, SORBS1, TSC2
Fc gamma R-mediated phagocytosis	6.37 × 10-04	PRKCA, PIK3CD, AKT1, ARPC4, DNM1, LAT, PIK3R1, PIK3R2, AKT3, ASAP2, ASAP3, CRKL, DNM3, INPP5D, LYN, MARCKSL1, MYO10, PIK3CA, PIK3R5, PIP5K1A, PIP5K1B, PIP5K1C, PLA2G4D, PLCG2, PRKCB, RAC2, RPS6KB1, RPS6KB2, WAS, WASF2, WASF3
Small cell lung cancer	8.41 × 10-04	RXRB, PIK3CD, AKT1, COL4A1, PIK3R1, PIK3R2, TRAF4, TRAF5, AKT3, BCL2, BIRC2, CCND1, CDK2, E2F2, IKBKG, LAMA3, LAMA4, LAMA5, LAMB4, NFKB1, NFKBIA, PIAS3, PIK3CA, PIK3R5, PTK2, RXRA, TRAF1, TRAF2
T cell receptor signaling pathway	8.41 × 10-04	PTPN6, TNF, PIK3CD, MAP3K14, AKT1, CARD11, CD4, FYN, LAT, PIK3R1, PIK3R2, ZAP70, AKT3, CBL, CD28, CD40LG, CTLA4, GRAP2, IKBKG, IL5, ITK, MAP2K7, MAPK12, NFATC1, NFKB1, NFKBIA, PAK6, PDCD1, PIK3CA, PIK3R5, PRKCQ, RASGRP1, RHOA
Acute myeloid leukemia	8.41 × 10-04	BAD, PIK3CD, STAT5A, AKT1, PIK3R1, PIK3R2, PIM2, PPARD, AKT3, CCND1, EIF4EBP1, IKBKG, NFKB1, PIK3CA, PIK3R5, PIM1, RPS6KB1, RPS6KB2, STAT3, STAT5B, TCF7
Apoptosis	8.41 × 10-04	TNF, BAD, PIK3CD, CAPN2, MAP3K14, AKT1, BID, IRAK4, PIK3R1, PIK3R2, AKT3, ATM, BCL2, BIRC2, CASP7, CASP8, CFLAR, ENDOG, IKBKG, NFKB1, NFKBIA, NGF, NTRK1, PIK3CA, PIK3R5, PRKAR2B, TRADD, TRAF2
Toxoplasmosis	1.43 × 10-03	TNF, BAD, PIK3CD, HLA-DOA, MAPK10, AKT1, HLA-DPB1, HSPA1L, IRAK4, MAP2K3, PIK3R1, PIK3R2, AKT3, BCL2, BIRC2, CASP8, CCR5, CD40LG, GNAI2, HLA-DMB, IKBKG, IL10RA, IL12A, JAK1, LAMA3, LAMA4, LAMA5, LAMB4, LDLR, MAPK12, NFKB1, NFKBIA, PIK3CA, PIK3R5, PLA2G12B, SOCS1, STAT3
Phosphatidylinositol signaling system	1.43 × 10-03	DGKZ, PRKCA, PIK3CD, INPP4B, PI4KB, PIK3R1, PIK3R2, CDS2, DGKI, INPP4A, INPP5A, INPP5D, INPP5E, ITPK1, ITPR2, PIK3C2G, PIK3CA, PIK3R5, PIP5K1A, PIP5K1B, PIP5K1C, PLCG2, PLCZ1, PRKCB, SYNJ2
Endocytosis	1.43 × 10-03	ARAP1, HLA-F, IGF1R, SMURF1, EHD1, ERBB3, HLA-E, AP2A2, DNM1, EHD2, HLA-B, HSPA1L, IL2RB, IQSEC3, SMAP1, ACAP1, ADRBK1, AGAP2, ARAP3, ASAP2, ASAP3, CBL, CCR5, CHMP6, DNM3, FGFR2, GRK4, GRK6, HGS, HLA-A, HLA-C, IL2RA, LDLR, MET, NEDD4L, NTRK1, PARD6A, PDCD6IP, PIP5K1A, PIP5K1B, PIP5K1C, PRKCI, PSD, PSD4, RAB7A, RABEP1, RHOA, SH3GL1, SMAD6, SRC, TFRC
mTOR signaling pathway	2.95 × 10-03	RPTOR, PIK3CD, AKT1, PIK3R1, PIK3R2, AKT3, CAB39, EIF4E, EIF4E2, EIF4EBP1, INS, PIK3CA, PIK3R5, RPS6KA1, RPS6KB1, RPS6KB2, TSC2, VEGFA*

FDR: False discovery rate; mTOR: Mammalian target of rapamycin.

Discussion

The biological process of aging can be conceptualized as the interplay of genetics and intrinsic and extrinsic stress experienced by an individual beginning at conception and continuing throughout the lifespan [54]. These exposures accumulate across time and are reflected by the phenomenon of ‘epigenetic drift’ by which the human methylome can vary widely between individuals, even identical twins, as age increases [55]. In this study, we focus on human naive CD4⁺ T cells and identify age-associated DNA methylation changes and their relationship to cellular functions.

We observed that over half of the age-sensitive methylation sites identified by our study increased in methylation with age. This observation is consistent with other studies that report similar proportions of the age-associated CpG sites [13,14,42,43], although general tendency toward global age-associated hypomethylation has been noted [43,44]. These hypermethylated CpG sites were enriched in CpG islands and evolutionarily conserved regions of the human genome. CpG islands are CpG-rich homogeneous regions of the genome that are associated with transcriptional regulation [56]. Consistent with previous observations, our results support the notion that CpG sites hypermethylated with age are enriched in PRC2 signature, while hypomethylated CpGs are enriched in enhancers [17].

Consistent with the H3K27me3 enrichment, we identified an enrichment of binding sites of EZH2 and SUZ12, two members of PRC2 complex, as being enriched in CpG sites hypermethylated with age. Conversely, CpG sites hypomethylated with age were significantly depleted in EZH2 and SUZ12. Although it is tempting to speculate that the entire PRC2 complex preferentially binds at CpG sites hypermethylated with age, we observed enrichment of EED, another component of the PRC2 complex, in CpG sites both hypermethylated and hypomethylated with age. A prior study by our group suggests that EZH2 might play a role in epigenetic remodeling in naive CD4⁺ T cells that favors T cell activation, and might predispose to lupus flares [57]. Our observations warrant further studies of the role of PRC2 in autoimmunity, focusing on the complex interplay of the PRC2 components.

Despite this contrasting difference in EZH2 and SUZ12 binding sites enrichment, CpG sites hypermethylated or hypomethylated with age were similarly depleted in PHF8 and KDM5B transcription factor binding sites. Both PHF8 and KDM5B are members of the Jumonji family involved in histone demethylation and cell cycle progression [51,52]. This observation positions Jumonji family of proteins as ‘protected’ from age-associated methylation changes, suggesting their binding sites are preserved during aging. These results suggest that age-associated methylation differences affect distinct transcription factor binding site signatures, while methylation status of cell cycle and Jumonji family binding sites are largely unaffected during aging.

We localized the areas of methylation associated with aging to specific chromosomal bands. Chromosomal regions enriched with CpG sites of increasing DNA methylation levels with age included the top five most significant regions: 2q36.2, 2q24.1, 2q31.1, 6q16.3 and 3q24. These regions have been previously associated with aging or age-associated disorders like Alzheimer's disease and hearing loss in genome-wide association studies [58–61]. Chromosomal regions enriched with DNA methylation sites that decrease in methylation levels with age include: 6p21.32, 17q23.1, 3p21.2, 6p21.33 and 11q13.1. These regions have been associated with disease like amyotrophic lateral sclerosis, and immune-mediated diseases, such as multiple sclerosis and Crohn's disease [62–65]. These results suggest that CpG sites hypermethylated and hypomethylated with age are concentrated in distinct genomic locations, and might regulate functionally different genes by altering distinct epigenomic signatures.

To identify functional networks related to age-sensitive CpG sites, we first identified genes proximal to these sites. Notably, our results identified age-associated hypermethylation in the ELOVL2 gene, which encodes a protein involved in the synthesis of long polyunsaturated fatty acids, and previously reported to be strongly associated with age [14,18,42]. ELOVL2 methylation has been shown to be a biomarker for aging in whole blood by Garagnani et al. [66], particularly site cg16867657, which had the strongest association with age in our study (Pearson coefficient = 0.81). We observed a second CpG site associated with ELOVL2 (cg24724428) with a slightly lower, but significant association with age (Pearson coefficient = 0.72). A recent study of Elovl2 ^(-/-) mice has shown increased production of the inflammatory cytokines IFN-γ and IL-17, a sign of increased Th1 and Th17 activity, and a reduction in the number of Foxp3+ T_reg cells compared with wild-type mice [67]. As hypermethylation is generally associated with transcriptional silencing, progressive age-dependent hypermethylation of ELOVL2 in naive CD4⁺ T cells might suggest a proinflammatory T-cell epigenotype with age. Other genes, such as APOE, FOXO3, NEFM, CCDC102B, MBP and CAPN2 have also been reported as age-associated in different cell- and tissue types (Supplementary Table 1) [14,18,42,68]. These observations suggest that some genes may show age-associated methylation changes across multiple tissues, including naive CD4⁺ T cells. The functional role of these methylation changes in T-cell maturation and activity and how it changes with age remains to be investigated.

Gene ontology and pathway analyses of genes near or containing age-sensitive CpG sites revealed several age-related functional groups. Hedgehog signaling pathway genes were enriched for CpG sites hypermethylated with age as well as signaling pathways for MAPK and the Wnt pathways. Gene ontologies for RNA polymerase II binding sites and system and structural development pathways were also enriched for in these sites. In contrast, CpG sites hypomethylated with age were enriched for pathways related to T-cell activity including T cell receptor, mTOR and chemokine signaling, as well as insulin signaling and apoptosis. Gene ontologies were enriched for cellular responses and immune regulation.

The age-associated methylation pattern we observed in naive CD4⁺ T cells corroborate in many ways findings observed in T cells isolated from lupus patients. Lupus is a prototype autoimmune disease characterized by T cell DNA methylation defect resulting in T-cell autoreactivity [24]. Indeed, lupus T cells are characterized by defective MAPK signaling [69], and our findings showed that MAPK signaling pathway appeared significantly hypermethylated with age and therefore likely silenced at the epigenetic level in aging naive CD4⁺ T cells. CD40LG and CD11a (encoded by ITGAL) which are regulated by DNA methylation and are hypomethylated and overexpressed in lupus CD4⁺ T cells and normal CD4⁺ T cells treated by DNA methylation inhibitors [70,71], were also hypomethylated with age in our study. Indeed, hypomethylation of ITGAL and CD11a overexpression has been previously demonstrated with aging in total T cells [26], consistent with our naive CD4⁺ T cell data. Lupus T cells are characterized by mTOR activation, which contributes to T cells autoreactivity and proinflammatory phenotype [72]. Similarly, mTOR signaling pathway appeared progressively poised for activation at the epigenetic level in naive CD4⁺ T cells from healthy individuals with increased age in our study. T cells in lupus patients and in aging are characterized by increased apoptosis, presumably providing the source of autoantigens for T-cell autoreactivity. We show that apoptosis-related pathways were hypomethylated in naive CD4⁺ T cells with age. Taken together, our data support that epigenetic dysregulation at the DNA methylation level in aging naive CD4⁺ T cells favor T-cell autoreactivity and increased apoptosis. These findings might contribute to increased risk of autoimmunity with age.

Previous enhancer-mapping studies have reported that active cis-regulatory modules in gene promoters are CpG-depleted [38], suggesting CpG sites negatively associated with age may be located in such enhancer-like cis-regulatory modules. These observations further strengthen our observation that CpG sites hypermethylated or hypomethylated with age may have different epigenomic signatures. Notably, CpG sites hypermethylated with age showed overall higher regulatory similarity despite the fact these sites were identified in different cell types (Figure 3), indicating similar location and regulatory impact of age-associated hypermethylated CpGs across cell types. These observations suggest age-associated hypermethylated regions may be conserved among cell and tissue types and are highly distinct from the cell type specific age-associated hypomethylated regions.

Conclusion

We demonstrate age-dependent DNA methylation changes in naive CD4⁺ T cells, and distinct genomic and regulatory enrichment patterns between loci hypermethylated and hypomethylated with age. Our findings suggest a progressive age-associated shift in T-cell epigenomes toward proinflammatory and T cell activating epigenotypes. These findings support a role for age-dependent DNA methylation changes in explaining increased autoimmunity with age. Future studies that examine longitudinal samples collected over time from the same individuals, and studies to examine other specific immune cell types will allow for a more comprehensive understanding of age-related autoimmune epigenotypes.

Summary points.

• The aim of this study was to characterize age-associated DNA methylation changes in human naive CD4⁺ T cells and their potential disease relevance.

• Seventy four healthy female individuals (age 19–66 years) were recruited, and naive CD4⁺ T cells were collected from peripheral blood.

• Genome-wide DNA methylation was assessed using the Infinium HumanMethylation450 BeadChip array (Illumina).

• Age-related DNA methylation changes were defined using regression analysis and characterized for genomic and epigenomic enrichment patterns, and gene set functional enrichment.

• We identified 11,431 age-associated CpG sites, 57% of which were hypermethylated with age in naive CD4⁺ T cells.

• We observed distinct genomic locations and epigenomic enrichment patterns of CpG sites hypermethylated or hypomethylated with age.

• CpG sites hypermethylated with age were enriched in EZH2 and SUZ12 binding sites, comprising the well-known polycomb repressive complex 2 signature.

• CpG sites hypomethylated with age showed T cell specific enrichment in active enhancers marked by H3K27ac and H3K4me1 histone marks.

• A total of 3022 genes associated with CpG sites hypomethylated with age were enriched in immune-related pathways, such as T-cell receptor signaling and the mammalian target of rapamycin pathway.

• A total of 3183 genes associated with CpG sites hypermethylated with age were enriched in Hedgehog signaling, pathways in cancer, focal adhesion and Wnt signaling. The MAPK signaling pathway, which is defective in autoimmune diseases such as lupus, was also significantly hypermethylated with age.

• Our findings extend the role of age-associated hypermethylation of polycomb repressive complex 2 binding signature to naive CD4⁺ T cells, and strengthen the notion that age-associated methylation changes may alter distinct regulatory mechanisms and signaling pathways that predispose to autoimmunity.