Age-Related Decrease in FKBP5 Methylation Is Not Confounded by Blood Cell Type Heterogeneity and Occurs in Purified Immune Cell Subtypes.

Peripheral blood cell counts change along the lifespan (31), raising the possibility that heterogeneity in blood cell type composition could be confounding our results (32), despite the consistent inverse relation we observed between aging and FKBP5 methylation after adjustment for calculated cell types in the regression models (Fig. 1A and Dataset S2). To rule out this possibility, we first performed a series of sensitivity analyses in our cohorts. The inverse relation between age and methylation of the two FKBP5 CpGs (cg20813374 and cg00130530) was consistent across the GTP, KORA, and MPIP cohorts (all P values <10⁻⁷; Fig. 1A), and there was no consistent relation between calculated blood cell subtypes (the potential confounder) and either age or FKBP5 methylation levels (our variables of interest), suggesting that strong confounding by cell subtypes was not present (Dataset S3). This was further validated using an additional dataset of male and female subjects (n = 213) with both 450K data and differential complete blood counts; methylation of the age-related CpGs did not significantly correlate with any of the counted blood cell types (Dataset S3) and was again robustly associated with age after adjustment for sex and all cell types (β = −0.0077, SE = 0.0009, P = 5.2 × 10⁻¹⁵).

To further understand how aging influences FKBP5 methylation in specific immune cell types, we analyzed publicly available DNA methylation data in whole blood, as well as FACS-sorted CD4 T cells and neutrophils, from male subjects with a broad age range (33). We again observed an inverse relation between age and methylation of the two FKBP5 CpGs in whole blood (n = 184, r = −0.30, P = 3.6 × 10⁻⁵). Importantly, the same effect size was present in purified CD4 T cells (n = 46, r = −0.32, P = 3.3 × 10⁻²), whereas this effect was in the same direction but nonsignificant in purified neutrophils (n = 48, r = −0.20, P = 1.7 × 10⁻¹; SI Appendix, Fig. S3). Together, these findings show that increasing age is associated with lower FKBP5 methylation in T cell (and likely other distinct immune cell) subtypes and that this effect is not solely the result of age-related changes in blood cell type composition.

Early Life Stress and Depressive Phenotypes Accelerate the Age-Related Decrease in FKBP5 Methylation.

FKBP5 responds to stress and glucocorticoids and can undergo decrease in DNA methylation at distinct CpGs (26–28, 34, 35). Therefore, it is plausible that higher stress burden throughout life could induce lasting epigenetic changes, potentially accelerating decrease in methylation of the two age-related FKBP5 CpGs. To investigate this hypothesis, we examined the combined effects of age and stress-related phenotypes on average methylation of the two CpGs. As information on current depressive symptoms was available in all three cohorts, we first investigated this phenotype. After adjusting for all covariates (SI Appendix, Supplementary Methods), depressive phenotypes consistently accelerated the age-related decrease in FKBP5 methylation (total n = 2,249, meta-analysis interaction P = 2.6 × 10⁻²; Fig. 1B). This association remained significant after further adjusting for education, as a measure of socioeconomic status, in all cohorts (meta-analysis interaction P = 3.7 × 10⁻²). Because early life trauma is among the strongest risk factors for developing MDD (5), we further examined whether the effect of depression on age-related decrease in FKBP5 methylation is moderated by childhood trauma severity as measured with the childhood trauma questionnaire (CTQ) in the GTP. This stratified analysis yielded a significant age–depression interaction in the higher-CTQ (interaction P = 4.6 × 10⁻²) but not the lower-CTQ group (interaction P = 3.3 × 10⁻¹) and no main effect of childhood trauma severity (P = 3.7 × 10⁻¹). Finally, to examine whether exposure to a severe and prolonged early childhood stressor itself is associated with lasting decrease in methylation of the age-related CpGs, we compared elderly individuals that had prolonged early life separation from their parents with sex- and age-matched nonseparated controls in a fourth cohort, the Helsinki Birth Cohort Study (n = 160, age range 58 to 69 y, demographics in Dataset S1). In this cohort, early life separation was associated with reduced methylation of the age-related CpGs (P = 7.4 × 10⁻³; Fig. 1C and SI Appendix, Fig. S1B).

These findings suggest that childhood trauma and depressive phenotypes together accelerate the age-related decrease in FKBP5 methylation in peripheral blood.

The Effects of Both Aging and Stress on FKBP5 Methylation Are Recapitulated in Vitro.

The findings presented above identify two FKBP5 CpGs (cg20813374 and cg00130530) that show a consistent association of lower methylation levels with aging and stress-related phenotypes; however, these findings are inherently limited by the use of human subjects where experimental manipulation is not feasible. To experimentally support these associations, we used a cellular model of replicative senescence (IMR-90 fibroblasts) to test whether replicative aging and stress―which is commonly modeled in the dish with the stress (glucocorticoid) hormone receptor agonist dexamethasone (DEX) (26, 36)―influences FKBP5 methylation at these sites. Population doubling level (PDL) was calculated as previously (37), and FKBP5 methylation was measured with targeted bisulfite pyrosequencing and compared between cells of young (PDL = 22) and old (PDL = 42) replicative age treated for 7 d with either vehicle (DMSO) or 100 nM DEX. In accordance with our in vivo findings, in vitro aging and DEX additively decreased DNA methylation at the two FKBP5 CpGs (interaction P = 4.6 × 10⁻², DNA methylation decrease in old vs. young cells = 10.1%, and additional methylation decrease in old cells treated with DEX vs. vehicle = 3.4%; Fig. 1D). Together with our observations in human cohorts (Fig. 1 A–C), these convergent findings show that aging and stress may influence FKBP5 methylation across different cohorts, distinct cell types, and contexts.

Decreased Methylation at the Age/Stress-Related FKBP5 CpGs Is Associated with FKBP5 Up-Regulation in Peripheral Blood.

DNA methylation patterns can shape gene expression, thereby contributing to cellular function and phenotypic expression (38, 39). The age/stress-related FKBP5 CpGs identified above lie proximally (<500 bp) upstream of the TSS for all highly expressed isoforms of FKBP5 and are intronic only for the minimally expressed variant 2 of the gene (UCSC Genome Browser, UCSC Gens Track; GTEx portal). Integrative analysis of chromatin states using ChromHMM (40) showed that the two CpGs colocalize with signatures that are consistent with either enhancers or flanking active TSS in a large number of cell types (Dataset S4). In immune cells, the CpGs are commonly mapped to either an enhancer or flanking active TSS (SI Appendix, Fig. S4). Further, in most cell types the two sites show intermediate levels of methylation and colocalize with H3K4me1 and H3K27me3 signatures (Roadmap Epigenome Browser; shown for immune cell proxy in SI Appendix, Fig. S5). This landscape is most consistent with a poised enhancer (41) that upon transcription factor binding could interact and regulate the downstream FKBP5 TSS.

To examine whether DNA methylation of these CpGs influences gene transcription, we used FKBP5 mRNA data measured in the GTP cohort with Illumina HumanHT-12 Expression BeadChip arrays (n = 355). DNA methylation levels of the age/stress-related sites were inversely associated with FKBP5 mRNA levels (P = 1.6 × 10⁻²; Fig. 2A). We found similar negative correlations in publicly available data from breast tissue samples of control female subjects (n = 84, r = −0.26, P = 1.6 × 10⁻²; SI Appendix, Fig. S6) (42). Since FKBP5 transcription is robustly induced by glucocorticoids (28, 43) and given that the CpGs are located in predicted poised enhancers (SI Appendix, Fig. S5), we speculated that methylation at the age/stress-related FKBP5 CpGs could moderate the effect of cortisol on FKBP5 levels. After confirming a robust positive correlation between cortisol and FKBP5 mRNA (r = 0.41, P = 1 × 10⁻¹²), we found that the cortisol–FKBP5 relationship was significantly stronger in individuals with below- compared with above-median methylation levels in the GTP (interaction P = 1.4 × 10⁻³; Fig. 2B). In addition, the phenotypes associated with lower methylation levels moderated the relationship between cortisol and FKBP5 mRNA; specifically, this relationship was significantly stronger in older subjects as defined with a median split of age (interaction P = 2.4 × 10⁻⁵; Fig. 2C) and in individuals with higher severity of depression and childhood trauma (interaction P = 7.3 × 10⁻⁵; Fig. 2D). These findings are in line with previous observations that stressors can induce lasting epigenetic effects on other sites of the FKBP5 locus (26–28) and suggest that the effects of aging and stress converge at distinct CpGs to epigenetically up-regulate FKBP5 in human blood.

Open in a separate window

Fig. 2.

Aging and stress-related phenotypes are associated with epigenetic up-regulation of FKBP5 in peripheral blood in the GTP (n = 355). (A) FKBP5 mRNA levels are negatively associated with average methylation of the age/stress-related CpGs (β = −0.3835, SE = 0.1585, P = 1.6 × 10⁻²). (B and C) The cortisol–FKBP5 relationship is stronger at lower methylation levels of the age-related FKBP5 CpGs: interaction P = 1.4 × 10⁻³, β_cortisol for lower methylation = 0.0299 (SE = 0.0044) vs. β_cortisol for higher methylation = 0.0069 (SE = 0.0039). The cortisol–FKBP5 relationship is stronger in older ages: interaction P = 2.4 × 10⁻⁵, β_cortisol for older subjects = 0.0376 (SE = 0.0050) vs. β_cortisol for younger subjects = 0.0075 (SE = 0.0035). (D) Higher levels of depressive symptoms are associated with stronger cortisol–FKBP5 relationship in subjects with higher levels of childhood trauma (cortisol–depression interaction P = 7.3 × 10⁻⁵) but not in subjects with lower levels of childhood trauma (cortisol–depression interaction P = 1.4 × 10⁻¹). In all panels, FKBP5 mRNA residuals are after controlling for all covariates (SI Appendix, Supplementary Methods).

FKBP5 Up-Regulation Promotes NF-κB–Related Peripheral Inflammation and Chemotaxis.

To examine potential functional effects of FKBP5 up-regulation in an unbiased manner, we used genome-wide gene expression data from peripheral blood in the GTP cohort (n = 355) to identify genes that are coregulated with FKBP5. After FDR correction for multiple comparisons (FDR-adjusted P < 0.05), FKBP5 correlated significantly with a total of 3,275 genes (Dataset S5). Using these transcripts as input and the unique array genes expressed above background in blood (except FKBP5) as reference (9,538 genes), we performed pathway and disease association analysis in WebGestalt. The strongest enrichment was observed for inflammation and was conferred by a total of 123 inflammation-related genes (FDR-adjusted P = 8.1 × 10⁻⁶; Fig. 3A and Dataset S6). Notably, FKBP5 showed robust positive relation with a host of proinflammatory genes, such as interleukin and toll-like receptors (Dataset S7). Furthermore, FKBP5 levels were positively associated with the granulocyte proportion (r = 0.22, P = 5.8 × 10⁻⁵) and the granulocyte-to-lymphocyte ratio (r = 0.31, P = 7.4 × 10⁻⁹; SI Appendix, Fig. S7), an inflammation marker that is associated with increased cardiovascular risk and mortality (44, 45), but not with the proportions of CD4 T cells (r = −0.05, P = 3.6 × 10⁻¹). These associations suggest that FKBP5-related inflammation could be driven by enhanced chemotaxis of granulocytes and other proinflammatory cells. As plausible mediator of this effect, we focused on interleukin-8 (IL-8), a major chemokine that recruits and activates granulocytes and other proinflammatory cells (46). Although FKBP5 down-regulation has been found to suppress IL-8 production in melanoma cells (19), no studies have examined whether FKBP5 up-regulation influences IL-8 secretion by immune cells. To address this possibility, we overexpressed FKBP5 in Jurkat cells, a human T cell line that allowed efficient and reproducible transfection with FKBP5 expression vectors (~3.2-fold induction; Fig. 3B), and measured their potential to secrete IL-8. FKBP5 overexpression nearly doubled IL-8 secretion upon immune stimulation (P = 4.4 × 10⁻⁷; Fig. 3C), supporting that increased FKBP5 in T cells could drive chemotaxis of proinflammatory cells.

Open in a separate window

Fig. 3.

FKBP5 up-regulation promotes NF-κB–driven peripheral inflammation. (A) FKBP5-related genes in peripheral blood show enrichment for inflammation-related genes and NF-κB gene targets. The number of genes for each analysis is shown in parentheses. Statistical details are provided in Datasets S5–S9. (B) Western blotting confirming FKBP5 overexpression in Jurkat T cells transfected with FKBP51-FLAG vs. cells transfected with the control vector. (C) FKBP5 overexpression nearly doubles IL-8 secretion by Jurkat T cells stimulated overnight with 25 ng/mL of phorbol-12-myristate-13-acetate and 375 ng/mL of ionomycin (PMA/I). The bar graph depicts IL-8 secretion in stimulated cell supernatants measured with ELISA from two independent experiments (t = 8.8, P = 4.4 × 10⁻⁷, n = 8 per condition). For each experiment, fold ratios of IL-8 secretion were calculated relative to stimulated cells expressing the control vector. IL-8 was not detectable in nonstimulated cells. (D) FKBP5 overexpression increases NF-κB activity in stimulated Jurkat T cells. The bar graph depicts NF-κB reporter activity in stimulated cells measured with dual-luciferase reporter assays from three independent experiments (t = 3.2, P = 5.5 × 10⁻³, n = 9 per condition). For each experiment, fold ratios of NF-κB activity were calculated relative to nonstimulated cells expressing the control vector. (E) FKBP5 expression changes are associated with extensive alterations in the NF-κB coexpression network in peripheral blood. The circles depict genes encoding molecular partners of the NF-κB pathway. Continuous lines (edges) represent positive and dotted lines negative pairwise correlations corrected for expression levels of all other genes in the pathway (details in SI Appendix, Supplementary Methods). Edge widths are proportional to the absolute value of the respective correlation coefficient. The gene pair with the most robust difference in correlation between the two groups (CHUK-MAP3K14) is highlighted in orange. Statistical details for all gene pairs are provided in Dataset S10. Error bars depict the SE around the group mean. **P < 10⁻²; ***P < 10⁻³.

To further examine whether the effects of FKBP5 on the immune system may be driven by distinct transcription factors, we performed transcription factor target analysis in the GTP cohort using the same input and reference gene sets (3,275/9,538). The strongest enrichment was observed for NF-κB (FDR-adjusted P = 4.5 × 10⁻³; Fig. 3A and Dataset S8), a master immune regulator that has been linked to FKBP5 (13, 17), and this was driven by a total of 75 NF-κB gene targets (Fig. 3A and Dataset S9). To experimentally confirm that FKBP5 up-regulation promotes NF-κB signaling in immune cells, we performed dual-luciferase reporter assays comparing NF-κB activity between Jurkat cells overexpressing FKBP5 and cells transfected with control vector. FKBP5 overexpression led to increased NF-κB activity in response to immune stimulation (P = 5.5 × 10⁻³; Fig. 3D). Together these findings support that FKBP5 up-regulation in immune cells promotes NF-κB–dependent peripheral inflammation accompanied by the release of proinflammatory cytokines, such as IL-8. Therefore, our subsequent analyses sought to better characterize the mechanisms through which FKBP5 impacts the NF-κB pathway.

Changes in FKBP5 Levels Are Associated with Extensive Alterations in the NF-κB Coexpression Network.

To determine the network-level effects of FKBP5 deregulation on NF-κB signaling, we used the gene expression data in the GTP cohort (n = 355) to calculate the pairwise correlations between genes encoding molecules that directly interact within the NF-κB pathway, as defined in the KEGG Pathway Database. These pairwise correlations were adjusted for the expression levels of all other genes in the pathway and compared between subjects above vs. below the median split for FKBP5 expression levels. As shown arithmetically in Dataset S10 and schematically in Fig. 3E, several partial pairwise correlations within the NF-κB pathway differed between the two groups, but the strongest and only significant effect after multiple test correction was noted for the MAP3K14–CHUK pair (r_{lower FKBP5} = 0.13 vs. r_{higher FKBP5} = −0.28, FDR-adjusted P = 1.1 × 10⁻², permutation P = 2.6 × 10⁻³). This effect remained robust after controlling for sex, age, cortisol, and Houseman-calculated blood cell proportions (FDR-adjusted P = 1.3 × 10⁻⁴, permutation P = 7.1 × 10⁻³), indicating that the effects of FKBP5 on NF-κB signaling are not confounded by cortisol levels or blood cell composition.

FKBP5 Up-Regulation Promotes NF-κB Signaling by Strengthening the Interaction of Key Regulatory Kinases.

Since FKBP5 is involved in scaffolding of regulatory protein complexes, it could enhance NF-κB signaling by influencing protein–protein interactions between regulators of the NF-κB pathway. Intriguingly, MAP3K14 and CHUK, the transcript pair most influenced by FKBP5 levels (Fig. 3E), respectively encode the NF-kappa-B-inducing kinase (NIK) and the antagonist of nuclear factor kappa-B kinase subunit alpha (IKKα), two key regulatory kinases of the alternative NF-κB pathway. NIK interacts with and phosphorylates IKKα at serine 176 (pIKKα^S176), thereby activating IKKα and facilitating NF-κB signaling (47, 48).

To examine whether FKBP5 modulates the NIK–IKKα protein complex, we performed a series of protein–protein binding experiments in human Jurkat cells and peripheral blood mononuclear cells (PBMC). These experiments showed binding of FKBP5 with both NIK and IKKα and binding between NIK and IKKα (Fig. 4A). We then examined whether glucocorticoid treatment and FKBP5 up-regulation can influence the FKBP5–NIK–IKKα complex. Both cell types were stimulated with DEX that robustly induces FKBP5 expression (28, 43). After confirming the DEX-induced up-regulation of FKBP5 (~2.2-fold), we found that DEX treatment significantly increased the binding between FKBP5, NIK, and IKKα in both Jurkat cells and PBMC; this increase was abolished by concomitant treatment with the selective FKBP5 antagonist SAFit1 (49) in both cell types (Fig. 4 A and B). Accordingly, these effects on protein binding were accompanied by an increase in pIKKα^S176, whereas pIKKα^S176 induction was abolished by treatment with SAFit1 (Fig. 4C). This effect on pIKKα^S176 was recapitulated by FKBP5 overexpression and again blocked by concomitant treatment with SAFit1 in Jurkat cells (Fig. 4D). Additionally, in this cell line deletion of the FKBP5 gene with CRISPR/Cas9 abolished the effect of DEX on pIKKα^S176 levels but did not influence vehicle-treated cells (SI Appendix, Fig. S8), thus mimicking the effects of SAFit1. In line with these functional effects on the NIK–IKKα complex, FKBP5 overexpression nearly doubled NF-κB activity in Jurkat cells, whereas this effect was again prevented by concomitant treatment with SAFit1 (Fig. 4E). As schematically summarized in Fig. 4F, these convergent findings show that FKBP5 up-regulation strengthens NIK–IKKα binding, increases pIKKα^S176, and in effect promotes NF-κB signaling.

Open in a separate window

Fig. 4.

FKBP5 up-regulation promotes NF-κB signaling by strengthening the binding of key regulatory kinases, and these effects are prevented by selective FKBP5 antagonism. (A and B) Immunoprecipitation (IP) for either FKBP5 or NIK followed by Western blotting and binding quantification in lysates from Jurkat cells or PBMC treated with the stress hormone (glucocorticoid) receptor agonist DEX, which robustly induces FKBP5 expression, and/or the selective FKBP5 antagonist SAFit1. IgG: control IP without primary antibody. Jurkat: FKBP5 to NIK binding, DEX × SAFit1 F_1,8 = 9.3, interaction P = 1.6 × 10⁻²; IKKα to FKBP5 binding, DEX × SAFit1 F_1,8 = 4.7, interaction P = 6.2 × 10⁻²; IKKα to NIK binding, DEX × SAFit1 F_1,8 = 5.8, interaction P = 4.3 × 10⁻². PBMC: FKBP5 to NIK binding, DEX × SAFit1 F_1,8 = 5.7, interaction P = 4.4 × 10⁻²; IKKα to FKBP5 binding, DEX × SAFit1 F_1,8 = 11.2, interaction P = 1 × 10⁻²; IKKα to NIK binding, DEX × SAFit1 F_1,8 = 3.9, interaction P = 8.4 × 10⁻². n = 3 biological replicates per condition. (C) Western blotting of Jurkat cell and PBMC lysates (n = 3 replicates per condition) showing increase in the functional phosphorylation of IKKα at serine 176 (pIKKα) by DEX that is prevented by SAFit1 (Jurkat: DEX × SAFit1 F_1,8 = 12.9, interaction P = 7 × 10⁻³; PBMC: DEX × SAFit1 F_1,8 = 0.6, interaction P = 4.6 × 10⁻¹). (D) Similar effects are observed when Jurkat cells are transfected with an expression construct encoding FKBP5 (ect. FKBP5) and treated with SAFit1 (ect. FKBP5 × SAFit1 F_1,12 = 6.6, interaction P = 2.5 × 10⁻², n = 4 replicates per condition). (E) FKBP5 overexpression increases NF-κB activity in Jurkat cells stimulated overnight with 25 ng/mL of phorbol-12-myristate-13-acetate and 375 ng/mL of ionomycin, and this increase is prevented by concomitant treatment with SAFit1 (ect. FKBP5 × SAFit1 F_1,32 = 4.5, interaction P = 4.2 × 10⁻², n = 9 replicates per condition). NF-κB reporter activity was measured with dual-luciferase reporter assays in three independent experiments. (F) Scheme summarizing the results from protein–protein binding and reporter gene experiments. All treatments with DEX and SAFit1 were performed for 24 h at 100 nM concentration. All data are shown as fold changes compared with the control-vector vehicle-treated cells. All statistical comparisons were performed with two-way ANOVA, using either DEX treatment or FKBP5 overexpression as the first factor and SAFit1 treatment as the second factor. Statistically significant effects were followed with Bonferroni-corrected pairwise comparisons, shown as follows: *P < 5 × 10⁻², **P < 10⁻², ***P < 10⁻³, statistically significant pairwise comparisons for control vs. DEX or ect. FKBP5; ^#P < 5 × 10⁻², ^##P < 10⁻², ^###P < 10⁻³, pairwise comparisons for vehicle vs. SAFit1 treatment (shown only for significant interaction terms of the respective two-way ANOVAs). Error bars depict the SE around the group mean.

NF-κB Signaling Promotes FKBP5 Expression via an NF-κB Response Element Containing the Age/Stress-Related CpGs.

Notably, the above-identified age/stress-related FKBP5 CpGs flank an NF-κB response element (SI Appendix, Fig. S9), raising the possibility that NF-κB signaling could itself modulate FKBP5 expression in immune cells via this site. To address this possibility, we performed dual-luciferase reporter gene assays using a CpG-free vector (50). We inserted into this vector the FKBP5 sequence that surrounds the NF-κB response element and includes the two CpG sites of interest but completely lacks any other CpGs (SI Appendix, Fig. S9). Immune stimulation induced expression of this reporter construct in monocyte-derived human cell lines (THP-1) (Fig. 5A), thus supporting functionality of this response element in immune cells. Furthermore, in vitro DNA methylation of the age/stress-related FKBP5 CpGs within this reporter construct resulted in statistically significant reduction (~40%) of baseline expression levels and nearly abolished the induction seen with immune stimulation (Fig. 5A). To further examine whether these functional effects are mediated by alterations in NF-κB binding, we used a biotinylated oligonucleotide-mediated chromatin immunoprecipitation (ChIP) method (51) (Fig. 5B and SI Appendix, Fig. S9). After confirming immune stimulation-driven NF-κB binding in THP-1 cells to the enhancer, in vitro DNA methylation essentially abolished NF-κB binding to the age/stress-related enhancer site (Fig. 5 C and D). Together, these findings demonstrate that NF-κB signaling—which, as we showed above, is promoted by FKBP5 (Figs. 3 and ​and4)—can4)—can in turn trigger FKBP5 expression in immune cells, thereby forming a positive feedback loop that can potentiate FKBP5–NF-κB signaling. This positive feedback can thus be accentuated by decreased methylation of the NF-κB–responsive FKBP5 enhancer, which can occur as a consequence of aging and stress.

Open in a separate window

Fig. 5.

NF-κB signaling drives FKBP5 expression via a response element gated by the age/stress-related CpGs. (A) Data from dual-luciferase reporter gene assays using a CpG-free luciferase reporter construct, which includes the FKBP5 sequence surrounding the NF-κB response element and the age/stress related CpGs (insert sequence shown in SI Appendix, Fig. S9) but completely lacks other CpGs. This reporter construct was in vitro-methylated and transfected into monocyte-derived human cell lines (THP-1). Cells were then stimulated overnight with 25 ng/mL phorbol-12-myristate-13-acetate and 375 ng/mL ionomycin (PMA/I), a combination that robustly induces NF-κB signaling. Data are derived from two independent experiments (n = 12 replicates per condition). Comparison was performed using two-way ANOVA with methylation and treatment as factors (F_1,44 = 59.5, interaction P < 10⁻³), and statistically significant effects were followed with Bonferroni-corrected pairwise comparisons. (B–D) The effect of in vitro DNA methylation on PMA/I-induced NF-κB binding to the NF-κB response element was examined using biotinylated oligonucleotide-mediated ChIP in THP-1 cells (oligonucleotide sequence shown in SI Appendix, Fig. S9). Schematic summary of the experimental setup is shown in B. After ChIP, NF-κB/p65 binding was quantified by Western blotting using antibodies specific for NF-κB (C: example blots; D: quantifications). CTRL (control) 1: magnetic beads lacking conjugated streptavidin; CTRL 2: cells transfected with nonbiotinylated oligonucleotide. Bar graph shows data derived from four independent experiments (t = 2.5, P = 4.4 × 10⁻², n = 4 per condition). Statistical t test compared cells carrying the unmethylated probe treated overnight with vehicle or PMA/I. Binding was not quantifiable for cells carrying the methylated probe. Data are always shown as fold changes compared with the vehicle-unmethylated cells. Error bars depict the SE around the group mean. P values for pairwise comparison are shown as follows: ***P < 10⁻³, statistically significant pairwise comparisons for methylated vs. unmethylated. ^#P < 5 × 10⁻²; ^###P < 10⁻³, statistically significant pairwise comparisons for vehicle vs. drug treatment.

This work was supported by a Marie-Sklodowska Curie fellowship (H2020 Grant 653240) to A.S.Z., a grant from the National Institute of Mental Health (MH071538) to K.J.R., a European Research Council starting grant within the FP7 Framework to E.B.B. (Grant 281338, GxE molmech), a grant from the National Institute of Mental Health (U19 MH069056), a grant by the German Federal Ministry of Education and Research (BMBF) through the Integrated Network IntegraMent (Integrated Understanding of Causes and Mechanisms in Mental Disorders) under the auspices of the e:Med Programme (Grant 01ZX1314J) to E.B.B., and by the Academy of Finland (Grants 284859, 2848591, and 312670). Part of this work has been included and presented in the PhD dissertation of A.S.Z. (63).