The Signature of Maternal Rearing in the Methylome in Rhesus Macaque Prefrontal Cortex and T Cells

Abstract

Early-life adversity is associated with a broad scope of life-long health and behavioral disorders. Particularly critical is the role of the mother. A possible mechanism is that these effects are mediated by “epigenetic” mechanisms. Studies in rodents suggest a causal relationship between early-life adversity and changes in DNA methylation in several “candidate genes” in the brain. This study examines whether randomized differential rearing (maternal vs surrogate–peer rearing) of rhesus macaques is associated with differential methylation in early adulthood. The data presented here show that differential rearing leads to differential DNA methylation in both prefrontal cortex and T cells. These differentially methylated promoters tend to cluster by both chromosomal region and gene function. The broad impact of maternal rearing on DNA methylation in both the brain and T cells supports the hypothesis that the response to early-life adversity is system-wide and genome-wide and persists to adulthood. Our data also point to the feasibility of studying the impact of the social environment in peripheral T-cell DNA methylation.

Introduction

Early-life adverse social experiences in humans are associated with an increased risk for developing psychiatric disorders, such as anxiety and major depression (Kaufman et al., 2000; McEwen, 2000; Heim and Nemeroff, 2001), as well as increasing vulnerability to developing chronic disease during adulthood (Power et al., 2007). The broad range of phenotypes suggests a system-wide response to early-life adversity. The main question is, what is the mechanism that registers the response to early-life adversity and has a life-long system-wide impact on multiple phenotypes?

It is now well established that genes are regulated by epigenetic information, including DNA methylation. There is an association of childhood abuse with DNA methylation levels in the promoters of ribosomal RNA genes (McGowan et al., 2008) and exon 1f of the glucocorticoid receptor gene promoter (McGowan et al., 2009). Exposure of infant rats to stressed caretakers produced persisting changes in DNA methylation of the brain-derived nerve growth factor gene promoter in the adult prefrontal cortex (PFC) (Roth et al., 2009). Early-life stress (Gräf et al., 2007) in mice caused sustained DNA hypomethylation in the arginine vasopressin gene (Murgatroyd et al., 2009). Although these data support the hypothesis that DNA methylation is involved in registering early-life adversity in the DNA, they do not explain the range of phenotypes affected by child adversity.

There are critical challenges in expanding the initial findings in rodents, in which causality could be tested, to humans. It is impossible to randomize early-life stress in humans, and therefore it is hard to tell whether the differences observed are driven by innate genetic differences or whether they are environmentally driven. Nonhuman primates offer several advantages in addressing these challenges, including a close evolutionary and phenotypical relationship with humans. In the model examined here, rhesus macaque monkeys are randomly assigned at birth to differential rearing conditions by either their mother or an inanimate, cloth-covered surrogate. The surrogate rearing serves as a controlled and relevant early-life stressor. In nonhuman primate models, maternal deprivation with some form of social contact disrupts the mother–infant relationship, leading to important emotional and social disturbances and behavioral abnormalities, such as motor stereotypes (Suomi, 1991; Champoux et al., 2002; Barr et al., 2003). Peer-reared macaques develop inadequate social skills, are highly reactive and aggressive, and, as adults, show increased voluntary alcohol consumption and typically rank at the bottom of the social dominance hierarchy (Suomi, 1991; Barr et al., 2003). We examined here in parallel genome-wide promoter methylation profiles from isolated T cells and from the PFC of adult rhesus males subjected to maternal and surrogate rearing conditions after birth using the method of methylated DNA immunoprecipitation (MeDIP) with comprehensive genome-wide promoter microarray hybridization, statistical analysis, and false discovery rate (FDR) correction. We provide evidence that the impact of the social environment on DNA methylation is genome-wide and can be seen in brain as well as peripheral T cells.

Materials and Methods

Animals and rearing procedures

Samples were obtained from eight male rhesus monkeys (Macaca mulatta) aged 7 years old (born in 2000) and 11 male monkeys aged 14–30 d old (born between 2010 and 2011) that were housed born and reared at the Laboratory of Comparative Ethology, National Institute of Child Health and Human Development breeding facility of the Animal Center (Poolesville, MD). The monkeys were randomly divided into two groups at birth, resulting in different early-life social and rearing experience. The “mother-reared” (MR) monkeys were raised by their biological mother in a social group, whereas “surrogate peer-reared” (SPR) monkeys were reared with an inanimate surrogate as well as daily socialization periods with age-matched peers. For the first month of life, the SPR monkeys were placed in a nursery until they were able to drink milk from a bottle by themselves, at which point they were transferred to a cage with their surrogate mother. At ∼7 months of age, the 7-year-old group of animals were socially housed in large, mixed-sex peer groups. Both rearing conditions were described in detail previously (Shannon et al., 1998). All samples were processed and analyzed by experimenters blind to rearing variables. All environmental conditions, procedures, and handling of animals were in strict compliance with the Institutional Animal Care and Use Committee, and all experimental procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals.

T-cell DNA preparation

For the adult analysis, 20 ml of blood was drawn in EDTA-coated tubes and stored at 4°C overnight 2 weeks before necropsy. For the 1-month-old infants, 3 ml of blood was perfused in EDTA-coated tubes. PBMC (whole mononuclear cells from peripheral blood) and T-cell isolation procedures were adapted from previously published protocols (Cooper and Paterson, 1997). Briefly, PBMC isolation was done by centrifugation with Ficoll-Paque (GE Healthcare) and washed twice with HBSS (Invitrogen). T cells were isolated from the PBMCs by immunomagnetic isolation using CD3 Dynabeads (Dynal). The beads were washed three times and incubated with the PBMCs for 45 min on a rotator at 4°C. Coated CD3⁺ cells with the Dynabeads were isolated using a strong magnet (Steam Cell Technology) and washed five times with PBS/FBS. CD3⁺ cells coated with the Dynabeads were then frozen at −80°C until DNA extraction. T-cell DNA was extracted with Wizard Genomic DNA Purification kit (Promega) following the protocol of the manufacturer.

PFC DNA and RNA preparation

Animals were sedated with ketamine and brought to a deep surgical plane of anesthesia with pentobarbital administered intravenously, given to effect. A craniotomy was performed to expose the cerebral cortex and cerebellum, followed by a thoracotomy and perfusion through the left ventricle of the heart for 1.5 min with a chilled, oxygenated buffer solution. The rest of the brainstem was then exposed, and the brain was removed within 5 min. Prefrontal tissue was flash frozen in isopentane at −55°C within 15 min of death and kept at −80°C until DNA/RNA extraction. The tissue consisted mostly of dorsolateral and ventrolateral PFC rostral to the caudal end of the arcuate sulcus, dorsal to the cingulate sulcus and lateral to the lateral orbital sulcus. PFC DNA was extracted using the Qiagen DNeasy kit and RNA extraction using Trysol (Invitrogen), both following the protocol of the manufacturer.

MeDIP and hybridization to microarrays

The MeDIP analysis was adapted from Keshet et al. (2006). Briefly, 2 μg of each of the T cells and PFC DNA were sonicated, and methylated DNA was immunoprecipitated with 10 μg of anti-5-methyl-cytosine (Calbiochem). For the infant samples, the amount of T-cell DNA collected from the original 3 ml of blood was insufficient to perform one MeDIP per sample. Therefore, equal amounts of DNA from the six MR infants (187.5 ng each) and the five SPR infants (300 ng each) were pooled to perform the MeDIP with 1.5 μg DNA of starting materials in each group. Before sonication, two control plasmids were added to the DNA (6 pg each): an unmethylated GFP plasmid and an in vitro methylated Luciferase plasmid. The DNA–antibody complex was immunoprecipitated with 5 mg of protein A, and the methylated DNA was eluted with 150 μl of TE at 1.5% SDS. The input and bound fractions were then purified, and specificity for methylated DNA and absence of nonspecific binding were validated by PCR analysis for two control genes, H19 (methylated control) and GAPDH (unmethylated control), and the two added plasmids, GFP (unmethylated control) and Luciferase (in vitro methylated control) using the following primers: H19, forward, 5′-TTGGTGGAACACGCTGTGATCA-3′; H19 reverse, 5′-GAGCCGCACCAGGTCTTCAG-3′; GAPDH, forward, 5′-TTTCTTTCCTTTCGCGCTCTG-3′; GAPDH reverse, 5′-CCATTCATTTCCTTCCCGGTT-3′; GFP forward, 5′-CAAGGGCGAGGAGCTGTT-3′; GFP reverse, 5′-CGGCCATGATATAGACGTTG-3′; methylated Luciferase forward, 5′-AGAGATACGCCCTGGTTCC-3′; and methylated Luciferase reverse, 5′-CCAACACCGGCATAAAGAA-3′. The input and bound fraction were then amplified in triplicate using the Whole Genome Amplification kit (Sigma). The amplified input and bound fractions were labeled for microarray hybridization with either Cy3–dUTP or Cy5–dUTP (PerkinElmer Life and Analytical Sciences), respectively, using the CGH labeling kit (Invitrogen) following the instructions of the manufacturer.

MeDIP microarray design, hybridization, scanning, and analysis

A custom 244K promoter tiling array design was used for this study (Agilent Technologies). Microarray probe sequences of 45–60 bp were designed using software developed in our laboratory based on published recommendations available at the time of design (May 2008) (Hughes et al., 2001; Matveeva et al., 2003; He et al., 2005; Li et al., 2005; Reilly et al., 2006; Gräf et al., 2007; Sharp et al., 2007). Microarray probe sequences were selected to tile all gene promoter regions defined as the genomic interval from −1000 bp upstream to 300 bp downstream of each transcription start site as defined for the rhesus macaque by the Ensembl database (version 46.34p) (http://www.ensembl.org/) or within 250 bp of a microRNA annotated in miRBase (Griffiths-Jones et al., 2006). All the steps of hybridization, washing, scanning, and feature extraction were performed following the Agilent Technologies protocol for chip-on-chip analysis. Two replicate microarrays were hybridized and analyzed for each DNA sample. After microarray scanning, probe intensities were extracted from scan images using Agilent's Feature Extraction 9.5.3 Image Analysis Software. The extracted intensities were then analyzed using the R software environment for statistical computing (R Development Core Team, 2007). Log-ratios of the bound (Cy5) and input (Cy3) microarray channel intensities were computed for each microarray, and then microarrays were normalized to one another using quantile normalization (Bolstad et al., 2003) under the assumption that all samples have identical overall methylation levels.

List of differentially methylated regions between groups of samples was determined in several stages to ensure both statistical significance and biological relevance. In the first stage, linear models implemented in the “limma” package (Smyth, 2005) of Bioconductor (Gentleman et al., 2004) were used to compute a modified t statistic from the normalized intensities of the probes across all samples between the two groups. An individual probe was called “differentially methylated” if the significance of its t statistic was at most 0.05 (uncorrected for multiple testing) and the associated difference of means between the groups was at least 0.25. For each gene promoter (−1000 to +300 bp of the transcription start site), we calculated the significance of enrichment for high or low probe t statistics of all probes within the promoter (typically 10). Significance was determined using the Wilcoxon's rank-sum test comparing t statistics of these probes against those of all the probes on the microarray. The resulting p values for each gene were then corrected for multiple testing by calculating their FDR. A gene promoter was then called differentially methylated if its FDR was at most 0.2 and one of its probes was called differentially methylated.

The methylation level of a probe or site, when estimated from microarray data, was obtained by applying a Bayesian deconvolution algorithm (Down et al., 2008) to normalized probe intensities and corresponding sequence information indicating the locations of CpG dinucleotides. These estimated methylation levels are shown in Figures 1A and 5A.

Figure 1.

Pyrosequencing, Q-MeDIP, and HRM analyses of differentially methylated sequences mRNA microarrays and qRT-PCR analyses of expression differences in PFC between MR (n = 4) and SPR (n = 4) animals. A, Inverse correlation between PFC promoter methylation and gene expression levels estimated by microarray data (r = −0.24, p < 2E10⁻¹⁶). Genes are divided into 20 levels by expression percentiles (0–5, 5–10, …, 95–100). Shown are the distributions of estimated promoter methylation levels for each expression percentile. The distributions show that genes with low or no expression (represented in green) tend to have highly methylated promoters, whereas genes with high expression (represented in red) tend to have lower promoter methylation. B, Q-MeDIP analysis of DNA methylation differences between the rearing groups in six promoters predicted to be more methylated and four predicted to be less methylated in the SPR animals (gray) by MeDIP microarray analysis. Relative bound fraction concentrations obtained in triplicate by qRT-PCR are shown for the 10 genes (see Materials and Methods). C, DNA methylation differences (percentage) between the rearing groups in the MMP7 and A2D681 genes as determined by pyrosequencing. CpG sites near the transcription start sites were analyzed in triplicate. For each gene, the mean methylation per rearing group per CpG with the SEM are shown in the bar graph. The rightmost bar indicates the mean methylation levels of all CpG sites analyzed in the region. A map of the sites relative to the transcription start site is shown above the bar graph. Each line with a circle represents a CpG site. The location of probes whose fold difference is significantly different between rearing groups is identified by a gray square. Red arrows delimit the region analyzed by pyrosequencing. D, Relative expression levels of MMP7 and A2D681 as determined by qRT-PCR performed in triplicate is shown for each group. E, Correlations (negative) between methylation levels obtained by pyrosequencing for MMP7 CpG4 and A2D681 CpG1 (B) and expression levels obtained by qRT-PCR for both genes (C). MR animals are represented by black dots and SPR by gray dots. Correlation statistics were performed using Pearson's correlation. F, Correlation between methylation levels estimated by HRM and the microarray data for 21 amplicons. HRM methylation levels were estimated after normalization with HRM results obtained for 0 and 100% methylated DNA standards. Each amplicon was analyzed in triplicate per sample. Microarray estimates were obtained by applying a Bayesian deconvolution algorithm to the DNA sequence and microarray data (see Materials and Methods). Correlation statistics were obtained using Pearson's correlation. All error bars represent SEM. ^#p ≤ 0.1 and *p ≤ 0.05 from Mann–Whitney U test.

Figure 5.

Promoter methylation differences between T cells and PFC across the chromosomes of the rhesus macaque genome. A. Overall promoter methylation levels estimated from MeDIP microarray data for T cells (n = 8) and PFC (n = 8) across the entire genome. B. An expanded view from the UCSC genome browser of chromosome 6 is depicted. The average probe fold differences between PFC and T cells estimated from microarray data are shown. In green are promoters whose probes indicate higher methylation in PFC and in red are those more methylated in T cells. The black and gray squares identify significantly differentially methylated regions (DMRs) between PFC and T cells where black indicate higher methylation in PFC and gray in T cells. The highlighted region identifies the protocadherin cluster found to be more methylated in T cells. C, Heat map showing relative methylation levels of the 1.5% most variable promoters across all samples, from both T cells and PFC. Unsupervised clustering coincides with cell type but not with rearing or animal.

In Figures 3D and 4D, plotted values are average probe t statistics of all probes within 10 Mb regions covering the entire genome. The t statistics correspond to differences between rearing groups as described above in PFC (see Fig. 3D) and T cells (see Fig. 4D).

Figure 3.

DNA methylation differences between rearing groups in PFC. A, Bar heights indicate the degree to which each chromosome contains an unexpectedly high number of differentially methylated promoters as calculated by Fisher's exact test followed by adjustment for multiple testing by converting p values to FDRs. Bar heights are −log10 values of the FDRs, so bars higher than the dashed line have FDRs below 0.1. B, Heat map depicts normalized intensities of microarray probes contained in promoters (at most 1 per promoter) that best differentiate between rearing groups in PFC. Rows correspond to promoters and columns to animals. Red indicates higher methylation in a row, and green indicates lower methylation. C, Expanded views from the UCSC genome browser of chromosomes 4 (top) and X (bottom) are depicted. The average MeDIP probe fold differences (Log₂) between MR and SPR groups are shown for PFC. In green are promoters whose probes are hypomethylated, and in red are those that are hypermethylated in the SPR animals. The black and gray squares identify megabase regions significantly differentially methylated (DMRs) between PFC and T cells where black indicates higher methylation in PFC and gray in T cells. Highlighted in yellow are megabase regions significantly differentially methylated between rearing conditions. The histone gene cluster is located in one of the two highlighted regions on chromosome 4, whereas the histone 2B gene family is enriched in one region on chromosome X. D, Methylation differences in PFC between MR and SPR animals as predicted by microarray data across the entire genome are shown. Highlighted in yellow are the chromosomes with the most significant overall decreased methylation in SPR animals compared with MR animals (FDR < 0.01). E, Significantly reduced CpG frequencies in promoters differentially methylated in PFC. The bar labeled “all” refers to all promoters tiled on the microarray, and “SPR < MR” and “SPR > MR” refer to promoters with decreased, and increased, methylation levels in SPR animals compared with MR animals, respectively. Significance was calculated using the Wilcoxon's rank-sum test.

Figure 4.

DNA methylation differences between rearing groups in T cells. A, Bar heights indicate the degree to which each chromosome contains an unexpectedly high number of differentially methylated promoters. This was calculated by Fisher's exact test followed by adjustment for multiple testing by converting p values to FDRs. Bar heights are −log10 values of the FDRs, so bars higher than the dashed line have FDRs below 0.1. B, Heat map depicts normalized intensities of microarray probes contained in promoters (at most 1 per promoter) that best differentiate between rearing groups in T cells. Rows correspond to promoters and columns to animals. Red indicates higher methylation in a row, and green indicates lower methylation. C, An expanded view from the UCSC genome browser of chromosome 19 is depicted. The average MeDIP probe fold differences (Log₂) between MR and SPR groups are shown for T cells. In green are promoters whose probes are hypomethylated, and in red are those that are hypermethylated in the SPR animals. The black and gray squares identify megabase regions significantly differentially methylated (DMRs) between PFC and T cells where black indicates higher methylation in PFC and gray in T cells. Highlighted in yellow are the megabase regions significantly differentially methylated between rearing conditions across the chromosome. Two of the three ZNF clusters located on chromosome 19 are found to be enriched with higher methylation levels in SPR. D. Methylation differences in T cells between MR and SPR animals as predicted by microarray data across the entire genome are shown. Highlighted in yellow are the chromosomes with the most significant overall decreased methylation in SPR animals compared to MR animals (FDR < 0.01). E. Significantly reduced CpG frequencies in promoters more methylated in SPR group in T cells. The bar labeled “all” refers to all promoters tiled on the microarray and “SPR < MR” and “SPR > MR” refer to promoters with decreased, and increased, methylation levels in SPR animals compared to MR animals, respectively. Significance was calculated using the Wilcoxon's rank-sum test.

Figures 3B and 4B depict heat maps of probe intensities. One probe with the most extreme t statistic was selected for each gene promoter called differentially methylated with respect to rearing groups, as described above. Heat maps are colored so the median values on each row are gray, high values are red, and low values are green. Clustering was performed using Ward's hierarchical clustering algorithm with Pearson's correlation distance as the distance metric.

Figures 3C, 4C, 5C, and 6 were obtained using the University of California, Santa Cruz (UCSC) genome browser (http://genome.ucsc.edu/). They depict whole-chromosome views of tracks composed of average normalized probe intensity differences between groups.

Figure 6.

High inter-individual variation in the state of DNA methylation of the promoters of KCTD16 and IL-20 genes in PFC. Graphs represent methylation percentage per sample obtained by pyrosequencing. Five CpG sites located in chr6:140637150-140637400 were analyzed in the KCTD16 promoter (top), and three CpG sites located in chr1:163490500-163490700 were analyzed in the IL-20 promoter (bottom). Error bars represent the SD of three replicates.

Figures 3E and 4E provide bar plots depicting normalized CpG frequencies across all promoters profiled, as well as promoters with increased and decreased methylation levels in the SPR monkeys. The normalized CpG frequency of a sequence is the frequency of CpG sites in the sequence divided by the expected frequency of CpG sites given the GC content of the sequence, i.e., the frequency of G nucleotides times the frequency of C nucleotides. The most common definition of CpG island requires that the DNA sequence have a normalized CpG frequency of at least 0.6.

All functional analysis was based on gene sets obtained from GO (Ashburner et al., 2000), KEGG (Kanehisa and Goto, 2000), and mSigDB (Subramanian et al., 2005) after mapping rhesus macaque genes to human genes using Biomart (http://www.biomart.org/). Functional significance was determined by applying the hypergeometric to the overlap between defined gene sets and differentially methylated genes. FDRs were obtained by adjusting these significance levels over all gene sets and pathways considered.

cDNA microarray hybridization and analysis

Total RNA from the PFC of all eight male monkeys was quantified by NanoDrop Spectrophotometer ND-1000 (NanoDrop Technologies), and the quality was confirmed by 2100 Bioanalyzer (Agilent Technologies). Double-stranded cDNA was synthesized from 250 ng of total RNA, and in vitro transcription was performed to produce biotin-labeled cRNA using Affymetrix Gene Chip 3′ IVT Express reagent kit according to the instructions of the manufacturer (Affymetrix). After fragmentation, 12.5 μg of cRNA was hybridized with GeneChip Rhesus Macaque Genome Array (Affymetrix) containing 47,000 genes. GeneChips were then scanned with the GeneChip Scanner 3000 (Affymetrix).

Microarray probe intensities were normalized to each other using RMA (Irizarry et al., 2008). Expression differences between groups were then obtained by applying linear models implemented in the “limma” package (Smyth, 2005) of Bioconductor (Gentleman et al., 2004) to obtain modified t statistics and corresponding p values. p values were adjusted for multiple testing by converting them to FDRs.

Validation

Quantitative real-time PCR of immunoprecipitated DNA samples.

Gene-specific quantitative real-time PCR validation of microarray (Q-MeDIP) was performed on the amplified-bound fraction for the same subjects used for microarray experiments. Relative enrichment of triplicate reactions were determined as a ratio of the crossing point threshold (Ct) of the amplified specific gene over the Ct of four amplified controls according to the following formula: bound (Ct)/average controls (Ct). We used four control regions for the PFC analysis and one for the T-cell day 14–30 analysis, which were not different between the groups as determined by the microarray analysis and with a small variance between the samples.

High-resolution melting and pyrosequencing of the bisulfited DNA.

The high-resolution melting (HRM) method distinguishes between amplified products of bisulfite-converted methylated DNA containing CG base pairs and unmethylated DNA containing TA base pairs based on differences in melting properties of AT and CG base pairs (Wojdacz et al., 2008). Regions whose normalized intensities were significantly different between rearing groups were chosen for validation (see MeDIP microarray design, hybridization, scanning, and analysis). Primers were designed to generate amplicons <250 bp corresponding to the sequence containing differentially methylated regions in the microarray analysis. PCR amplifications were performed in two steps: 50 ng of bisulfite DNA was first amplified with “outside” primers, and then 2 μl of the “outside” primers amplification reaction mixture was then amplified with nested primers. The outside PCR amplification was performed using TaqDNA polymerase (Fermentas), whereas the nested PCR was performed using hot-start Taq polymerase (Roche).

To obtain controls for the HRM reactions, we amplified the same regions from completely unmethylated (0%) or fully methylated (100%) rhesus DNA as described previously (Borghol et al., 2012). These 0 and 100% and mixture controls were used for each set of primers to calibrate the assay and to control for amplification bias. The PCR and melting analyses were conducted using Roche LightCycler 480 System. The LightCycler 480 Gene scanning software (version 1.5) was used to provide accurate analysis of HRM curves to determine the melting temperature (Bell et al., 2010) for each amplicon. Methylation levels were estimated from HRM curves by obtaining HRM curves for the corresponding 0 and 100% methylated amplicons using the following formula: (Tm amplicon − Tm 0%)/(Tm 100% − Tm 0%). We then determined the correlation between the methylation levels derived from HRM analysis and the values derived from the microarray analysis (Batman scores).

For pyrosequencing analysis, 25 μl of the nested bisulfite PCR products [amplified as above for HRM but, in this case, TaqDNA polymerase (Fermentas) was used in both the outside and nested reactions] were processed according to the standard protocol of the manufacturer (Biotage). Sequencing reactions were performed with a PyroMark Gold Q24 Reagent Kit (Biotage, Qiagen) according to the instructions of the manufacturer. The percentage methylation at each CpG site was calculated from the raw data using PyroMark Q24-CpG Software (Biotage).

For Q-MeDIP validation, 10 probes were tested (9 of 10 positively validated the microarray at p < 0.05 and the other one at p < 0.1). Six genes were analyzed by pyrosequencing. Four of them positively validated the microarray differences between the groups (two at p < 0.05 and two at p < 0.1), and the two other genes positively validated the high inter-individual variation predicted by the microarray but not the differences observed between the groups at p < 0.05. For HRM, 21 probes were tested (10 of 21 validated the microarray differences between rearing condition at p < 0.05, and all of them validated the overall methylation levels estimated by the microarray). The lower success rate with HRM may be attributable to the fact that HRM is not sufficiently accurate to detect small methylation differences.

Quantitative RT-PCR

cDNA synthesis was performed using random hexamer primers (Invitrogen) according to the instructions of the manufacturer. Ribosomal protein 13A (Rpl13a) was used as the reference gene because it has been shown to have consistent expression in different tissues in rhesus macaques (Ahn et al., 2008). SYBR green quantitative RT-PCR (qRT-PCR) was performed using the LightCycler 480 system (Software 3.5; Roche Molecular Biochemicals). The presence of a single melting peak followed by analysis on 1.5% agarose gel confirmed product specificity. Reactions were performed in triplicate. As a control for contaminating DNA, reactions were also performed in the absence of reverse transcriptase. To determine the relative concentration of mRNA expression, we used a standard curve of 10-fold serial dilutions of a control cDNA .The concentration of the test mRNA was divided by the concentration of Rpl13a to derive the normalized concentration of the mRNA per subject, and an average normalized concentration was calculated for the two rearing groups. Mediation analysis was used to determine whether or not there was evidence that differences in DNA methylation between the rearing groups obtained by pyrosequencing mediated the effect of rearing on gene expression levels. The mediate package (Imai et al., 2010) in the R statistical environment (R Development Core Team, 2007) was used for this analysis with default settings. Hence, percentage DNA methylation per region was used as the mediator variable and expression levels as the dependent variable.

Rationale for the MeDIP microarray hybridization approach

MeDIP was selected among many other methods for methylation profiling because it is one of the few methods that is feasible for studying genome-wide methylation differences between groups of subjects, it has been successfully applied in many published studies (Feber et al.; Weber et al., 2005; Keshet et al., 2006; Novak et al., 2006; Zhang, 2006; Cheng et al., 2008; Down et al., 2008; Tomazou et al., 2008; Liu et al., 2009; Murphy et al., 2009; Takeshima et al., 2009; Bell et al., 2010; Cheung et al., 2010; Flanagan et al., 2010; Guerrero-Bosagna et al., 2010; Günther and Grundhoff, 2010; Hiura et al., 2010; Movassagh et al., 2010; Tsui et al., 2010; Lempiäinen et al., 2011; Morris et al., 2011), and it has been found to be competitive with the other high-throughput profiling methods that are in use (Irizarry et al., 2008; Jia et al., 2010; Jin et al., 2010; Robinson et al., 2010; Nair et al., 2011; Rajendram et al., 2011; Yang et al., 2011). Moreover, MeDIP detects DNA methylation exclusively from hydroxymethylation and is not confounded by chemical conversion and biased amplification of converted sequences. Indeed, the MeDIP data in our study lived up to expectations. Technical replicates have higher association with one another than with those from other samples, such that according to the eigenR2 algorithm (Chen and Storey, 2008), differences between samples account for 60% of the variation in the data. Interestingly, eigenR2 also shows that rearing conditions explain 8% of the variance, more than any other division of the samples into equal-sized groups. Using a Bayesian deconvolution method to estimate methylation levels from the data (Down et al., 2008), we found promoter methylation levels to be significantly anticorrelated with mRNA expression levels obtained from cDNA microarrays (r = −0.24, p < 2 × 10⁻¹⁶; see Fig. 1A). We also found that these DNA methylation levels significantly correlated with methylation levels obtained using bisulfate-based methods (see Fig. 1E).

Results

We created and analyzed genome-wide promoter methylation profiles of PFC and T-cell DNA from eight adult, male rhesus macaques. Four of the animals were raised in nurseries without mothers (SPR), and the other four were controls raised by their biological mothers (MR). Methylation profiles were created using the method MeDIP with microarray hybridization (for rationale for using MeDIP and microarrays, see Materials and Methods).

Microarray validation

Microarray data was validated using four approaches: (1) comparison with gene expression profiles to see whether there was an overall inverse correlation between promoter methylation and gene expression levels, (2) Q-MeDIP of the bound fraction to validate the MeDIP enrichment, (3) HRM to estimate methylation levels from regions covering a few hundred bases, and (4) pyrosequencing to precisely determine methylation levels of individual CpG sites.

qRT-PCR was applied on the bound fraction of the MeDIP to determine the enrichment levels of 10 promoters called differentially methylated in PFC. Relative enrichment of 10 regions were determined as a ratio of the Ct of the amplified specific gene over the Ct of four amplified controls (see Materials and Methods). As expected, the normalized bound fraction shows significant enrichment or depletion in the SPR group as predicted from the microarray data (Fig. 1B).

Pyrosequencing was applied to CpG sites upstream of MMP7 and A2D681 genes, two regions found enriched by the array analyzed in the PFC of the SPR group (Fig. 1C). As expected, two CpG sites for MMP7 and one for A2D681 were found significantly more methylated in the SPR monkeys as well as the average methylation of the four CpGs analyzed for MMP7. mRNA expression of these genes in PFC were validated by qRT-PCR (Fig. 1D). As expected, average group expression differences are anticorrelated with average group methylation differences, although the difference is only statistically significant for MMP7. We used mediation analysis to determine whether there was evidence that DNA methylation of these specific CG sites mediated the effect of rearing on gene expression. We found evidence for mediation involving CG4 of MMP7 (p = 0.06; mediation effect = −0.4 with 95% CI = −0.8 to 0.0) and for CG3 of NR3C1 (p = 0.04; mediation effect = 4.5 with 95% CI = 0.6 to 9.7) (for expression and methylation levels for MMP7 at CpG site 4 and A2D681 at CpG site 3, see Fig. 1E). Both PCR sequences were analyzed for possible single nucleotide polymorphisms (SNPs) that could affect the results of the pyrosequencing analysis. For example, a sequence change of a C to a T could be interpreted as DNA demethylation if only bisulfate-converted DNA is analyzed. We therefore subjected unconverted DNA to sequencing. Our results show that the amplicons contained no SNPs that could have affected the analysis of pyrosequenced bisulfate-converted DNA (Fig. 2). We found five SNPs in the MMP7 gene promoter and five SNPs in the A2D681 gene promoter. None of the SNPs identified were associated with either rearing or methylation of the CpG sites analyzed for both regions. MMP7 encodes a matrix metalloproteinase (MMP) involved in the breakdown of extracellular matrix in normal physiological processes, such as embryonic development, reproduction, and tissue remodeling, as well as in disease processes, such as arthritis and cancer. MMPs are also expressed in the CNS and are known to play a role in normal brain development and physiology, as well as in several diseases of the nervous system, including Alzheimer's disease, HIV dementia, and multiple sclerosis (Yong et al., 2001; Amălinei et al., 2010).

Figure 2.

DNA sequences of MMP7 (A) and A2D681 (B) gene promoters in the eight monkeys. Each line represents the sequence for an individual animal. The primers used for pyrosequencing are underlined with an arrow representing their 5′-3′ direction. The dashed line indicates 89 bases that were not sequenced between two amplicons.

To follow up on the gene expression differences identified by PCR in MMP7 and A2D681, we hybridized PFC mRNA from each of the monkeys to gene expression microarrays. As shown in Figure 1A, overall expression levels were inversely correlated with promoter methylation levels. Unfortunately, however, no differentially expressed gene was found to be significantly associated with rearing. All FDRs were well above the commonly used threshold of 0.2, and no gene with an association p value <0.05 had a fold change >2. The fact that expression differences between the rearing groups were found by qRT-PCR but not by microarray suggests that our arrays were not sufficiently sensitive to detect the expression differences that were detected by qRT-PCR between the rearing groups. One possibility is that microarrays have a noisier signal than PCR. These data suggest that DNA methylation arrays provide a more sensitive readout of programming differences between the differential rearing groups than transcription arrays. This has implications on the future validation of behaviorally relevant human biomarkers. Future studies investigating gene expression differences will either need to include larger numbers of monkeys or use more sensitive technologies, such as deep RNA sequencing to such differences at a genome-wide scale.

Finally, HRM was applied to 21 regions called differentially methylated in PFC. We estimated methylation levels from HRM results by interpolation with HRM results from 0 and 100% methylated DNA (see Materials and Methods). These HRM-derived methylation levels correlated significantly with methylation levels estimated from the microarray data using a Bayesian deconvolution algorithm (Down et al., 2008) (Fig. 1F; r = 0.57, p = 0.01).

Rearing-associated promoter methylation in PFC

We tested whether early-life rearing conditions would have an impact on the methylome of PFC. PFC is one of the brain tissues known to sustain long-term effects of early rearing conditions, e.g., aggressive behavior (Miczek et al., 2007) and addictive behavior (Uekermann and Daum, 2008). Both behaviors are enhanced in the SPR group (Harlow and Harlow, 1965; Suomi et al., 1976; Higley et al., 1991). Moreover, it is well known that PFC is an important modulator of the hypothalamic–pituitary–adrenal (HPA) axis and stress response (Dedovic et al., 2009).

DNA methylation differences identified in the PFC MeDIP microarray data between rearing groups were found distributed across the genome (Fig. 3A) with some enrichment inside a few chromosomes (chromosomes 4, 5, 12, and 18; p < 0.01; hypergeometric). In all, we found 1981 probes from 1373 distinct gene promoters whose normalized intensities were significantly associated with rearing conditions (for criteria, including multiple testing correction, see Materials and Methods, MeDIP microarray design, hybridization, scanning, and analysis). Of these promoters, 538 were more methylated in the control group, and 835 were more methylated in the SPR group. This differentially methylated list of promoters included four microRNA sites more methylated in SPR (mml-mir-125b-1, 125b-2, 194-1, and 30b) and two more methylated in MR (mml-mir-135a-2 and 153-2) of the 102 included on the microarray. A heat map of the clustering analysis of the probes that best differentiate between rearing groups is shown in Figure 3B. The top 50 significant gene promoters differentially methylated between the groups are shown in Table 1. We further confirmed differential methylation for several of these probes by Q-MeDIP, pyrosequencing, and HRM curve analysis described in Materials and Methods (Fig. 1).

Table 1.

Top 50 gene promoters differentially methylated between MR and SPR animals in PFC

Ensembl gene ID	Gene name	More methylated in	p value
ENSMMUG00000009091	MTTP	SPR	8E−05
ENSMMUG00000008428	APEX2	SPR	8E−05
ENSMMUG00000010174	INTS7	SPR	9E−05
ENSMMUG00000030219	DUS4L	SPR	1E−04
ENSMMUG00000005010	JTB	SPR	1E−04
ENSMMUG00000030452		MR	1E−04
ENSMMUG00000011810		MR	2E−04
ENSMMUG00000006481	GNS	SPR	2E−04
ENSMMUG00000011949	ACTR1A	SPR	2E−04
ENSMMUG00000004396	SUFU	SPR	2E−04
ENSMMUG00000005718	HYAL4	MR	2E−04
ENSMMUG00000030058		MR	3E−04
ENSMMUG00000022770	SERINC4	SPR	3E−04
ENSMMUG00000029692		SPR	3E−04
ENSMMUG00000008419	GNL2	SPR	3E−04
ENSMMUG00000015704	ZBTB7C	MR	4E−04
ENSMMUG00000029791	KCTD16	SPR	4E−04
ENSMMUG00000009460	OBFC2A	SPR	4E−04
ENSMMUG00000017211	PPAN	SPR	5E−04
ENSMMUG00000013604	POF1B	MR	5E−04
ENSMMUG00000015744	ZNF302	SPR	5E−04
ENSMMUG00000010963	MTMR9	MR	5E−04
ENSMMUG00000014393		MR	5E−04
ENSMMUG00000000209		SPR	5E−04
ENSMMUG00000012492		MR	5E−04
ENSMMUG00000003028		SPR	6E−04
ENSMMUG00000023248	STK38L	SPR	6E−04
ENSMMUG00000016535	LRRIQ1	SPR	6E−04
ENSMMUG00000022408	ST6GALNAC1	MR	7E−04
ENSMMUG00000023588	CCR9	MR	7E−04
ENSMMUG00000005833	CLEC9A	MR	7E−04
ENSMMUG00000012634	GTF2F2	SPR	7E−04
ENSMMUG00000019117	DNAJB7	MR, SPR	7E−04
ENSMMUG00000023675	HYI	SPR	7E−04
ENSMMUG00000013897	NINJ2	MR	8E−04
ENSMMUG00000001697	C2orf40	SPR	8E−04
ENSMMUG00000015996	LOC698165	SPR	8E−04
ENSMMUG00000010860	FAM179B	SPR	8E−04
ENSMMUG00000017678	KLHL28	SPR	8E−04
ENSMMUG00000022277	HOXD3	SPR	8E−04
ENSMMUG00000003826	ZNF780A	SPR	8E−04
ENSMMUG00000007234	ATP11C	SPR	8E−04
ENSMMUG00000021576	ROBO3	SPR	8E−04
ENSMMUG00000003855	ZNF385B	MR	9E−04
ENSMMUG00000013356	C18orf26	SPR	9E−04
ENSMMUG00000011252	EFHA1	MR	9E−04
ENSMMUG00000013993	HCCS	SPR	9E−04
ENSMMUG00000004154	TCEAL5	SPR	1E−03
ENSMMUG00000005816	CYB561D1	SPR	1E−03
ENSMMUG00000005834		MR	1E−03

Despite the fact that differences in DNA methylation between rearing conditions were distributed uniformly across the chromosomes, probes with similar differences tended to appear in clusters within chromosomes. To measure the strength of this clustering, we partitioned the genome into 500 kb regions and asked whether each region contained a surprisingly high number of differentially methylated probes. Of the ∼6000 regions, we found that 55 were significantly enriched in differentially methylated probes (FDR < 0.01 for each region). Repeated permutation of probe location 100 times failed to produce even one single significantly enriched region (see Materials and Methods). Figure 3C depicts two such regions, including a histone cluster with lower methylation in the SPR group relative to the MR group, and a H2B histone cluster containing higher methylation in the same group.

We subjected gene promoters containing probes associated with rearing in PFC to enrichment analysis with respect to various gene set databases including GO (Ashburner et al., 2000), KEGG (Kanehisa and Goto, 2000), and mSigDB (Subramanian et al., 2005). Pathways associated with nervous system development, immune function, RNA metabolism, and response to stimulus were most prominently enriched, but axonogenesis, leukocyte differentiation, regulation of transcription, nucleosome assembly, and response to stress were also significantly enriched.

We asked whether DNA sequence properties were predictive of changes in DNA methylation. Indeed, CpG frequencies were unexpectedly low in differentially methylated promoters (p < 10⁻¹⁵, Wilcoxon's rank-sum test; Fig. 3E); however, CpG frequency was not predictive of methylation increases or decreases.

Rearing-associated promoter methylation in T cells

We tested whether early-life rearing conditions would have an impact on the methylome beyond the brain. Several lines of evidence predict bilateral crosstalk between the brain and immune system, especially pertaining to the HPA axis and stress response. Blood cells play an important role in controlling the immune response to the bioenvironment, and previous studies have shown that whole blood cell (WBC) DNA methylation is associated with environmental exposure (Kinnally et al., 2011; Uddin et al., 2010, 2011). Because WBCs comprise heterogeneous cell populations that could mask cell-type-specific methylation differences, we isolated and examined CD3⁺ T cells because they have been shown previously to be involved in the stress response (Stiller et al., 2011) and the HPA axis (Viveros-Paredes et al., 2006). We note that PFC is also composed of a heterogeneous cell population. However, given the small amount of PFC sample available for the study, we were unable to get sufficient amount of DNA from an isolated specific cell type; we therefore used whole PFC tissue.

As in PFC, methylation differences between rearing conditions were scattered throughout the genome (Fig. 4A) with some enrichment in chromosomes 3 and 18 (FDR < 0.1; hypergeometric). In all, we found 227 probes in 138 different gene promoters for which normalized intensities associated with rearing conditions. Of these promoters, 46 were more methylated in the control group, and 92 were more methylated in the peer-reared group. Of the 102 microRNAs tiled on the microarray, only one microRNA site was found more methylated in SPR, mml-mir-34a. A heat map of the clustering of the probes that best differentiate between MR and SPR groups is shown in Figure 4B. The top 50 significant gene promoters differentially methylated between the groups are shown in Table 2.

Table 2.

Top 50 gene promoters differentially methylated between MR and SPR animals in T cells

Ensembl gene ID	Gene name	More methylated in	p value
ENSMMUG00000028813	ZNF724P	SPR	5E−05
ENSMMUG00000020902	NP_001028108.1	MR	1E−04
ENSMMUG00000016191	ABCA13	SPR	4E−04
ENSMMUG00000031177	XM_001089243.1	SPR	5E−04
ENSMMUG00000008650	TPK1	SPR	7E−04
ENSMMUG00000016190	TAF4B	SPR	7E−04
ENSMMUG00000014330	ORC1L	SPR	1E−03
ENSMMUG00000012841	KLK3_MACMU	SPR	2E−03
ENSMMUG00000009295	RIBC1	SPR	3E−03
ENSMMUG00000017029	CPB2	SPR	3E−03
ENSMMUG00000011584	CHD7	SPR	3E−03
ENSMMUG00000012144	LOC713885	SPR	3E−03
ENSMMUG00000015035		SPR	4E−03
ENSMMUG00000015031	EDA2R	MR	4E−03
ENSMMUG00000006755	TTLL4	SPR	4E−03
ENSMMUG00000021106	XM_001098336.1	SPR	5E−03
ENSMMUG00000006526	C6orf134	SPR	5E−03
ENSMMUG00000000222	GLHA_MACMU	SPR	6E−03
ENSMMUG00000018325	GPR3	MR	7E−03
ENSMMUG00000006164	LOC693771	SPR	7E−03
ENSMMUG00000000717	CD3EAP	MR	8E−03
ENSMMUG00000004974	TRIM21	SPR	8E−03
ENSMMUG00000007507	CLCN2	MR	8E−03
ENSMMUG00000030267		SPR	8E−03
ENSMMUG00000031665		SPR	8E−03
ENSMMUG00000012241	C4H6orf182	MR	8E−03
ENSMMUG00000022516	AEN	SPR	9E−03
ENSMMUG00000006983		SPR	9E−03
ENSMMUG00000006850	LYZL6	SPR	9E−03
ENSMMUG00000021231	MYO16	SPR	9E−03
ENSMMUG00000008726	NACAP1	SPR	9E−03
ENSMMUG00000031881		SPR	9E−03
ENSMMUG00000018831	SLC24A5	SPR	1E−02
ENSMMUG00000018273	PARP11	MR	1E−02
ENSMMUG00000019875	KIAA0391	MR	1E−02
ENSMMUG00000021743		SPR	1E−02
ENSMMUG00000021653	AARS	SPR	1E−02
ENSMMUG00000031216		SPR	1E−02
ENSMMUG00000017283	DEFB129	SPR	1E−02
ENSMMUG00000026899	mml-mir-34a	SPR	1E−02
ENSMMUG00000031065	XM_001094582.1	SPR	1E−02
ENSMMUG00000010999	HDHD2	SPR	1E−02
ENSMMUG00000018217	KRT84	SPR	1E−02
ENSMMUG00000016803	SCRN1	SPR	1E−02
ENSMMUG00000029879		SPR	1E−02
ENSMMUG00000009494	CPA5	SPR	1E−02
ENSMMUG00000000925	HXC10_MACMU	MR	1E−02
ENSMMUG00000030675		SPR	1E−02
ENSMMUG00000004292	FAM180B	MR	1E−02
ENSMMUG00000012637	XM_001082471.1	SPR	2E−02

Given the dramatically smaller number of differences found in T cells compared with PFC, it was surprising to find that there was even more clustering of probes with similar differences between rearing conditions within chromosomes. In T cells, 68 regions each covering 500 kb were found enriched (FDR < 0.01). Again, permutation analysis repeated 100 times did not produce a scenario with even a single region significant clustering (see Materials and Methods). Two enriched regions are shown in Figure 4C corresponding to two of the three zinc finger protein (ZNF) clusters located on chromosome 19. Both regions have higher methylation in the SPR group than in the MR group. Such methylation differences would be expected to have an important effect on genome function in T cells because genes in the ZNF family typically act as transcription factors and bind primarily to regulatory genomic regions. Pathway analysis of the differentially methylated genes in T cells identified pathways associated with transcription, fatty acid oxidation, and immune response most prominently, but proteolysis, chromatin modification, and response to stimulus were also enriched with genes having differentially methylated promoters.

As for methylation differences in PFC, we asked whether CpG frequencies were related in any way to DNA methylation. Indeed, we found that CpG frequencies were unexpectedly low in promoters more methylated in SPR compared with promoters less methylated in SPR (p < 10⁻¹⁵, Wilcoxon's rank-sum test; Fig. 4E).

Methylation differences between T cells and PFC

The methylation differences described so far suggest that the response to rearing is quite different in T cells compared with PFC. To further compare the responses to rearing in these two cell types, we first define the background against which these different responses occur. Importantly, within our sample population, most of the methylation variation is explained by differences between cell types. In particular, cell-type differences are detected by ∼15% of the probes on the promoter-tiling microarray. Plotting methylation estimates across the genome for each cell type shows that, in general, T cells have more promoter methylation than PFC cells (Fig. 5A). The number and strength of these differences is illustrated in Figure 5B where the 1.5% of probes having the most variation across the samples clusters the samples perfectly by cell type. These observations are consistent with numerous other studies (Lister et al., 2009; Yuen et al., 2011), including studies of other species (Razin and Szyf, 1984). The functions of the genes associated with the affected promoters are also consistent with the cell types compared, T cells (immune), and PFC (brain). For example, differences were highly enriched in the promoters of genes involved in neurotransmission (p = 0.0035, hypergeometric test), immune function (p = 1.6E10⁻¹³, hypergeometric test), and Ig function (p = 0.0036, hypergeometric test).

We then removed probes with the strongest associations with cell type (p < 0.05, modified t test) and then clustered by the remaining 1.5% most variable probes to see whether these probes clustered according to some other variable. Surprisingly, even these probes clustered the samples by tissue type with only three errors. We observe both large as well as small differences in DNA methylation between T cells and PFC. Large changes in DNA methylation are consistent with classical thinking that DNA methylation is involved in on/off switching of gene expression during cellular differentiation. The fact that we observe changes in DNA methylation that are small suggests that important cellular processes, such as differentiation, could involve not only large but also small changes in DNA methylation. Small changes in DNA methylation imply that only a small fraction of the cells within the sample change their DNA methylation state. The rearing-associated methylation differences are in most cases of small amplitude, but nevertheless they might play significant roles in organismal development, as well.

Despite the relative high frequency of cell-type methylation differences compared with rearing differences, cell-type differences exhibit the same tendency to form clusters in the genome. In particular, given the same partitioning of the genome into 500 kb regions used to identify clusters among the rearing methylation differences, we identified 691 regions that contained an enrichment of probes associated with cell-type differences, a significantly large number because repeating the analysis on 100 random permutations of the probe locations produced only a single enriched region. Figure 5B highlights one such region containing the protocadherin family of genes with higher methylation in T cells than in PFC. Protocadherins are a superfamily of genes that encode proteins involved in cell–cell adhesion, communication, and synaptogenesis and might have specific neuronal function that is consistent with their demethylation in PFC. These results show that, despite the fact that rearing and cell-type methylation differences arise in response to very different cellular processes, the responses exhibit evidence both of a high-level organization throughout the genome as well as having a modulating effect on cell function rather than an extreme on/off effect.

As expected, there are inter-individual methylation differences between samples that could not be explained by neither rearing nor cell type, and some of these were quite large. We confirmed two examples in PFC of such variation by pyrosequencing (Fig. 6). In the KCTD16 gene promoter, methylation levels at five CpG sites were extremely variable, with differences between individuals as high as 65%. Similar variation was also observed in the IL-20 gene promoter for one CpG site (largest difference of 68%), whereas two other nearby CpG sites have much smaller differences (∼20%). Although it is perhaps surprising to see such large variation in the same tissue among monkeys raised in similar environments, others have reported similar inter-individual variations (Elango et al., 2009; Farcas et al., 2009).

Tissue-specific methylation and response to rearing

In PFC, a significantly large proportion of the sites (25%) differentially methylated between rearing groups coincide with sites that are also differentially methylated between T cells and PFC (p < 2E10⁻¹⁶, hypergeometric). This large overlap suggests a functional connection between tissue-specific methylation and response to rearing. Providing additional evidence for this connection, we found an overall significant association between differential probe statistics corresponding to tissue differences and rearing differences in PFC (r = −0.15, 95% CI = −0.147 to −0.16). In other words, the connection between tissue-specific methylation and rearing-associated methylation can be seen not only within sets of statistically significant methylation differences but also throughout genome-wide methylation profiles. Interestingly, this connection was not seen in the T cells because neither of the above tests applied to T-cell methylation differences was statistically significant (at p = 0.05).

Overlapping rearing-associated methylation in T cells and PFC

Despite the extremely different responses in T cells and PFC to rearing, there were some similarities. At the level of the whole genome, there was a weak but significant correlation between rearing-associated differential probe statistics in PFC and T cells (r = 0.069, 95% CI = 0.065–0.73). At the chromosomal level, we identified chromosomes in each cell type with significantly increased or decreased methylation in the SPR animals compared with MR animals. Surprisingly, the chromosomes with the most significant results were found to have lower methylation in SPR animals, and the set of chromosomes was almost identical in each cell type. In PFC, the chromosomes were 16, 20, and X, and in T cells, the chromosomes were 15, 16, and 20 (Figs. 3D, 4D, highlighted in yellow; FDR < 0.01 for each chromosome). Figures 3D and 4D suggest that chromosome X is also less methylated in SPR animals in both cell types but at a somewhat lower level of significance in T cells (FDR < 0.2). Within chromosomes, we noted above that 55 and 68 regions each covering 500 kb were enriched with rearing-associated methylation differences in PFC and T cells, respectively. Of these, 10 are enriched with differences in both tissues, a statistically significant overlap (p < 7.4E10⁻⁸, hypergeometric). One such region contains a cluster of olfactory receptor genes on chromosome 14 that is generally more methylated in the SPR animals in both tissues (Fig. 7; significance of enrichment in each cell type and between cell types was at FDR < 0.01). At the promoter level, 16 gene promoters contain probes significantly different between rearing groups in both T cells and PFC (p < 0.08; Table 3), eight of which are more methylated in the SPR group in both T cells and PFC. Finally, at the individual probe level, at least five probes identified significant methylation differences between rearing groups in both tissues (Table 3, gray). Such an overlap is significantly larger than expected by chance (p < 0.044, hypergeometric test). Of note is a site upstream of A2D681, the homolog of NR3C1 in human, which is more methylated in SPR in both tissues. This gene encodes the glucocorticoid receptor protein. This differentially methylated site aligns in the human genome to a location just upstream of exon 1D. In rats, exon 1F of this gene has higher methylation in pups receiving low maternal care (Weaver et al., 2004). Similar pathways in both tissues were also found to be enriched with genes differentially methylated between the groups, such as the immune response, transcription, and response to stimulus (Table 4).

Figure 7.

Hypermethylation of olfactory receptor genes in SPR animals in both PFC and T cells. An expanded view from the UCSC genome browser of the olfactory receptor cluster located on chromosome 14 is depicted. The average MeDIP probe fold differences (Log₂) between MR (n = 4) and SPR (n = 4) are shown for T cells (top) and PFC (bottom). In green are promoters whose probes indicate lower methylation, and in red are those more methylated in the SPR animals. Highlighted in yellow are megabase large regions significantly differentially methylated (DMRs) between MR and SPR.

Table 3.

Gene promoters containing probes predicting differential methylation between rearing conditions in both PFC and T cells

Location	Closest gene	Description	Distance from TSS (bp)		Log2 (SPR/MR)		p value		q value
			PFC	T cells	PFC	T cells	PFC	T cells	PFC	T cells
Chr2:94462881-94463166	TRAK1a	Similar to the Homo sapiens TRAK1 (trafficking protein, kinesin binding 1), ENSMMUG00000031665	−106	−106	0.7	0.8	0.04	0.04	0.14	0.192
				−333		1.9		8E−3
Chr6:139843104-139843160	A2D681_MACMU	NR3C1 fragment (glucocorticoids receptor isoform 3), ENSMMUG00000000421	−2358	−2358	1.6	2.7	7E−3	7E−4	0.181	0.239
Chr8:110846732-110846884	TTC35	Tetratricopeptide repeat domain 35, ENSMMUG00000019260	−549	−549	2,0	−1.3	0.02	0.03	0.131	0.242
			−657		0.7		0.02
Chr19:22258146-22258769	ZNF724P	Zinc finger protein 724 (pseudogene), ENSMMUG00000028813	−320	−146	−2.1	2.5	0.03	0.05	0.186	0.192
			−433	−433	−1.3	1.8	0.03	0.03
			−716	−542	−1.9	3.3	9E−3	5E−5
Chr20:27727928-27728003	ZG16	Zymogen granule protein 16 homolog, ENSMMUG00000009686	−807	−807	−1.7	1.2	3E−3	0.05	0.232	0.243
			−882		−1.1		0.04
Chr1:12202352-12203602	mml-mir34A	mml-mir-34a, ENSMMUG00000026899	483	−678	1.8	1.2	0.02	0.01	0.174	0.242
Chr1:55240406-55240569	PRPF38A	PRP38 pre-mRNA processing factor 38 domain containing A, ENSMMUG00000014333	251	355	1.4	1.2	0.03	1E−3	0.234	0.192
Chr2:100700723-100701381	Q8WMQ0_MACMU	Transformer-2β fragment (TRA2B), ENSMMUG00000001185	−873	−262	0.7	−1.6	0.04	0.05	0.229	0.233
Chr3:182078245-182079046	TPK1	Thiamin pyrophosphokinase 1, ENSMMUG00000008650	−938	−183	1.1	2.3	0.04	7E−4	0.17	0.243
Chr4:154793465-154794734	C4H6orf182b	cep57-related protein-like, ENSMMUG00000012241	−1397	−183	2,0	−1.6	0.02	8E−3	0.213	0.232
Chr6:103809444-103810021	ENFB2a,c	Similar to the Homo sapiens ENFB2 (EF hand domain family, member B isoform 2), ENSMMUG00000029879	−783	−418	1,0	1.8	0.04	0.01	0.242	0.207
Chr15:13151901-13152038	OR1B1	Olfactory receptor, family 1, subfamily B, member 1, ENSMMUG00000012074	−115	−201	1.2	1.8	0.01	0.02	0.104	0.245
Chr18:19238404-19238865	TAF4B	TAF4b RNA polymerase II, TATA box binding protein (TBP)-associated factor, ENSMMUG00000016190	238	−23	−1.8	3,0	0.01	0.01	0.242	0.242
Chr18:56967360-56967855	SERPINB13	Serpin peptidase inhibitor, clade B (ovalbumin), member 13, ENSMMUG00000004565	−685	−325	−0.8	1.3	0.01	0.02	0.188	0.239
Chr13:126497622-126498034	RALB	V-ral simian leukemia viral oncogene homolog B, ENSMMUG00000000462	−184	−552	−1.6	0.8	0.02	0.05	0.048	0.2
Chr19:50829494-50830008	ZNF229a	Similar to the Homo sapiens ZNF229 (zinc finger protein 229), ENSMMUG00000012810	−199	201	2.9	1.8	0.01	0.09	0.218	0.272

Highlighted in bold are promoters in which the same probe in both tissues was found to be differentially methylated between rearing condition. In the other gene promoters listed, the specific differentially methylated probes within the promoters are different between T cells and PFC.

^aNucleotide alignment was performed using BLASTN 2.20.23 with >93% sequence match to human gene name.

^bThe probes found to be differently enriched are also located upstream of the transcription start site of Sestrin 1 (p53-regulated protein PA26, ENSMMUG00000018232).

^cThe probes found to be differently enriched are also located in intron 1 of ENFA5 gene (ephrin-A5, ENSMMUG00000018398).

Table 4.

Common functional categories enriched with rearing condition differentially methylated genes in both PFC and T cells

Pathways and categories	Fold enrichment	p value	Genes name more or less methylated in SPR
PFC
Immune response	1.62	2.0E−03	C8A, CCR9, CD8A, CIITA, CSF3, DMBT1, FCGR3A, ICOS, IFI35, IFNB1, IL15, IL1A, IL20, IL23R, IL31RA, IL7, IRF7, LTB, SEMA4D, SH2D1A, ZAP70, CCL11, CD55, CXCL3, DEFB127, HLA-DOB, HLA-DQB1, IFIT1, IL2, IL8, MASP1, MNDA, PGLYRP3, RBM4, SPN, TBK1, TCF7, TNFSF12
Transcription from RNA polymerase II promoter	1.82	1.0E−04	FOXO4, GLI2, IRF7, MNAT1, MTA2, MYEF2, MYF5, POLR2A, RUNX1, TAF7, ASCL1, BLZF1, BRD8, CCNT1, ETV3, GTF2E1, GTF2F2, GTF2I, HOXC6, JAZF1, MAML2, MITF, MLL, MSC, MTF1, NFYC, PPARGC1A, SMAD1, SUFU, TAF5L, TARBP1, TCF7, TFAP2A, TRAK1, ZBTB32, ZNF238
Response to stimulus	1.66	1.0E−04	ANXA5, AOC3, C8A, CYBB, DUSP10, ERCC1, FABP4, FBLN5, FGB, FGF7, FOXN3, FOXO4, GP5, IL1A, IL20, IL23R, IL25, IRF7, NINJ2, RUNX1, ADRB1, AHSA2, CCL11, CD55, CMA1, CXCL3, EDN1, EIF2AK3, EIF2S1, HSP90AA1, HSPB7, HSPBAP1, IL8 KPTN, MAPKAPK5, MASP1, MTF1, PDIA5, PPARGC1A, RBM4, SERP1, SMAD1, SPN, TAOK2, TP53I11, TXNDC4, VPS45, WDR33
T cells
Transcription	2.30	1.0E−03	POLR2B, EDA2R, NFKB2, HOXC10, TBL1XR1, CD3EAP, SSRP1, CHD7, ZNF724P, TAF4B, ZNF236, TAF6L, ZNF569, APBB1, ABCA13, ZBTB39, A2D681
Immune response	3.16	1.0E−02	CXCL9, CD3EAP, IL16, PRG2, IGLL1, PDCD1LG2
Response to stimulus	1.83	2.0E−02	IFNGR2, CXCL9, CD3EAP, IL16, SSRP1, PRG2, SLC24A5, IGLL1, CNGA2, PDCD1LG2, OR1B1, A2D681, DEFB129, SERPINB13

Adult rearing-associated promoter methylation is also observed in 14- to 30-d-old monkey infants

To determine whether the changes in DNA methylation associated with rearing conditions in adult monkeys are seen early in life, we used preliminary data on methylation analysis in T cells of 14- to 30-d-old male monkeys (n = 11) from an ongoing longitudinal study in the laboratory. We confirmed by Q-MeDIP that some of the differences associated with rearing in adult monkeys are also observed in 14- to 30-d-old infants with differential rearing, both MR (n = 6) and nursery reared (n = 5). More specifically, we identified nine gene promoters that, by microarray, were differentially methylated between rearing groups in the T cells of both the infants and the adults. We confirmed by qRT-PCR of the MeDIP DNA that these promoters were differentially methylated between infant rearing groups (Fig. 8), and in six of these promoters, the direction of differential methylation was the same as found in the adult animals.

Figure 8.

Q-MeDIP analysis in day 14–30 infant T cells of nine promoters that are differentially methylated in 7-year-old monkeys between MR (n = 5) and nursery-reared (NR)/SPR (n = 6) groups. DNA methylation differences between the rearing groups in T-cell 14- to 30-d-old infants analyzed by Q-MeDIP are shown. Relative bound fraction concentrations obtained in triplicate by Q-PCR are shown for nine genes (see Materials and Methods); five of these gene promoters were more methylated, and four were less methylated in the 7-year-old SPR animals as determined by the microarray analysis. Black bars represent the MR infants and gray bars the nursery reared. All error bars represent SEM. ^#p ≤ 0.1 and *p ≤ 0.05 determined by Mann–Whitney U test.

Discussion

It is well established that the social environment early in life and specifically the quality of maternal care have a long-lasting impact on the offspring, but the mechanisms that biologically embed the response to early-life social environment in the offspring are still unclear. We addressed this question in this paper by comparing DNA methylation in young adult rhesus monkeys that were exposed to two extreme rearing conditions: MR versus SPR. One of the longstanding issues in deriving a causal relationship between an environmental intervention and a differential DNA methylation is the possibility that the epigenetic difference is driven by genetic polymorphism rather than environmental causes. Association of genetic and epigenetic variations was documented previously (Kang et al., 2008; Kerkel et al., 2008). Because there was no known bias in assigning the monkeys to differential infant rearing conditions, it is unlikely that group differences in DNA methylation are attributable to genetic polymorphisms rather than differential rearing conditions. Our sequencing data suggest that the differences illustrated here in DNA methylation between the rearing groups are not immediate consequences of change in sequence.

Previous research over the past five decades has provided ample demonstration that the long-standing model of early childhood adversity (nursery-rearing) produces an array of behavioral, physiological, and neurobiological deficits that parallel those identified in human studies of early adversity (Harlow and Harlow, 1965; Sackett, 1965, 1984; Kaufman and Rosenblum, 1969; Kraemer, 1992; Suomi, 1997; Machado and Bachevalier, 2003; Bennett, 2010). The animals studied here were reared under identical conditions to those animals in previous noninvasive, nonterminal studies that produced evidence of significant rearing group differences in behavior. The study reported here, using brain and peripheral tissue, is from a relatively small number of subjects and was not designed or powered to provide extensive behavioral characterization. Ethical and practical considerations inherent in primate studies that require terminal measures precluded the larger sample size that would have been required for behavioral assessment. However, both the strength of previous findings that reliably identify long-lasting effects of early rearing condition, as well as the significant rearing group differences identified in the current study, provide no indication that this set of subjects differs from the previous larger groups in which early rearing group differences in behavior were identified.

Previous studies have reported gene–environment interaction between rearing conditions in rhesus monkeys and the serotonin transporter polymorphism (5HTTLPR) (Champoux et al., 2002). In our sample, seven of eight animals were homozygous for the l/l alleles. No difference in methylation and expression was found in the 5HTT locus in our samples. The only monkey with l/s heterozygosity did not display differential methylation compared with the other genotype. Thus, polymorphism in the rh5HTTLPR gene is not an explanation for the widespread DNA methylation differences between the rearing conditions.

It has been a long-standing theory that the principal role of DNA methylation is in cell differentiation as predicted by Waddington (1959), as well as in programming “allele-specific” expression during development (Sapienza, 1990). The data presented here expand the role of DNA methylation beyond the innate predetermined process of embryonal cell-type differentiation to a possible role in response to changing environments. This study tests the relationship between these two “epigenetic” roles of DNA methylation. We see both functional and structural organization of differential DNA methylation between PFC and T cells. Similarly, the response to “rearing conditions” is not stochastically distributed across the genome but shows a high level of structural organization targeting specific cellular functions. This long-range organization of the DNA methylation response suggests that there are multiple modes at which DNA methylation exerts its biological effects above and beyond the site-specific effect on gene regulatory regions. These must be examined to understand how differential methylation responds to different biological challenges.

The changes in DNA methylation that we observe between tissues and in response to differential rearing conditions are numerous and significant, but each individual effect is small (i.e., per-probe fold change is small). Although it is possible to brush off these subtle changes as biologically irrelevant, their consistency and statistical significance point to the possibility of an important biological role whereby the epigenome is modulated by a combination of small changes across functional pathways and chromosomal regions. It is important to note in this respect that DNA methylation is a binary signal, that is, a site is either methylated or unmethylated in a given cell. Therefore, a partial methylation such as is observed in our study indicates that a small but statistically significant subpopulation of cells is differentially methylated. A challenge for future experiments is to identify the cellular populations that exhibit these changes in DNA methylation and to understand their biological role.

Understanding the role of DNA methylation in human behavior in live subjects and assessing the impact of different environmental exposures and interventions on DNA methylation will require studying peripheral tissues other than the brain. We examined this possibility by mapping the impact of different rearing conditions on DNA methylation in T cells and in the brain in the same individuals. We report here a limited overlap in sites that are differentially methylated in different rearing conditions in T cells and PFC. However, the scope of changes in DNA methylation in T cells in response to rearing conditions involves genes that are not affected in PFC. This is consistent with the hypothesis that the response to early-life adversity has an immune component that we would expect to be missing from brain tissue.

There is strong evidence of a crosstalk between the immune system and the brain. The best-studied example is the HPA–immune system bilateral relationship. It is well known that increase in glucocorticoids in response to activation of the HPA axis results in a profound silencing of gene expression of proinflammatory proteins and cytokines. It was shown recently that early-life social class can affect the expression of genes bearing response elements to CREB/ATF, nuclear factor-κB, and the glucocorticoid receptor (Miller et al., 2009). Posttraumatic stress was shown recently to be associated with DNA methylation changes in blood cells in genes that are involved in immune function (Uddin et al., 2010). Thus, the immune system might play a role in the overall response to social adversity.

One of the characteristics of the DNA methylation response is clustering across gene families. Remarkably, several such clusters are identified to be affected in both T cells and PFC. Thus, it is possible that single probes are not capable of univariately detecting a significant response but require the combined multivariate response across a genomic region to achieve significance. This might have not only statistical implications but biological implications as well. A combination of small changes across a cluster could result in an extremely significant change in the total output of the cluster. The fact that we see a response that covers an entire cluster suggests coregulation of the DNA methylation response to rearing conditions either through common cis-elements across the promoters in the cluster or through a cluster response element, such as the CTCF locus control element. Future experiments will be needed to uncover the mechanism that guide cluster-wide response in DNA methylation. An interesting cluster that appears to respond to rearing in both tissues is the olfactory receptor cluster. The selective expression of these genes is known to be responsible for odor-specific reception in olfactory neurons, and it is surprising that they are differentially methylated in T cells and PFC in response to rearing conditions. Future experiments need to examine the possibility that these receptors have other yet unknown biological roles.