DNA methylation signatures of early life adversity are exposure-dependent in wild baboons

Abstract

The early life environment can profoundly shape the trajectory of an animal's life, even years or decades later. One mechanism proposed to contribute to these early life effects is DNA methylation. However, the frequency and functional importance of DNA methylation in shaping early life effects on adult outcomes is poorly understood, especially in natural populations. Here, we integrate prospectively collected data on fitness-associated variation in the early environment with DNA methylation estimates at 477,270 CpG sites in 256 wild baboons. We find highly heterogeneous relationships between the early life environment and DNA methylation in adulthood: aspects of the environment linked to resource limitation (e.g., low-quality habitat, early life drought) are associated with many more CpG sites than other types of environmental stressors (e.g., low maternal social status). Sites associated with early resource limitation are enriched in gene bodies and putative enhancers, suggesting they are functionally relevant. Indeed, by deploying a baboon-specific, massively parallel reporter assay, we show that a subset of windows containing these sites are capable of regulatory activity, and that, for 88% of early drought-associated sites in these regulatory windows, enhancer activity is DNA methylation-dependent. Together, our results support the idea that DNA methylation patterns contain a persistent signature of the early life environment. However, they also indicate that not all environmental exposures leave an equivalent mark and suggest that socioenvironmental variation at the time of sampling is more likely to be functionally important. Thus, multiple mechanisms must converge to explain early life effects on fitness-related traits.

Fig. 1.. Socioenvironmental predictors of DNA methylation depend on early life habitat quality.

(A) Study design: we investigated (i) habitat quality at birth (lefthand triangle: high/post-shift versus low/pre-shift), (ii) cumulative exposure to each of five individual sources of early adversity (lefthand box), and the effects of (iii) age and (iv) dominance rank at the time of sample collection (right hand box). (B) The absolute value of cumulative early adversity effects estimated for individuals born in high-quality habitat (purple) versus those born in low-quality habitat (peach) for sites passing a 20% FDR in one or both conditions (n=12,872 CpG sites; Model 2). Solid and dashed lines show the mean and 95% intervals, respectively, for each distribution. (C) Standardized betas, from Model 2, comparing the effect of cumulative early adversity for individuals born in low- versus high-quality habitats, across the same set of sites (n=12,872). Each line connects the two effect sizes for one CpG site (one effect size estimate from samples of individuals born in the high-quality habitats and the second estimated for those born in low-quality habitats).

Fig. 2.. Early life adversity is associated with DNA methylation in adulthood for baboons born in low-quality habitat.

(A) The number of CpG sites associated with each tested predictor (<10% FDR) in Model 3. The x-axis is shown on a log₁₀ scale. (B) Reaction norms for two example CpG sites (chr12_111013997 and chr11_430191) that were significantly associated with early life drought, but only for baboons born in low-quality habitat (peach; 10% FDR). Colored bars indicate standard errors. (C) Distributions of the absolute value of standardized effect sizes across tested sites for each of five individual-level sources of early adversity. In all cases, effect sizes are systematically larger for individuals born into low-quality habitat (peach) environments than those born into high-quality environments (purple). (D) UpSet plot of the number of CpG sites associated with habitat quality, each individual source of adversity (within low-quality habitat), and their overlap. Each bar represents the number of sites associated with the source(s) of adversity indicated in the matrix beneath the bar graph. To avoid calling sites “unique” due to small differences in FDR values, overlaps show sites that are significant at a 10% FDR threshold for at least one predictor variable and ≤20% FDR for the other predictor variable(s).

Fig. 3.. Genomic distribution of CpG sites associated with age, rank, and early life adversity.

(A) Enrichment of the top four predictors of DNA methylation levels in functional compartments across the genome. Color indicates log₂(Odds Ratio) from a Fisher’s exact test, with the brightest colors indicating highest and lowest odds ratios. (B) Enrichment of the same four sets of age, rank, or early environment-associated CpG sites, across 15 distinct chromatin states, based on annotation in human peripheral blood mononuclear cells with coordinates lifted over to Panubis1.0. States are ordered roughly by their association with active gene regulation, from left (active) to right (repressed/quiescent). Opaque dots indicate p<0.05 for enrichment based on Fisher’s exact test.

Fig. 4.. Early life habitat quality can be accurately predicted from DNA methylation, but this signal attenuates over time.

(A) Known early life habitat quality (x-axis) versus predicted early life habitat quality from an elastic net regularization model (y-axis). More negative values correspond to cases in which the model predicted that the individual was born in high-quality habitat (the post-habitat shift environment); more positive values correspond to cases in which the model predicted that the individual was born in low-quality habitat (the pre-shift environment). (B) Receiver operating characteristic (ROC) curve for early life habitat quality predictions (AUC=0.926; dashed line denotes the y=x line). (C) Predicted habitat quality (y-axis) versus the time since habitat shift in days (x-axis) for animals born in low quality habitat (linear model p=0.0084). 0 days since habitat shift indicates a sample from an animal still in the low-quality environment.

Fig. 5.. CpG sites associated with drought and male dominance rank are enriched in functional regions of the genome based on a high-throughput reporter assay.

(A) Workflow for the mSTARR-seq experiment and an example of read pileups at a regulatory window that exhibits methylation-dependent regulatory activity and overlaps a drought-associated CpG site in the observational data from Amboseli. Summed read counts are shown for methylated (blue) and unmethylated (yellow) experimental replicates. In the highlighted methylation-dependent regulatory region, unmethylated treatments drive substantial expression (yellow RNA counts) compared to methylated treatments (blue RNA counts), even though the amount of input DNA (overlapping yellow and blue DNA counts) was near-identical across treatments. (B) Enrichment of regulatory regions from mSTARR-seq across 15 chromatin states lifted over to the baboon genome from human peripheral blood mononuclear cells (27). Regions with empirically identified regulatory activity are enriched in regions orthologous to putative enhancer and promoter regions in human PBMCs, and depleted in states associated with regulatory quiescence/repression. (C) Enrichment statistics for male dominance rank- (blue), drought- (red), and age-associated CpGs (gray) in regions capable of regulatory activity in mSTARR-seq. The x-axis shows the FDR threshold for identifying age, drought, or rank-associated CpG sites; the y-axis shows the log₂(OR) for enrichment in mSTARR-seq putative regulatory elements (all identified at FDR = 10%). Opaque points indicate significant FET enrichment (p<0.05).