PMC

Diagram of biological system interconnectivity. Adverse childhood experience including physical injury (trauma) are hypothesized to alter DNA methylation and lead to negative outcomes. Biological systems most likely to be impacted by trauma include hormonal regulation and the immune and neural systems, all of which are inter-related and affect each other; and all of which are developing in early childhood.

Flow chart of computational processing of Illumina BeadChip data. Workflow showing data sourcing/acquisition through site by site differential methylation assessment, starting with data acquisition and formatting, pre-processing steps, quality control, exploratory analysis, including heatmaps, and differential methylation tables with statistically significant values reported. Output from intermediate steps allows for global data trends to be reviewed, corrected and re-assessed along the way. Differential methylation output may be used to identify gene ontology with other software.

The ACE pyramid. This diagram shows relationships between Adverse Childhood Experience (ACE) and diseases, discovered by Felitti et al. in their landmark 1998 paper [], modified in Bearer et al. [] and further re-designed here. These associations have been confirmed by 100s of subsequent studies. More information is posted on the website for the U.S. Centers for Disease Control (https://www.cdc.gov/violenceprevention/acestudy/). The biological basis by which ACE increases risk for disease across the lifespan is unknown (indicated by unknown link and gray arrows), but epigenetic events, such as changes in DNA methylation levels of specific genes involved in immunity, hormone regulation and neurosystem development, are expected to occur.

Gene Ontologies of significant differentially methylated sites in saliva DNA after cell type deconvolution. Differential methylation was analyzed in RnBeads after adjusting the methylation level for each site in each sample according to the estimated cell composition of that sample, using as references the keratinocyte and whole blood methylation patterns shown in Figs. (8-10). Gene names of regions associated with each gene surviving statistically significant differentially methylated site between non-ASD and ASD cohorts (237 sites) were updated in DAVID (https://david.ncifcrf.gov) and submitted for associations in the DAVID v6.7 database in November 2017. Results of associations with cellular components, biological processes, and metabolic functions are shown in pie charts generated in Excel from the DAVID output. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

Comparison of sites identified by MeDIP or Illumina BeadChip aligned on a template human genome. Shown are screen shots from the Integrative genomics viewer (IGV) showing alignments for four children at two loci on chromosome 10 analyzed by two different methods: Illumina BeadChip HG450 and methyl-binding pulldown (MeDIP) with bisulfite sequencing. In (A) the centromeric region of ch10 is shown. Note that alignment of the Illumina BeadChip data shows no hits over this centromeric region and only a few sites adjacent to it (red arrow). In contrast, sequences from the MeDIP in the centromeric region are numerous (green arrow). Only one of the four children is shown for the MeDIP, but all showed the same pattern of hits. In (B) the Illumina BeadChip identified two sites with significant differences between the children (red arrows). These sites are not detected in the MeDIP for any of the four children, even when results for each of the four children are separately aligned. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

Diagram for comparing assay methods for detecting genome-wide methylations. Diagram comparing two methods for identifying differences in methylation levels at specific sites across the genome, with high-density array on the left and MeDIP on the right. In both cases the DNA is fragmented, the fragments are sized and adaptors ligated. Both protocols use bisulfite conversion. Fragmentation may occur before or after bisulfite conversion for the BeadChip analysis but must be done before conversion when using MeDIP since this method depends on the 5mC to pull-down fragments. For the BeadChip, single-stranded DNA is hybridized to known sequences on the chip that encode the unconverted or converted cytosine methylation site. Fluorescently labeled nucleotides are added for single nucleotide extension [], and the ratio of red or green fluorescent methylated probes intensity gives the ratio of methylated to un-methylated fragments at that site. For MeDIP, many more steps are required and computational analysis is complicated by the multiple methylation sites in many of the fragments that are altered after bisulfite conversion is performed. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

SNP distance heatmap. We obtained 8 IDAT files of methylation data from GEO, four for keratinocytes and four for whole blood, performed on the Illumina 450k chip. The Illumina 450k chip contains 65 genotyping probes that determine ancestry, and that can take one of three possible b-value levels: low (homozygous for one allele), high (homozygous for the other allele), or intermediate (heterozygous for both alleles). An individual should have the same b -values at all sites regardless of cell type, since this is based on somatic genomic sequence. The heatmap is produced by unsupervised hierarchical clustering of the intensity signals for 65 SNPs. The dendrogram above the heatmap gives a global picture of genotype-related sample grouping and similarities. Shown is a heatmap for methylation of the 8 different cell lines posted on GEO (GSM2260732; GSM2260731; GSM2260730; GSM2260729; GSM1936951; GSM1936939; GSM2071075; GSM2071074) [, -]. Cell lines numerically coded (1-8) and indicated below the heatmap. The SNP heatmap shown here demonstrates that two of these cell lines numbers 1 and 2, were obtained from the same individual, known to have been performed as technical duplicates. As more buccal keratinocyte data is obtained, the database for cell composition will expand. Heatmaps for the other samples demonstrated different individual SNPs. Such clustering of SNP heatmaps can also be used to detect sample mix-ups.

Biochemical basis of DNA Methylation. DNA Methylation is one of the biochemical events that are changed by experience and affect gene expression. A. Methyl groups are added to the 5' position on cytosine in DNA from a methyl donor (i.e. SAM) by at least three enzymes, DNMT 1, 2 and 3. Removal of methyl groups is initiated by oxidation mediated by any of the TET enzymes converting the 5' methyl cytosine to 5' hydroxyl-cytosine. The hydroxyl-cytosine is subsequently removed by other enzymes. Some investigators argue that additional enzymes may initiate the removal process, or that removal may involve excision of the methylated cytosine from the DNA strand and replacement with an un-methylated cytosine via DNA repair enzymes. Importantly, methylation for some sites appears to be dynamic. This dynamic reversibility makes DNA methylation sites promising biomarkers of experience, and as targets for theoretical interventions. 
B. Methylation in promoter and enhancer regions influences gene expression by recruiting methyl-binding proteins that may interfere with transcription. Methyl-binding proteins may also recruit histones and lead to other mechanisms of transcriptional repression, including chromatin packaging. Methyl-binding domains from these proteins are useful in isolating methylated DNA from a mixture of DNA fragments useful for reduced representation whole genome methylation analysis.

Micrographs of saliva smears. To determine the types of cells in saliva, we first performed histopathology on saliva smears collected from 6 healthy volunteers either directly onto a glass slide (A) or first into Oragene DNA collection cups which contain a buffer to preserve the DNA and then onto a glass slide (B). Slides were stained with haematoxylin and eosin according to normal pathology procedures and cover-slipped in mounting media. Slides were reviewed and types of cells counted. Keratinocytes (an example indicated by the orange arrow) and WBCs (example indicated by green arrow) were present. Preparations differed in the relative number of each type of cell. After treatment with the Oragene buffer (B), keratinocytes appeared ghosted and lacked nuclear staining, consistent with DNA extraction (orange arrow). Some small round blue dots remaining apparently represent un-extracted nuclei or non-human microbes. These results convinced us that methylation patterns obtained from saliva need to be corrected for relative amounts of DNA from keratinocytes versus white blood cells. Analysis of our 5mC data demonstrated that children with no stress-related cortisol elevation had a lower ratio of keratinocytes (“skin cells”, aka cheek or buccal cells) to blood cells than children with a history of trauma (lower table). Because 5mC levels at specific sites differs between cell types [], 5mC levels can be used as a surrogate for quantifying cell types in saliva samples []. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

Quality controls of cell type methylation data from keratinocytes and whole blood. A. Original probe distribution. Note similar methylation level distribution in both cell types. B. Distribution of probes in genomic locations: Open sea (between genes); Shelf (2kb flanking genes); Shore (region where methylation levels are highly variable, usually close to promoters, likely to encode enhancers); Island (CpG island, typically at least 200 bp with a CpG ratio greater than 50%). Note sites in all genomic locations. C. Histogram comparing the removed and retained sites after quality controls and filtering. Note the relatively sparse numbers of sites removed for both datasets. D. Principal Component Analysis (PCA) of data from three cell types: Buccal keratinocytes, foreskin keratinocytes and whole blood from 8 different cell culture samples. Shown is a graph of principal components 1 and 2. Three components were identified that predict >95% of all variance as based on all sites remaining after filtering. The keratinocytes (orange and green circles) are widely separated from the whole blood, as has also been shown for mouse cell types assayed by reduced representation bisulfite sequencing (RRSS) [, ]. The range of the x-axis is -80 to +80, and the range for the y-axis is -30 to +30. DNA was from primary cultures of keratinocytes from foreskin (green, Kerat.f) or from buccal (red, Kerat.b) and whole blood (purple, WB). E. Scatter plot of group-wise mean DNA methylation levels for all keratinocyte and blood samples across all promoter sites. Sites with significant differences between cell types are colored red, as determined by RnBeads via a three-part metric. Sites with similar methylation levels in both samples lie on the diagonal and are colored blue. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

Cell composition de-convolution: Adjusting methylation data for varying contributions of keratinocytes and nucleated blood cells. (A) The average ratios of cell types in four cohorts of children’s saliva. The top two pie-charts are for a cohort of children ages 4-8 years old, and the bottom two from children ages 18 months to 4 years old. Red indicates blood cells and blue keratinocytes. Two of the pie-charts (the 1^st and 3^rd down) are children without reported trauma and the other two are from children with high cortisol levels and reported traumatic experience (2^nd) or an autism spectrum diagnosis (4^th down). (B) Comparing profiles for individual children in the datasets, even though the pie-charts for the cohorts appear relatively similar, there is wide variation between individual samples of cell composition as detected by covariate inference []. In the first step the reference methylomes were used to estimate the association of each CpG position to each of the cell types. The strength of association was measured using an F-test. To decrease the computational load, only 50,000 most variable CpGs were considered. Finally, only 500 CpGs with the lowest F-test p-value were used in the contribution estimation. Selecting the most informative CpGs is equivalent to applying an F statistic cut-off of 1.788. (C) Scatterplots showing the significant sites between ASD and non-ASD (control) after cell composition de-convolution for promoters (top) or over all sites (bottom). Non-significantly different sites are colored blue, and those that met the automatically generated rank cutoff are red. Compare these scatterplots with the one shown in Fig. (9E) comparing 5mC patterns from cultured keratinocytes and whole blood, where many more significantly different sites appear. De-convolution decreases the complexity of the sample, removes the confounder of variation in cell type ratios, and improves identification of trait-related differences. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)

Exploratory analyses of methylation patterns. A. The degree of methylation across all sites in each sample of keratinocyte and whole blood is displayed as a heatmap of the number of sites in each sample with that percent of methylation, ranging from 0 to 100% (see color key in upper left). Pure red is 0% and blue is 100% methylated, with blended red/blue for partial methylation and no color for equally methylation/demethylated (50%). Note that cell types cluster together according to the dendrogram above the heatmap, with blood samples indicated by orange and keratinocyte samples by green bars above the heatmap. Note that even at this low level of analysis, the samples cluster according to cell type. B. Heatmaps and unsupervised hierarchical clustering for all sites across all samples of both cell-types, correlation-based, with complete agglomeration strategy (linkage), visualizing the 1000 most variable loci, demonstrates large differences between cell types, where some sites are highly methylated in blood and not much methylated in keratinocytes and others are vice versa. These dramatic differences represent the outer corners of the mean difference plot shown in Fig. (9E). The column to the left is color coded for the 5mC location, whether in open sea (blue), shelf, (turquoise), shore (purple), or CpG island, (red) for each site. C. Heatmap from the same analysis visualizing only promoter sites in the 1000 most variable loci, with the dissimilarity metric set for correlation-based and agglomeration strategy (linkage) average. Note the large difference in 5mC patterns between the two cell types. D. Children’s 5mC patterns from saliva clusters with either skin or blood cells. Heatmaps of hierarchical clustering of methylation data from brain (green), blood (purple) and keratinocytes (orange) together with 45 saliva samples from healthy children (color coded in gray above the heatmap). Some children were sampled at two different time points, and some were run in duplicate. Note that these saliva samples fall into two clusters as shown in the dendrogram above the heatmap. One group clusters with blood and the other with keratinocytes. Brain clusters with all samples at a greater distance []. Examination of the samples revealed that a few children sampled at two different time points clustered differently: one time point with keratinocytes and the other time point with whole blood. This result demonstrates the critical need to adjust 5mC results for cell composition. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this paper.)