Hippocampal and peripheral blood DNA methylation signatures correlate at the gene and pathway level in a mouse model of autism

Abstract

Autism spectrum disorders (ASD) are polygenic multifactorial disorders influenced by environmental factors. ASD-related differential DNA methylation has been found in human peripheral tissues, such as placenta, paternal sperm, buccal epithelium, and blood. However, these data lack direct comparison of DNA methylation levels with brain tissue from the same individual to determine the extent that peripheral tissues are surrogates for behavior-related disorders. Here, whole genome methylation profiling at all the possible sites throughout the mouse genome (>25 million) from both brain and blood tissues revealed novel insights into the systemic contributions of DNA methylation to ASD. Sixty-six differentially methylated regions (DMRs) share the same genomic coordinates in these two tissues, many of which are linked to risk genes for neurodevelopmental disorders and intellectual disabilities (e.g. Prkch, Ptn, Hcfc1, Mid1, and Nfia). Gene ontological pathways revealed a significant number of common terms between brain and blood (N = 65 terms), and nearly half (30/65) were associated with brain/neuronal development. Furthermore, seven DMR-associated genes among these terms contain methyl-sensitive transcription factor sequence motifs within the DMRs of both tissues; four of them (Cux2, Kcnip2, Fgf13, and Mrtfa) contain the same methyl-sensitive transcription factor binding sequence motifs (HES1/2/5, TBX2 and TFAP2C), suggesting DNA methylation influences the binding of common transcription factors required for gene expression. Together, these findings suggest that peripheral blood is a good surrogate tissue for brain and support that DNA methylation contributes to altered gene regulation in the pathogenesis of ASD.

autism, Cntnap2, whole genome methylation sequencing, epigenetics, epigenomics

Introduction

Autism spectrum disorders (ASD) are considered multifactorial hereditary disorders, resulting from contributions from numerous genes and environmental factors. While epigenome-wide association studies have identified locus-specific differential DNA methylation signatures associated with ASD in the brain, several lines of evidence indicate that ASD is a systemic disorder, affecting tissues outside the brain, including the immune system [1, 2], gastrointestinal tract [3], and hormone imbalances [1, 4, 5]. Consistent with these findings, ASD-related differential DNA methylation levels have been found in human peripheral tissues, such as placenta, paternal sperm, buccal epithelium, and blood [6–9]. These data support the potential of peripheral DNA methylation levels to capture the complex multivariate ASD etiology but lack direct comparison of DNA methylation levels between brain and blood tissues from the same individual to determine the extent that peripheral tissues are surrogates for brain tissue in behavior-related disorders.

It has been challenging to find robust DNA methylation signatures in blood for idiopathic ASD; however, syndromic forms of ASD offer promise for comparisons of DNA methylation levels in brain and blood. Mice harboring a homozygous deletion of Cntnap2 exhibit parallels to the core ASD-related deficits in humans, including altered vocalizations, repetitive movements, and decreased social interactions [10]. In addition, major neuropathological features were observed in Cntnap2 homozygous knockout mice (i.e. Cntnap2^−/−), such as deficits in neuronal migration of cortical projection neurons and a reduction in the number of striatal GABAergic interneurons [10]. A recent multi-omics study integrating data from Cntnap2 homozygous knockout mice and ASD patients found common molecular networks that highlight mitochondrial dysfunction, axonal impairment, and synaptic activity, suggesting a strong association with ASD pathophysiology in both CNTNAP2-deficient humans and mice [11]. Previously, we showed significant disruptions in DNA hydroxymethylation (5-hydroxymethylcytosine) in the brain of Cntnap2 homozygous knockout mice on a significant number of genes associated with ASD in humans, including genes linked to epigenetic mechanisms and synaptic plasticity, suggesting a role for DNA methylation modulation in the pathogenesis of ASD [12]. Here we sought to test the independence of DNA methylation signatures in brain and blood of Cntnap2 homozygous knockout mice by conducting whole genome methylation sequencing to compare the DNA methylation levels at all possible sites throughout the mouse genome (>25 million) in both tissues from the same mouse. Differentially methylated sites that are either common or unique between brain and blood reveal novel insights into the systemic contributions of DNA methylation to the complex phenotype of ASD that could be used to develop blood-based biomarkers of ASD to improve early diagnosis, facilitate early interventions, and guide planned parenting for these disorders.

Materials and Methods

Mice

Cntnap2 homozygous knockout male and female mice [13] were purchased from the Jackson Laboratories (Bar Harbor, ME) and maintained on C57BL/6J background for more than 20 generations. Mice were housed under uniform conditions in a pathogen-free mouse facility with a 12-h light/dark cycle. Food and water were available ad libitum. All experiments were approved by the University of Wisconsin-Madison Institutional Animal Care and Use Committee (M005144).

Genotyping

Cntnap2 homozygous knockout and WT littermates were genotyped using the following primers: homozygous knockout Rev: CGCTTCCTCGTGCTTTACGGTAT, Common: CTGCCAGCCCAGAACTGG, WT Rev 1: GCCTGCTCTCA GAGACATCA. PCR amplification was performed with one cycle of 95°C for 5 min and 35 cycles of 95°C for 30 s, 56°C for 30 s, and 68°C for 30 s, followed by 68°C for 10 min. The homozygous knockout and WT alleles were obtained with 350-base pairs and 197-base pairs PCR products, respectively.

Tissue and DNA extraction

Three-months old female Cntnap2 homozygous knockout and wild-type (WT) mice (N = 3 per group) were anesthetized with isoflurane (2 hours after lights on) and ∼500–700 μl of whole blood was collected by cardiac puncture and stored in −20°C. Mice were then perfused with 25 ml of 1× PBS in order to remove the remaining blood from the body and the hippocampus was extracted and immediately flash-frozen in 2-methyl butane and dry ice. The whole hippocampus (∼30 milligrams of tissue) was homogenized and DNA was extracted using the All Prep DNA/RNA mini kit following the manufacturer protocol (Qiagen, Cat. No. 80204). DNA extraction from whole blood was performed by using the protocol from the Gentra Puregene Blood Kit (Qiagen, Cat. No. 158389).

Library preparation and high-throughput sequencing

Whole genome methyl sequencing (NEBNext Enzymatic Methyl-Seq, EM-seq™) was employed to profile the methylomes (all the potentially methylated sites, >25 million) of 12 genomic DNA samples extracted from hippocampus and blood collected from Cntnap2 homozygous knockout and WT female mice. The EM-seq™ approach uses TET2 and Oxidation Enhancer to protect methylated cytosines from deamination and APOBEC to convert unmethylated cytosines to uracil. Subsequent sequencing of the treated DNA provides single base-pair resolution of all methylated sites in the mouse genome. To process the samples, five hundred nanograms of high molecular weight genomic DNA was forwarded to the University of Illinois at Urbana-Champaign Roy J. Carver Biotechnology Center for DNA sequence library construction and whole genome sequencing on a Next-Generation sequencer (Illumina NovaSeq6000).

Quality control and alignment to the genome

An average of 565 million 150 base pairs paired-end reads were sequenced per sample. Raw FASTQ files were assessed for quality using FastQC (v0.11.8, https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) to interrogate read-wide base quality scores, base content (to ensure depletion of cytosine quantity), and GC content. Following assessment, raw paired-end FASTQ files were trimmed for adapters and quality using Trim_Galore (v0.6.7, https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) with the following parameters: quality score cutoff set to 30, minimum read length set to 30 base pairs, run in paired-end mode. After trimming, paired-end files were aligned to the UCSC mouse genome (mm10) using Bismark (v0.20.0) [14] in conjunction with bowtie2 (v2.3.4.3) [15]. During alignment, default parameters were used with the exception of: minimum score set to L,0, −0.3, and the maximum insertion length increased to 1000 base pairs. Following alignment, uniquely mapped reads were deduplicated to avoid PCR artefacts, and methylation estimates were extracted using Bismark. During methylation extraction, three base pairs from the 5′ and 3′ end of R1 and R2 reads were ignored to avoid methylation bias as indicated by M-bias plots. Following extraction, methylation estimates were collapsed across strands to report methylation at only one CpG from the sense strand.

Identification of differentially methylated regions

R package dmrseq [16] was employed to identify differentially methylated regions (DMRs). Samples were sequestered based on tissue (hippocampus or blood). Biological replicates (N = 3) from Cntnap2 homozygous knockouts were compared to WT (N = 3) mice for each tissue. CpGs were filtered to only those that had a coverage of at least 1× from all samples (~21 million CpGs). Candidate regions were identified in dmrseq using the following parameters: minimum number of CpGs in the region set to 5 CpGs, a smoothed methylation difference of >5% between groups, and a maximum of 10 permutations performed. Following candidate region identification and permutation testing, significant DMRs were retained for further analyses if their P-value was <0.05 for each tested tissue. Chromosomal enrichment/depletion analysis of DMRs relative to background regions was performed using the Locus Overlap Analysis (LOLA) [17].

Pyrosequencing

Ten DMRs comprising 112 CpGs were validated by pyrosequencing using the same 6 genomic DNA samples extracted from hippocampus and used for whole genome methyl sequencing. Genomic DNAs The Pyrosequencing approach utilizes sodium bisulfite modification by the EZ-96 DNA MethylationDirect Kit™ (ZymoResearch; Irvine, CA; Cat. No. D5023). To process the samples, one microgram of high molecular weight genomic DNA was forwarded to EpigenDx, Inc for DNA sequence library construction and pyrosequencing on the Ion S5™ sequencer using an Ion 530™ sequencing chip (Cat. No. A27764). FASTQ files from the Ion Torrent S5 server were aligned to a local reference database using the open-source Bismark Bisulfite Read Mapper program (v0.12.2) [14] with the Bowtie2 alignment algorithm (v2.2.3) [15]. Methylation levels were calculated in Bismark by dividing the number of methylated reads by the total number of reads. An R-squared value (RSQ) was calculated from the controls set at known methylation levels to test for PCR bias.

Functional testing of DMRs

DMRs were annotated to genes (±2 kilobases from transcription start sites) and genomic structures using R packages ChIPseeker, org.Mm.eg.db, and TxDb.Mmusculus.UCSC.mm10.knownGene [18]. In order to study the DMRs that share the same genomic coordinates in hippocampus and blood, R package regioneR [19] was used to perform permutation testing of genomic coordinate-based overlap enrichment of DMRs in both tissues, using a region randomization method over 100 000 permutations. R package clusterProfiler [20] was used for ontological analyses for the common DMR-associated genes between hippocampus and blood, using an adjusted P-value cutoff of 0.05 to identify significant terms. Redundant ontological terms were removed with clusterProfiler based on adjusted P-values and on a similarity value of 0.7.

Gene ontology and methyl-sensitive sequence motif analysis

For the study of the biological role of hyper-, hypo- and all DMR-associated genes, we performed gene ontology and functionally group network of terms analysis using Cytoscape (version 3.9.1) [21] and ClueGO Plug-in (version 2.5.7) [22]. If the list contains around 200 genes, the selection criteria of representative pathways were used by default. If the list contains more than 1000 genes, the gene ontology hierarchical tree was modified by the following criteria: 6 minimum level and 12 maximum level GO tree interval, a minimum of 2 genes from the uploaded list found to be associated with a term, and these genes represent at least 5% of the total number of associated genes. In order to have a complete representation of all the significant gene ontology terms (biological process), they have been associated with functional groups based on shared genes (Kappa score is used, set by default). Only groups containing significant terms (Bonferroni correction P-value < 0.01) are shown. The most significant term of a functionally grouped network is considered to be the leading term and represents the name of the functional group in the pie chart. The pie chart shows the percent of terms that belong to each significant functional group. All the terms and the groups are listed in Dataset S3,S4,S7 and S8 following the same color code used to represent the groups in the pie charts.

R package memes (v1.4.1) [23,24] was employed for the analysis of methyl-sensitive motif enrichment of DMR sequences, relative to background regions, against the Human Methylcytosine database and mm10 sequences. For functional testing of methyl-sensitive transcription factor sequence, we searched for motifs residing in DMRs located in the regulatory region (promoter and intron 1) of genes, which was first determined by the annotation of DMRs to genes and genomic features, and confirmed using the UCSC Genome Browser (GRCm38/mm10).

Overlap of DMR-associated genes and GO terms

A hypergeometric test was used to test the significance of the overlap of the following: 1) DMR-associated genes in the hippocampus and blood; 2) DMR-associated genes with high-confidence orthologous human autism genes; and 3) the common GO terms between tissues. The GO terms were defined “neurological terms” if they include any of the following: nervous system, brain, neuron, synapse, dendrite, dendritic spine, axon, or being included in a neurological functional group based on shared genes by ClueGO.

Results

Disruption of hippocampal whole genome DNA methylation levels in Cntnap2 homozygous knockout mice

To identify susceptibility loci related to ASD, hippocampal genomic DNA was isolated from female Cntnap2 homozygous knockout (i.e. Cntnap2^−/−) and WT mice (N = 3, for each genotype). NEBNext Enzymatic Methyl-seq (EM-seq) was employed for the detection of whole genome DNA methylation levels using a next-generation sequencer (Methods). This approach generated an average of 589.2 million raw sequence reads for each sample and, following filtering for quality (Methods), an average of 495 million sequence reads uniquely mapped to the mouse genome (mm10), providing an average genomic coverage of 44× for 21.1 million CpGs (Table S1). Candidate differentially methylated regions were identified and permutated, revealing a total of 3052 autism-related differentially methylated regions (DMRs, permutation P-value < 0.05; >5% differential), of which 2630 were hypermethylated (more methylated in Cntnap2 knockout compared to WT) and 422 were hypomethylated (less methylated in Cntnap2 knockout compared to WT, Fig. 1A; Dataset S1). Hierarchical clustering using the DNA methylation levels of the CpGs contained within each DMR showed stratification of each replicate by genotype (Fig. 1B). The differential DNA methylation levels of 10 hippocampal DMRs comprising 112 CpGs were confirmed by pyrosequencing (R-squared = 0.972, P-value = 2.2e−16; Fig. S1, Table S2).

Figure 1

Distribution of DMRs in the hippocampus of Cntnap2 homozygous knockout (Cntnap2^−/−) mice. (A) Circos plot depicts genomic distribution of differentially methylated regions (DMRs) across the mouse genome. (Outer ring) Each chromosome is shown as different rectangles where the relative chromosome size is represented by the rectangle length. (Middle and inner rings) Represent the relative genomic location of hyper- (middle) and hypo-methylated (inner) DMRs across each chromosome. (B) A heatmap depicts the mean DNA methylation level for each CpG (y-axis) contained in the DMRs (N = 3052) for each sample (x-axis; N = 3 for each genotype; WT [wild-type] and Cntnap2^−/− [Cntnap2 homozygous knockout]). The heatmap colors correspond from 0% methylation (blue) to 100% methylation (red), as indicated by the color key. The dendrogram (top) indicates the relatedness of each sample. (C) A Sankey plot depicts the abundance of ASD-related hippocampal DMRs across each standard genomic structures comprising 1–5 kilobases (kb) upstream of the transcription start site (“promoter”), the 5′ untranslated region (“5’ UTR”), any gene exon (“exonic”), and gene intron (“intronic”), the 3′ untranslated region (“3’ UTR”), 300 base pairs downstream of the transcription stop site (“downstream”), and 5 kb or more from any gene structure (“other”). (D) and (E) A pie chart showing the functional groups obtained with the ASD-related hypermethylated (N = 1417) and hypomethylated (N = 242) DMR-associated genes in hippocampus, respectively. Gene ontology (GO) terms (biological process) have been associated into functional groups based on shared genes (Methods). The percentage of GO terms per group is shown in each wedge of the pie chart, as well as the most significant term of a group (considered the leading term). Only groups containing significant terms (Bonferroni correction, P-value < 0.01) are shown.

Since specific regions of the genome are differentially methylated based on the biological functions of the genes contained within that region, we determined whether DMRs (N = 3052) were enriched or depleted on each chromosome (Methods). This analysis revealed chromosomes 7, 14, 17, and X have disproportionately less DMRs and chromosome 18 has disproportionately more DMRs than expected by chance alone (Fisher’s exact test, P-value < 0.05; Fig. S2A). These results indicate that some chromosomes are more sensitive to DNA methylation changes and these changes are not randomly distributed across the genome.

Hippocampal DMR-associated genes

Annotation of the DMRs to standard genomic features and genes, obtained from the UCSC mouse genome (mm10), showed enrichment in intronic and distal intergenic regions (Fig. 1C). This approach also revealed 1417 and 242 genes with hyper- and hypo-DMRs, respectively (Dataset S2). Two-hundred and eighty-eight genes have more than one DMR, and 57 genes contain both hyper- and hypo-DMRs (Dataset S2). To determine the biological relevance of DMR-associated genes, gene ontological analysis was separately performed on hyper and hypo DMR-associated genes (N = 1417 and N = 242, respectively), which revealed an enrichment of terms related to neuronal functions and processes from both DMR-associated gene sets (Fig. 1D and E). More than 80% of all hypermethylated DMR-associated terms were linked with neuronal processes (e.g. neuron generation, differentiation, maturation, development, and communication). Similarly, 36% of hypomethylated DMR-associated genes were associated with neuron morphogenesis and development, as well as axon guidance and glial cell differentiation (Fig. 1D and E; Datasets S3 and S4). Together, these data suggest that the absence of CNTNAP2 results in widespread dysregulation of genes in neuronal pathways and that DNA methylation plays a role in this process.

Disruption of peripheral blood whole genome DNA methylation levels in the Cntnap2 homozygous knockout mice

To test the possibility that peripheral tissues are surrogates for brain tissue in ASD, whole blood genomic DNA was isolated from the same animals profiled above (i.e. female Cntnap2 homozygous knockout and WT mice [N = 3, for each genotype]) and whole genome DNA methylation data was generated using the EM-seq method. This approach generated an average of 542 million raw sequence reads for each sample and an average of 451.9 million sequence reads uniquely mapped to the mouse genome (mm10), which provided an average genomic coverage of 41× for 21 million CpGs (Methods, Table S1). A density plot comparing global DNA methylation levels in peripheral blood and hippocampal tissue found a greater proportion of fully methylated CpGs in peripheral blood, regardless of genotype (Fig. S3A), indicating that DNA methylation abundance alone can segregate peripheral blood from hippocampal tissue (Fig. S3B). Candidate differentially methylated regions were identified and permutated, revealing a total of 1870 autism-related DMRs (permutation P-value < 0.05), of which 287 were hypermethylated and 1583 were hypomethylated (Fig. 2A; Dataset S5). Similar to the hippocampal data, hierarchical clustering using the DNA methylation levels of the CpGs contained within each DMR showed stratification of each replicate by genotype (Fig. 2B). Characterization of DMR enrichment or depletion on chromosomes revealed chromosomes 14 and X have disproportionately less DMRs and chromosomes 6, 10, and 11 have disproportionately more DMRs than expected by chance alone (Fisher’s exact test, P-value < 0.05; Fig. S2B).

Figure 2

Distribution of DMRs in the blood of Cntnap2 homozygous knockout (Cntnap2^−/−) mice. (A) Circos plot depicts genomic distribution of differentially methylated regions (DMRs) across the mouse genome. (Outer ring) Each chromosome is shown as different rectangles where the relative chromosome size is represented by the rectangle length. (Middle and inner rings) Represent the relative genomic location of hyper- (middle) and hypo-methylated (inner) DMRs across each chromosome. (B) A heatmap depicts the mean DNA methylation level for each CpG (y-axis) contained in the DMRs (N = 1870) for each sample (x-axis; N = 3 for each genotype; WT [wild-type] and Cntnap2^−/− [Cntnap2 homozygous knockout]). The heatmap colors correspond from 0% methylation (blue) to 100% methylation (red), as indicated by the color key. The dendrogram (top) indicates the relatedness of each sample. (C) A Sankey plot depicts the abundance of ASD-related blood DMRs across each standard genomic structures comprising 1–5 kilobases (kb) upstream of the transcription start site (“promoter”), the 5′ untranslated region (“5′ UTR”), any gene exon (“exonic”), and gene intron (“intronic”), the 3′ untranslated region (“3′ UTR”), 300 base pairs downstream of the transcription stop site (“downstream”), and 5 kb or more from any gene structure (“other”). (D) and (E) A pie chart showing the functional groups obtained with the ASD-related hypermethylated (N = 166) and hypomethylated (N = 1065) DMR-associated genes in blood, respectively. Gene ontology (GO) terms (biological process) have been associated into functional groups based on shared genes (Methods). The percentage of GO terms per group is shown in each wedge of the pie chart, as well as the most significant term of a group (considered to be the leading term). Only groups containing significant terms (Bonferroni correction, P-value < 0.01) are shown.

Peripheral blood DMR-associated genes

Annotation of DMRs to standard overlapping genomic features showed that DMRs were mostly distributed in the promoter regions (33%, Fig. 4C), suggesting a functional link to gene expression for ASD-related DNA methylation changes in blood. This approach also revealed 166 and 1065 genes with hyper- and hypo-DMRs, respectively (Dataset S6). One-hundred forty-one genes have more than one DMR, and 22 genes contain both hyper- and hypo-DMRs (Dataset S6). Separate gene ontological analyses of the hyper and hypo DMR-associated genes (N = 166 and N = 1065, respectively) revealed terms related to the immune system and blood functions (e.g. leukocyte differentiation, blood vessel development, interleukin production) as well as pathways associated with neuronal ontological terms, including synaptic assembly, cytoskeleton organization, neurogenesis, and dendritic spine development (Fig. 2D and E; Datasets S7 and S8), suggesting a link between blood and brain DMR-associated genes.

Common DMRs and DMR-associated genes in hippocampus and peripheral blood

We next examined the occurrence of hippocampal and peripheral blood DMRs to overlap each other and found a significant number of DMRs (N = 66, permutation P-value < 0.0001) share overlapping genomic coordinates in these two tissues. Ten of these 66 hippocampal DMRs, comprising 112 CpGs, have been validated by pyrosequencing (Fig. S1). Thirty-two of these overlapping DMRs are associated with genes, including genes previously linked to neurodevelopmental disorders and intellectual disabilities (e.g. Prkch, Ptn, Hcfc1, Mid1, and Nfia; Dataset S9) [25–29]. Hierarchical clustering using the DNA methylation levels of the CpGs contained within each overlapping DMR (N = 66) showed stratification of each sample by genotype, regardless of tissue (Fig. 3A; Fig. S4), suggesting similar DNA methylation levels between hippocampus and blood at these CpGs. In fact, DNA methylation levels were altered in the same direction (i.e. increases or decreases) in both tissues (Fig. 3B) in approximately 80% (N = 51/66) of the overlapping DMRs. The gene ontological terms of the overlapping DMR-associated genes (N = 32) are associated with neurological features with relevance to autism spectrum disorders (i.e. dendritic arborization and learning; Fig. 3C). Together, these findings suggest that ASD-related DMRs in blood mirror those in the brain.

Figure 3

ASD-related overlapping DMRs and common DMR-associated genes in hippocampus and blood. (A) A dendrogram generated using the mean DNA methylation levels of the CpGs contained in the 66 DMRs found to have genomic overlap between both tissues indicates the relatedness of each sample. Sample genotype and source (x-axis; N = 3 for each genotype; WT [wild-type] and Cntnap2^−/− [Cntnap2 homozygous knockout]; hippocampus [Hipp] and blood) are indicated. (B) A heatmap depicts the difference of DNA methylation levels between Cntnap2^−/− and WT (Cntnap2^−/—specific DMRs) in the 66 DMRs found to be common between hippocampus and blood. Each row (y-axis) of the heatmap indicates the name of the common ASD-related DMR-associated gene (if DMR is located within 2 kilobases of a gene) or that it is located >2 kilobases from any gene (“distal intergenic”). The rows of the heatmap are ordered according to the difference in mean DMR methylation levels between tissues (smallest to largest). (C) A bar plot showing the significance (x-axis) of gene ontology terms (y-axis, biological process) obtained with the overlapping DMR-associated genes (N = 32) between hippocampus and blood. (D) A Venn diagram depicts the overlap of hippocampus and blood DMR-associated genes (N = 214) and the overlap of hippocampal (N = 80) and blood (N = 62) DMR-associated genes with high-confidence orthologous human autism genes (N = 393). The asterisk (^*) denotes a significant overlap (hypergeometric enrichment, P-value < 0.05; Methods).

Since a genomic overlap of brain and blood DMRs is not required to identify tissue-independent genes contributing to the phenotype, we also compared common DMR-associated genes between hippocampus and blood that may not share the same genomic positions in the genome. This approach revealed a significant number of DMR-associated genes in common between both tissues (N = 214, P-value < 0.01, Fig. 3D; Dataset S10). To examine whether the DMR-associated genes were enriched for ASD-related genes, we compared them to a list of orthologous human autism genes from the Simons Foundation Autism Research Initiative (SFARI) database (N = 393, syndromic and high confidence genes) [30] and found a significant overlap of orthologs in both hippocampus (N = 80/393) and blood (N = 62/393; Fig. 3D; Dataset S11). These results suggest that DMRs in both brain and blood have a molecular role in the disruption of neurodevelopment following a homozygous loss of Cntnap2. Indeed, this finding is similar to our previous data in the striatum of Cntnap2 homozygous knockout mice [12].

Distinct DMR-associated genes contribute to convergent pathways in brain and blood

Next, we compared the gene ontological terms obtained from both tissues (Dataset S12 and S13) and found a significant number of common terms between hippocampus and blood (N = 65 terms, P-value < 0.001), of which nearly half (30/65) were neuronal terms (e.g. brain/neuron development, Methods), such as dendritic spine organization, synapse assembly, and neurogenesis (Fig. 4A, Dataset S14). While 524 DMR-associated genes comprise these 30 common terms, only 11% of these genes (N = 60/524) have DMRs in both tissues. The majority of genes (88%) exhibited disruptions in only one tissue (hippocampal-specific: N = 289, blood-specific: N = 175; Fig. 4B, Dataset S15). These findings suggest that disruptions in convergent gene pathways related to ASD are driven by distinct DMR-associated genes in brain and blood.

Figure 4

Distinct DMR-associated genes from hippocampus and blood tissue contribute to convergent pathways. (A) A Venn diagram shows the overlap of common gene ontology (GO) terms (N = 65 terms, biological process) obtained using the total number of ASD-related hippocampus and blood DMR-associated genes (N = 3052 and N = 1870, respectively). A subset of these GO terms (N = 30) were associated with neurological GO terms (Methods) that comprise 524 DMR-associated genes. The asterisks (^***) denotes a significant overlap (hypergeometric enrichment, P-value < 0.001). (B) A pie chart depicts the number and percentage of the 524 DMR-associated genes that are found in both hippocampus and blood (“hippocampal/blood DMR-associated genes, N = 60, 11.4%), only found in the hippocampus (hippocampal specific DMR-associated genes, N = 289, 55.1%) and only found in blood (blood-specific DMR-associated genes, N = 175, 33.4%).

Methyl-sensitive transcription factors may regulate DMR-associated genes in the hippocampus and blood

Finally, we tested for significant enrichments of known methyl-sensitive transcription factor binding sequence motifs in the promoter region of the common DMR-associated genes (N = 60) from the 30 convergent neurological terms (Methods). This approach identified seven DMR-associated genes containing methyl-sensitive transcription factor binding sequence motifs within the DMRs of both tissues (Dataset S16). Four of these seven genes contained the same methyl-sensitive transcription factor binding sequence motif in the DMRs of both the hippocampus and blood, suggesting DNA methylation could be affecting the binding of common transcription factors required for gene expression in both tissues (Fig. 5). Notably, all seven of these genes have strong links to neurodevelopmental disorders (i.e. autism, intellectual disability, and epilepsy) [6, 31–36]. In most cases (85%) the DMRs are hypermethylated in the hippocampus and hypomethylated in blood, suggesting that common DMR-associated genes in brain and blood may have opposing functions with compensatory molecular effects. Together, these data support that ASD-related outcomes are driven by molecular contributions from tissues other than brain.

Figure 5

Genes containing identical methyl-sensitive transcription factors binding motifs in hippocampal and blood DMRs. Gene schematics of the 5′ end of Fgf13, Mrtfa, Knip2 and Cux2 are shown and indicate the relative location of the hippocampal (“Hipp-”) and blood-DMRs (black brackets). The relative location of each methyl-sensitive transcription factor binding site is depicted with triangles: HES1 (green), HES2 (yellow), HES5 (orange), TBX2 (pink), TFAPC2 (blue; see key). The orientation of each triangle indicates the DNA methylation status of each DMR (i.e. upside-down/above the gene = hypermethylated, right-side-up/below the gene = hypomethylated). Exons are represented with blue boxes and numbered. The transcription start site is indicated with an broken arrow. The relative size of each gene shown is indicated by the scale bar for each gene. If the relative distance between DMRs is too long to be represented, the length of the region not shown is indicated between slashes. Full gene structures are not shown. kb = kilobases.

Discussion

Here, we show evidence that a mouse model of autism (Cntnap2 homozygous knockout) exhibits stable changes in hippocampal DNA methylation levels that also are present in peripheral blood, and DMR-associated genes obtained from both tissues significantly overlap with well-known autism genes in humans. In addition, while disruptions in convergent neuronal pathways were found, these pathways were revealed by tissue-specific DMR-associated genes. Together, these findings may provide novel peripheral biomarkers of developmental brain disorders by providing new insight into the link between brain and blood epigenetic signatures that support blood as a promising surrogate for brain tissue.

The hippocampus is a brain region involved in the pathology of social behavior with high expression of Cntnap2 and was previously shown to be correlated with autism-related disorders (i.e. altered sociability and repetitive behaviors) in the Cntnap2 homozygous knockout mice [10]. We previously reported that Cntnap2 male homozygous knockout mice have a genome-wide disruption of 5-hydroxymethylcytosine (5hmC) in the striatum [12]. Comparison of these striatal data with the current hippocampal methylation (5mC) data identified a significant overlap of differentially hydroxymethylated and methylated genes (N = 407, hypergeometric enrichment, P-value < 0.05), suggesting the methylation findings described here may reflect a composite disruption of both 5mC and 5hmC, and that there are common DNA methylation changes in different brain regions of the Cntnap2 homozygous mouse model.

Differences in DNA methylation levels are expected between brain and blood tissues since they exhibit distinct characteristics and functions. The Cntnap2 homozygous knockout DMRs found in the hippocampus were predominantly (>85%) hypermethylated, whereas only 15% of the DMRs found in blood were hypermethylated, suggesting decreased DNA methylation in blood. These findings are consistent with previous studies reporting several hypermethylated genes in brain tissue of children with ASD [37–39], and decreased DNA methylation in cord and peripheral blood of children with ASD [40, 41]. While a previous study has shown less than 10% of DMRs are consistent between brain and blood in neurodevelopmental disorders [42], it and others examined less than 2% of the CpGs present in the genome [43], underscoring the power of using the whole genome methylation sequencing method employed here. This approach provided a deeper understanding of tissue-specific and genotype-specific DNA methylation levels. For example, global DNA methylation levels clustered samples by tissue type, whereas overlapping DMRs (N = 66) separated samples by genotype, and nearly 80% of them were altered in the same direction. We also found a significant number of DMR-associated genes in common between brain and blood tissues (N = 214), further demonstrating similarities between brain and blood and suggesting that Cntnap2 deficiency has a systemic molecular effect. Convergent neuronal gene ontology terms further implicate blood as a surrogate tissue for brain-related disorders. While differentially methylated genes associated with neuronal development have been previously observed in brain tissues from ASD individuals [44], finding blood-specific DMR-associated genes related to the same neuronal terms strengthens its claim as an ideal biological surrogate tissue.

Syndromic forms of neurodevelopmental disorders can involve the disruption of transcription factor function via epigenetic modifications, including DNA methylation [45, 46]. We report multiple methyl-sensitive transcription factor binding sequence motifs within DMR-associated genes in common between brain and blood tissues. All of these genes (Adgrb1, B3gnt2, Cux2, Fgf13, Knip2, Mrtfa, and Nrn1) have been associated with behavior-related alterations, including social deficits, seizure occurrence, intellectual disability, ASD, and schizophrenia [6, 31–36]. Moreover, some are differentially expressed (B3gnt2 and Kcnip2), or differentially methylated (Nrnr1) in surrogate tissues (i.e. peripheral blood, skin fibroblast, and placenta) derived from individuals with ASD [6, 33, 35]. Four of these seven genes contain the same methyl-sensitive transcription factor binding sequence motifs (i.e. HES1/2/5, TBX2, and TFAPC) in both tissues. These transcription factors play an important role in brain formation during embryogenesis, such as neural ectoderm development, neural stem cell maintenance, and astrocyte growth [47–49]. Together, these data suggest the loss of Cntnap2 influences the binding of key transcription factors and contributes to the observed ASD-like phenotype.

Blood-epigenetic signatures of autism and other neurodevelopmental disorders are of great interest [9, 40–42, 50, 51]. Although CNTNAP2 may have low expression in blood, the increased blood brain barrier permeability in Cntnap2 homozygous rodents [52] coincides with new lines of evidence that show altered pathways in brain are also predictors in blood, providing support for the importance of studying the peripheral-brain axis in the pathogenesis of brain related disorders [53, 54]. Thus, while the potentially low expression of CNTNAP2 in blood may not be driving blood-specific changes in DNA methylation levels, the altered DNA methylation levels observed in blood may be the consequence of Cntnap2-related neuronal disruptions that ultimately impact tissues other than brain. While it is well established that Cntnap2 plays an important role in nervous system development, by mediating interactions between neurons and glia, its function in blood (i.e. the immune system) has not been fully explored and should be the subject of future studies. Notably, the majority (>20%) of the gene ontological terms, associated with the hypomethylated DMR-associated genes in blood, are associated with lymphocyte differentiation. For example, 34 DMR-associated genes have a role in B-cell proliferation and differentiation, such as Bcl2, Cd27, and Lef1, supporting a role for Cntnap2-related differential methylation in blood. Furthermore, while the role of Cntnap2 on epigenetic mechanisms (if any) is unclear, it has been reported that DNA methylation is an important epigenetic mark for activity-dependent gene expression in neurons [55]. Accordingly, here and previously we have shown that the loss of Cntnap2 results in the modulation of DNA methylation levels on genes directly associated with epigenetic mechanisms (e.g. Dnmt3a, Dnmt3b, Tet3, and Mecp2) and synaptic plasticity (e.g. Grin2b, Reln, Homer, Gria1, Gabra2, and Shank3) [12]. Together, these DNA methylation level discrepancies between tissues may shed light on the different roles of Cntnap2 in brain and blood.

This study is an extension of our published findings in a female Cntnap2 mouse model of autism [56]. Here we report a large number of female blood-specific DMR-associated genes (N = 995) with a significant overlap with ASD-related genes (N = 47). Some of these blood-specific DMR-associated genes are also differentially methylated in the blood of individuals with ASD (e.g. Lcp1, Gabrb3, and Mecp2) [40, 41]. In fact, 114 of the blood-specific DMR-associated genes described here are also differentially methylated in the cord blood of female children with ASD [40]. Also, 175 of the blood-specific DMR-associated genes contributed to common neuronal pathways that were similarly altered in the brain. This finding is consistent with recent reports suggesting that epigenomic signatures in both brain and surrogate tissues provide specific clues about convergent gene pathways in the molecular pathogenesis of syndromic disorders [57]. Finally, blood may be particularly sensitive to changes on the X-chromosome, as there were 3.5 times more X-linked DMRs in the blood compared to hippocampus. With the high prevalence of X-linked disorders in males, this finding urges future studies in males to determine sex-specific correlations of DNA methylation signatures in brain and blood pertaining to the pathogenesis of ASD.

Conclusion

We employed an unbiased whole genome approach that has revealed an extensive list of blood-specific epigenetic signatures that may provide new insight into the systemic pathogenesis of ASD. We report common epigenetic signatures in brain and blood that influence convergent neuronal pathways. However, the majority of the DMR-associated genes that comprise the common neuronal pathways are tissue-specific. These findings reinforce and support the paradigm that a single mutation (i.e. loss of Cntnap2) leads to DNA methylation changes throughout the entire genome in multiple tissues. While it remains unclear whether blood-specific changes in DNA methylation are a cause or consequence of the observed autism-related outcome in this model, these data suggest that blood has molecular signatures associated with autism that are both unique and common to those in the brain, and further supports that ASD is a systemic disorder.

Acknowledgements

We thank members of the University of Wisconsin-Madison Research Animal Resources and Compliance and Phillip Bergmann for manuscript edits.

Conflict of interest statement

None declared.

Funding

This work was supported in part by the University of Wisconsin-Madison Department of Neurological Surgery and the American Association of University Women (AAUW).

Data availability

We declaredall the data is available and present in SRA (Sequence Read Archive) and the accession number is located on our lab website (https://www.neurosurgery.wisc.edu/research/alisch/).