Comparative DNA methylation among females with neurodevelopmental disorders and seizures identifies TAC1 as a MeCP2 target gene

Genomic DNA from females with RTT or seizure disorder (SEZ) was obtained from the Coriell Cell Repository[11]. Genomic DNA from age-matched females with autism (AUT) or controls (CTLs) was selected from the NIMH Center for Collaborative Genetic Studies on Mental Disorders Autism Pedigrees v5.0[12]. All 60 DNA samples (Additional file1: Table S1) used for methylation assays originated from LCLs. Fresh-frozen postmortem brain samples were obtained through the Autism Speaks-supported Autism Tissue Program at the Harvard Brain Tissue Resource Center[13] or the Eunice Kennedy Shriver National Institute of Child Health and Human Development Brain and Tissue Bank at the University of Maryland School of Medicine[14]. Tissue samples of the striatum, cingulate and temporal cortices were obtained from five RTT females and five CTL females (Additional file1: Table S3). Additional samples of cingulate and temporal cortices were obtained from five AUT females and three additional CTL females.

Genomic DNA was bisulfite modified using the EZ DNA Methylation Kit (Zymo Research, Orange, CA, USA) and hybridized to the Illumina (San Diego, CA, USA) Infinium HumanMethylation27 BeadChip by the USC Epigenome Center as described elsewhere[15, 16]. To evaluate array fidelity, DNA aliquots for six samples per clinical group from the discovery cohort were included in the replication cohort. The β values (0 to 1.0, where 0 represents no methylation and 1 represents complete methylation) for each CpG site were calculated as the methylated signal intensity divided by the sum of the methylated and unmethylated signals. Loci that possibly contained SNPs and/or repetitive elements were removed, leaving 21,583 probes representing CpG sites in 12,936 genes for analysis. Two AUT samples were excluded due to sex discrepancy (Additional file1: Figure S1). No additional major chromosomal abnormalities were apparent. Copy number variants were previously reported for 11 samples included in this study (Additional file1: Table S4).

The validation experiment included 56 samples from the Illumina discovery and replication cohorts. Another LCL DNA aliquot was bisulfite modified using the EZ DNA Methylation Kit (Zymo Research) by the USC Epigenome Center, according to the manufacturer’s protocol. Genomic DNA was isolated from gray matter dissected from the cingulate cortex using the Wizard Genomic DNA Purification Kit (Promega, Madison, WI, USA) and was bisulfite modified. Sequenom primers (Additional file1: Table S5) were designed to target the region that overlapped with Illumina probe location using the EpiDesigner BETA software[17]. The Sequenom (San Diego, CA, USA) MALDI-TOF mass spectrometry EpiTYPER Assay was performed by the Vanderbilt University Center for Human Genetics Research DNA Resources Core according to the manufacturer’s protocol.

Total RNA was isolated from LCLs using the Illustra TriplePrep Kit (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, USA) according to the manufacturer’s protocol. Total RNA was isolated from gray matter-dissected brain tissue or HEK cell pellets using the RNeasy Plus Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s protocol. RNA quality was determined using an Agilent (Palo Alto, CA, USA) Bioanalyzer 2100 system; the RNA integrity number was >8 for all samples. cDNA was synthesized using the SuperScript III First Strand cDNA Synthesis Kit (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s protocol using oligo dT primers. Quantitative RT-PCR was performed in triplicate for each sample using the Power SYBR Green PCR master mix (Life Technologies Corp., Carlsbad, CA, USA) and target-specific primers (Additional file1: Table S4) according to the manufacturer’s recommendations. Assays were analyzed using a CFX96 Real-Time PCR detection system (Bio-Rad, Hercules, CA, USA). The delta cycle threshold (ΔCt) of target relative to ACTB was averaged per sample.

The EZ-Magna ChIP A/G Chromatin Immunoprecipitation Kit (EMD Millipore Corporation, Billerica, MA, USA) was used with slight modifications to the manufacturer’s protocol, as described in Additional file1.

lllumina probes with mean |Δβ| ≥0.15 between pairs of clinical groups within the discovery cohort were first selected for analysis (n = 391). Analysis of variance (ANOVA) and the false discovery rate (FDR) were performed to test the association between clinical group and DNA methylation for the selected probes. Illumina probes with P <0.01 and FDR <0.10 in the discovery cohort were evaluated in the replication cohort and in a pooled analysis that included all samples. Hierarchical clustering analysis (Ward’s method) using dissimilarity matrices (1 – Pearson’s correlation) and principal component analysis were performed to identify subjects with similar DNA methylation patterns. Statistical analyses were performed using R software[18].

Tukey post-hoc analysis was performed to identify pair-wise associations between clinical groups and DNA methylation for the probe that was significantly differentially methylated among clinical groups (cg14224417; TAC1). ANOVA and Tukey post-hoc tests were performed for each TAC1 promoter CpG site measured by the Sequenom assay. t tests were used to assess the significance of TAC1 expression (ΔCt) stratified by clinical group for each brain region and to assess the treatment effect in cell lines. To investigate the association between DNA methylation and mRNA expression in the cingulate cortex, stratified by clinical group, we fitted linear regression models. Statistical analyses were performed using SPSS version 18.0.3 software (IBM, Somers, NY, USA).

Genome-wide DNA methylation profiling using LCLs from females diagnosed with a neurodevelopmental disorder and age-matched controls was performed in two independent subject cohorts: a discovery cohort (n = 20) and a replication cohort (n = 40). The discovery cohort contained LCL samples from females diagnosed with RTT (n = 10), AUT (n = 4), or CTL (n = 6) (Additional file1: Table S1). In discovery cohort samples, most probes had low DNA methylation levels (median = 0.04 to 0.06; mean = 0.21 to 0.24; range = 0.01 to 0.99). Correlations among discovery cohort samples were similar across probes (Pearson’s r >0.98), consistent with previous reports[19‐24].

To identify robust DNA methylation differences (Δβ), we further analyzed all probes with pair-wise mean |Δβ| ≥0.15 among clinical groups within the discovery cohort. This threshold generated a normally distributed dataset containing 391 probes (Additional file1: Table S2). Among these probes, 127 were differentially methylated between RTT and CTL, 104 between RTT and AUT, and 261 between AUT and CTL (Figure 1A). No probes were differentially methylated among all three pair-wise comparisons, but 53 probes were differentially methylated in either RTT or AUT compared with CTL. Interestingly, 42 probes that were differentially methylated between RTT and AUT and between AUT and CTL were opposite in magnitude. Twenty-six probes hypermethylated in RTT compared with AUT were hypomethylated in AUT compared with CTL, and 16 probes hypomethylated in RTT versus AUT were hypermethylated in AUT versus CTL. The distribution of hypermethylation and hypomethylation loci among comparison pairs was highly significant (χ²(2-df) = 14.17, P = 0.0008). RTT was hypermethylated compared with CTL and AUT at 102 (80%) and 74 (72%) loci, respectively. AUT was also hypermethylated compared with CTL (62%). The distribution of nonoverlapping loci hypermethylated (72%) or hypomethylated (54%) in AUT compared with CTL was significantly different than loci hypermethylated (82%) or hypomethylated (18%) between RTT and CTL (χ²(1-df) = 13.96, P = 0.0002).

We performed ANOVA for 391 differentially methylated probes identified in the discovery cohort (Additional file1: Table S2). Hierarchical clustering using the top 40 significant probes (P <0.01, FDR <0.10) suggested discovery samples could be classified based on their DNA methylation profiles (Figure 1B). Principal component analysis further suggested three distinct clusters using the same 40 probes (Figure 1C). DNA methylation among the 40 probes suggested more homogeneity in RTT (mean Pearson’s r = 0.93) than among the samples in comparison groups (mean Pearson’s r = 0.74 and 0.83 for AUT and CTL groups, respectively).

Because seven of the 10 RTT females in this study were reported to have abnormal brain activity (seizures (n = 5) or subclinical abnormal EEG activity (n = 2)], a common feature of RTT[25], we included an additional subject cohort in the replication experiment. We reasoned that this would address the possibility that AEDs, which can alter DNA methylation[26, 27], might underlie some of the differential DNA methylation observed in the discovery cohort. The replication cohort contained LCL samples from females with RTT (n = 8), AUT (n = 13), SEZ (n = 6), or CTL (n = 13) (Additional file1: Table S1). As measured in the discovery cohort, most probes had low DNA methylation levels in these samples (median = 0.06 to 0.12; median = 0.21 to 0.26; range = 0.01 to 0.99). Correlations among samples were high (Pearson’s r >0.90), but less similar than samples within the discovery cohort. To ensure technical fidelity across assay runs, replicates from the discovery cohort were included in the DNA methylation analysis of the replication cohort, with nearly perfect correlation between replicate pairs (Pearson’s r >0.99) (Additional file1: Figure S2). The 40 probes that distinguished clinical groups in the discovery cohort analysis similarly distinguished technical replicates, with high correlation between replicate pairs (Pearson’s r >0.98) (Additional file1: Figure S3). Duplicate samples were excluded from the replication cohort analysis.

In the replication cohort, hierarchical clustering using the 40 probes identified in the discovery cohort did not produce clustering of DNA methylation patterns by clinical group (Additional file1: Figure S4A). Principal component analysis also failed to reveal distinct clinical clusters among the replication samples (Additional file1: Figure S4B), suggesting greater molecular heterogeneity among the clinical samples included in the replication cohort.

To select particular loci with differential DNA methylation among clinical groups, we performed ANOVA for the 40 probes from the discovery cohort analysis using DNA methylation values from the replication cohort samples and in a pooled analysis (Table 1). We then selected probes below the conservative threshold of P <0.001 and FDR <0.05 from each analysis. In the discovery cohort, five loci showed significant differential DNA methylation (HOXA11, TCN1, PTCD2, TAC1, ETNK2). In the replication cohort, no locus was significant. In the pooled analysis, four loci were highly significant (PTCD2, TAC1, FLJ44881, SPAG7; P <0.0005, FDR <0.005). Overall, no locus emerged across these three analyses. However, TAC1 differential methylation approached significance in the replication analysis (P = 0.003, FDR = 0.13), and was highly significant in both discovery and pooled analyses (P <0.0005, FDR <0.05). TAC1 was previously implicated in RTT pathophysiology through convergent findings in RTT patients and mouse models[28‐32]. We thus selected TAC1 to target for validation.

A Sequenom EpiTYPER assay was designed to quantify DNA methylation within the TAC1 promoter CpG island (Figure 2A). First, we determined whether DNA methylation levels detected at specific CpG sites within the Sequenom target region validated the Illumina assay results. Sequenom DNA methylation levels at two CpG sites located closest to the Illumina probe were highly correlated within samples (Additional file1: Figure S5). Next, we evaluated whether significant differential methylation occurred among clinical groups using the Sequenom CpG methylation values. The clinical group had a significant main effect on DNA methylation at TAC1 CpG site 5 (ANOVA F_(3,52) = 5.70, P = 0.002). DNA methylation levels at this site had the strongest correlation with Illumina probe methylation values and validated the TAC1 results obtained from the genome-wide analysis.

To determine which clinical groups exhibited TAC1 differential methylation, we performed Tukey post-hoc pair-wise comparisons for discovery, replication, and pooled cohorts using Illumina and Sequenom data (Figure 2B). In the Illumina dataset, TAC1 was hypomethylated in RTT compared with AUT (P = 0.003) in the discovery cohort, and compared with SEZ (P = 0.008) in the pooled analysis. AUT was significantly hypermethylated relative to CTL in discovery (P = 0.0003) and pooled analyses (P = 0.004). SEZ was significantly hypermethylated versus CTL in replication (P = 0.001) and pooled analyses (P = 0.0001). Sequenom data confirmed significant AUT hypermethylation versus CTL in discovery (P = 0.02) and pooled analyses (P = 0.04), and SEZ hypermethylation versus CTL in replication (P = 0.01) and pooled analyses (P = 0.002). Overall, SEZ hypermethylation relative to CTL emerged with the most significant differential DNA methylation in both Illumina and Sequenom datasets.

Since DNA methylation at CpG islands proximal to gene promoter regions is an important mediator of gene transcription[33], we evaluated whether TAC1 promoter DNA methylation differences were associated with differential expression. Because TAC1 is only weakly expressed in lymphocytes[34, 35], evaluating expression differences in LCLs among clinical groups was uninformative (Additional file1: Figure S6). However, if altered DNA methylation observed in LCLs occurred early in development, then differential DNA methylation detected in LCLs might exist in additional tissues. TAC1 is expressed in brain throughout development[36, 37], and the brain is the primary tissue affected in neurodevelopmental disorders. We thus examined whether TAC1 DNA methylation status from a peripherally derived tissue correlates with central gene expression. DNA methylation analysis revealed significant TAC1 hypermethylation in SEZ versus CTL. Given that seizures commonly occur in RTT[25], we obtained postmortem brain samples from RTT females. TAC1 expression was assayed in three brain regions from age-matched RTT and CTL females (Additional file1: Table S3) using quantitative RT-PCR (Figure 3A). TAC1 expression was significantly reduced in RTT cingulate cortex (t(6) = 2.98, P = 0.03), with a trend for reduced expression in temporal cortex (t(8) = 2.22, P = 0.06). No differences in TAC1 expression were detected in striatum (t(7) = 0.54, P = 0.61). Given these results, we examined TAC1 expression in additional postmortem brain samples obtained from female subjects diagnosed with AUT and for which the patient’s cause of death was ascribed to SEZ[38] (Figure 3B). Indeed, TAC1 expression was also reduced in AUT+SEZ cingulate cortex, approaching significance (t(5) = 2.29, P = 0.07). No significant difference in TAC1 expression was detected in AUT temporal cortex (t(7) = 0.98, P = 0.36). A combined analysis using all brain samples and comparing TAC1 expression in the cingulate cortex suggested a significant TAC1 reduction in affected brains (t(12) = 3.67, P = 0.003) (Figure 3C).

Several studies have reported differential DNA methylation among brain regions in postmortem tissue[39, 40]. To determine whether DNA methylation at the TAC1 promoter varied among brain regions, we selected Illumina DNA methylation data (cg14224417) assayed in four brain regions from female samples in a previous study[39]. The TAC1 promoter displayed significant differential DNA methylation among certain brain regions in females (Additional file1: Figure S7).

To determine whether the reduced TAC1 expression observed in cingulate cortex of seizure phenotype (RTT and AUT+SEZ) brains may be associated with differential DNA methylation, we extracted DNA from the same brain samples used for the expression studies and evaluated TAC1 DNA methylation using the Sequenom EpiTYPER assay. As expected, an inverse correlation between TAC1 promoter DNA methylation and expression was observed for both seizure phenotype and CTL cingulate cortex samples (Figure 3D). However, no significant differential DNA methylation was detected at TAC1 CpG site 5 between seizure phenotype and CTL cingulate cortex samples (t(16) = −0.71, P = 0.49), nor for any of the other CpG sites in the TAC1 promoter locus (data not shown). DNA methylation at TAC1 CpG site 5 was also not a significant predictor of TAC1 expression in the cingulate cortex of CTL brains (β = −6.09, P = 0.78). However, in this small sample, there was a trend suggesting DNA methylation at TAC1 CpG site 5 may predict TAC1 expression in the cingulate cortex of seizure phenotype brains (β = −6.40, P = 0.09).

RTT is mostly caused by MECP2 coding mutations[6]. More recently, increased MECP2 expression was detected in temporal cortex of patients with intractable temporal lobe epilepsy[41]. To determine whether TAC1 is a MeCP2 target in vivo, we performed chromatin immunoprecipitation assays in HEK cells, which express high levels of TAC1[34, 35]. Anti-MeCP2 specifically precipitates TAC1 DNA, while control IgG yields no enrichment of binding to the TAC1 sequence (Figure 4A), supporting direct binding of MeCP2 to the TAC1 promoter region (Figure 2A). We further examined the possibility that DNA hypermethylation detected at the TAC1 promoter in seizure-associated phenotypes was due to AED treatment. VPA is an anticonvulsant, commonly administered for seizure control in RTT, and a potent histone deacetylase inhibitor that can directly alter DNA methylation in vitro[26]. Following VPA treatment of HEK cells, anti-MeCP2 failed to precipitate TAC1 DNA, whereas anti-acetylated histone 3 precipitates more DNA, as expected (Figure 4A, B).

DNA methylation is one of several key mechanisms through which gene expression is controlled. Analysis of DNA methylation patterns has been the focus of comparative studies in normal and malignant tissues[43], biopsied samples from peripheral organs[44‐46], and in white blood cells from control and disorder-diagnosed populations[21, 47, 48]. This strategy is a challenge particularly with neurological disorders, because molecular assays of brain are often not possible in patients. Several studies have examined DNA methylation patterns in postmortem brain tissue from control subjects and those with a psychiatric disorder diagnosis[49, 50], but the relationship between peripheral and central measures of DNA methylation are only beginning to emerge[45]. Moreover, there is variability in the fidelity of different assays used in studies, and few studies use secondary methods to validate alternations in specific DNA methylation sites. The 27K chip has been shown to be highly reliable and sensitive across assay runs[19, 23], and we showed in the present study that replicate samples run at different times, independent of diagnostic group, are >99% identical in terms of global DNA methylation patterns. One caveat of our study was the use of the 27K chip, the only validated chip at the time we initiated the studies. This chip has high fidelity and samples CpG sites across the genome, although it only probes approximately one-half of the genes in the genome. Future studies using assays with greater coverage will therefore possibly reveal new patterns of DNA methylation not recognized here. An additional technical caveat is the use of patient LCLs. These cells provide an opportunity to assess patient-derived samples that are commonly accumulated in repositories in sufficient numbers for the study of rare neurological disorders. However, the transformation and growth processes of these cell lines may lead to epigenetic changes not present in the primary patient cells[51]. These challenges notwithstanding, the present study emphasizes the importance of controlling for complex clinical phenotypes, such as seizures. The data presented here suggest that the state-dependent status of DNA methylation patterns, as well as central and peripheral differences, may make it challenging to detect disorder-specific patterns that consistently reflect categorical differences across samples.

We initially focused on RTT due to the importance of MeCP2 in regulating transcription through binding to methylated DNA. The possibility of altered DNA methylation in RTT had not previously been investigated in clinical samples. We hypothesized that the altered MeCP2 function in RTT might lead to compensatory changes in the DNA methylation status of certain genes. In fact, the initial sample illustrated that using LCLs from patients, analyzed using cluster and principle component methods of the top differentially methylated genes, distinguished the clinical groups from CTL. More than double the number of CpG sites showed differential DNA methylation between AUT and CTL (n = 261) than between RTT and CTL (n = 127). At first glance, these results were surprising given our hypothesis that RTT would display the greatest differential DNA methylation from CTL because of the involvement of MeCP2 in this disorder. However, the more limited differences in the first RTT cohort are not unprecedented. For example, in another neurodevelopmental disorder, Cornelia de Lange syndrome, fewer CpG sites demonstrated differential methylation than the number of genes with expression differences[52, 53]. Expression profiling in AUT has been more compelling than in RTT. Hundreds of differentially expressed genes in LCLs and brain from AUT subjects have been reported[54, 55], while global gene expression analyses have failed to reveal any dramatic differences in RTT compared with control subjects using a variety of tissues[56‐61]. The present data showing an inverse correlation between DNA methylation and TAC1 expression, despite the reduced TAC1 expression in cortex, may be a reflection of the small sample size and/or the complex relationship between transcriptional regulation and gene expression.

The initial promise of categorizing clinical groups by DNA methylation pattern from the first cohort did not replicate in a second, larger sample that included LCLs from individuals with SEZ. DNA methylation status is highly influenced by nonheritable factors, including physiological state[62], medications[26] and even diet[63]. We reasoned that since the presence of seizures in neurodevelopmental disorders is common and a hallmark of RTT, it could serve as a unifying feature in the clinically diagnosed populations. In fact, using LCLs from subjects with SEZ and postmortem brain tissue from AUT cases in which seizures were noted in the medical history, or was the cause of death, we show that DNA methylation patterns of specific genes may be more related to seizure status, a finding that has not been previously reported. The repositories that provided the LCLs and postmortem brain samples do not collect sufficient information regarding medication status. We can therefore only suggest that contributions to DNA methylation status at certain sites may be due in part to drug treatment effects. To investigate this possibility, VPA-treated HEK cells showed dramatic changes in DNA methylation of the TAC1 gene; additional sites were probably also impacted by VPA. Interestingly, VPA treatment altered MeCP2 binding to the TAC1 promoter, which suggests that this putative regulatory interaction may be responsible in part for the consistent reduction of TAC1 expression in cortical brain regions that we assayed, although not in striatum.

The TAC1 DNA methylation and gene expression changes were not expected, although this gene has been a candidate target in RTT due to the finding of reduced expression in Mecp2 mutant mice[31, 32]. The present study is the first to identify molecular changes of TAC1 in brain samples from RTT patients. TAC1 gene expression changes were statistically significant in the cingulate cortex, with a trend that almost reached significance in the temporal cortex given the limited sample size; other brain regions should be evaluated in the future. It is noteworthy that TAC1 expression changes were not observed in the striatum, indicating that seizure pathophysiology and AEDs do not produce global, nonspecific effects throughout the brain. The alterations in the cingulate cortex are of functional interest, due to the role of this cortical region in executive function and complex multisensory processing[64‐66], both of which are disrupted in AUT and SEZ[67, 68]. The TAC1 gene encodes neurokinins and substance P. These neuromodulators are expressed in nociceptive primary sensory neurons, and serve as modulators of pain perception and inflammation[69]. In the neocortex, substance P is an excitatory neuromodulator of projection neuron activity[70]. Furthermore, there are several convergent findings related to expression of this peptide and RTT. First, substance P is reduced in the cerebrospinal fluid of RTT patients[28]. Second, substance P immunoreactivity is reduced in brains of RTT patients compared with age-matched controls[30], although not in the bowels (enteric nervous system)[29]. Third, Tac1 expression is reduced in the hypothalamus of juvenile male Mecp2 mutant mice[31] and in adult male mice with postnatal Mecp2 loss[32]. Finally, respiratory abnormalities are a clinical hallmark of RTT[71] and neurons in the rhythm-generating center of the respiratory network are dynamically regulated by substance P[72]. The discovery of TAC1 deficiency in RTT thus adds to other convergent lines of evidence implicating the neuropeptides in RTT pathophysiology. Further consideration of substance P and other neuropeptides transcribed from the TAC1 locus as therapeutic targets in RTT and AUT+SEZ is warranted.