Functional Genomics of Serotonin Receptor 2A (HTR2A): Interaction of Polymorphism, Methylation, Expression and Disease Association
- First Online:
- Cite this article as:
- Falkenberg, V.R., Gurbaxani, B.M., Unger, E.R. et al. Neuromol Med (2011) 13: 66. doi:10.1007/s12017-010-8138-2
- 991 Downloads
Serotonergic neurotransmission plays a key role in the pathophysiology of neuropsychiatric illnesses. The functional significance of a promoter polymorphism, −1438G/A (rs6311), in one of the major genes of this system (serotonin receptor 2A, HTR2A) remains poorly understood in the context of epigenetic factors, transcription factors and endocrine influences. We used functional and structural equation modeling (SEM) approaches to assess the contributions of the polymorphism (rs6311), DNA methylation and clinical variables to HTR2A expression in chronic fatigue syndrome (CFS) subjects from a population-based study. HTR2A was up-regulated in CFS through allele-specific expression modulated by transcription factors at critical sites in its promoter: an E47 binding site at position −1,438, (created by the A-allele of rs6311 polymorphism), a glucocorticoid receptor (GR) binding site encompassing a CpG at position −1,420, and Sp1 binding at CpG methylation site −1,224. Methylation at −1,420 was strongly correlated with methylation at −1,439, a CpG site that is dependent upon the G-allele of rs6311 at position −1,438. SEM revealed a strong negative interaction between E47 and GR binding (in conjunction with cortisol level) on HTR2A expression. This study suggests that the promoter polymorphism (rs6311) can affect both transcription factor binding and promoter methylation, and this along with an individual’s stress response can impact the rate of HTR2A transcription in a genotype and methylation-dependent manner. This study can serve as an example for deciphering the molecular determinants of transcriptional regulation of major genes of medical importance by integrating functional genomics and SEM approaches. Confirmation in an independent study population is required.
KeywordsChronic fatigue syndrome Serotonergic neurotransmission Serotonin receptor 2A (HTR2A) Polymorphism Gene expression DNA methylation Cortisol Structural equation modeling
Regulatory genetic and epigenetic variations constitute major determinants of differential gene expression via the loss or gain of transcription factor binding or DNA methylation sites. Case–control studies have identified a number of functional polymorphisms, such as single-nucleotide polymorphisms (SNPs) and variable number tandem repeats (VNTRs), in the promoter regions of medically important genes including a group of genes involved in serotonergic signaling, the serotonin transporter (SLC6A4), monoamine oxidase A (MAOA) and serotonin receptor 2A (HTR2A). Functional polymorphisms in all three of these genes have been associated with major depression, chronic fatigue syndrome (CFS) and post-traumatic stress disorder (PTSD), as well as a variety of other mood disorders (Li and He 2008; Smith et al. 2008; Uher and McGuffin 2008).
Current studies to evaluate functional polymorphisms have generally been restricted to either in vitro reporter gene assays, gel-shift assays to assess DNA–protein interactions, or allele-specific analysis of mRNA (Knight 2005). DNA methylation analyses are usually carried out independently from analyses of genetic polymorphisms and gene-environmental interactions (Gopalakrishnan et al. 2008). Because gene expression is a multi-factorial process, isolated genomic studies rarely result in a defined molecular mechanism or integrated biological model to explain these genetic associations with disease. Simultaneous examination of genetic and epigenetic variations in the context of environmental and developmental responsive signaling molecules (for example, cortisol secretion in response to stress) may provide a better assessment of their influence on gene expression and clarify their pathogenic role. We employed an integrated experimental and computational approach to assess the influence of genetic polymorphisms and DNA methylation on HTR2A expression in peripheral blood mononuclear cells (PBMC). HTR2A is implicated in the regulation of serotonergic neurotransmission and the hypothalamic-pituitary-adrenal (HPA) axis (Lanfumey et al. 2008; Lister-Williams et al. 1998; Porter et al. 2004). Subjects were derived from a population-based study of CFS, a complex disease of unknown etiology but with major etiologic roles hypothesized for both HPA-axis-mediated cortisol signaling (hypocortisolism) and serotonergic neurotransmission (Cleare 2003; Cleare et al. 1995; Heim et al. 2000; Smith et al. 2008).
To integrate and evaluate the interactions between polymorphisms, methylation, stress hormones (e.g. cortisol) and gene expression, we applied structural equation modeling (SEM), a statistical approach that has been widely used in the psychosocial sciences (Schreiber 2008). SEM allows the user to design latent variables that can be connected to various observed data points. The algorithm then does model fitting to find the path coefficients and error variances that best describe the observed data based on the latent variables and pathways in the model. This allows the user to assess the contribution of various observed data points to the outcome in the best-fit model.
Our results reveal that changes in HTR2A expression are associated with allele-specific expression differences modulated by both genetic polymorphisms and DNA methylation at critical transcription factor binding sites (TFBS). This computational approach to integrate regulatory genetic polymorphisms, epigenetic variations and gene expression changes is a novel approach to decipher the molecular determinants of disease and will have applications to other complex illnesses.
This study adhered to human experimental guidelines of US Department of Health and Human Services and the Helsinki Declaration. The CDC Human Subjects committee approved the study protocol, and all subjects gave written informed consent.
Subjects and Illness Classification
Subjects from a population-based CFS surveillance study in Wichita, KS, USA were recruited for a previously described 2-day in-hospital case–control study (Vernon and Reeves 2006). Subjects were classified according to the 1994 research case definition as evaluated by standardized questionnaires (Reeves et al. 2005). Of the initial 227 subjects recruited, 13 did not have genotype information for rs6311 (−1438G/A) and were dropped from this study leaving 214 enrollees (rs6311 genotypes: 83 GG, 95 GA, and 36 AA). Methylation analysis was done on 185 subjects with sufficient residual DNA (71 GG, 84 GA, 30 AA). Mean age and body mass index (BMI) of these 185 subjects were 50 ± 8.8 years and 29 ± 4.8, respectively. Ninety-four percent of subjects were Caucasians, and 82% of subjects were female in this analysis.
Demographic analysis of CFS and NF subjects in the study
NF (n = 49)
CFS (n = 39)
Age (mean ± SD)
50.53 ± 8.3
50.59 ± 9.2
BMI (mean ± SD)
28.4 ± 4.7
29.8 ± 4.5
Blood collection and separation of PBMCs from this study population were described previously. Both DNA and RNA were extracted from the same PBMC sample by Trizol DNA/RNA extraction protocol (Invitrogen, Carlsbad, CA). Methods to assess the quality and quantity of DNA and RNA preparations were described earlier (Sorensen et al. 2009).
Genotyping and Methylation Analyses
Primer sets for quantitation of site-specific HTR2A CpG methylation by pyrosequencing
−988, −983, −949
−314, −310, −253
Gene Expression Analysis
The −1438G/A (rs6311) and C102T (rs6313) polymorphisms are in complete linkage disequilibrium. This information was used to design real-time reverse-transcription PCR (RT-PCR) and pyrosequencing assays encompassing the C102T exonic marker. Real-time RT-PCR was used to determine total HTR2A expression in 117 subjects of whom 31 were CFS and 15 were NF subjects. Real-time RT-PCR was conducted as described previously using Light Cycler (Sorensen et al. 2009) except that it generated a 197 bp HTR2A product using the biotinylated forward (AGCTCAACTACGAACTCCCTAATG) and reverse (TCCAGTTAAATGCATCAGAAGTGT) primers. HTR2A primers were annealed at 62°C resulting in PCR efficiency of 1.96. The primers and annealing temperature for peptidylpropylisomease B (PPIB), internal reference gene for normalization, were reported previously (Sorensen et al. 2009). The biotinylated LightCycler product was used directly for pyrosequencing with the sequencing primer ATCAGAAGTGTTAGCTTCTC to analyze the sequence CRGAGTTAAAGTCATTACTGTAGAGCC where R corresponds to the allelic ratio in mRNA for the HTR2A C102T marker. Pyrosequencing was used to determine allele-specific expression in all 22 subjects from the disease-specific expression analysis that was heterozygous for the −1438G/A marker (14 CFS and 8 NF).
Urinary 24-h free cortisol was measured as reported previously (Smith et al. 2009).
All statistical analyses were done with the Graphpad Prism program (Graphpad Software, CA). Age, BMI and expression data were normally distributed and were analyzed by student’s t tests, and sex and race were analyzed by Chi-square tests. Mann–Whitney U test was used to compare methylation levels between groups. Spearman correlations were used to assess correlation between methylation levels of different CpG sites. Primarily, we used the more conservative Bonferroni correction for multiple hypothesis testing, and associations that did not withstand the Bonferroni correction were tested again by the less conservative false discovery rate (FDR; Reiner et al. 2003). SEM was performed using SSI’s LISREL 8.8 software program. Analysis used 46 carriers of the A-allele (13 AA, 33 GA) for rs6311, who have data on HTR2A expression and methylation (at least two of −1,439, −1,420, and −1,224 CpG sites) as well as urinary 24-h free cortisol values. Total expression of HTR2A was used for the outcome variable “HTR2A expression”. SEM was performed on z-score transformed data with missing data points imputed using the PRELIS software supplied in the LISREL software suite (15 missing values/330 data points). Path diagrams were designed and run using SIMPLIS syntax, and to increase the degrees of freedom, the error variances of the methylation data were set equal to the calculated technical variance of the methylation assay after z-score normalization of the data (0.1). The P value of the presented model, 0.98 is close to the best-fit value of 1.0 indicating that this model fits the observed data very well.
DNA Methylation Varies Along HTR2A Promoter
Methylation levels varied along the HTR2A promoter (Fig. 2) with extremely low methylation (mean 6.5%; range 2.9–9.4%) at seven CpG sites (−1,065 to −949). Methylation levels increased in both directions from this region. CpG sites −1,439, −1,420 and −1,224 in the distal region had average methylation levels of 53, 63 and 60%, respectively. CpG sites at −817, −753 and −721 had average methylation levels of 27, 34 and 47%, respectively. The four most proximally located CpG sites, spanning a 190 bp region, were the most highly methylated sites in the HTR2A promoter with average methylation levels of 80% at −314, 73% at −310, 69% at −253 and 89% at −125 (Fig. 2).
Methylation levels at three out of 17 CpG sites (−1,224, −983, and −314) differed significantly between CFS and NF control subjects (Fig. 2). At −1,224, NF subjects had significantly higher methylation than CFS subjects (P = 0.0036), whereas at CpG sites −983 and −314, NF subjects showed lower methylation than CFS subjects (P = 0.04 for both sites). None of these associations remained significant after the conservative Bonferroni correction for multiple testing (Bonferroni corrected alpha = 0.003). However, multiple hypothesis testing correction by false discovery rate (FDR 33%, alpha = 0.058) indicated that while one of these three associations could have arisen by chance the other two are likely to be true positives.
Since CpG site −1,439 is lost as a result of G to A sequence variation (SNP rs6311) at −1,438, −1438G/A genotype-dependent methylation at CpG sites −1,439, −1,420 and −1,224 was examined in all subjects for whom data were available (Fig. 3 top panel). Methylation at −1,439 was very strongly dependent upon the −1438G/A genotypes (GG 87.5%, GA 43%, AA 0%; r2 = 0.999), as expected. Methylation of the adjacent CpG site −1,420 was also slightly genotype-dependent, but the association was significant only in comparison with the AA genotype (P < 0.00085, Bonferroni corrected alpha = 0.0006). No genotype-dependent effect on methylation was observed on the remaining 15 CpG sites.
Genotype-dependent methylation also differed between CFS and NF subjects (Fig. 3 bottom panel). CFS subjects with GG genotype had significantly higher methylation at −1,420 (a CpG site 19 bp downstream from the SNP site) than NF subjects with GG genotype (P = −0.038). The association between methylation at −1,420 between GG and AA genotypes was significant only for CFS subjects (P = 0.0007, Bonferroni corrected alpha = 0.003). To further investigate CFS-associated methylation differences, pairwise Spearman correlation matrices were created for all 17 CpG sites for CFS and NF subjects separately. Methylation levels at positions −1,439 and −1,420 were correlated significantly in CFS subjects (P = 0.000000076, Bonferroni corrected alpha = 0.00003) but not in the NF subjects (P = 0.37) reflecting the differential genotype-dependent methylation that is associated with CFS.
Allele-Specific Expression of HTR2A
When analysis was restricted to those classified as CFS or NF (31 CFS and 15 NF) and stratified by disease status, the relative expression of HTR2A was 1.56-fold higher in CFS compared to NF subjects (P = 0.01; Fig. 4b). For the 22 heterozygous subjects (rs6311 GA) in this group (14 CFS, 8 NF), the allele-specific expression is shown in Fig. 4c. The A-allele contributed a higher percentage of total expression than the G-allele in CFS subjects (P = 0.019), whereas the opposite was true (though not significant) for NF subjects (P = 0.064). This disease-dependent differential expression of the A-allele was also seen in the overall expression of HTR2A, in that CFS subjects with the AA genotype had significantly higher expression of HTR2A than NF subjects with AA (P = 0.036, 6 CFS, 5 NF; data not shown). This disease-specific difference may contribute to the large standard error for the AA genotype in the analysis with all 117 subjects.
Modeling Transcriptional Regulation of HTR2A
Alternative models were also examined, and fitting the data to a model that includes the link between GRE and E47 without the direct link between GRE and HTR2A expression has slightly better statistics, e.g. the AIC = 20.58, further strengthening the suggested role for GRE. In order to decrease the risk of overfitting of the alternative model discussed earlier and the model shown in Fig. 6, the residual variance (i.e. the measured variance not contributing to the interaction of the latent variables) of four of the five measured variables (not cortisol) was manually set to 0.10, in keeping with a conservative estimate of the performance of the assay. Although estimating the residual variance of 24-h urinary free cortisol resulted in a relatively smaller contribution to the latent variable GRE (path coefficient = 0.30), the contribution is significant with a t value = 1.96 (Gaussian equivalent P value = 0.05).
This study suggests that the HTR2A gene may be up-regulated in CFS through allele-specific expression. An integrated analysis of genetics, methylation, gene expression and clinical measurements using SEM reveal that this up-regulation may be dependent on the cis-regulatory transcription factors E47, GR and Sp1. We previously showed that the minor allele A of the promoter polymorphism rs6311 (−1438G/A) in HTR2A was more common in CFS than NF subjects (Smith et al. 2008). This SNP, which has also been associated with other complex disorders including depression, PTSD and schizophrenia, results in the creation of a binding site for E47 (Smith et al. 2008) and also results in the loss of CpG methylation site at −1,439. Our current study examined HTR2A methylation in subjects from a population-based clinical study of CFS and identified two CpG sites, −1,224 and −1,420 that showed differential methylation between CFS and NF subjects and dependence on sequence variation at position −1,438. We recently demonstrated the first experimental evidence for the binding of GR at CpG site −1,420 (Falkenberg and Rajeevan 2010), whereas binding of Sp1 at CpG site −1,224 and the genotype-dependent binding of E47 at −1,438 were reported earlier (Smith et al. 2008; Zhu et al. 1995). Changes at these cis-regulatory elements, two of which are potentially heritable (methylation at −1,439 and −1,420), may have contributed to increased expression of A-allele and to the overall up-regulation of HTR2A in CFS.
While the specific mechanism for the association of allele-specific expression with CFS is not known, we present qualitative (Fig. 5) and quantitative (Fig. 6) models that may account for the potential influence of E47, GR and Sp1 on HTR2A expression. As a result of reduced cortisol production in some CFS subjects (Heim et al. 2000), GR binding would also be reduced. This state of hypocortisolism is included in the model as leading to inhibition of GR-mediated transcriptional repression (Sorensen et al. 2009). The qualitative model suggests that in situations of high methylation at −1,420 and low GR activity as in CFS subjects, the A-allele is over-expressed in relation to the G-allele. On the other hand, when methylation at −1,420 is low and GR activity is high as in NF subjects, the G-allele is over-expressed. The joint contributions of transcription factor activation and promoter methylation in the context of sequence variation may explain the lack of consensus between previous investigations that used only isolated functional studies to examine the role of the rs6311 promoter polymorphism in the regulation of HTR2A (Myers et al. 2007; Norton and Owen 2005; Parsons et al. 2004; Polesskaya et al. 2006). These joint contributions of several regulatory factors may also explain the lack of simple direct correlations between methylation and expression levels or instances of deviations from simple genetic models as revealed by some results (Fig. 4a) in this study.
Quantitative analysis by SEM supports two significant aspects of the qualitative model of HTR2A transcriptional regulation. First, the most important contribution of GRE to the transcription of HTR2A is indirect, mediated through competition with E47 (direct pathway coefficient is only −0.054, whereas GRE to the E47 pathway coefficient is 0.49). Second, there is a weaker but suggestive interaction between Sp1 binding and HTR2A expression (although the path coefficient of 0.28 is not significant, this study is underpowered for determining coefficients in this range).
Although the model posits that GR binding to −1,420 would be important for the inhibition of E47 binding, only a small amount of the variance in the 24-h urinary free cortisol contributes to the model (about 10%, vs. the unaccounted for variance of 90%). However, compared to methylation at −1,420, variance in cortisol contributes a relatively substantial amount to the latent variable GRE, (0.32 vs. 0.93, or about 25% of the total). In addition, excluding cortisol from the model drives the model (Chi-squared) P value down to 0.12 and the P value of the RMSEA up to 0.15, both indicating a worse fit. This indirect contribution of GRE to transcriptional regulation also raises the possibility that other GR family members with different kinetics and tissue distributions, such as mineralocorticoid receptor, may be more important for regulating HTR2A expression. As illustrated by the use of SEM in this study to provide a quantitative evaluation of the interaction of different TFBS on HTR2A allele-specific expression, SEM is an additional computational tool for analyzing complex transcriptional regulation paradigms.
The significances of our observed association of methylation levels between CpG sites at −1,439 and −1,420 in CFS but not in NF subjects are presently unknown. It is not clear how commonly an association of methylation between sites occurs, nor is it clear what accounts for a disease-specific association. Correlation of methylation between adjacent CpG sites in the distal region of the serotonin transporter promoter has been reported (Philibert et al. 2008), so further investigation into this phenomenon is clearly needed. Further investigation is also needed to assess the impact of other SNPs in this region on HTR2A expression.
There are several limitations to this study. Since we enforced strict quality control on CpG site-specific methylation analysis, information on methylation levels of all 17 CpG sites on all 185 subjects was not available. This contributed to small sample sizes in certain stratified analyses of genotype-dependent methylation levels. On the other hand, since methylation changes being close to the action of genome and a quantifiable molecular phenotype, it is likely that polymorphic CpG methylation at −1,439 may exhibit higher effect size than complex disease phenotypes that often involve several biological pathways. As expected, the nearly perfect concordance (r2 = 0.99, P = 0.0261–0.0392) between methylation levels at −1,439 and −1,438 genotypes (Fig. 3 bottom panel) suggests that our conclusions are unlikely to be influenced by sample size. With respect to the allele-specific expression, the data presented included all 22 CFS and NF subjects heterozygous for the −1,438 G/A promoter polymorphisms. As required by the case definition, this analysis (Fig. 4c) excluded those with exclusionary medical/psychiatric conditions. However, we examined this finding in a larger group of individuals (n = 35) by including subjects who reported exclusionary medical and psychiatric conditions but otherwise met criteria for CFS and NF in this study. The results with the larger group of subjects (data not shown) support the original finding (Fig. 4c) of allele-specific expression of HTR2A in PBMCs that differ with the fatigue status of the subjects. We recognize that this allele-specific expression finding, although statistically significant, should be considered as suggestive because of the relatively small sample size. Future studies are needed to replicate this finding.
Analytically, the close fit between the SEM model and the data does not mean that the model shown is the one true model, only that it is statistically consistent with the data. The better P values for this model, given what is typical for SEM in the literature, may be reflecting relative paucity of the data and simplicity of the model instead of overfitting of the data. On the other hand, the small sample size likely contributed to some type II errors in path identification, e.g. the study is underpowered to identify the link between Sp1 and HTR2A if indeed the path coefficient is in the range shown. Also, the average 24-h urinary free cortisol used in the SEM is not an ideal measurement for modeling the contribution of cortisol to HTR2A gene expression in PBMCs.
Potential influences of E47, GR and Sp1 binding sites were considered in our qualitative and quantitative models to provide a mechanistic understanding of the allele-specific expression. In silico computational analyses and competitive EMSA provide evidence for direct binding of transcription factors E47, GR and Sp1 in the HTR2A promoter. While EMSA is an accepted method to validate the possibility of functional binding, in vivo binding assays such as chromatin immunoprecipitation (Chip) should be considered in future studies. We believe that evidence for direct binding to specific sites is sufficient for modeling and integrating genetic and epigenetic risk in complex traits.
As depression is a recognized comorbid condition in CFS, the specificity of this model for CFS versus depression will require further exploration as serotonin signaling is recognized to be important in depressive disorders. To address the contribution of depression, we tested the association of methylation at each site by stratifying this study population based on Zung depression scores and did not identify an association (data not shown). However, methylation at one of the CpG sites (−1,224) was negatively correlated (r = −0.4315, P = 0.001) with increasing Zung scores and thus a potential association between HTR2A methylation and depression cannot be totally excluded from this study.
Finally, it should be kept in mind that the expression findings were made in PBMCs. Methylation and gene expression are tissue-dependent, and HTR2A expression in PBMCs may only be an indirect reflection of CFS pathogenesis. However, the model presented for HTR2A promoter methylation and expression in PBMCs is relevant for understanding mechanisms that may be playing a role in HTR2A expression in other tissues like the brain (Le-Niculescu et al. 2009).
This study underscores the importance of the coordinated influence of both genetic and epigenetic variations in determining an individual’s susceptibility to disease, and the value of integrated functional and computational genomics for simultaneous identification and quantitative evaluation of multiple cis-regulatory elements in major genes of medical interest. To our knowledge, this is the first study to apply SEM to promoter analysis. Novel findings in this study include disease-specific correlations of methylation at adjacent CpG sites and disease-associated allele-specific expression of HTR2A that is influenced by both CpG methylation and genotype-dependent transcription factor binding. These regulatory mechanisms of HTR2A expression may play significant roles in the pathology of common complex diseases with central nervous system abnormalities such as CFS, depression and PTSD. Confirmation of these findings in an independent study population is required.
Support for VR. Falkenberg was provided by the research participation program at the Centers for Disease Control and Prevention (CDC), National Center for Zoonotic, Vector-Borne Enteric Diseases, Division of Viral and Rickettsial Diseases, administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the US Department of Energy and the CDC. The authors acknowledge I. Dimulescu and MM. Khin for their laboratory support and WC Reeves for his support and encouragement with this research project.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.