Serotonin system gene variants and regional brain volume differences in pediatric OCD

Abstract

Obsessive-compulsive disorder (OCD) is phenotypically heterogeneous and genetically complex. This study aimed to reduce heterogeneity using structural brain imaging to study putative intermediate phenotypes for OCD. We hypothesized that select serotonin gene variants would differ in their relationship with brain volume in specific regions of the cortico-striato-thalamo-cortical (CSTC) circuits between OCD patients and controls. In a total of 200 pediatric subjects, we genotyped candidate serotonin genes (SLC6A4, HTR2A, HTR1B, and HTR2C) and conducted structural magnetic resonance imaging (sMRI) to measure regional brain volumes within CSTC circuits. In males and females separately, we first tested the association between serotonin gene variants and OCD and the effect of serotonin gene variants on brain volume irrespective of diagnosis. We then carried out a series of analyses to assess the effect of genotype-diagnosis interaction on brain volume. In females, but not in males, we identified a statistically significant genotype-diagnosis interaction for two single nucleotide polymorphisms (SNPs) in HTR2C, rs12860460 (interaction term estimate of 5.45 cc and interaction P value of 9.70e-8) and rs12854485 (interaction term estimate of 4.28 cc and interaction P value of 2.07e-6). The tested allele in each SNP was associated with decreased anterior cingulate cortex (ACC) volume in controls and with increased ACC volume in OCD patients. Our findings suggest that, in females, sequence variation in HTR2C influences ACC volume in pediatric OCD. The variants may contribute to differences in ACC volume and to OCD in a sex-specific manner when acting together with other genetic, biological, and/or environmental factors.

Introduction

Obsessive-compulsive disorder (OCD) is a neuropsychiatric disorder characterized by intrusive, repetitive thoughts and behaviors that are typically distressing to the patient (National Institutes of Health [NIH] 2008; American Psychiatric Association [APA] 2013). The disorder is heritable, as evidenced by twin studies and direct estimation of heritability from genome-wide common variant data (Davis et al. 2013; Pauls et al. 2014). Furthermore, OCD is genetically complex and phenotypically heterogeneous (Arnold and Richter 2007; Grados 2010). Various approaches have been used to identify genetic variants associated with OCD. These include studies of variants within candidate serotonin system genes. The best-studied serotonin candidate genes for OCD include SLC6A4 (solute carrier family 6 member 4), HTR2A, HTR1B, and HTR2C, which correspond with the serotonin transporter, or SERT, and serotonin receptors 5-HT_2A, 5-HT_1B, and 5-HT_2C respectively (Taylor 2013; Sinopoli et al. 2017). Within SLC6A4, the most extensively studied polymorphism, 5-HTTLPR (serotonin transporter-linked polymorphic region), lies within the promoter region of the gene (Heils et al. 1996; Lesch et al. 1996). The polymorphism exists in two major forms, one long (L) variant which consists of 16 sets of 20–23 bp tandem repeats and one short (S) variant which consists of 14 sets of 20–23 bp tandem repeats. In addition, an intrinsic A/G single nucleotide polymorphism (SNP), rs25531, occurs within the L variant and influences SERT expression. Taken together, the 5-HTTLPR and rs25531 is a triallelic system (L_A, L_G and S variants) with the L_A variant resulting in greater SERT expression compared with the S and L_G variants. These variants appear to be additive in their effect (Hu et al. 2006).

Strong evidence supports an association between OCD and this commonly studied polymorphic region of SLC6A4 (Taylor 2013; Walitza et al. 2014; Taylor 2016; Sinopoli et al. 2017); however, findings have been somewhat mixed with respect to which variant of 5-HTTLPR is associated with OCD. These mixed results may be because of heterogeneity in the OCD phenotype that has not been adequately addressed in previous studies. Sources of heterogeneity include multiple overlapping symptom dimensions in OCD (Bloch et al. 2008; Alvarenga et al. 2015), subtypes of OCD that depend on factors such as age of onset or comorbidity with tic disorders (Grados 2010; Leckman et al. 2010; APA 2013; Williams et al. 2013) and sex differences (Alvarenga et al. 2015; Mak et al. 2015).

Biologically salient, quantitative measures that mediate genotype and phenotype, often called “intermediate phenotypes”, are arguably less heterogeneous than clinical diagnostic syndromes like OCD (Meyer-Lindenberg and Weinberger 2006). For example, brain differences measured by neuroimaging techniques like magnetic resonance imaging (MRI) may provide intermediate phenotypes that can be tested for their association with a particular set of candidate genes or used in a genome-wide association study (GWAS). There are two putative advantages of studying intermediate phenotypes compared with psychiatric diagnoses like OCD: 1) Reduced heterogeneity of such measures may lead to increased power to detect genetic association due to increased effect sizes (Leite et al. 2015), although recent imaging genetics studies suggest such effect sizes may be smaller than originally assumed (Carter et al. 2017), and 2) Imaging and other biologically salient measures may help delineate underlying biological mechanisms mediating gene variation and clinical symptoms (Meyer-Lindenberg and Weinberger 2006; Leite et al. 2015). OCD has been repeatedly linked to dysfunction in cortico-striato-thalamo-cortical (CSTC) circuits. Evidence includes volumetric and functional abnormalities in various regions of CSTC circuits including the striatum (caudate and putamen), thalamus, anterior cingulate cortex (ACC), and orbitofrontal cortex (OFC), with some studies reporting differences in the directionality/magnitude of the effects between pediatric and adult OCD patients (MacMaster 2010; Brem et al. 2012; Pauls et al. 2014; Boedhoe et al. 2017). Both pharmacotherapy with a serotonin reuptake inhibitor (SRI) and cognitive behavioral therapy for OCD also alter metabolic activity in CSTC-related brain regions (Pauls et al. 2014).

Only three studies have been published which focus on the association between serotonin system gene variants and brain structure or function as measured by neuroimaging in individuals (predominantly adults) with OCD (Atmaca et al. 2011; Hesse et al. 2011; Honda et al. 2017). Upon examining the ACC, OFC, and thalamus using MRI, Atmaca et al. (2011) reported a significant genotype-diagnosis interaction whereby the effect of the 5-HTTLPR polymorphism on OFC volume was dependent on OCD diagnosis. S variant carriers showed a significantly smaller OFC volume in individuals with OCD, but not in healthy controls. Hesse et al. (2011) used positron emission tomography (PET) with a SERT-selective radiotracer and examined both 5-HTTLPR and STin2 VNTR (variable number of tandem repeats polymorphism in intron 2), another commonly studied polymorphism in SLC6A4. There was no significant effect of the interaction between presence of OCD and 5-HTTLPR genotype on SERT availability. When examining OCD patients alone, however, the authors reported an overall trend towards increased SERT availability in patients with the genotype S/S of 5-HTTLPR (reaching significance in the midbrain), a significant effect of STin2 VNTR genotype on SERT availability in the ACC and putamen, and a significant effect of the interaction between 5-HTTLPR and the STin2 VNTR on SERT availability in the ACC, putamen, and hypothalamus. Neither the study by Atmaca et al. (2011) nor the study by Hesse et al. (2011) corrected for multiple comparisons. The most recent study examined association between 5-HTTLPR and regional brain volumes measured using MRI, while accounting for the differential functionality imparted by rs25531. Authors reported a nominally significant genotype-diagnosis interaction whereby the effect of the 5-HTTLPR polymorphism on the right frontal pole was dependent on OCD diagnosis. They also noted a trend for OCD patients carrying L_A to have reduced gray matter volume in the right frontal pole relative to healthy controls carrying L_A (Honda et al. 2017). Overall, serotonin system gene variants may be associated with brain changes in OCD but, given the limited number of studies (which used different imaging modalities and different genotyping methods), further and more extensive studies are required.

To date, no studies examining serotonin system gene variants and their relationship to potential neuroimaging phenotypes in OCD have been conducted in pediatric samples. Early-onset OCD (onset typically in childhood or early adolescence) is considered a distinct subtype that differs from late-onset OCD (onset typically in late adolescence or early adulthood) with regard to a number of demographic and clinical features (MacMaster 2010). To reduce heterogeneity, we have focused on children and adolescents with OCD (therefore predominantly capturing individuals with early-onset OCD) (Mattina and Steiner 2016). There is also evidence from a recent meta-analysis suggesting that serotonin system gene variants differ in their association with OCD in males versus in females. Specifically, the S variant may be more associated with OCD in females than in males (Mak et al. 2015). To further reduce heterogeneity in our study, we therefore also chose to stratify by sex (Mattina and Steiner 2016). This decision is further supported by the fact that one of our genes of interest (HTR2C) is located on chromosome X and may therefore be associated with sex differences.

In our study, we examined selected genetic variants in four candidate serotonin system genes, SLC6A4, HTR2A, HTR1B, and HTR2C. Using brain volume in regions previously implicated in OCD, we aimed to identify associations between gene variants and neuroimaging phenotypes, specifically volumes of the ACC and other brain regions that have been implicated in OCD. We focused on children and adolescents, and we examined males and females separately to reduce heterogeneity. If the serotonin system is involved in OCD pathogenesis and if dysfunctional CSTC circuitry is involved in OCD pathogenesis, we hypothesized that we would find serotonin gene variants that differ in their association with brain volume in OCD patients versus in controls. Furthermore, given previously reported sex differences in OCD (Yücel et al. 2008; Mattina and Steiner 2016), we hypothesized that our findings would differ between males and females in our pediatric population.

Materials and methods

Subjects

A total of 411 participants 6–19 years of age were initially recruited at two sites, which included Wayne State University in Detroit, Michigan and the University of Michigan in Ann Arbor, Michigan, as part of a continuing collaborative study (Arnold et al. 2009; Wu et al. 2013). Written and informed consent and assent was obtained for all participants and, when applicable, from their parents. Each site received study approval from its respective Human Investigation Committee prior to recruitment of participants into the study. OCD participants were required to have a lifetime OCD diagnosis and a current, minimum total score of 10 on the Children’s Yale-Brown Obsessive-Compulsive Scale (CY-BOCS) (Scahill et al. 1997). OCD cases and healthy controls were assessed for lifetime diagnosis of Axis I psychiatric disorders, using DSM-IV-TR criteria (APA 2000) based on information obtained through two semi-structured interviews, the Schedule for Affective Disorder and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL) (Kaufman et al. 1997) and the pediatric version of the Schedule for Obsessive-Compulsive and Other Behavioral Syndromes (SOCOBS) (Hanna 2010), administered to both the participant and/or both parents. Composite diagnoses were based on agreement between the two interviews regarding 1) specific diagnoses and 2) level of diagnosis-related interference. OCD cases and healthy controls were also assessed using the Pubertal Development Scale (PDS) (adapted from Petersen et al. 1988).

For both OCD cases and healthy controls, participants were excluded if neurological disorders, such as seizure disorder, were identified. Participants were also excluded if they had a history of head injury with sustained loss of consciousness, chronic medical illness, history of substance abuse/dependence, history of psychosis, bipolar disorder, primary or recurrent major depressive disorder (MDD) (unless onset of MDD was after the onset of OCD), history of conduct disorder, an IQ less than or equal to 80, lifetime diagnosis of autism spectrum disorder (ASD) (including a score of 15 or higher on the Social Communication Questionnaire, lifetime version) (Rutter et al. 2003), lifetime diagnosis of eating disorder, if they were adopted, or if they were living away from both biological parents at the time of the study. In addition to the exclusions above, healthy controls were also screened for lifetime diagnosis of Axis I psychiatric disorders of any kind (APA 2000). For this study, we included only unrelated participants and only participants with both parents of European Caucasian descent to minimize possible population stratification.

Imaging

All participants had T₁-weighted structural magnetic resonance imaging (sMRI) data collected at the Children’s Hospital of Michigan Imaging Center. Detailed methods are described elsewhere (Szeszko et al. 2004; Arnold et al. 2009; Wu et al. 2013). Briefly, a Sigma 3.0-Tesla unit (Horizon HDX software, General Electric Medical Systems, Milwaukee, WI) was used to gather four separate contiguous three-dimensional (3D) MRI volumes. A 3D magnetization-prepared 180 degrees radio-frequency pulses and rapid gradient-echo (MP RAGE) sequence was utilized. Acquisition parameters included echo time of 1.6 ms, repetition time of 2200 ms, acquisition matrix of 512 × 512 × 204, pixel dimensions of 0.5 × 0.5 × 0.8 mm³, field of view 256 × 256 mm², 204 axial slices, flip angle of 13 degrees, and inversion times of 776 ms, 780 ms, 794 ms, and 808 ms for each respective 3D MRI volume. All MRIs were reviewed by a pediatric neuroradiologist to rule out clinically significant abnormalities and magnetic field inhomogeneities. A manual tracing technique was then carried out by trained operators, blind to case-control status, using BRAINS2 image processing software (Magnotta et al. 2002) to measure striatum, caudate, putamen, and thalamus, and using Analyze Direct 11.0 (AnalyzeDirect, Inc., Overland Park, KS) for the ACC and the OFC. This measured regional brain volumes in select structures with a priori evidence of their involvement in OCD, along with the total intracranial volume (ICV) for each subject, in cubic centimeters (cc). Overall, the brain regions we examined were the caudate, the putamen, the striatum (caudate + putamen combined), the thalamus, the ACC, and the OFC. As a result, a total of six regions of interest (ROIs) were examined, with respective parameters for the brain regions outlined further in our past studies (Szeszko et al. 2004; Arnold et al. 2009; Wu et al. 2013). In this study, we combined left and right hemisphere and used the total volume for each ROI and we report intraclass correlation coefficient (ICC) values ranging from 0.970–0.999.

DNA collection and extraction

Either saliva or blood was collected from each participant. Oragene•DNA collection kits were used to extract DNA from saliva using the accompanying, recommended protocol (DNA Genotek). An Autopure LS automated DNA extractor (QIAGEN) was used to extract whole EDTA blood using Puregene chemistry (Gentra). Saliva DNA was quantified using the Quant-iT^™ PicoGreen^® dsDNA Assay Kit (Invitrogen) and blood DNA was quantified using an FLx800 Fluorescence Reader (Biotek) and Hoechst 33258 dye (Sigma). DNA quality was verified using gel electrophoresis. Samples with DNA concentrations below 60 ng/μl (if genotyped on the HumanOmni2.5–4 v1.0 or HumanCoreExome-12 v1.0 microarrays (Illumina)) or samples with poor DNA quality were excluded.

Selection of candidate genes

The serotonin system has been the focus of numerous candidate gene studies in OCD, given the efficacy of SRIs in treating the disorder (Sinopoli et al. 2017). We therefore examined variants across serotonin system genes most commonly studied in OCD including SLC6A4, HTR2A, HTR1B, and HTR2C (Taylor 2013) to further understand the relationship of these genes with neuroimaging and OCD. Evidence suggests that the most promising of these genes in OCD are SLC6A4 (and more specifically polymorphism 5-HTTLPR) and HTR2A. There is more limited evidence to suggest the involvement of HTR1B in OCD, with studies indicating that this gene may be specific to certain OCD subgroups (Taylor 2013; Sinopoli et al. 2017). Though there is the least amount of evidence supporting the role of HTR2C in OCD, this gene has been implicated in a number of psychiatric disorders and is of interest to our structural neuroimaging study given that it is almost exclusively expressed in the central nervous system (Werry et al. 2008; Fagerberg et al. 2014).

Genotyping

5-HTTLPR

Given that variation in 5-HTTLPR is not captured using microarrays, we directly genotyped this region of SLC6A4 in a two-step process (Wendland et al. 2006) as previously outlined (Sinopoli et al. under review). This was carried out with The Centre for Applied Genomics (TCAG) at the Hospital for Sick Children (Toronto, Canada). First, we amplified 5-HTTLPR using forward primer 5’-ATGCCAGCACCTAACCCCTAATGT-3′ (5′-labeled with HEX fluorescent dye) and reverse primer 5’-GGACCGCAAGGTGGGCGGGA-3′ (Gelernter et al. 1997). Variants were identified as L or S based on variant length. Second, DNA was digested at the rs25531 SNP location using restriction enzyme MspI (Fermentas, Life Technologies), which only cuts when G is present but not when A is present. Peak Scanner Software v1.0 (Thermo Fisher Scientific) was used to analyze results. Depending on the outcomes of the two steps, variants were identified as L_A, L_G, or S. Based on their known functionality, we further categorized variants into two allelic groups: L_A and [L_G + S] which yield increased or decreased SERT expression respectively (Hu et al. 2006).

Quality control (QC) was applied using the standard settings in the Peak Scanner Software v1.0 (Thermo Fisher Scientific). Individuals were genotyped for 5-HTTLPR with a 95% completion rate. Reported siblings were removed. In our final sample, 1 5-HTTLPR-genotyped individual was carrying the previously noted rare S_G variant (Wendland et al. 2006; Voyiaziakis et al. 2011). After all QC and removal of ensuing individuals, we had a final N of 160 individuals with the 5-HTTLPR polymorphism genotyped (77 OCD cases and 83 healthy controls).

SNPs across SLC6A4, HTR2A, HTR1B, and HTR2C

Samples were originally genotyped to gather GWAS data for our continuing collaborative study (Arnold et al. 2009; Wu et al. 2013) on the HumanOmni2.5–4 v1.0 or HumanCoreExome-12 v1.0 microarrays (Illumina). We focused on data from SNPs across SLC6A4, HTR2A, HTR1B, and HTR2C (UCSC Genome Browser, hg19).

SNPs with call rate < 0.97 were excluded. Samples were excluded on the basis of sex misspecification and ambiguity and cryptic relatedness of half sibling or above. We verified European ethnicity via principal components analysis (PCA) using EIGENSTRAT version 3.2.10 (Price et al. 2006) and using HapMap populations CEU and TSI to ensure that the genetic data of our participants clustered with that of Europeans, relative to other ethnicities, and then via another PCA to identify and remove outliers within our own population. An additional PCA was run without outliers. A Tracy-Widom test of the resulting principal components (PCs) showed no significant PCs, therefore none were retained to control for population structure in our analyses. After all quality control and removal of ensuing individuals, we had a final N of 154 individuals with SNPs genotyped across our four genes of interest (74 OCD cases and 80 healthy controls).

Chromosomes 6, 13, 17 and X were pre-phased using SHAPEIT2 version 2.790 (Delaneau et al. 2014). SNP data was imputed using IMPUTE2 version 2.3.1 (Howie et al. 2009) for each of the candidate genes, including a 50 kb flanking region on either side of each gene. 1000 Genomes Phase 3 reference data was used for both pre-phasing and imputation. The non-pseudoautosomal region from 1000 Genomes chromosome X was used for chromosome X pre-phasing and imputation. After imputation, SNPs were removed for low quality imputation (information score < 0.8). SNPs with MAF ≥ 0.05 were kept in the sample. Data were kept as dosage calls for all analyses. Table 1 shows SNP counts across SLC6A4, HTR2A, HTR1B, and HTR2C.

Table 1.

SNP counts across 4 candidate genes: All common genotyped and imputed SNPs (MAF ≥ 0.05) are shown for each gene, and its flanking regions, included in the analyses for males and in the analyses for females

Gene	Number of SNPs in Males	Number of SNPs in Females
SLC6A4	117	102
HTR2A	301	313
HTR1B	241	235
HTR2C	401	482

SNP, single nucleotide polymorphism; SLC6A4, solute carrier family 6 member 4 (serotonin transporter gene); HTR2A, 5-HT_2A receptor gene; HTR1B, 5-HT_1B receptor gene; HTR2C, 5-HT_2C receptor gene

Statistical analyses

T-tests were used to analyze demographics for statistically significant differences in age, in total ICV, and in PDS for all male cases versus male controls and for all female cases versus female controls. The following analyses were conducted for 5-HTTLPR and for all SNPs across SLC6A4, HTR2A, HTR1B, and HTR2C, using the program, R, Version 3.0.1 (https://www.R-project.org).

Assessing the association of serotonin gene variants with OCD

Logistic regressions were conducted in males and in females separately, using the expected number of tested alleles for each SNP (designated [L_G + S] for 5-HTTLPR) against OCD case versus control status, in an additive model with age included as a covariate. The Genetic Type 1 error calculator (GEC) developed to account for dependent SNPs (Li et al. 2012) was used to calculate the P value thresholds for the male and female analyses in the SNPs. The resulting significance P value threshold for SNP analyses was 2.23e-4 for males and 2.15e-4 for females.

Assessing the effect of serotonin gene variants on brain volume

A linear regression was fit for each SNP using the expected number of tested alleles (designated [L_G + S] for 5-HTTLPR) against volume for each ROI and across the combined set of cases and controls (i.e., assessing the effect of SNP on brain volume irrespective of OCD diagnosis), in males and in females separately. An additive model was used including age and total ICV as covariates. Analyses involving sMRI data used the GEC-generated values that were then corrected for the number of brain regions under investigation. This resulted in a brain region-adjusted significance P value threshold of 3.72e-5 (2.23e-4 to account for dependent SNPs / 6 ROIs) for SNP analyses in males and of 3.58e-5 (2.15e-4 to account for dependent SNPs / 6 ROIs) for SNP analyses in females.

Assessing the effect of genotype-diagnosis interaction on brain volume

An interaction model was then fit for all individuals (cases and controls) with a main effect for both SNP (tested allele) and OCD (though we only report on SNP main effect, given our gene-driven hypotheses), along with an interaction term of OCD*[tested allele]. This allowed us to assess whether OCD status significantly changed the association of the tested allele with brain region volume (having corrected for age and total ICV by including these variables as covariates in the model). Brain region-adjusted significance P value thresholds were used to assess significance in this model.

Results

Demographic characteristics for all 200 individuals in the study are reported in Table 2. There were no significant differences to report in age, in total ICV, or in PDS between total male cases and total male controls. Between total female cases and total female controls, there was a significant difference in age (absolute t-statistic = 2.3, P value = 0.02) and in PDS (absolute t-statistic = 2.2, P value = 0.03).

Table 2.

Patient and control demographics: Group characteristics for all individuals combined, for OCD cases, and for controls

All Individuals
	Males	Females
Total N	97	103
N with 5-HTTLPR Genotyped	75	85
N with SNPs Genotyped Across SLC6A4, HTR2A, HTR1B, and HTR2C	75	79
OCD Cases
	Males	Females
Total N	46	51
Mean Age (SD)	13.2 (3.1)	13.6 (3.8)
Age Range	6.5- <20.0	6.2- <20.0
Mean Total ICV (cc) (SD)	1684.4 (152.7)	1534.8 (133.3)
Mean PDS Score (SD)	11.6 (3.4)	12.3 (5.6)
PDS Score Range	5–17	1–19
Mean CY-BOCS Score (SD)	27.0 (6.3)	28.4 (7.0)
CY-BOCS Score Range	13–37	10–39
MDD, N (%)	2 (4.3)	6 (11.8)
ADHD, N (%)	9 (19.6)	11 (21.6)
Anxietyi, N (%)	20 (43.5)	26 (51.0)
Ticsii, N (%)	16 (34.8)	8 (15.7)
Controls
	Males	Females
Total N	51	52
Mean Age (SD)	14.2 (3.4)	15.4 (3.8)
Age Range	6.2–19.9	6.8–19.8
Mean Total ICV (cc) (SD)	1676.6 (135.1)	1542.7 (120.9)
Mean PDS Score (SD)	12.0 (3.5)	13.9 (3.4)
PDS Score Range	4–18	5–19

N, sample size; 5-HTTLPR, serotonin transporter-linked polymorphic region; SNP, single nucleotide polymorphism; SLC6A4, solute carrier family 6 member 4 (serotonin transporter gene); HTR2A, 5-HT_2A receptor gene; HTR1B, 5-HT_1B receptor gene; HTR2C, 5-HT_2C receptor gene; OCD, obsessive-compulsive disorder; SD, standard deviation; ICV, intracranial volume; PDS, Pubertal Development Scale; cc, cubic centimeters; CY-BOCS, Children’s Yale-Brown Obsessive-Compulsive Scale; MDD, major depressive disorder; ADHD, attention-deficit/ hyperactivity disorder

ⁱIncludes Separation Anxiety Disorder, Panic Disorder (with or without Agoraphobia Specified), Specific Phobia, Social Phobia, Generalized Anxiety Disorder (GAD)

ⁱⁱIncludes Tourette’s Disorder, Chronic Motor Tic Disorder, Chronic Vocal Tic Disorder, Transient Tic Disorder (Motor and/or Vocal), Lifetime History of Motor/Vocal Tic

Assessing the association of serotonin gene variants with OCD

There was no significant association between 5-HTTLPR and OCD in males or in females. In the SNP analyses, the top 38 SNPs for OCD in males all corresponded with gene HTR1B (odds ratio, OR, ranging from 0.37 to 0.43; P value ranging from 1.27e-2 to 1.98e-2) and 24 of the 25 top SNPs for OCD in females all corresponded with gene SLC6A4 (OR ranging from 0.37 to 3.35; P value ranging from 1.37e-3 to 4.89e-3). The SNP findings did not remain significant after correction for multiple comparisons.

Assessing the effect of serotonin gene variants on brain volume

There were no significant associations between 5-HTTLPR and brain volume for any of the ROIs in males or in females. For the candidate gene SNPs, in males, the top 20 SNPs were all associated with reduced caudate volume, where 17 of the SNPs corresponded with HTR2A (SNP effect ranging from −0.17 to −0.15 cc; P value ranging from 2.68e-4 to 6.53e-4) and 3 of the SNPs corresponded with SLC6A4 (SNP effect of −0.15 cc; P value of 6.23e-4) (Table 3). In females, the top 20 SNPs were all associated with reduced volume of the striatum, where 8 of the SNPs corresponded with HTR1B (SNP effect ranging from −0.18 to −0.15 cc; P value ranging from 2.70e-4 to 6.47e-4) and 8 of the SNPs corresponded with HTR2C (SNP effect ranging from −0.17 to −0.15 cc; P value ranging from 3.64e-4 to 6.49e-4) (Table 4). None of the SNP findings remained significant after correction for multiple comparisons.

Table 3.

Effect of serotonin gene variants on brain volume in males: Top 20 SNP findings

Brain Region	SNP	Gene	Tested Allele	MAF	SNP Effect (cc)	P value
Caudate	rs7330368	HTR2A	A	0.22	−0.16	2.68E-04
Caudate	rs9316235	HTR2A	A	0.23	−0.17	2.92E-04
Caudate	rs2770297	HTR2A	T	0.26	−0.16	3.87E-04
Caudate	rs61948331	HTR2A	A	0.13	−0.16	4.56E-04
Caudate	rs731779	HTR2A	C	0.11	−0.16	4.60E-04
Caudate	rs9562687	HTR2A	A	0.10	−0.16	4.60E-04
Caudate	rs9562688	HTR2A	A	0.10	−0.16	4.60E-04
Caudate	rs61948356	HTR2A	A	0.14	−0.15	5.11E-04
Caudate	rs61948354	HTR2A	C	0.14	−0.15	5.13E-04
Caudate	rs17287961	HTR2A	T	0.07	−0.16	5.49E-04
Caudate	rs61948357	HTR2A	C	0.14	−0.15	5.61E-04
Caudate	rs73477746	HTR2A	G	0.14	−0.15	5.62E-04
Caudate	rs8078900	SLC6A4	A	0.25	−0.15	6.23E-04
Caudate	rs33980254	SLC6A4	T	0.25	−0.15	6.23E-04
Caudate	rs73268099	SLC6A4	G	0.25	−0.15	6.23E-04
Caudate	rs61647933	HTR2A	C	0.17	−0.15	6.38E-04
Caudate	rs61948335	HTR2A	G	0.11	−0.15	6.42E-04
Caudate	rs61948318	HTR2A	T	0.13	−0.16	6.44E-04
Caudate	rs9534501	HTR2A	T	0.13	−0.16	6.53E-04
Caudate	rs9567743	HTR2A	C	0.13	−0.16	6.53E-04

SNP, single nucleotide polymorphism; MAF, minor allele frequency; cc, cubic centimeters; SLC6A4, solute carrier family 6 member 4 (serotonin transporter gene); HTR2A, 5-HT_2A receptor gene

Table 4.

Effect of serotonin gene variants on brain volume in females: Top 20 SNP findings

Brain Region	SNP	Gene	Tested Allele	MAF	SNP Effect (cc)	P value
Striatum	rs9343614	HTR1B	G	0.32	−0.18	2.70E-04
Striatum	rs9361228	HTR1B	A	0.30	−0.18	2.96E-04
Striatum	rs540285	HTR2C	T	0.27	−0.17	3.64E-04
Striatum	rs12836749	HTR2C	A	0.13	−0.16	4.07E-04
Striatum	rs1777761	HTR1B	G	0.25	−0.16	5.12E-04
Striatum	rs4543330	HTR1B	C	0.08	−0.15	5.18E-04
Striatum	rs2798949	HTR1B	G	0.24	−0.16	5.22E-04
Striatum	rs1891752	HTR1B	G	0.25	−0.16	5.22E-04
Striatum	rs199602838	HTR2C	T	0.11	−0.15	5.39E-04
Striatum	rs184155834	HTR1B	G	0.08	−0.15	5.48E-04
Striatum	rs9902340	SLC6A4	G	0.44	−0.16	5.72E-04
Striatum	rs12861270	HTR2C	G	0.10	−0.16	6.05E-04
Striatum	rs11797988	HTR2C	A	0.10	−0.16	6.34E-04
Striatum	rs9567745	HTR2A	C	0.06	−0.15	6.41E-04
Striatum	rs3945573	HTR2A	A	0.06	−0.15	6.41E-04
Striatum	rs2226183	HTR1B	A	0.07	−0.15	6.47E-04
Striatum	rs3764450	SLC6A4	A	0.26	−0.16	6.48E-04
Striatum	rs12855533	HTR2C	A	0.18	−0.16	6.49E-04
Striatum	rs12847225	HTR2C	C	0.18	−0.16	6.49E-04
Striatum	rs12842363	HTR2C	A	0.18	−0.16	6.49E-04

SNP, single nucleotide polymorphism; MAF, minor allele frequency; cc, cubic centimeters; SLC6A4, solute carrier family 6 member 4 (serotonin transporter gene); HTR2A, 5-HT_2A receptor gene; HTR1B, 5-HT_1B receptor gene; HTR2C, 5-HT_2C receptor gene

Assessing the effect of genotype-diagnosis interaction on brain volume

When we assessed whether OCD diagnosis significantly changed the relationship between 5-HTTLPR and brain volume, there were no significant findings to report for males or for females. When assessing whether OCD diagnosis significantly changed the relationship between SNPs across our four candidate genes and brain volume in males (Table 5), there were no findings that remained significant after correction for multiple comparisons. In females (Table 6), OCD diagnosis significantly changed the relationship between two HTR2C SNPs (rs12860460 and rs12854485) and ACC volume. Both rs12860460 and rs12854485 were associated with a significant SNP main effect in the interaction model (SNP main effect of −4.37 cc and − 3.26 cc respectively; SNP P value of 1.75e-7 and 5.45e-6 respectively). There was a significant genotype-diagnosis interaction for both SNPs (rs12860460, interaction term estimate of 5.45 cc and interaction P value of 9.70e-8; rs12854485, interaction term estimate of 4.28 cc and interaction P value of 2.07e-6). Fig. 1 shows the change in the direction/magnitude of effect of rs12860460 (Fig. 1-a) and rs12854485 (Fig. 1-b) on ACC volume in female OCD patients versus in controls. For both SNPs, the tested allele (versus what we observed for the alternate allele) was associated with reduced ACC volume in female controls and with increased ACC volume in female OCD patients.

Table 5.

Effect of genotype-diagnosis interaction on brain volume in males: Top 20 SNP findings

Brain Region	SNP	Gene	Tested Allele	MAF	SNP Main Effect (cc)	SNP P value	Interaction Term Estimate (cc)	Interaction P value
OFC	rs1777764	HTR1B	G	0.31	−0.97	3.52E-01	5.78	1.58E-03
OFC	rs28656158	HTR1B	C	0.31	−0.71	4.81E-01	5.43	2.23E-03
Thalamus	rs17069005	HTR2A	G	0.20	−0.74	4.24E-02	1.47	2.32E-03
Caudate	rs4942577	HTR2A	T	0.38	−0.89	3.02E-04	1.06	3.06E-03
OFC	rs2223832	HTR1B	C	0.14	−2.51	1.17E-01	7.20	3.61E-03
OFC	rs11757592	HTR1B	C	0.14	−2.51	1.17E-01	7.20	3.61E-03
Caudate	rs9534488	HTR2A	T	0.46	−0.69	3.04E-03	0.97	4.44E-03
Caudate	rs9526236	HTR2A	T	0.46	−0.69	3.22E-03	0.98	4.73E-03
OFC	rs9343614	HTR1B	G	0.20	−1.93	1.11E-01	6.22	4.77E-03
Caudate	rs12110491	HTR1B	T	0.12	0.46	2.37E-01	−1.72	4.90E-03
Caudate	rs55636038	HTR1B	T	0.12	0.46	2.37E-01	−1.72	4.90E-03
Caudate	rs11755194	HTR1B	C	0.12	0.46	2.37E-01	−1.72	4.90E-03
Caudate	rs1228798	HTR1B	G	0.12	0.46	2.37E-01	−1.72	4.90E-03
Caudate	rs1228797	HTR1B	T	0.12	0.46	2.37E-01	−1.72	4.91E-03
Caudate	rs2207056	HTR1B	T	0.12	0.46	2.37E-01	−1.72	4.92E-03
Caudate	rs10806097	HTR1B	T	0.12	0.48	2.34E-01	−1.74	5.06E-03
Caudate	rs1228806	HTR1B	C	0.11	0.50	2.36E-01	−1.76	5.47E-03
Caudate	rs17069005	HTR2A	G	0.20	−0.86	7.41E-03	1.17	5.50E-03
Caudate	rs1228805	HTR1B	G	0.11	0.50	2.36E-01	−1.77	5.56E-03
Caudate	rs1228804	HTR1B	G	0.11	0.51	2.38E-01	−1.77	5.74E-03

SNP, single nucleotide polymorphism; MAF, minor allele frequency; cc, cubic centimeters; OFC, orbitofrontal cortex; HTR2A, 5-HT_2A receptor gene; HTR1B, 5-HT_1B receptor gene

Table 6.

Effect of genotype-diagnosis interaction on brain volume in females: Top 20 SNP findings. (Findings significant after correction for multiple comparisons are in bold)

Brain Region	SNP	Gene	Tested Allele	MAF	SNP Main Effect (cc)	SNP P value	Interaction Term Estimate (cc)	Interaction P value
ACC	rs12860460	HTR2C	C	0.18	−4.37	1.75E-07	5.45	9.70E-08
ACC	rs12854485	HTR2C	G	0.23	−3.26	5.45E-06	4.28	2.07E-06
Striatum	rs9567737	HTR2A	T	0.42	0.94	8.56E-04	−1.45	3.40E-04
Striatum	rs6561333	HTR2A	C	0.42	0.95	8.77E-04	−1.45	3.42E-04
Striatum	rs1923886	HTR2A	T	0.42	0.90	1.53E-03	−1.43	4.37E-04
Striatum	rs1885884	HTR2A	G	0.13	1.21	1.15E-03	−1.78	6.59E-04
ACC	rs35031982	HTR2C	G	0.21	−2.81	2.53E-04	3.28	6.64E-04
Thalamus	rs1738504	HTR1B	C	0.26	0.42	1.30E-01	−1.43	7.04E-04
Thalamus	rs1777764	HTR1B	G	0.34	0.50	5.10E-02	−1.34	8.82E-04
ACC	rs12861270	HTR2C	G	0.10	−5.46	2.19E-03	6.20	1.01E-03
ACC	rs2192371	HTR2C	G	0.26	−2.36	5.44E-04	2.86	1.06E-03
ACC	rs12862598	HTR2C	C	0.10	−5.40	2.57E-03	6.17	1.14E-03
Thalamus	rs2207055	HTR1B	A	0.33	0.46	7.12E-02	−1.30	1.16E-03
ACC	rs12851998	HTR2C	T	0.10	−5.49	1.85E-03	6.00	1.23E-03
ACC	rs12854729	HTR2C	G	0.10	−5.46	2.25E-03	6.10	1.23E-03
ACC	rs3813928	HTR2C	A	0.10	−5.46	2.25E-03	6.09	1.24E-03
ACC	rs3813929	HTR2C	T	0.10	−5.46	2.25E-03	6.09	1.24E-03
ACC	rs11798441	HTR2C	G	0.10	−5.46	2.25E-03	6.09	1.24E-03
ACC	rs17260565	HTR2C	G	0.10	−5.46	2.25E-03	6.09	1.24E-03
ACC	rs17326429	HTR2C	A	0.10	−5.46	2.25E-03	6.09	1.24E-03

SNP, single nucleotide polymorphism; MAF, minor allele frequency; cc, cubic centimeters; ACC, anterior cingulate cortex; HTR2A, 5-HT_2A receptor gene; HTR1B, 5-HT_1B receptor gene; HTR2C, 5-HT_2C receptor gene

Fig. 1.

Brain volume versus genotype in females graphed: a) HTR2C SNP rs12860460, showing females with genotype and corresponding ACC volume (cc) for controls (green) and for OCD cases (purple); ACC volume decreases by 4.37 cc in controls (with 0, 1, or 2 tested C alleles) and ACC volume increases by 1.08 cc in OCD cases (with 0, 1, or 2 tested C alleles). b) HTR2C SNP rs12854485, showing females with genotype and corresponding ACC volume (cc) for controls (green) and for OCD cases (purple); ACC volume decreases by 3.26 cc in controls (with 0, 1, or 2 tested G alleles) and ACC volume increases by 1.02 cc in OCD cases (with 0, 1, or 2 tested G alleles). OCD, obsessive-compulsive disorder; ACC, anterior cingulate cortex; cc, cubic centimeters

Discussion

Studies to date have indicated that serotonin system genes may play a role in OCD, but the associated variants reported are inconsistent between studies (Sinopoli et al. 2017). Like other complex disorders, OCD is thought to result from multiple genes of small effect interacting with environmental factors (Pauls et al. 2014). In this study, we adopted a candidate gene approach to examine common potential serotonin system gene polymorphisms hypothesized to be involved in OCD. We aimed to address some of the limitations of previous studies including failure to account for sex or age of onset. To attempt to reduce phenotypic heterogeneity in our study, we used an exclusively pediatric group of individuals and stratified by sex. We also attempted to reduce heterogeneity by using structural brain imaging in order to study putative intermediate phenotypes for OCD, specifically using CSTC circuitry relevant to OCD (MacMaster 2010; Pauls et al. 2014).

In our case-control analyses, the top SNPs nominally associated with OCD in males were located in a region 5′ of HTR1B. These findings did not remain significant after correction for multiple comparisons. In females, the majority of our top SNPs were nominally associated with OCD and corresponded with SLC6A4. Again, these findings did not remain significant after correction for multiple comparisons, but the difference in trends in males and females is in line with previous literature showing that risk variants for OCD may differ by sex (Mattina and Steiner 2016). We did not find any significant association in males or in females between OCD and the most commonly studied candidate gene polymorphism, 5-HTTLPR, which has been associated with the disorder in previous literature (Taylor 2013). Though our case-control analyses were underpowered, our findings suggest a need to analyze candidate serotonin system genes in males and in females separately.

For our brain region analyses, we first examined the effect of candidate serotonin gene variants on brain volume in cases and controls combined (irrespective of diagnosis). None of the candidate variants were significantly associated with regional brain volumes in males or in females after correction for multiple comparisons. The results from our analyses assessing the effect of genotype-diagnosis interaction on brain volume supported our hypothesis that the relationship between some serotonin system gene variants and brain volume differs between OCD patients and controls and that these differences vary between males and females. Our findings in males did not survive correction for multiple comparisons. In females, we identified a significant genotype-diagnosis interaction for two SNPs in HTR2C, rs12860460 and rs12854485. We also note that both significant SNPs were genotyped and not imputed, further supporting the reliability of our findings. Specifically, these HTR2C SNPs were differentially associated with ACC volume in female children/adolescents in such a way that the tested allele (versus what we observed for the alternate allele) in each SNP was associated with decreased ACC volume in healthy individuals and with increased ACC volume in OCD patients. The effect of gene variant on ACC volume was also smaller in OCD cases than it was in controls. We cannot refer to one of the variants in either SNP as a genetic risk factor, however, since we did not identify any to be significantly associated with OCD in our first set of analyses (though underpowered).

Our study implicated the ACC, consistent with prior findings of increased ACC volume in pediatric OCD (Rosenberg and Keshavan 1998; Szeszko et al. 2004) and findings of ACC abnormalities being more pronounced in female OCD patients than in male OCD patients (Yücel et al. 2008). Our study highlights a possible mechanism in OCD where the allelic variants in the implicated HTR2C SNPs act differently in cases versus in controls with regard to the direction and magnitude of effect on ACC volume.

Our diagnosis-specific findings are similar to what has previously been reported. Two prior studies that examined 5-HTTLPR in OCD showed a nominally significant genotype-diagnosis interaction effect on brain volume (Atmaca et al. 2011; Honda et al. 2017). Atmaca et al., (2011) noted a stronger effect of 5-HTTLPR genotype on OFC in OCD patients versus in controls, with reduced volume in the OFC of S carriers with OCD (versus individuals of genotype L/L with OCD). Honda et al., (2017) noted a stronger effect of 5-HTTLPR genotype on the right frontal pole in OCD patients versus in controls, with reduced gray matter volume in the right frontal pole in L_A carriers with OCD versus L_A carriers without OCD (Honda et al. 2017). We similarly identified significant genotype-diagnosis interaction effects on brain volume for serotonin system gene variants and our findings survived correction for multiple comparisons. Unlike these previous studies, though, we did not observe a significant genotype-diagnosis effect for 5-HTTLPR. Instead, our strongest findings were within HTR2C. Furthermore, the effects of the significant HTR2C SNPs were more marked in controls as opposed to in OCD cases, opposite to what was observed for 5-HTTLPR in previous studies.

The 5-HT_2C receptor is widely distributed throughout the central nervous system, plays a number of roles in cell signalling, and is important in mood, sex, and appetite regulation (Molineaux et al. 1989). Studies in psychiatry have also suggested the receptor’s involvement in anxiety, schizophrenia, depression, and suicide (Werry et al. 2008), with a group of studies reporting significant differences in RNA editing for HTR2C in post-mortem brain tissue of suicide victims (Niswender et al. 2001; Gurevich et al. 2002; Iwamoto and Kato 2003; Dracheva et al. 2008; Lyddon et al. 2013; Di Narzo et al. 2014; Weissmann et al. 2016).

Rs12854485 and rs12860460 are adjacent to one another (~4.67 kb apart) in intron 2 of HTR2C. Nearby the SNPs, and overlapping the same region as HTR2C intron 2, are 4 microRNA (miRNA) genes (MIR764, MIR1912, MIR1264, MIR1298) which yield non-coding RNAs involved in gene expression and post-transcriptional regulation, and an H/ACA box small nucleolar RNA (snoRNA) gene (SNORA35), part of a class yielding non-coding RNAs which facilitate rRNA or spliceosomal RNA modification (Lestrade and Weber 2006; UCSC Genome Browser, hg19). SNORA35 is predominantly expressed in the brain (UCSC Genome Browser, hg19; GTEx, Release V6). It could be the case that our significant SNPs lie in potential regulatory regions of HTR2C that affect gene expression, or more specifically RNA editing or alternative splicing, depending on additional risk factors (Werry et al. 2008; Lu et al. 2012; Wang et al. 2015). The SNPs may also be tagging variants in another region of the gene(s) that are implicated in HTR2C expression or functionality or in the expression or functionality of the small regulatory genes overlapping HTR2C. Our significant genotype-diagnosis interaction shows us that the effect of each SNP variant on female ACC volume differs in OCD cases versus in controls. HTR2C variation is likely to be only one contributing factor in the complex etiology of OCD, which interacts with additional genetic, epigenetic, biological, and/or environmental risk factors to produce disease-relevant changes in ACC volume in female children and adolescents.

Though our study provides evidence of the value of studying the genetics of neural mechanisms for OCD in homogenous subgroups reflective of sex and age (children/adolescents versus adults), it had a number of limitations. First, subgrouping by sex resulted in reduced sample size for each analysis, which could have reduced our power to detect associations in the absence of strong sex effects. To address this issue, we performed post hoc analyses for all autosomal variants (excluding HTR2C which is on chromosome X) combining males and females. There were no significant findings for these combined analyses after correction for multiple comparisons. The second limitation of our study was that our sample was too small to consider genome-wide data and all potential brain regions, both of which would have conferred a stricter significance threshold to account for multiple comparisons. We therefore elected to focus on identifying associations of interest confined to select candidate serotonin system genes and a priori brain regions pertinent to OCD.

Future studies are warranted to replicate our findings in a larger sample and, ideally, using a more comprehensive, genome-wide approach. Should our findings be replicated, more extensive studies would be necessary to identify additional pathological factors driving the development of OCD and that are specific to the relationship between ACC and HTR2C in pediatric females. Future genetic studies of OCD should similarly stratify by sex and age group, given that we have evidence supporting genetic association in a pediatric sample that would not have been identified without stratifying based on sex. Age has been shown to have an effect on brain volume, particularly in subcortical brain regions, with puberty status describing such developmental changes even better than age (Wierenga et al. 2018). Given that the age range of our sample traverses adolescence, we included age as a covariate in our study to control for its effects on brain volume. Future studies in larger samples may benefit from more detailed analyses of the influence of age and pubertal development (i.e., by stratifying based on these variables) given the effect of puberty on brain structure. Our sample size was insufficient to account for symptom dimensions in OCD, but future studies with larger samples should test for genetic associations based on symptom dimensions. This will help us to better understand the phenotypic heterogeneity of OCD and the complexity of the disorder at the level of gene systems and brain morphology.

In summary, although limited by a small sample size and by its focus on single candidate genes, our study is important given the novelty of our approach. Ours is one of the first studies of its kind (and the largest to date) to look across several candidate serotonin genes in OCD, using brain imaging as a potential mediating biological factor in OCD and sex-based analyses to reduce phenotypic heterogeneity. This approach has not yet been implemented on a genome-wide scale in OCD. Our findings, therefore, help lay methodological groundwork and will help guide future genome-wide studies aimed at understanding the genetic and neural mechanisms driving OCD. Consistent with a precision medicine approach, identification of potential genetic and neurobiological mechanisms in OCD, combined with information regarding key demographic features (sex, age of onset), may allow us to target OCD treatment to specific subgroups of patients.

Highlights.

• Significant effect of genotype-diagnosis interaction on ACC volume in females.

• Effect of HTR2C SNPs on ACC volume differs between OCD patients and controls.

• No significant genotype-diagnosis interaction found in males