Cortical Abnormalities Associated With Pediatric and Adult Obsessive-Compulsive Disorder: Findings From the ENIGMA Obsessive-Compulsive Disorder Working Group

Disease models of obsessive-compulsive disorder (OCD) propose that abnormalities in the cortico-striato-thalamo-cortical circuits are key to the pathophysiology of OCD. Recent findings also implicate the involvement of fronto-limbic and fronto-parietal regions in pediatric and adult OCD (1–3). An important limitation of brain imaging research is the typically small samples that limit sensitivity and presumably contribute to the lack of reproducibility and reliability of findings (4). This issue may be partially addressed by the use of meta- and mega-analysis of multiple study samples. We therefore initiated the OCD working group within the Enhancing Neuro-Imaging Genetics Through Meta-Analysis (ENIGMA) consortium (5), in which researchers around the world collaborate to boost statistical power, with the aim of elucidating brain abnormalities in OCD.

We recently performed meta- and mega-analyses on data from 3,589 individuals and reported (6) subcortical volume differences between OCD patients and healthy controls that were related to clinical characteristics. Distinct subcortical volume abnormalities were detected in adults and children with OCD. Adult OCD patients had significantly smaller hippocampal and larger pallidal volumes. The smaller hippocampal volume seemed to be driven by comorbid depression and an adult illness onset. The larger pallidal volume was more pronounced in adult OCD patients who had a childhood illness onset. Children with OCD had larger thalamic volumes compared with control children.

With regard to the cortex, previous MRI studies have consistently shown abnormalities in dorsomedial prefrontal and anterior cingulate cortices (7–10), findings that are supported by mega-analyses from the OCD Brain Imaging Consortium (OBIC) (11, 12). Abnormalities in the fronto-parietal and temporo-parietal regions have also been reported (10, 12–14). Findings regarding the orbitofrontal cortex (8, 11, 15) and the operculum have been inconsistent (8, 10, 11, 13, 16). These inconsistencies may be partially explained by differences in processing protocols, limited statistical power, and clinical heterogeneity related to variation in disease profile and developmental stage.

Most of these studies were predominantly based on volumetric measures using voxel-based morphometry (VBM). Volumetric measures, however, depend on a combination of changes in gray matter thickness and surface area (17). Fewer studies have used surface-based methods to generate detailed maps of cortical thickness and surface area. These measures represent distinct features of cortical morphometry that are somewhat genetically independent and are driven by different neurobiological processes (18). Studying these properties independently will make it easier to interpret the cortical abnormalities reported in OCD in the context of the postulated neurodevelopmental basis for OCD (19) (see Supplementary Section S1 in the data supplement that accompanies the online edition of this article).

Here we performed the largest coordinated worldwide study to date of cortical measures in patients with OCD compared with healthy controls. We extracted cortical thickness and surface area estimates of 1,905 patients with OCD and 1,760 healthy control subjects, using harmonized data processing and analysis strategies across 27 sites. We also aimed to establish the potential modulating effects of demographic and clinical characteristics. Based on the literature, we expected lower cortical thickness in the anterior cingulate cortex, orbitofrontal cortex, dorsomedial prefrontal cortex, and parietal regions in OCD patients compared with healthy controls. In addition, we explored the cortical surface area profile in the OCD sample.

Method

Samples

The ENIGMA-OCD working group includes 38 data sets from 27 international research institutes, with neuroimaging and clinical data from OCD patients and typically developing healthy control subjects (i.e., free of psychopathology), including both children and adults (for a map showing site location, see Figure S1 in the online data supplement). Six of these 38 data sets (the entire OBIC sample) were identical to those included in the OBIC mega-analyses using VBM (11) and vertex-based FreeSurfer (12). We defined adults as individuals age 18 years and older and children as individuals under age 18 years. The split at age 18 followed from a natural selection of the age ranges used in these samples, as most samples used age 18 as a cutoff for inclusion. The samples’ respective demographic and clinical characteristics are detailed in Tables S1 and S2 in the data supplement. In total, we analyzed data from 3,665 subjects, including 1,905 OCD patients (407 children and 1,498 adults) and 1,760 control subjects (324 children and 1,436 adults). All local institutional review boards permitted the use of measures extracted from the anonymized data for mega-analyses.

Image Acquisition and Processing

Structural T₁-weighted MRI brain scans were acquired and processed locally. Image acquisition parameters for each site are listed in Table S3 in the data supplement. All cortical parcellations were performed with the fully automated segmentation program FreeSurfer, version 5.3 (20), following standardized ENIGMA protocols to harmonize analyses and quality control procedures across multiple sites (see http://enigma.usc.edu/protocols/imaging-protocols/). Segmentation of 68 (34 left and 34 right) cortical gray matter regions based on the Desikan-Killiany atlas (21), and two whole-hemisphere measures were visually inspected and statistically evaluated for outliers. Details on image exclusion criteria and quality control are presented in Supplementary Section S1 in the data supplement.

Statistical Analysis

We performed both a meta-analysis (i.e., using group statistics from the independent studies) and a mega-analysis (i.e., pooling extracted measures from individual subjects across sites, while adjusting for site effects) to be consistent with our previous report. In this report, we focus on the mega-analysis; for methods, results, and discussion of the meta-analysis, see Supplementary Section S3 in the data supplement.

We examined differences between OCD patients and controls within a mega-analytical framework by pooling the extracted cortical thickness and surface area measures from each site. Each of the 70 cortical regions of interest (68 regions and two whole-hemisphere average thickness or total surface area measures) served as the outcome measure and a binary indicator of diagnosis as the predictor of interest in multiple linear regression models. All cortical thickness models were adjusted for age and sex; cortical surface area models were corrected for intracranial volume (see Supplementary Section S1 in the data supplement), age, age-squared, sex, age-by-sex interaction, and age-squared-by-sex interaction, to account for any higher-order effects on cortical surface area of age and sex as well as head size, which do not appear to be detectable for cortical thickness measures (22). Additionally, all models were also adjusted for site, coded by using dummy variables. Effect size estimates were calculated using Cohen’s d, computed from the t statistic of the diagnosis indicator variable from the regression models. Similarly, for models testing interactions (i.e., sex-by-diagnosis interaction and age-by-diagnosis interaction), a multiplicative predictor was the predictor of interest, with the main effect of each predictor included in the model. The effect size was calculated using the same procedure.

To detect potentially different effects of disease with age, we performed all analyses separately for pediatric and adult participants. We performed stratified analyses comparing the medicated and unmedicated groups of OCD patients separately to controls and to each other. Likewise, stratified analyses were performed to investigate the effect of comorbid major depressive disorder, comorbid anxiety disorders, and OCD symptom dimensions (using the adult and child versions of the Yale-Brown Obsessive Compulsive Scale [YBOCS] [23, 24] symptom checklist; see Supplementary Section S2 in the data supplement). To study the neurodevelopmental aspects of illness within the adult samples, we performed separate stratified analyses comparing childhood-onset OCD patients (onset before age 18) and adult-onset OCD patients (onset at age 18 or later). Furthermore, we examined associations with age at onset, illness duration, and illness severity (using the total severity score from the YBOCS) as continuous variables. In these analyses, effect sizes were expressed as partial-correlation Pearson’s r after removing nuisance variables (age, sex, site, and intracranial volume). In all tables (see the data supplement), regions are listed in order of effect size (strongest to weakest). Throughout, we report p values corrected for multiple comparisons using the Benjamini-Hochberg procedure to ensure a false discovery rate limited to 5% (q=0.05) for 70 cortical measures.

Results

An overview of the demographic and clinical characteristics of the pooled samples is provided in Table 1.

Mega-Analysis

Cortical thickness and surface area differences between OCD patients and controls.

Adults:

Lower cortical thickness was observed in adult OCD patients (N=1,498) compared with controls (N=1,436) in the inferior parietal cortex bilaterally (effect size [Cohen’s d]=−0.14) (Figure 1; see also Table S4 in the data supplement). A lower surface area was observed in the left transverse temporal cortex (Cohen’s d=−0.16) (see Table S5 and Figure S2 in the data supplement). None of the regions showed significant sex-by-diagnosis or age-by-diagnosis interaction effects.

Children:

We found significantly thinner cortices in pediatric OCD patients (N=407) compared with controls (N=324) in the left and right superior parietal and the left inferior parietal cortices (Figure 2) and the left lateral occipital cortex (Cohen’s d values between −0.24 and −0.31) (see Table S6 in the data supplement). None of the regions analyzed showed significant differences in cortical surface area or evidence of sex-by-diagnosis or age-by-diagnosis interaction effects (see Table S7 in the data supplement).

Influence of medication on cortical thickness and surface area.

Adults:

Left and right hemisphere cortical thickness was lower in medicated OCD patients (N=646) compared with controls (N=1,436). Regionally, we found significantly thinner cortices in frontal, temporal, parietal, and occipital regions of adult medicated OCD patients (Cohen’s d values between −0.10 and −0.26) (Figure 3; see also Table S8a in the data supplement). We did not detect significant differences in cortical thickness in unmedicated OCD patients (N=831) compared with controls (see Table S8b in the data supplement). Medicated OCD patients compared with unmedicated patients showed lower cortical thickness in frontal, temporal, and parietal regions (Cohen’s d values between −0.13 and −0.21) (see Table S8c and Figure S3 in the data supplement). Similar to the main group comparison, we found a lower surface area for the left transverse temporal cortex in medicated OCD patients compared with controls (Cohen’s d=−0.20) (see Table S9a and Figure S4 in the data supplement). We did not detect differences in surface area in unmedicated OCD patients compared with controls and when we compared medicated and unmedicated patients directly (see Table S9b,c in the data supplement).

Children:

Compared with controls (N=324), medicated children with OCD (N=183) showed lower cortical thickness of the inferior parietal and superior parietal cortices bilaterally and the left lateral occipital cortex (Cohen’s d≈−0.31) (see Table S10a and Figure S5 in the data supplement). We did not detect significant differences in cortical thickness in unmedicated pediatric OCD patients (N=222) compared with controls or when we compared medicated with unmedicated patients (see Table S10b,c). More widespread surface area differences were detected when we compared medicated pediatric OCD patients and controls, mainly in several frontal regions (Cohen’s d values between −0.27 and −0.33) (Figure 4; see also Table S11a in the data supplement). No differences in surface area were observed when we compared unmedicated patients and controls (see Table S11b). We did observe a lower surface area for the right lingual (Cohen’s d=−0.34) and pericalcarine (Cohen’s d=−0.40) cortices in medicated compared with unmedicated pediatric OCD patients (see Table S11c and Figure S6 in the data supplement).

Influence of comorbidities on cortical thickness and surface area.

We did not detect any associations between cortical thickness or surface area and current comorbid depression or anxiety disorder in adults (N=167 and N=224, respectively) or in children (N=29 and N=132, respectively). These numbers, however, are too small because of the lack of systematic assessment of comorbidities in some samples, and they reflect an underestimation of comorbidity because of exclusion of comorbid cases in other samples. For full details, see Supplementary Section S4 in the data supplement.

Influence of symptom dimensions on cortical thickness and surface area.

Adults:

Regression analyses on OCD patients’ symptom dimensions (N=1,214) showed no associations between the presence of a particular symptom dimension and cortical thickness or surface area within any of the regions.

Children:

In pediatric OCD patients with ordering and symmetry symptoms (N=181), the surface area of the left cuneus was higher (Cohen’s d=0.49) (see Table S20 and Figure S7 in the data supplement). None of the regions analyzed showed significant differences in cortical thickness or evidence of associations with the other symptom dimensions.

Influence of age at onset and illness duration on cortical thickness and surface area.

Adult OCD patients who had an adult illness onset (N=775), compared with controls (N=1,436), showed thinner cortices in the left and right hemisphere overall. Regionally, we observed thinner cortices in frontal and temporal regions of adult-onset patients (Cohen’s d values between −0.11 and −0.16) (see Table S21a and Figure S8 in the data supplement). We also found lower surface areas for the left transverse temporal cortex (Cohen’s d=−0.17) and the left pars opercularis (Cohen’s d=−0.14) in OCD patients who had an adult illness onset (see Table S22a and Figure S9 in the data supplement). We did not detect significant differences in cortical thickness or surface area in adult OCD patients who had a childhood illness onset (N=646) compared with controls, or when we compared adult-onset and childhood-onset patients directly (see Table S21b,c and Table S22b,c in the data supplement).

Furthermore, we did not observe any significant linear (see Tables S23–S26 in the data supplement) or quadratic (see Tables S37–S40 in the data supplement) associations between age at onset or illness duration as continuous variables and cortical thickness or surface area changes in the adult (N=1,419) or pediatric (N=708) OCD groups.

Association between illness severity and cortical thickness and surface area.

We did not detect any significant linear (see Tables S27 and S28 in the data supplement) or quadratic (see Tables S41 and S42 in the data supplement) associations in either the adult (N=1,453) or the pediatric (N=404) OCD patients between illness severity (YBOCS) and cortical thickness or surface area.

Meta-Analysis

Decreased cortical thickness of the inferior parietal cortex was present in adult patients with OCD compared with healthy controls, but at a less stringent significance threshold (Cohen’s d≈−0.14; p<0.01, uncorrected). The meta-analysis did show significant widespread effects of medication on cortical thickness and a lower surface area for the transverse temporal cortex in adult OCD patients. The pediatric meta-analysis, also at a less stringent significance threshold (Cohen’s d≈−0.31; p<0.05, uncorrected), showed decreased cortical thickness of the inferior and superior parietal cortex in children with OCD. In addition, scanner field strength did not significantly explain the effect size estimates of cortical thickness or surface area differences in adult and pediatric OCD patients compared with controls (see Supplementary Section S3 in the data supplement).

Discussion

Cortical Thickness

This is the largest neuroimaging study conducted on cortical measures in OCD to date. We found that the parietal cortex was consistently implicated in both adult and pediatric OCD, which is consistent with previous VBM and FreeSurfer studies (12, 25). Lower cortical thickness of the inferior parietal cortex in adult OCD patients compared with controls is in accordance with results reported by Kühn et al. (10) and the OBIC consortium (12). Lower cortical thickness of the inferior and superior parietal cortex in children with OCD is a novel finding. The only other study of cortical thickness in pediatric OCD that we are aware of found lower cortical thickness for another parietal region, the supramarginal gyrus (26). Other imaging studies have reported lower gray matter volume in the parietal lobe, especially in the angular gyrus of the inferior parietal lobe, in children and adults with OCD (25, 27).

In contrast with previous mega-analyses from the OBIC consortium, we did not find cortical thickness abnormalities in the orbitofrontal cortex, anterior cingulate cortex, or dorsomedial prefrontal cortex. Six of our 38 data sets (the entire OBIC sample) were identical to those included in the OBIC mega-analysis. Apart from a much larger sample size and the inclusion of samples from more countries, no discrepancies between demographic and clinical characteristics could be found between this sample and the OBIC sample. Thus, these inconsistencies likely reflect differences in analytical methods and the overall sample size. While FreeSurfer measures thickness and surface area separately, it segments whole structures based on probabilistic information from a predefined atlas (20), compared with VBM’s voxel-wise registration (28). It is mainly global or regional differences in structure that can be inferred from these atlas-based FreeSurfer analyses, as opposed to local morphology, as with VBM. Moreover, the OBIC consortium’s FreeSurfer mega-analysis (12) was conducted using vertex-based analyses rather than the atlas-based approach we used in the present study. It is thus possible that certain abnormalities on the vertex level are not detectable when data are averaged across whole regions (29). Notably, the OBIC sample included only 1.5-T scans and was processed using an earlier version of FreeSurfer (version 4.5). Further research using higher-resolution parcellation (such as that described in reference 30) is necessary to validate our results.

In this study, we had sufficient statistical power to detect subtle cortical abnormalities in OCD (Cohen’s d values, −0.15 to −0.31) (see Supplemental Section S5 in the data supplement). Large-scale studies such as ours are well powered to distinguish consistent, generalizable findings from false positives. Structural MRI provides a crude and indirect measure of putative alterations at the molecular level, but these subtle abnormalities in the parietal cortex may still be relevant from a pathophysiological perspective (31). These results provide insight into what systems are affected, which can promote further research to evaluate specific pathways implicated in the pathophysiology of OCD.

Neuroimaging studies of normal brain maturation demonstrate a continuous increase in parietal thickness, reaching peak values around age 12, followed by a steady decrease over subsequent decades (32). In terms of neurodevelopmental abnormalities, our results may be cautiously interpreted as evidence for a relationship between the expression of OCD and disturbances in factors influencing radial cortical expansion, which influences gray matter thickness, rather than factors influencing the tangential expansion that determines the overall surface area (33). In this context, our results could indicate an altered cortical maturation in OCD resulting in a thinner parietal cortex in early childhood that persists into adulthood, although further confirmatory work using longitudinal samples is needed.

Cognitive studies in OCD suggest that the parietal cortex plays a significant role in accounting for the cognitive deficits seen in OCD patients. Parietal lobe activation may be related to attention, set shifting, planning, and response inhibition, which are also reported to be impaired in OCD patients (34) and reflect a lack of cognitive flexibility that may be related to the repetitive nature of OCD symptoms and behaviors. The inferior parietal cortex is an important node in both the fronto-parietal network and the default mode network. Several studies have reported altered connectivity within these networks in patients with OCD (35–37). The phenomenology of the disorder is consistent with the idea of a disrupted relationship between ongoing internal thought and external information, in that patients often focus excessively on internally generated fears that are inconsistent with evidence present in the external environment (38).

We reported lower cortical thickness of numerous regions throughout the brain in medicated adult OCD patients. These medication effects partially overlap with those reported in previous research (12). Although these findings need to be interpreted with caution, it has been suggested that antidepressants may modulate plasticity in the brain (39). Additionally, post hoc analyses suggest that these medication effects are strongest in those patients taking antidepressants with adjuvant antipsychotics (see Table S36a–c and Supplementary Section S4 in the data supplement). Alternatively, those patients taking medication could represent a more clinically severe cohort that manifests these morphometric abnormalities. The results may have been confounded by a higher illness severity and a higher percentage of comorbid depression in the medicated adult OCD group (see Table S35 in the data supplement). However, results of post hoc analyses comparing the most severe (YBOCS score >30) unmedicated OCD patients with controls did not show the same pattern of medication effects. Nevertheless, the cortical abnormalities in currently medicated OCD patients could reflect persistent abnormalities related to greater OCD severity before treatment. In addition, medication effects persisted after adding a covariate correcting for illness severity (data not shown). The lack of association between severity according to the YBOCS and cortical measures could be due to medication reducing the symptom severity. Additionally, current symptom severity may not be an optimal measure for capturing the long-term disease severity.

With regard to retrospectively ascertained age at onset in adult patients, the lack of inferior parietal abnormalities in the adult sample with childhood onset might be explained by insufficient power. When looking at the effect sizes, decreased cortical thickness of the inferior parietal cortex was present in adult patients with a childhood onset compared with healthy controls, but at a less stringent significance threshold. The effect size was even slightly larger than that of the main group comparison, suggesting a power issue rather than a lack of inferior parietal abnormalities. In contrast, adult illness onset was associated with widespread thinner cortices. The adult-onset group is older than the childhood-onset group but also has a higher percentage of medicated patients. Post hoc analyses showed that these effects mostly disappear when medication status is corrected for, suggesting that these findings are driven mainly by medication.

Cortical structural deficits were not associated with comorbid depression or anxiety. The effect sizes of these small subgroups with comorbid anxiety or depression indicate insufficient statistical power to address this issue with certainty. From a clinical point of view, comorbid Tourette’s syndrome and attention deficit hyperactivity disorder are more relevant to study in children with OCD. Because of the lack of systematic clinical investigation of comorbidities, we were unable to investigate this. Common comorbidities may be more aptly termed interacting variables, as they interact in complex ways. Therefore, excluding comorbid conditions will ignore complex interactions that are often integral to the disorder.

Surface Area

The transverse temporal cortex surface area deficit was consistent across analyses in adult OCD. This region belongs to the primary auditory cortex and has not been implicated in OCD pathophysiology before. Lower cortical thickness and lower volume of this region have been associated with auditory hallucinations in schizophrenia (40). Previous attempts to detect structural alteration in this region may have been hampered by small samples or the modest sensitivity of conventional volumetric approaches. The advantage of high statistical power allows us to examine abnormalities throughout the brain without the need to prespecify regions of interest and thus to identify new regions putatively associated with the disorder. Further research is necessary to understand the involvement of the transverse temporal cortex in OCD.

Medicated children with OCD had smaller left and right hemisphere total surface area, reflecting a diffuse pattern of frontal surface area deficits. These findings cannot be explained by differences in illness severity, comorbidity, or age at onset (see Table S35 in the data supplement). This may indicate delayed cortical maturation, although longitudinal studies are needed to prove that. The surface area of these frontal regions matures over a more prolonged interval during adolescence (41) and may be especially prone to a maturational delay in pediatric OCD, possibly affected by medication status. Such delayed maturation may alter functional connections with other regions through decreases in growth and branching of dendritic trees and the number of synapses associated with gray matter volume (42), which may persist into adult OCD even if surface area measures normalize after the transition to adulthood. The absence of cortical surface area abnormalities in the adult OCD patients who had a childhood onset could indicate such normalization.

Limitations

We used existing data across samples worldwide, and the data collection protocols were not prospectively harmonized. Imaging acquisition protocols and clinical assessments therefore differed across studies, which limits analysis of sources of heterogeneity. We note also that the T₁-weighted scans were not collected with direct measures of head motion, which may have introduced potential motion-induced bias in cortical measures (43).

In addition, FreeSurfer measurements may benefit from manual edits if they are made consistently across all scans. Although we had an extensive standardized protocol for quality checking, the individual sites did not perform manual editing, as this could have resulted in increased variation in the data across sites because of the high number of sites involved.

We reported widespread medication effects in both adults and children with OCD. However, the present study did not allow a reliable investigation of medication effects because of its cross-sectional design and a lack of detailed information on history, type, dosage, and duration of psychotropic medication treatment. Our results must therefore be interpreted with caution, and we cannot make any conclusions about the effect of anti-OCD medication. Further efforts, such as intervention studies with comparisons before and after medication, are required to draw valid conclusions on the impact of medication use on cortical structure.

Several studies using a symptom dimensional approach suggest that symptom dimensions may be mediated by partially distinct neural systems (44, 45). Except for the association between a higher surface area of the left cuneus and ordering/symmetry dimension in children, we did not detect thickness or surface area effects of the other symptom dimensions in children and adults. An explanation may be that symptom subtype differences are more focal and remain undetected in this atlas-based analysis. On the other hand, variance in use of instruments across the participating sites may have led to suboptimal harmonization of the symptom dimension scores, which in turn may explain the absence of associations with symptom dimensions in our study. During harmonization, we defined symptom dimensions in a binary manner as absent or present for each participant based on the YBOCS symptom checklist, whereas previous studies have correlated the dimension scores with cortical measures.

As a minor limitation, we note that we followed the pediatric-adult age cutoff of the study samples to split the data into adult (age 18 or over) and pediatric (age under 18) groups, a cut-off that may not be an optimally related to the onset and evolution of OCD (46, 47). In addition, the pediatric sample represents a wide age range, including puberty. We did not have enough data on pubertal stage to take pubertal development into account. Given the role of hormonal influence on cortical structures, this will be useful to pursue in future research.

Conclusions

The parietal cortex was implicated in both adult and pediatric OCD. These results support the hypothesis that the pathophysiology of OCD cannot be explained solely by alterations of the classical cortico-striato-thalamo-cortical regions and emphasize the importance of parietal regions. Widespread cortical thickness abnormalities were found in medicated adult OCD patients, while more pronounced surface area deficits were found in medicated pediatric OCD patients. Cortical thickness and surface area represent distinct features of the cortex and may be differentially affected by OCD and possibly moderated by medication status. Further work using longitudinal designs and incorporating genetic and environmental variables will be useful in understanding the precise mechanisms underlying the structural abnormalities preceding the onset of the illness and occurring during the course of the illness.