Main findings

This FreeSurfer mega-analysis of multisite T ₁-weighted MRI scans showed that patients with OCD exhibit decreased cortical thickness in a number of frontoparietal regions. These included the superior and inferior frontal gyri findings similar to previous VBM meta-analyses and cortical thickness work. In addition to other temporoparietal regions such as the precentral, posterior cingulate, middle temporal, inferior parietal and precuneus gyri. There was also limbic involvement as evidenced by decreased hippocampal volume compared with healthy controls. Another finding was a group × age interaction effect for left parietal thickness suggesting that there is exaggerated cortical thinning of the parietal gyri with advancing age and/or illness duration in patients compared with controls.

The most notable finding in this FreeSurfer study, consistent with the findings of our VBM mega-analysis, ^{Reference de Wit, Alonso, Schweren, Mataix-Cols, Lochner and Menchón5} is the involvement of the inferior frontal gyrus/operculum and superior frontal gyrus/dorsomedial prefrontal cortex in OCD. This is important, insofar as findings from smaller studies have been contradictory. ^{Reference van den Heuvel, Remijnse, Mataix-Cols, Vrenken, Groenewegen and Uylings22,Reference Pujol, Soriano-Mas, Alonso, Cardoner, Menchón and Deus23} This finding in both the mega-analyses suggests that cortical thinning and grey matter volume could relate to a similar underlying pathological process responsible for abnormalities in these regions. This finding was even more pronounced when excluding patients on medication and with lifetime comorbid anxiety and depression, indicating that it may be specific to OCD. Cortical thinning in the inferior frontal gyrus and superior frontal gyrus is also consistent with the findings of some previous VBM studies ^{Reference Radua, van den Heuvel, Surguladze and Mataix-Cols7,Reference van den Heuvel, Remijnse, Mataix-Cols, Vrenken, Groenewegen and Uylings22,Reference Yoo, Roh, Choi, Kang, Ha and Lee24} and one cortical thickness study. ^{Reference Shin, Yoo, Lee, Ha, Lee and Lee11} Decreased cortical thickness of the inferior frontal gyrus and superior frontal gyrus may be involved in impairments of cognitive and emotional control in OCD; the inferior frontal gyrus together with the anterior insula is part of a circuit implicated in behavioural inhibition and disgust-related responses. ^{Reference Tops and Boksem8} The inferior frontal gyrus is also part of a network implicated in working memory and emotion regulation, ^{Reference Tops and Boksem8} together with the interconnected precentral and superior (including dorsomedial) frontal gyri, that both show decreased cortical thickness in the present study. The decreased thickness in these regions could also be related to the abnormal connectivity evident in the default-mode and attention network of people with OCD. Previous investigations have demonstrated that there is an increased functional connectivity between the prefrontal and insula/cingulate regions, suggesting greater cognitive control, but weaker connectivity between precentral gyrus and prefrontal cortex (suggesting lower motor inhibition) in OCD. ^{Reference Li, Li, Dong, Luo, Han and Xiong25} This incongruence between the dorsal and ventral networks has also been demonstrated in another study which investigated the structural covariance of cortical thickness networks in OCD. ^{Reference Kim, Jung, Kim, Jang and Kwon26}

Both our data and previous work suggest at least some involvement of parietal and temporal regions in OCD, ^{Reference Rotge, Langbour, Guehl, Bioulac, Jaafari and Allard4,Reference Radua, van den Heuvel, Surguladze and Mataix-Cols7,Reference Shin, Yoo, Lee, Ha, Lee and Lee11} consistent with the suggested involvement of the fronto-parietal and limbic circuits in the pathophysiology of OCD. The results also held true after exclusion of patients on medication and with lifetime comorbid anxiety and/or depression. Evidence from functional MRI and cognitive testing has demonstrated that parietal and temporal regions are involved in impairment of inhibitory control ^{Reference Bannon, Gonsalvez, Croft and Boyce27,Reference Penadés, Catalán, Rubia, Andrés, Salamero and Gastó28} and executive planning ^{Reference van den Heuvel, Veltman, Groenewegen, Cath, van Balkom and van Hartskamp29,Reference Chamberlain, Fineberg, Blackwell, Clark, Robbins and Sahakian30} in OCD, whereas the hippocampus together with amygdala and anterior cingulate cortex may comprise affective components of the OCD fronto-cortical circuit. ^{Reference Menzies, Chamberlain, Laird, Thelen, Sahakian and Bullmore31} Resting-state studies confirm abnormal connectivity of frontoparietal networks, with increased connectivity of motor networks, ^{Reference Stern, Fitzgerald, Welsh, Abelson and Taylor32} but decreased connectivity in frontotemporal networks, especially in the posterior cingulate (an important hub of the default-mode network). ^{Reference Zhang, Wang, Yang, Wu, Li and Chen33,Reference Meunier, Ersche, Craig, Fornito, Merlo-Pich and Fineberg34}

Although previous VBM studies have found greater striatal volume, ^{Reference Rotge, Langbour, Guehl, Bioulac, Jaafari and Allard4,Reference Radua, van den Heuvel, Surguladze and Mataix-Cols7} the same findings could not be replicated in this mega-analysis. A possible reason for the absence of this finding is as a result of the manner in which Freesurfer segments subcortical structures. Whereas VBM performs segmentation of the striatum based on contrast differences between grey and white matter, Freesurfer segments whole structures based on probabilistic information from a predefined atlas. ^{Reference Fischl, Salat, Busa, Albert, Dieterich and Haselgrove35} Therefore, only global changes in subcortical structure can be inferred from Freesurfer, rather than local morphology as with a voxel-based method such as VBM. It is thus possible that these volume increases in the striatum are not detectable when averaging across the whole structure. Also, because the MRI scans were 1.5 T, there are limitations to the signal-to-noise ratio of the data, as well as to the resolution. ^{Reference Zhou, Zambelli and Heredia36}

Findings from the group × age interaction analyses

Group × age interaction analyses show that the inferior and superior parietal regions demonstrate relatively exaggerated age-related cortical thinning in the OCD group compared with controls. Cortical thinning with age is normal, ^{Reference Sowell, Thompson and Toga37,Reference Grieve, Clark, Williams, Peduto and Gordon38} but this exaggerated decrease in the parietal lobes of our patient cohort might explain some characteristics typical of OCD: the parietal lobes are involved in planning and response inhibition, both processes that are impaired in OCD and potentially aggravated with age and disease duration. ^{Reference Menzies, Chamberlain, Laird, Thelen, Sahakian and Bullmore31} Although we did not find any significant association of disease severity or disease duration with grey matter volume ^{Reference de Wit, Alonso, Schweren, Mataix-Cols, Lochner and Menchón5} or thickness, the possibility exists that the more pronounced thinning in parietal regions in older patients, compared with older controls, may partly be explained by longer disease duration. Additionally the association with age was squared, indicating that rapid parietal cortical thinning occurred at set age intervals, rather than linear progressive thinning as patients became older. Even in healthy individuals, grey matter density decline in parietal regions is non-linear, with the majority of loss occurring between 50 and 70 years of age. ^{Reference Sowell, Thompson and Toga37} To further disentangle the effect of ageing and the effect of disease chronicity a longitudinal design is needed.

Comparison with findings from other VBM studies

There is some consistency of this study with previous VBM meta-analyses. Most notably previous VBM studies have also found grey matter volume decreases in inferior frontal, orbitofrontal and dorsolateral prefrontal regions. ^{Reference Rotge, Langbour, Guehl, Bioulac, Jaafari and Allard4,Reference van den Heuvel, Remijnse, Mataix-Cols, Vrenken, Groenewegen and Uylings22} However in contrast to previous VBM work, there were no cortical thickness changes in the anterior cingulate and supramarginal gyri. In addition this mega-analysis described decreased cortical thickness in regions such as the middle temporal, inferior parietal, precentral, precuneus and lateral occipital regions, for which grey matter volume changes were not evident in previous VBM meta-analyses. ^{Reference Rotge, Guehl, Dilharreguy, Tignol, Bioulac and Allard3,Reference Rotge, Langbour, Guehl, Bioulac, Jaafari and Allard4}

The inconsistencies in findings likely reflect, at least partly, the differences in the methodologies of the analyses techniques. Segmentation of structures is performed in fundamentally different ways across the techniques, i.e., FreeSurfer uses a surface-based morphometry approach compared with VBM's voxel-wise registration. ^{Reference Clarkson, Cardoso, Ridgway, Modat, Leung and Rohrer39} The inconsistencies might also indicate that for some brain regions differences in grey matter volume are not explained by differences in cortical thickness. ^{Reference Palaniyappan and Liddle40} Grey matter volume differences as found by VBM can result from (a) a change in cortical thickness in the absence of surface area changes, (b) a change in surface area in the absence of cortical thickness changes, or (c) changes in both cortical thickness and cortical surface area. ^{Reference Palaniyappan and Liddle40} An increase in surface area together with a decrease in cortical thickness might mask the effect when expressed in grey matter volume. Conversely, minor increases or decreases in both surface area and cortical thickness may remain undetected when analysed separately, but might reach significance when combined with grey matter volume. Therefore, findings of cortical thickness and grey matter volume are complementary and the combined interpretation of both measures strengthens the conclusions that can be drawn from this data-set.

Comorbid anxiety and depression

Given that comorbid anxiety and depressive disorders are frequently observed in OCD, ^{Reference Ruscio, Stein, Chiu and Kessler41} we examined whether the lifetime presence of such disorders would partly explain or influence our main findings of decreased cortical thickness in patients with OCD. In the current study prefrontal involvement was most pronounced in the patients with comorbid anxiety disorders, whereas ventral frontal structures and the hippocampus were most affected in those with comorbid major depressive disorder. Previous work has also shown these regions to be affected in both anxiety and major depressive disorder, perhaps suggesting a shared mechanism underlying emotional and cognitive deficits across disorders. ^{Reference Radua, van den Heuvel, Surguladze and Mataix-Cols7,Reference van Tol, van Der Wee, van den Heuvel, Nielen, Demenescu and Aleman42,Reference Cardoner, Soriano-Mas, Pujol, Alonso, Harrison and Deus43} In addition, decreased cortical thickness of the inferior frontal regions, insula and dorsomedial prefrontal cortex was found in the OCD group compared with controls when excluding the confounding effects of comorbid anxiety and depression in this sample. This suggests that involvement of these cortical areas is unique to OCD. This is consistent with previous work, ^{Reference van Tol, van Der Wee, van den Heuvel, Nielen, Demenescu and Aleman42} together with findings from the OCD VBM mega-analysis. ^{Reference de Wit, Alonso, Schweren, Mataix-Cols, Lochner and Menchón5}

Impact of medication

We examined whether psychotropic medication use has a confounding effect on the changes observed in cortical thickness or subcortical volume. For the prefrontal, temporal and occipital regions there was greater cortical thinning in patients with OCD on medication compared with unmedicated patients. Although it has been suggested that selective serotonin reuptake inhibitors might modulate neuroplasticity in brain regions of people with OCD, ^{Reference Hoexter, de Souza Duran, D'Alcante, Dougherty, Shavitt and Lopes44} these findings show little internal consistency. Thus, the extent to which medication has consistent effects on brain structure remains open for future study. That being said, after excluding patients on medication, the main findings in the inferior and superior frontal, precentral, posterior cingulate and middle temporal regions still held true, indicating that irrespective of medication use, these regions are altered in patients with OCD. Indeed, the findings were somewhat more pronounced when patients on medications were excluded, emphasising the power that mega-analyses have to address confounding variables.

The negative association of lateral occipital thickness with the YBOCS severity score is also a unique finding, together with the increased cortical thickness of this region in patients with OCD that were not on medication. Although not part of the corticostriatal and limbic pathways thought to be affected in OCD, previous volumetric studies have found decreased grey matter volume in the occipital lobe of people with OCD. ^{Reference Togao, Yoshiura, Nakao, Nabeyama, Sanematsu and Nakagawa45,Reference Kim, Lee, Kim, Kim, Kim and Han46}

Findings from the symptom dimensions analysis

In addition, the finding that different symptom dimensions have distinctive neuroanatomical correlates is consistent with previous work. ^{Reference Radua, van den Heuvel, Surguladze and Mataix-Cols7,Reference Narayan, Narr, Phillips, Thompson, Toga and Szeszko16} Similar to our present cortical thickness findings, associations of temporal and occipital regional volumes with the harm/checking and symmetry/order dimensions have been shown before, ^{Reference van den Heuvel, Remijnse, Mataix-Cols, Vrenken, Groenewegen and Uylings22} as well as positive associations of the sexual/religious dimension and hoarding dimensions with prefrontal and lateral orbitofrontal volumes. ^{Reference Kim, Lee, Kim, Kim, Kim and Han46} Not all FreeSurfer findings of the symptom dimension analyses are consistent with previous VBM analyses, ^{Reference van den Heuvel, Remijnse, Mataix-Cols, Vrenken, Groenewegen and Uylings22,Reference Alvarenga, do Rosário, Batistuzzo, Diniz, Shavitt and Duran47} possibly as a result of clinical heterogeneity of participants and differences in quantification of the dimension variables for each study. For example, in this study, symptom dimensions were defined in a binary manner as absent or present for each participant, whereas the above-mentioned studies have correlated the dimension scores with grey matter volumes. Our seemingly incongruent findings of a positive association between cortical thickness and specific dimensions coupled with decreased cortical thickness in patients may reflect these methodological problems in measuring symptom dimensions. In contrast to a considerably clear image of the general structural brain changes in OCD, there is arguably a paucity of literature to support a detailed model of the underlying neuroanatomy of each particular symptom dimension and therefore replication of these findings is warranted. Also, because of the limitations of the clinical data-set, a cluster analysis of symptom dimensions was not possible. ^{Reference Lochner, Hemmings, Kinnear, Nel, Hemmings and Seedat48} Conclusions drawn from this analysis about the neurocircuitry of symptom dimensions must be tentative.

Strengths and limitations

The benefit of a large multisite study such as this is the greater sample size, which decreases the likelihood of type I and type II errors. Another benefit is the congruency of preprocessing and analytical methods in a mega-analysis compared with meta-analyses. Sample size and other clinical confounders such as comorbid anxiety/major depressive disorder and medication use have limited previous investigations of cortical thickness in OCD. By performing an analysis in this large group, these clinical confounders can be controlled for more easily.

The present study also has some important limitations. These include the potential confounders of scan sequence and other site differences. Therefore, variance in scan sequence was controlled for in the main and regression analyses. The measurement of symptom dimension scores was based on the symptom checklist of the YBOCS and was perhaps suboptimal because of variance in use of the instrument across the participating centres. Although the strength of this study is that clinical confounders could be controlled for with a stepwise regression and participant exclusion, it is still difficult to fully disentangle the confounding effects of medication and comorbidity. Future studies should employ instruments specifically designed and validated for this purpose. In addition, our group × age findings are weakened by the cross-sectional design of the study. Future longitudinal studies of brain imaging in OCD will help shed light on the neurodevelopmental aspects of this disorder. This involves both the altered neurodevelopment before the onset of the disease and the disease-related and treatment-related neuroplastic changes during the course of the disease. Also, we performed this analysis on 1.5 T brain scans, which do have inherent resolution limitations compared with 3 T MRI. Further work, using longitudinal designs, 3 T MRI and incorporating genetic and environmental variables, will be useful in understanding the precise mechanisms underlying the structural abnormalities preceding the onset of the disease and that occur during the course of the disease.

In summary, this is the largest cortical thickness and subcortical volume study of OCD to date. The results support the importance of structural abnormalities in fronto-striatal, fronto-parietal and limbic circuits in the pathophysiology of OCD. There is notable consistency of the present FreeSurfer findings with previous VBM meta-analyses and cortical thickness findings with regard to inferior frontal gyrus/operculum and superior frontal gyrus/dorsomedial prefrontal cortex, suggesting that cortical thinning as well as grey matter volume abnormalities could perhaps relate to a similar pathological process underlying the disorder.