The aim of this thesis was to get a better understanding of the neural correlates of obsessive-compulsive disorder (OCD), using structural neuroimaging. Structural T1-weighted magnetic resonance imaging (MRI) scans of the ENIGMA (Enhancing NeuroImaging Genetics through Meta-Analysis) OCD working group were analyzed with FreeSurfer. We obtained subcortical volume, cortical thickness and cortical surface area measures, to investigate morphometric alterations throughout the brain of patients with OCD in relation to development and aging, and clinical factors such as comorbid diagnoses, medication use and symptom dimensions. As a follow-up research question to the subcortical and cortical alterations, we investigated how these morphometric alterations relate to asymmetry in OCD. We also compared measures of structural brain connectivity, based on structural covariance networks, between OCD patients and controls, using a graph theoretical approach. We used two different approaches to analyze the data, i.e., using mega-analysis and meta-analysis. Next, we critically evaluated the methodology of, and results obtained by, meta- and mega-analyses to determine the optimal method. Finally, we investigated which morphological brain alterations are specific to OCD or non-specific and shared by other neurodevelopmental disorders such as attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD).
In this final chapter, I revisit the main findings per chapter, providing an overall summary. I discuss how these findings relate to each other and how this work complements and advances existing literature. I will subsequently discuss some methodological issues regarding the work presented in this thesis. Lastly, this chapter considers the potential clinical implications of our findings and recommendations for future research.
In Chapter 2, we reviewed the literature regarding gray matter (GM) and white matter (WM) morphological brain alterations in children and adults with OCD. These studies together suggested widespread morphological brain alterations across several different regions and circuits, and their inter-connectivity. What became clear from this review was that the pathophysiology of OCD cannot be explained solely by alterations in function and structure of the classical cortico-striato-thalamo-cortical (CSTC) regions exclusively and emphasizes the importance of other brain regions as well, such as the fronto-limbic and fronto-parietal areas and the cerebellum.
In Chapter 3, we assessed regional volume differences in subcortical GM volume between 1,830 OCD patients and 1,759 controls of 35 cohorts worldwide. We found a distinct pattern of subcortical differences in children and adults with OCD. Results showed a larger thalamus in children with OCD and a larger pallidum and smaller hippocampus in adults with OCD, with overall small effect sizes. The larger thalamus in children with OCD was driven by the unmedicated patients and seemed unrelated to comorbid anxiety and depression. The larger thalamic volume in children with OCD is in line with previous literature (Gilbert et al., 2000; Rosenberg et al., 2000) and suggests an altered neurodevelopment in children with OCD. The hippocampal effect seemed to be driven, at least partly, by adult OCD patients with comorbid depression and an adult disease onset of disease. Hippocampal abnormalities have been found in several other psychiatric disorders (Hibar et al., 2016; Schmaal et al., 2016; Van Erp et al., 2016), thus seem disease non-specific (Sala et a., 2004). In line with the literature we found a larger pallidum in adult patients with OCD compared to controls (Peng et al., 2012; Radua et al., 2009; Rotge et al., 2010; Hu et al., 2018). Radua et al. (2010) reported greater lenticular (i.e., putamen and pallidum) volume in OCD patients compared to controls and patients with other anxiety disorders. Several studies suggest a role for the pallidum in repetitive and disinhibited reinforced behavior (Laplane et al., 1989; Kretschmer, 2000; Bechard et al., 2016). Since repetitive behaviors differentiate OCD from other anxiety disorders, the increased pallidum volume in OCD may reflect these unique symptoms (Radua et al., 2010). Our findings also suggested that the child-onset adult OCD patients drove the pallidum effect. We therefore hypothesize that a larger pallidum in adult OCD patients could be the result of disease chronicity. Taken together, we postulate that the increased thalamus volume in children might be an early and transient marker of the disease, whereas the increased pallidum volume in adults might be a consequence of the disease related to disease chronicity.
In Chapter 4, we assessed cortical thickness and surface area differences between 1,905 patients with OCD and 1,760 healthy control subjects of 38 cohorts. Cortical thickness alterations of the parietal cortex were implicated in both adults and children with OCD. Effect sizes in children (≈ -0.3) were larger than in adults (≈ -0.16), but overall effect sizes were small. These results indicate an altered cortical neurodevelopment resulting in a thinner parietal cortex in early childhood, persisting in adulthood. However, confirmatory work using longitudinal samples is warranted. The structural alterations in the parietal cortex might be associated with deficits in functional networks such as the fronto-parietal network (FPN) related to cognitive functioning (attention, set shifting, and response inhibition) and default mode network (DMN) related to internal processing, which have also been reported to be impaired in OCD patients (Graybiel et al., 2000; Gursel et al., 2018). Goncalves et al. (2017) investigated the interactions between FPN and DMN in relation to OCD symptomatology and suggested that a disturbed interplay between self-referential processes, attributed to the DMN, and goal-directed behavior, related to the FPN, contributes to OCD symptomatology.
Widespread cortical thickness alterations were found in medicated adult OCD patients (versus controls), while more pronounced surface area alterations were found in medicated pediatric OCD patients (versus controls). Unmedicated OCD patients did not differ from controls. Cohen’s d effect sizes were slightly larger than the main group comparison, up to d=-0.24 in adults and d=-0.33 in children. These results must be interpreted with caution, since this study did not allow a reliable investigation of medication effects, due to lack of detailed information on dosage and duration of medication use, and medication history. Nevertheless, these findings highlight the importance of accounting and controlling for medication when assessing structural alterations in patients with OCD. Further efforts are required to draw valid conclusions on the impact of medication use on cortical structure, mainly involving longitudinal studies on long-term effects of medication on the developing brain over the lifespan. Overall results of this study support the hypothesis that the pathophysiology of OCD cannot be explained solely by alterations in the classical CSTC regions and emphasize the importance of parietal regions and the impact of medication status.
In Chapter 5, McKay and colleagues expressed their concerns about the minimum effect size at which one may declare imaging results to be substantively, specifically, and causally related to putative psychopathological states. The first issue concerns the extent of structural alterations seen in psychiatric disorders in general and in OCD in particular. The authors advocate a standard including a minimum threshold for implying a role for a brain area involved in psychiatric disorders relative to healthy controls, as well as a critical value or heuristic for making claims about this role. The second issue concerns the specificity of structural alterations in brain imaging studies. MacKay et al. mentioned the small effect size of the thalamus finding (Chapter 3) and that it may not be specific to OCD. Lastly the third issue was causality. The authors mention that findings are correlative and not demonstrably causative.
We responded to these three concerns raised by McKay et al. First, previous OCD imaging studies have been comprised of relatively small samples, resulting in low statistical power. Low statistical power negatively affects the likelihood that a statistically significant finding actually reflects a true effect (Button et al., 2013). With meta- and mega-analyses, we can put these results in context, and better estimate true effect sizes. Admittedly, our findings remain subtle. However, Cohen’s (Cohen, 1988) rules of thumb fail to address the point that even, a very small effect size can help in understanding the pathophysiology of a disorder. The second issue concerned the (non-)specificity of the findings. It is becoming increasingly clear that mental disorders share genetic risk factors and so it is not surprising that there also is overlap in the brain circuits involved. Nevertheless, results by other ENIGMA disease working groups do not show thalamus alterations in their patient groups indicating disease specificity at least to some level. Third, indeed structural MRI provides only a crude and indirect measure of putative alterations at the molecular level. However, taken together, a broad range of basic and clinical work has certainly provided mechanistic insights into how specific brain regions may contribute to OCD. Large-scale studies such as ours are well powered to distinguish consistent, generalizable findings from false positives. Our findings give us good insight as to what systems may be more affected than others and promote further research to evaluate specific pathways implicated in the causes and consequences of disease.
In Chapter 6, we assessed subcortical and cortical left-right asymmetry between 2,279 patients with OCD and 2,093 healthy controls of 46 cohorts. Children with OCD compared to controls had a more leftward asymmetry of the thalamus and less leftward asymmetry of the pallidum volume. These effects both had Cohen’s d values of around 0.2, which indicates their subtlety. In Chapter 3 we reported larger thalamus volumes in children with OCD. Chapter 6 revealed that this abnormality was largely left lateralized e.g. driven by a larger left than right thalamus. The abnormal asymmetry in the pallidum is less straightforward. In Chapter 3 we did not report volumetric alterations in pallidum volumes in children with OCD compared to controls. However, Chapter 6 did reveal abnormal pallidum asymmetry, which was driven by smaller left than right pallidum in children with OCD compared to controls. This abnormal asymmetry was not observed in adults with OCD. Then again Chapter 3 did show larger pallidum volumes in adult OCD patients with an early disease onset compared to controls. Taken together, these observations do not point to a simple model of structural alterations of the pallidum in OCD, but rather to a complex model involving disease stage-dependent decreases or increases relative to controls. In terms of cortical measures, we found no significant asymmetry of cortical thickness or surface area in the pediatric data and no evidence for asymmetry in either subcortical or cortical measures in the adult data. The subtle abnormal asymmetry of subcortical structures in pediatric OCD is not detectable in adults with the disorder. These findings may reflect altered neurodevelopmental processes in OCD affecting subcortical regions in the CSTC circuitries.
In Chapter 7, we assessed structural brain connectivity and hub organization based on structural covariance networks between 1,616 OCD patients and 1,463 healthy controls from 37 cohorts. Using a graph theory approach we demonstrated decreased clustering coefficient and decreased modularity as well as decreased small-worldness in the OCD structural network compared with healthy controls. Hub profiling, community detection and dice coefficient analysis indicated less modules and a different hub regions in the OCD network compared to the healthy control network. The global graph features were not associated with clinical characteristics such as illness severity, illness duration or age at onset.
Clustering coefficient measures how strongly connected brain regions are and decreased clustering may imply a subtle randomization of the OCD network, moving connections between brain regions away from clustered cliques, creating shortcuts between clusters and possibly affecting the information flow through hubs. Modularity provides a measure for the degree to which a network can be subdivided in separate modules and with community detection we can infer the organization of these modules. The decreased modularity is consistent with sparsity of modules and relative over-connectedness of certain brain regions, diminishing the ability of the network to flexibly adapt in patients with OCD. In addition, the reorganization of hub regions in the OCD network can be particularly devastating to network function, since hubs handle the majority of information traffic and are therefore vital for efficient network communication. To conclude, we showed disrupted modular organization and a different hub distribution in the OCD brain network. Our results support the hypothesis that OCD brain alterations are best described at the network level and involve alterations in the hierarchical structure of the brain.
A research question that followed from the first ENIGMA OCD projects (Chapter 3 and Chapter 4) was how to analyze this multicenter data most efficiently. In Chapter 8, we investigated which method had the greater sensitivity to detect subtle structural brain alterations related to OCD. We compared the inverse variance weighted random-effect meta-analysis model with a multiple linear regression mega-analysis model, but also with a linear mixed–effects random-intercept (random intercept for Site) mega-analysis model, using data from 38 cohorts including 1,905 patients with OCD and 1,760 healthy control subjects. Effect sizes were similar for the meta-analysis and linear regression mega-analysis. However, lower standard errors and narrower confidence intervals of both mega-analytical approaches compared to the meta-analysis suggest better performance of the mega-analytical approach over the meta-analytical approach. Additionally, Bayesian information criterion values indicated a better model fit of the linear mixed-effects random-intercept model compared to the linear regression mega-analytical model. Here we provided evidence that, in a multi-center study with a moderate amount of variation between cohorts, a linear mixed-effects random-intercept mega-analytical framework seems to be a more optimal approach for investigating structural neuroimaging data. Additionally, we advice researchers, when planning individual-participant data analyses in a multi-center setting, to pre-specify the choice and implementation of a mega-analysis or meta-analysis method as occasionally they lead to different conclusions.
Finally in Chapter 9, we assessed subcortical volumes, cortical thickness, and cortical surface area measures of in total 12,198 patients with ADHD, ASD, and OCD and controls from 151 cohorts worldwide, and investigated the overlap and specificity of structural brain alterations across neurodevelopmental disorders. We found ADHD-specific smaller intracranial volume (ICV) in adolescents and children, but not in adult ADHD patients. Overall, effect sizes were small. These results support the hypothesis that morphological alterations in ADHD may be due to a delay in brain maturation (Shaw et al., 2007), possibly normalized in adulthood. These results are also in line with the genetic correlation between risk for ADHD and smaller ICV (Klein et al., 2017). We also found ASD-specific thicker frontal cortices in adult patients, which has been found previously and may be linked to impaired cognitive control and executive dysfunction in ASD (Ecker et al., 2013). We did not find OCD-specific morphological alterations, nor alterations shared among all three disorders. Although previous meta-analysis showed significant morphometric overlap in two out of three disorders (Carlisi et al., 2017; Norman et al., 2016; Dougherty et al., 2016), in our study such findings did not survive multiple comparison correction. Overall, results showed a different pattern of subcortical and cortical alterations between the disorders across pediatric, adolescent, and adult patients, suggesting lifespan dynamics in morphological brain alterations between the disorders. Further efforts incorporating neural correlates, cognitive and genetic variables are necessary to understand the mechanisms underlying differing and overlapping deficits of these neurodevelopmental disorders.