Functional disruption of cerebello-thalamo-cortical networks in obsessive compulsive disorder

Abstract

BACKGROUND:

Obsessive-compulsive disorder(OCD) is characterized by intrusive thoughts and repetitive, compulsive behaviors. Neuroimaging studies have implicated altered connectivity among the functional networks of the cerebral cortex in the pathophysiology of OCD. There has, however, been no comprehensive investigation of the cross-talk between the cerebellum and functional networks in the cerebral cortex.

METHODS:

Forty-four adult participants with OCD and 43 healthy participants completed this functional neuroimaging study. We performed large-scale data-driven brain network analysis to identify functional connectivity patterns using resting-state functional magnetic resonance imaging data.

RESULTS:

OCD participants showed lower functional connectivity within the somatomotor network and greater functional connectivity among the somatomotor network, cerebellum and subcortical network(e.g. thalamus and pallidum, all Ps<0.005). Network-based statistics analyses demonstrated one component comprising connectivity within the somatomotor network that showed lower connectivity, and a second component comprising connectivity among the somatomotor network, and motor regions in particular, and the cerebellum that showed greater connectivity, in OCD relative to healthy participants. In OCD participants, abnormal connectivity across both network-based statistic-derived components positively correlated with OCD symptom severity(P=0.006).

CONCLUSIONS:

To our knowledge, this study is the first comprehensive investigation of large-scale network alteration across the cerebral, subcortical regions and the cerebellum in OCD. Our findings highlight a critical role of the cerebellum in the pathophysiology of OCD.

Introduction

Obsessive-compulsive disorder(OCD) affects 1–3% of the population(1,2) and is characterized by intrusive thoughts(obsessions) and repetitive behaviors(compulsions)(3,4). In OCD, alterations in habitual behaviors and thoughts(5–7) manifest in affect, motor planning, and cognition(8). Consistent with this range of symptoms, OCD is associated with alterations in several large-scale networks (e.g. the frontoparietal network(FPN), default-mode network(DMN), salience network(SN) and limbic network(LN,9,10). Because performance differences complicate interpretation of task-based functional magnetic resonance imaging (fMRI) studies, resting-state MRI methods are often employed to understand large-scale networks in OCD.

The FPN is critical to executive control and is comprised of the prefrontal cortex and temporo-parietal junction(11). Resting-state studies report lower functional connectivity in OCD versus healthy participants(10). These differences likely relate to dysfunctional inhibitory control in OCD(12). The DMN supports internally oriented attention and self-monitoring and includes the medial prefrontal cortex, posterior cingulate cortex, and hippocampus(13). The DMN has greater connectivity in OCD, specifically in the posterior cingulate and medial prefrontal cortices(14). These differences might be associated with disrupted self-monitoring and compulsive behaviors(15). The SN, comprised of the insula and anterior cingulate cortex, facilitates orienting toward salient external stimuli and internal events(16). Resting-state studies of OCD report greater connectivity in the SN(14,17), which may be associated with disrupted attention shifting(18–20). The LN, consisting of the orbitofrontal cortex and amygdala, is implicated in emotional processing and regulation(21). Hypothesis-driven studies using a priori seeds in the LN show disrupted functional connectivity between limbic and other(e.g., FPN, DMN) networks(22,23). There is evidence for alterations across several other networks in OCD, including the visual network(VN) and somatomotor network(SMN)(9,24). These studies indicate disrupted functional connectivity within and among cortical networks in OCD; more work is needed to elucidate the natureof these abnormalities.

The cerebellum provides output to the cortex, and “tunes” sensory input to facilitate behavioral (particularly motor) adjustment in response to feedback(25–27). The cerebellum is also involved in developing habits(28). Given these roles, and because the cerebellum provides output across the cortex(29), alterations in cerebellar-cortical network connectivity might underlie habitual behavior dysfunction in OCD that occurs across multiple domains(30). Yet, few studies have examined cerebellar-cortical network contributions in OCD. There is evidence for cerebellar dysfunction in OCD(31), however. A single photon emission computed tomography study revealed elevated regional cerebral blood flow in the cerebellum in OCD(32). A meta-analysis of task-based fMRI studies showed greater activity in anterior lobules I-VI of the cerebellum during task switching and response inhibition(33). Resting-state functional connectivity studies report greater global connectivity in the posterior cerebellum in OCD(34,35). OCD symptoms are also positively correlated with parietal cortex-cerebellar connectivity(36,37). These findings suggest that disrupted cerebellar-cortical connectivity is implicated in the pathophysiology of OCD, but they do not describe specific cerebellar-cortical connections and their relationship with symptom severity. A comprehensive investigation of cerebellar-cortical coupling would clarify the role of the cerebellum in the pathophysiology of OCD.

While the literature suggests that OCD symptoms are explained by aberrant functional connectivity among large-scale brain networks, including the FPN, SN, DMN and LN(9,10), these studies employed seed-based analyses(38–41). This approach might not be optimal when heterogeneous regions are implicated in a disorder. Inclusion of the cortex, subcortical regions, and the cerebellum in resting-state connectivity studies allows for a comprehensive investigation of network pathology. Advanced neuroimaging techniques, including graph-based connectome analysis, are used to explore the configuration of brain networks in psychiatric disorders(42,43). We applied a data-driven, graph-based network analysis to characterize resting-state functional connectivity using a multiband resting-state dataset. We hypothesized that OCD participants would show disrupted resting-state functional connectivity within and among large-scale functional networks in the cortex and cerebellum. We also hypothesized that the magnitude of any connectivity abnormalities within and between these networks would be associated with OCD symptom severity.

Material and methods

Participants

We report findings from an ongoing study that commenced in June, 2015. The study received institutional review board approval from the Department of Psychiatry, University of Pittsburgh School of Medicine. All participants were recruited from the department website or from affiliated outpatient clinics. Written informed consent was obtained from each participant. Eighty-seven participants(44 OCD, 43 healthy) were included (Supplemental). OCD diagnosis was confirmed by a trained clinician rater via the SCID-5. Symptoms were assessed with the Hamilton Rating Scale for Depression(17-items;HRSD-17)(44), Hamilton Anxiety Scale 14(HAMA-14)(45), and the Yale-Brown Obsessive-Compulsive Scale(Y-BOCS)(46), a self-report inventory broadly characterizing obsessional thoughts and compulsive behaviors.

Network construction

Details of MRI acquisition and preprocessing are in the Supplemental. 14 participants were excluded due to incomplete neuroimaging data, excessive head motion(>3 mm or 3°, 3 excluded), or structural image distortions. Preprocessed images were used to construct all networks. Each network was comprised of nodes and edges. To define the cortical nodes, we employed a fine-grained parcellation with 300 areas, generated from 1489 participants based on the gradient-weighted Markov Random Field model(47). We also used a cerebellum atlas with seven nodes(visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default-mode) for each hemisphere(14 nodes total), in the SUIT toolbox(http://www.diedrichsenlab.org/imaging/suit.htm). To define the subcortical nodes, we used the Harvard-Oxford Atlas with seven nodes for each hemisphere in FSL(https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases). Subcortical nodes included the thalamus, caudate, putamen, pallidum, hippocampus, amygdala, and accumbens. A total of 328 nodes were included in the network analysis. To define edges, we calculated Pearson correlation coefficients between each pair of nodes and converted these to z-scores using Fisher’s r-to-z transformation, resulting in a 328×328 correlation matrix for each participant. Because most network algorithms cannot interpret negative edges, we set all negative edges to zero and restricted network analysis to positive correlations(48,49).

Network analysis

Network connectivity strength:

Connectivity matrices were used to calculate average nodal strength per network. Nodal strength is a network index that represents the centrality of a node through the strength of its connectivity with other nodes in a given network. We computed nodal strength by using algorithms from the Brain Connectivity Toolbox(50). Specifically, we used weighted, undirected graphs thresholded at the 15% strongest positive connections (i.e., the sparsity of the graphs was 0.15) to focus on the most robust network connectivity for each participant, as in previous studies(51,52). We defined 9 specific networks covering the whole brain, including 7 widely-used networks in the cerebral cortex(53–55), a subcortical network(SCN) with all subcortical regions, and the cerebellum. The 7 cortical networks were defined by a previous network parcellation that used a clustering approach based on 300-area parcellations(47). This functional cortical network parcellation was generated from an fMRI dataset of 1,489 participants and employs a novel hybrid approach that integrates the local gradient approach for boundary-mapping with a global clustering approach to best reveal neurobiologically meaningful features of brain organization. This atlas of 7 cortical networks has been widely used in the literature(56,57). However, this cerebral parcellation does not cover subcortical and cerebellar structures. Therefore, we separately defined the SCN and cerebellum network by employing the well-known Harvard-Oxford Atlas and SUIT atlases. Thus, we employed a set of 9 parcellations that covered the entirety of the brain. Next, we averaged nodal strength of all nodes within each network in order to represent network connectivity strength. Finally, we used permutation testing(N=10,000) to test for differences in network strength between OCD and healthy participants(P_Bonferroni<0.05 for 9 comparisons), controlling for age, gender, and scanner(9,14,34).

Within- and between-network functional connectivity:

Although network connectivity strength provides an overall measure of network connectivity, it does not provide specific within-network or between-network connectivity profiles. We therefore analyzed within- and between-network connectivity for the networks showing between-group network connectivity strength differences. We used the original weighted connectivity graph. Within these individual subject maps, we retained correlation coefficients of P<0.01, FDR corrected. All other correlations were set to zero. In order to ensure sufficient power, we selected only the connections that remained following FDR correction for at least 50% of the participants for statistical analysis(48). Then, for each pair of networks and for each network, we computed the averaged z-score for each participant across all relevant edges. We next compared the OCD and HC group z-scores for each network pair and within each network using permutation testing. Group membership(OCD, healthy) was permuted 10,000 times to determine the significance level of the observed between-group difference relative to the between-group difference for each of the permuted groups. Permutation testing was performed to assess between-network connectivity for each network pair and within-network connectivity for each network. Results were Bonferroni corrected for the number of parallel analyses across all networks and across all pairs of networks, controlling for age, gender and scanner. To determine the specific edges within each network/network pair that showed between group differences in these analyses, we identified the significant between-group z-score differences for each edge in each networks/network pairs. This analysis was also Bonferroni corrected to control for the number of parallel tests.

Network-based statistics(NBS):

As described above, we identified network level between-group differences of network connectivity strength. We then identified edge level differences in the networks that showed group differences. We utilized NBS to localize edge-level group differences in both positive and negative connectivity using two sample t tests(P<0.001), controlling for age, gender, and scanner(58). This allowed us to identify a subset of suprathreshold edges. Then, we used permutation tests(N=10,000) to estimate the distribution of between-group differences for the most robust components(i.e., a group of edges). Here, for each permutation, all participants were randomly reallocated into two groups and two sample t-tests were computed independently for each edge. Then, the same primary threshold(P<0.001) was used to generate suprathreshold links among which the maximal connected component size was recorded. A family-wise error(FWE)-corrected p-value(P_FWE<0.05) for the network component was estimated by the proportion of permutations for which a component of greater size was identified.

Relationship between network connectivity and symptoms

To explore the relationship between abnormal connectivity(both within- and between-networks) and symptoms, we performed partial least squares(PLS) analysis in OCD participants. We used scores on the Y-BOCS, HRSD-17, and HAMA-14 as outcome variables for three parallel regressions. The independent variables were all of the components derived from the NBS analyses that showed significant between-group differences in connectivity(described above). For each participant, we extracted the original z-score for each edge in significant NBS-derived component identified in the previous analysis. Prior to PLS analysis, the effects of sex, gender and scanner were regressed out. We then used PLS to identify the most parsimonious set of connectivity measures that covaried with clinical scores. All connections among nodes in each PLS component derived from the above analyses were then assigned weights based on the amount of explained variance for each symptom score. The PLS components were considered significant at alpha=0.05 with 1,000 permutations, after permuting the clinical score labels assigned as each set of outcome variables. We then examined the ranking of the weights assigned to all edges in these PLS components to determine which specific edges were associated with clinical scores. We conducted a Pearson correlation analysis using the participant loadings on each significant PLS component and clinical variables to determine the direction of the relationship between connectivity and clinical scores. The statistical threshold for PLS analyses was set to P<0.05 for each PLS component, with Bonferroni correction for 3 parallel comparisons.

Validation analysis

To evaluate the robustness of the brain network results, we examined network sparsity and the effects of functional atlas, medication, head motion, and scanner. We assessed network strength at 11 different sparsities from 10 to 20%, at intervals of 1%. We repeated the primary analyses with the cerebral atlas parcellated into 200 versus 400 nodes(47). We also reran the brain network analysis with only the 26 unmedicated OCD participants and the 39 healthy participants to exclude for medication effects. We also performed a “scrubbing” procedure to exclude the effects of head motion in preprocessing(59). We replaced the volumes of frame-wise displacement > 0.5mm and their adjacent volumes(2 forward and 1 back) with the nearest neighbor interpolated data, and then repeated the network analysis(59). We also reanalyzed only data collected from the PRISMA scanner(38 OCD,38 healthy).

Results

Demographic and clinical characteristics

There were no significant between-group differences in age, gender or educational attainment(P>0.05). OCD participants had more severe depression(HRSD-17), anxiety(HAMA-14), and obsessive-compulsive symptoms(Y-BOCS,P<0.001;Table 1).

Table 1.

Demographic and clinical characteristics

	OCD (N = 44)	Healthy (N = 43)	P Value
Age (years)	23.61 ± 4.76	23.51 ± 4.05	0.92
Gender (male/female)	15/29	18/25	0.46
Education level (with/without college)	34/10	38/5	0.17
Scanner (Trio/Prisma)	6/38	5/38	0.78
HRSD-17	10.52 ± 5.80	1.30 ± 1.14	P < 0.001
HAMA-14	12.00 ± 6.40	1.05 ± 1.15	P < 0.001
Y-BOCS	20.34 ± 3.12	0.12 ± 0.76	P < 0.001
Illness duration	14.26 ± 7.19	N/A
Comorbid depression	24	N/A
Comorbid anxiety	29	N/A

Disrupted network connectivity in OCD

We observed lower connectivity in the SMN(T=−2.82,P=0.003) and greater connectivity in the cerebellum(T=2.87,P=0.003,Fig.1A,Table2) in OCD versus HC. We then analyzed within- and between-network connectivity in the SMN and cerebellum. We observed lower within-network SMN connectivity(T=−3.10,P=0.001), greater connectivity between the SMN and SCN(T=2.81,P=0.003), and between the SMN and cerebellum(T=3.26,P=0.001) in OCD(Fig.1B,Table2). We tested for between-group connectivity differences between the SMN and specific SCN and cerebellum nodes. The SMN showed greater connectivity with the thalamus(T=2.54,P=0.005) and pallidum(T=2.69,P=0.005) in the SCN(Fig.1C,Table2), and with the limbic(T=2.65,P=0.005), frontoparietal(T=3.45,P<0.001) and default-mode nodes(T=2.92,P=0.002) in the cerebellum(Fig.1D,Table2). Connectivity between the cerebellum and SMN was also greater in OCD(T=3.30,P=0.001;Fig.1E,Table2).

Figure 1. Group differences in functional brain network strength.

(A) Network strength was lower in the SMN and elevated in the cerebellum in OCD compared to healthy participants. (B) Lower SMN within-network connectivity and greater connectivity between the SCN and cerebellum in OCD participants. (C) Greater SMN connectivity with the thalamus and pallidum in OCD participants. (D) Greater SMN connectivity with limbic, frontoparietal and default-mode nodes in the cerebellum in OCD participants. (E) Greater cerebellum connectivity with the SMN in OCD participants. Asterisks (*) represent the networks or brain regions meeting significance criteria with Bonferroni correction. Abbreviation: VN: visual network; SMN: somatomotor network; DAN: dorsal attention network; VAN: ventral attention network; LN: limbic network; FPN: frontoparietal network; DMN: default-mode network; SCN: subcortical network; Tha: thalamus; Cau: caudate; Put: putamen; Pal: pallidum; Hip: hippocampus; Amy: amygdala; Acc: accumbens.

Table 2.

Between-group comparisons of functional network strength and network connectivity in OCD and healthy participants.

Functional network strength
Networks	OCD	HC	T	P
VN	25.38 ± 4.12	24.55 ± 4.88	0.79	0.2117
SMN	29.68 ± 9.68	36.52 ± 12.12	−2.82	0.0035
DAN	29.74 ± 5.21	29.34 ± 4.07	0.35	0.3652
VAN	33.98 ± 6.42	33.43 ± 6.94	0.44	0.3265
LN	16.43 ± 3.05	16.02 ± 3.37	0.65	0.2631
FPN	24.31 ± 3.62	23.87 ± 3.08	0.82	0.2138
DMN	26.83 ± 3.23	26.17 ± 3.69	0.89	0.1911
SCN	13.85 ± 4.68	12.64 ± 3.98	1.41	0.0799
CN	17.88 ± 4.42	15.40 ± 3.30	2.87	0.0030
Within- and between-network connectivity with SMN
Networks	OCD	HC	T	P
VN	0.16 ± 0.12	0.21 ± 0.12	−1.83	0.0341
SMN	0.37 ± 0.14	0.48 ± 0.17	−3.10	0.0009
DAN	0.29 ± 0.10	0.33 ± 0.09	−1.92	0.0303
VAN	0.32 ± 0.09	0.35 ± 0.09	−1.53	0.0653
LN	0.03 ± 0.10	0.05 ± 0.11	−0.75	0.2294
FPN	−0.15 ± 0.07	−0.17 ± 0.07	1.53	0.0642
DMN	−0.04 ± 0.06	−0.03 ± 0.07	−0.67	0.2471
SCN	0.22 ± 0.12	0.13 ± 0.17	2.81	0.0035
CN	−0.06 ± 0.13	−0.15 ± 0.14	3.26	0.0013
Within- and between-network connectivity with CN
Networks	OCD	HC	T	P
VN	0.30 ± 0.10	0.24 ± 0.11	2.43	0.0075
SMN	−0.06 ± 0.13	−0.15 ± 0.14	3.30	0.0011
DAN	0.18 ± 0.11	0.12 ± 0.11	2.25	0.0154
VAN	−0.03 ± 0.11	−0.07 ± 0.12	1.51	0.0653
LN	−0.06 ± 0.10	−0.01 ± 0.07	−2.53	0.0077
FPN	0.05 ± 0.09	0.05 ± 0.08	−0.51	0.3062
DMN	−0.10 ± 0.07	−0.06 ± 0.07	−2.62	0.0056
SCN	−0.20 ± 0.17	−0.17 ± 0.16	−0.67	0.2507
CN	0.43 ± 0.14	0.45 ± 0.12	−0.69	0.2557
Functional connectivity between SMN and SCN
Nodes	OCD	HC	T	P
Thalamus	0.35 ± 0.16	0.24 ± 0.21	2.54	0.0051
Caudate	−0.06 ± 0.12	−0.10 ± 0.15	1.45	0.0712
Putamen	0.14 ± 0.10	0.07 ± 0.17	2.09	0.0203
Pallidum	0.09 ± 0.12	0.00 ± 0.17	2.69	0.0046
Hippocampus	0.01 ± 0.13	0.07 ± 0.14	−2.15	0.0170
Amygdala	0.07 ± 0.10	0.07 ± 0.15	0.12	0.4571
Accumbens	−0.02 ± 0.10	−0.04 ± 0.12	0.75	0.2247
Functional connectivity between SMN and CN
Nodes	OCD	HC	T	P
Visual	0.33 ± 0.09	0.29 ± 0.12	1.74	0.0453
Somatomotor	0.12 ± 0.12	0.08 ± 0.14	1.33	0.0867
Dorsal attention	0.03 ± 0.15	−0.03 ± 0.18	1.67	0.0482
Ventral attention	0.05 ± 0.18	−0.04 ± 0.19	2.20	0.0174
Limbic	−0.04 ± 0.15	−0.14 ± 0.17	2.65	0.0053
Frontoparietal	−0.17 ± 0.18	−0.31 ± 0.18	3.45	0.0003
Default mode	−0.22 ± 0.17	−0.32 ± 0.16	2.92	0.0021

Abbreviations: VN: visual network; SMN: somatomotor network; DAN: dorsal attention network; VAN: ventral attention network; LN: limbic network; FPN: frontoparietal network; DMN: default-mode network; SCN: subcortical network; CN: cerebellum network.

NBS reveals components with disrupted functional connectivity in OCD

We applied NBS to study specific connectivity with functional abnormalities within and among the SMN, SCN and cerebellum. A component(Component 1) with 30 nodes and 48 connections showed significantly lower connectivity in OCD(Fig.2A). All these nodes were restricted to the SMN network, i.e., a within network component. There was one component with greater connectivity in OCD versus healthy participants(Component 2), which included 23 nodes and 28 connections linking regions in the motor cortex and cerebellum, including the ventral attention, frontoparietal and default-mode nodes of the posterior cerebellum(Lobules VI-VIII and Crus I and II), i.e., a between-network component(Fig.2B). There was a negative correlation between Components 1 and 2 across both groups(R=−0.73,P<0.001;Fig.2C).

Figure 2. Network-based statistics revealed significant alterations of functional connectivity among SMN, SCN and cerebellum in OCD.

Connections (lines) between nodes (nodes) exhibiting significant between-group differences in streamline count. Blue and red lines separately represent lower (A) and greater (B) functional connectivity in OCD compared with healthy participants. (C) Negative correlation between connectivity in each component across groups. (D) Positive correlation between abnormal functional connectivity and symptom severity in the OCD participants.

Correlation with clinical scores in OCD

We used PLS analysis to examine relationships between NBS-derived connectivity and symptoms. Connectivity correlated with Y-BOCS scores in OCD participants(P=0.006), but not with anxiety(HAMA-14,P=0.04) or depressive symptom severity(HRSD-17,P=0.03). Only one connectivity PLS component(the first,PLS1) explained more than 20% of the variance in Y-BOCS scores(PLS1: 44.2%). The edges with high weights(top 25%) contributing to Y-BOCS variance were in motor cortex, and cerebellar ventral attention, frontoparietal and default-mode nodes. There was a positive correlation between participant loadings on the PLS1 component and Y-BOCS scores(R=0.66,P<0.001;Fig.2D), indicating that more abnormal connectivity was associated with greater symptom severity.

Reproducibility of findings

Network strength findings in the SMN and cerebellum were replicated over the threshold range of 0.10≤Sparsity≤0.20(Fig. S1), and for area under the curve(SMN:T=−2.81,P=0.006;cerebellum:T=2.83, P=0.006). We repeated network strength, and within- and between-network connectivity, analyses replacing the functional parcellation with 228 nodes(TableS1) and 428 nodes(TableS2), excluding effects of head motion(TableS3), and medicated participants(TableS4) and TRIO scanner(TableS5). The overall pattern of findings persisted.

Discussion

Using large-scale network analysis, we observed disrupted functional connectivity among specific cortical, subcortical and cerebellar regions in OCD. Specifically, we showed lower within-SMN connectivity and greater cerebellar-SMN and SCN-SMN connectivity in OCD. We identified two components with disrupted functional connectivity showing lower within-SMN connectivity and greater connectivity among motor regions of the SMN and the cerebellum in OCD participants. The magnitudes of these abnormalities were positively associated with OCD symptoms, but not with anxiety or depressive symptoms. To our knowledge, this study is the first comprehensive investigation of cerebello-cerebral network connectivity in OCD. These findings implicate cortical and subcortical networks and the cerebellum in the pathophysiology of OCD.

The SMN integrates sensory perception and motor control(53). Using network connectivity analysis, we found lower functional coupling within the SMN in OCD. Previous findings indicate lower regional homogeneity, a measure of local synchronization(60), in the SMN in OCD(61), perhaps reflecting an impaired ability to modulate sensory information(62,63). Reduced sensorimotor gating might have pathophysiologic implications for both motor compulsions and cognitive dysfunction, both core OCD symptoms(62). Sensorimotor gating is a pre-attentive process by which irrelevant sensory information is filtered from attention. It is subserved by corticothalamic connectivity(65,66). Sensorimotor gating is impaired in OCD(67). Lower connectivity in the SMN and enhanced thalamic-SMN connectivity could be related to sensorimotor gating anomalies in OCD. SMN changes have also been associated with OCD treatment response, including greater somatomotor activity following cognitive behavior therapy(68). Repetitive low frequency transcranial magnetic stimulation to the somatomotor cortex also normalizes activity and improves symptoms(69). These findings suggest that the somatomotor cortex is a potential treatment target in OCD.

OCD participants showed greater SMN-cerebellum connectivity, specifically, the limbic, frontoparietal and default-mode nodes(Crus I and II), but not the somatomotor nodes(Lobules I-V). Functional abnormalities in the cerebellum likely disrupt the function of connected cortical regions. Cerebral-cerebellum connectivity disruption has been reported in several psychiatric disorders, including psychotic disorders(70), depressive disorder(71), autism(72) and attention-deficit/hyperactivity disorder(73). We observed differences in cerebellar nodes associated with executive control and attention(25,74,75). Greater connectivity between these regions and the cortex might be associated with intrusive thoughts and compulsive behaviors in OCD(5,22,76). Our findings are somewhat consistent with the OCD resting-state literature, although importantly, these studies employ seed-based approaches and do not include the cerebellum, in contrast to our large-scale network approach(17,77). Others have shown increased corticostriatal connectivity in OCD correlates with Y-BOCS scores(22). Task-based fMRI studies suggest increased functional connectivity between these regions in OCD may be related to internally-directed attention. Overall, our findings are consistent with an emerging model for OCD that suggests heightened internal focus mediated by increased between-network rsFC(63).

We observed greater connectivity between regions in the SMN and SCN, including the thalamus and the pallidum, in OCD. Consistent with our results, an OCD resting-state analysis demonstrated elevated connectivity between cerebellar and basal ganglia nodes that was associated with less cognitive flexibility(75). Others report elevated FPN-thalamic connectivity(17,77) and thalamic global connectivity in OCD(34). These studies did not assess specific thalamic connections, whereas we showed greater connectivity between the thalamus and SMN using whole-brain network analysis, thereby refining our understanding of thalamic connectivity in OCD. We also report altered SMN-pallidum connectivity. The pallidum is involved in habitual behavior and learning(78,79). Changes in SMN and pallidal activity are associated with motor automation(80) and pallidal damage produces compulsive behavior(5,81). Others have reported pallidal hyperactivity and structural deficits in OCD(82,83). Our findings of elevated SMN-thalamic and SMN-pallidal connectivity might underlie aberrant sensory integration and habit learning, respectively, in OCD.

We report elevated between-network connectivity in OCD participants, in regions that comprise the cerebello-thalamo-motor circuit. There are two subsystems in this circuit. The first, the cerebellar motor loop, sends output from the cerebellum to the motor cortex via the superior cerebellar peduncle and ventrolateral thalamus. The cerebellum also receives input from the motor cortex through the pons(25,26). The second, the basal ganglia motor loop, comprises reciprocal connections between motor cortex and basal ganglia(e.g. pallidum) via the ventrolateral thalamus(84,85). This circuit underlies motor control and learning via the integration of interoceptive, sensory, and motor information(86,87). Altered connectivity in these regions might underlie sensorimotor gating anomalies in OCD described in prepulse inhibition studies(88). Pallidal, thalamic, and motor cortex activity is also associated with checking symptoms in OCD(89). Our findings highlight elevated connectivity within the cerebellothalamo-motor circuit in OCD.

There was a significant correlation between the two NBS components comprising within-SMN connectivity and SMN-cerebellar connectivity in participants with OCD, such that participants with lower within-SMN connectivity had greater connectivity between the SMN and the cerebellum. This is consistent with theories of reciprocal network connectivity(90,91) in psychiatric disorders, which posits that lower localized information processing might be associated with greater information transfer between regions. Our network analysis revealed greater connectivity in the thalamus and pallidum in OCD participants that were not disrupted in the NBS analysis. This discrepancy can be explained by distinctions between these two methodologies. Based on the principles underpinning traditional cluster-based thresholding of brain maps with permutation tests, NBS identifies significant components(similar to “clusters”) with nodal connections(similar to “voxels”). Furthermore, in the NBS analysis, we applied a threshold(P<0.001) to the two-sample t-tests computed for each edge to identify a set of suprathreshold edges. Thus, we speculate that the subcortical connectivity findings identified in the network analysis might not have been sufficiently robust to form the components which met the permutation-based thresholds of NBS.

Using NBS components showing between-group differences in network connectivity, we observed a positive correlation between abnormal SMN-cerebellar connectivity and OCD symptom severity, as measured by the Y-BOCS, in OCD participants. These findings were localized to the motor cortex, as well as the ventral attention, frontoparietal and default-mode nodes of the posterior cerebellum(Lobules VI-VIII and Crus I and II). The posterior cerebellum is considered the “cognitive cerebellum”, because it links multiple regions of the cerebral cortex involved in executive function and attention switching(74,92,93). PLS analysis revealed associations between network connectivity and OCD, but not anxiety and depressive, symptom severity. Elevated posterior cerebellar connectivity might be involved in disrupted executive control and attention shifting that reflects the obsessional thoughts and compulsive behaviors in OCD(94). Others report correlations between connectivity in this circuit and Y-BOCS scores, although in those studies, findings were localized to the Crus I and thalamus(37,95).

Although we employed appropriate statistical correction approaches, replication studies with larger samples are needed to make generalizations about the overall OCD population. Particularly given the challenges with replication in neuroimaging, analysis of a larger independent sample is an important future direction. This study employed a cross-sectional approach and was not designed to assess effects of illness course or risk for the development of OCD. Future work with a longitudinal design can help identify relationships between connectivity alterations and the progression of OCD. Tests of attentional set-shifting should be employed in future studies to examine the relationship between network connectivity and cognitive dysfunction. Finally, due to heterogeneous functional subdivisions in the basal ganglia and thalamus, we were unable to pinpoint how cytoarchitectonically distinct subdivisions might be involved in circuit alterations in OCD. Future studies using high-resolution fMRI can delineate subsystems and explore their interactions in OCD.

We identified lower connectivity within the SMN and greater SMN-SCN and SMN-cerebellar connectivity in OCD. These connectivity patterns were also associated with OCD symptoms. To our knowledge, this study is the first to comprehensively examine cerebellar-cerebral connectivity in OCD. These findings extend our understanding of possible links between cerebellar function and OCD, and highlight cerebellar-cortical, particularly somatomotor, circuits as potential targets for novel treatments.