Specific Frontostriatal Circuits for Impaired Cognitive Flexibility and Goal-Directed Planning in Obsessive-Compulsive Disorder: Evidence From Resting-State Functional Connectivity

1Persistent intrusive thoughts and repetitive actions characterize the symptoms of obsessive-compulsive disorder (OCD). Over and above these often highly specific clinical symptoms is a general tendency toward poor performance on tasks requiring flexible behavior. Preclinical and clinical evidence indicates the integrity of the basal ganglia and their connections with the frontal cortex to be crucial in enabling the affective, cognitive, and motor flexibility necessary for goal-directed behavior. These data are thus broadly consistent with a leading neural model implicating frontostriatal circuits in the pathogenesis of OCD. However, this model has not been clearly linked to underlying cognitive or psychological processes mediated by this circuitry. This link has recently been provided by a major hypothesis concerning OCD, which postulates that OCD symptoms arise from impairments in goal-directed behavior, leading to autonomous habitual behavior as a consequence of functional imbalances in frontostriatal circuitry.

2Resting-state activity derived from functional magnetic resonance imaging (fMRI) enables inference of the strength of functional connectivity between different brain regions. OCD patients exhibit hyperconnectivity of ventrolimbic corticostriatal regions that correlates with symptom severity as well as hypoconnectivity of the caudate and putamen with cortical areas. However, the behavioral significance of this pattern of functional connectivity is yet to be established.

3Here, we focused on attentional set-shifting (cognitive flexibility) as well as goal-directed planning. Deficits in these executive functions represent candidate endophenotypes for the disorder and have been related to OCD symptoms. Planning is measured as the ability to attain a goal via a prepared series of actions controlled by a single instrumental contingency. OCD patients are impaired in this goal-directed capacity, especially at more demanding levels. Compared with control subjects, OCD patients exhibit decreased functional activation of the dorsolateral prefrontal cortex (PFC), caudate, and putamen during planning, consistent with other behavioral data suggesting an imbalance between the control exerted by goal-directed and habit systems over behavioral output in OCD.

4We hypothesized that impairment of frontostriatal circuitry entailing caudate and putamen and separate regions of PFC likely disrupts implementation of flexible goal-directed behavior in OCD patients. Notwithstanding overlap, the existence of topographic projections of different PFC regions to striatal regions in monkeys and humans suggests that separate frontostriatal circuits are implicated in regulating cognitive control and aspects of executive functioning. Thus, a lateral circuit including the ventrolateral PFC has been suggested to mediate switches in behavioral set with a dorsolateral PFC circuit being related to goal-directed planning. On the basis of its PFC inputs, the ventral striatum has alternatively been implicated in affective control and reward processing.

5We used resting state to investigate functional connectivity in OCD patients and to overcome limitations of task-related studies in which case-control differences in activation might be the result of task performance, effort, or strategy rather than the reflection of underlying core abnormalities. Moreover, the topography of networks arising from synchronous spontaneous functional activity is compatible with the underlying structural connectivity and consists of regions known to share common functions. We leveraged on a novel multiecho acquisition method with significantly increased signal-to-noise ratio and two complementary analytical approaches. First, we formulated specific a priori hypotheses and used seed-based analysis to test whether dysregulated resting-state connectivity arising from caudate and putamen accounted for impairments in cognitive flexibility and executive planning. We hypothesized that if functional abnormalities in these circuits were relevant to OCD, then we should find specific patterns of altered connectivity being associated with cognitive flexibility and goal-directed planning in patients. Consistent with previous evidence, we predicted that ventrolateral and dorsolateral PFC circuits would mediate cognitive flexibility and goal-directed planning, respectively. To overcome the limitation of exclusively focusing on a priori regions, the second approach applied a whole-brain data-driven graph theoretical analysis to identify novel features of abnormal brain network organization in OCD patients.

6The study included 87 participants, consisting of 44 OCD patients and 43 healthy control subjects matched for relevant demographic variables (Table 1, Supplement). The OCD sample included 27 medicated and 17 unmedicated patients; all but one of the medicated patients were taking selective serotonin reuptake inhibitors (Supplement). OCD patients reported higher levels of depressive symptoms and anxiety, although well below clinical threshold (Table 1). Unless otherwise reported, there were no differences between medicated and unmedicated OCD patients in the results presented.

7For resting-state data acquisition, we used multiecho planar sequence with improved signal-to-noise ratio (see Supplement for imaging variables). Participants were instructed to lie quietly with their eyes open and attend to a centrally presented white fixation cross on a black projection screen for 10 minutes; we monitored their degree of alertness by asking to complete the Stanford Sleepiness Scale ruling out differences in levels of arousal across participants (Supplement).

8To elucidate the behavioral significance of the functional abnormalities within frontostriatal circuits in OCD patients, in a separate session outside the scanner, the same participants were tested with objective and well-validated CANTAB paradigms. We used the intra-/extradimensional set shift (IED) and the One Touch Stockings of Cambridge (OTS) to measure cognitive flexibility and goal-directed planning, respectively. The IED is a nine-stage task, and the rule for correct responding is modified at the start of each stage. For the IED, crucial stages are the intradimensional shift (IDs) testing for the ability to generalize a rule to new stimuli and the extradimensional shift (EDs) testing cognitive flexibility as the ability to shift attention to a previously irrelevant dimension. On the OTS, planning abilities are tested at different difficulty levels with problem difficulty varying from 1 to 6 moves (see Supplement for description of the paradigms).

9Imaging data were preprocessed and analyzed using Analysis of Functional NeuroImages (AFNI) software. To denoise the data, we used a novel integrated procedure taking advantage of multiecho acquisition in combination with Multi-Echo Independent Component Analysis (ME-ICA) (AFNI tool meica.py, version 2.5 beta10). The rationale behind ME-ICA is to classify sources of variance in the fMRI time-series scaling linearly with echo-time and thus confidently regarded as indicative of blood oxygen level–dependent (BOLD) contrast (Supplement). The retained independent components, representing BOLD contrast, were optimally recomposed and visually inspected (see Supplemental Figure S1 for a representative subject). According to multiple, complementary indices, patients and control subjects did not differ for movement in the scanner; there were no significant groups differences in the number of high- versus low-motion subjects in each group or in the number of BOLD components retained or motion as measured by framewise displacement (Table 1).

10On the IED task, dependent measures at each stage were the number of subjects passing and the number of errors. Data were square root-transformed to stabilize variance and to reduce skewness in the distribution. On the OTS task, we measured the mean number of attempts made before obtaining the correct solution for easy (1–3 moves) and hard (4–6 moves) levels of difficulty (significance threshold, p <.025). Data were statistically analyzed using χ2, analysis of variance, Student’s t test, and the Mann-Whitney U test to detect group differences between control subjects and OCD patients. Only the 44 OCD patients were included for Pearson’s correlations between clinical scales and measures of tasks yielding significant group differences.

11We tested differences in functional connectivity strength between OCD patients and control subjects from a priori anatomical regions of interest based on the known neurobiological profile of OCD and previous findings in the literature. We examined connectivity from the dorsal caudate (DCd), putamen (PUT), and nucleus accumbens (NAc), defined in both hemispheres as 3-mm radial spheres located at Montreal Neurological Institute coordinates automatically provided by the AFNI-supplied atlas, namely, DCd: ± 12 6 14 [labeled as the body of the caudate and analogous to the dorsal striatal seed previously described in the literature]; PUT: ± 24 0 3; NAc: ± 12 8 −8.

12Based on evidence of caudate and putamen involvement in executive functions, we tested the a priori hypothesis that dysregulated functional connectivity from the DCd and PUT selectively accounted for impairment on relevant cognitive domains in OCD patients. For the OCD patients only, we used number of errors on the EDs stage and mean number of problems attempted at the hardest level of difficulty (6 moves) of the OTS task as a covariate of interest, to identify brain regions for which significant connectivity with DCd and PUT was significantly related to cognitive flexibility and goal-directed planning. Post hoc analyses investigated NAc connectivity relation with cognitive performance and clinical scores.

13ME-ICA denoised data were entered in 3dGroupInCorr to estimate functional connectivity: time-series were extracted from each dataset averaging locally per the seed’s radius and connectivity maps computed with Pearson’s correlation; Fisher’s r-to-z transform for the appropriate degrees of freedom (i.e., number of BOLD components identified for each subject) was used to derive standard scores. Whole-brain analyses were conducted in combination with cluster-based correction. We applied voxel-level height threshold of p <.01 and used 3dClustSim to determine the corrected p values that corresponded to the resulting clusters (Supplement).

14To perform data-driven network analysis, for each subject, time-series were extracted by averaging voxel time-series within each of equal-sized cortical and subcortical defined regions (nodes) (see Supplement for parcellation template and detailed procedure). Analysis focused on data at frequency interval 0.049–0.101 Hz at 10% cost, which is compatible with prior studies. We identified modular community structure, which is a feature of many complex networks, including nervous system. The identification of modules, subsets of nodes densely intraconnected (number of connections between nodes within the module) and sparsely interconnected with nodes in other modules, may uncover functional units. The Louvain algorithm as implemented in the Brain Connectivity Toolbox was used to identify modules. Default modularity resolution (gamma = 1) was used in the Louvain algorithm for data presented in the main text and further validated at different gamma levels (Supplement).

15Within-group striatal connectivity patterns overlapped with previously described neuro-functional maps. Both groups showed maps of connectivity consistent with models relating the caudate and putamen to cognitive and motor control and the nucleus accumbens to motivational and emotional responses (Supplement, Supplemental Figure S2).

16Compared with control subjects, OCD patients showed decreased connectivity strength from DCd and PUT to frontal and parietal regions, whereas ventral striatal-frontal connectivity was increased, in line with previous findings (Supplement, Supplemental Figure S3, Supplemental Table S1).

17OCD patients exhibited a profound impairment on cognitive flexibility as tested with the IED (Figure 1A). More OCD patients failed to complete all stages of the task (χ21 = 7.975, p =.005), with patients more likely than control subjects to fail selectively at the EDs stage. All subjects attempted the EDs stage (Supplemental Figure S4). There was a highly significant interaction of stage (IDs, EDs) and group (F1,84 = 7.128, p =.009) in the number of errors. Simple-effect analyses revealed significantly more errors at the EDs stage (t84 = −2.649, p =.01) in OCD patients than control subjects, but no difference at the IDs stage (t84 = 0.742, p =.460) (Figure 1B). Thus, OCD patients were able to form an attentional set and generalize to new stimuli as shown by intact performance up to the EDs stage, but they were selectively impaired when they had to shift attention to a previously irrelevant dimension. There was no significant correlation between the number of errors at the EDs stage and any of the clinical scales, including symptom subtypes and depression severity (all p >.203).

18To test whether functional connectivity in specific frontostriatal circuits predicted patients’ cognitive flexibility, we used number of errors at the EDs stage as a covariate of interest in the connectivity maps generated from DCd and PUT. In OCD patients, a higher number of errors at the EDs stage was associated with reduced functional connectivity between the left DCd and a set of brain regions, including caudate and putamen bilaterally, right medial frontal gyrus (Brodmann area [BA] 9), and a cluster peaking at BA 10 and extending to the left lateral PFC (BA 10/11/47) (Figure 1C). Specifically, for the left BA 10/11/47, lower connectivity with the left DCd was strongly associated with impaired cognitive flexibility in the OCD sample (Figure 1D, Supplement) and when including control subjects as well (Supplement). Covariation for age and verbal IQ did not alter the results (p <.001). Mean split of OCD patients according to the number of EDs stage errors showed that patients severely impaired in cognitive flexibility (high number of EDs stage errors) had significantly reduced functional connectivity between left DCd and left BA 10/11/47 compared with OCD patients with better performance (t42 = 5.338, p <.001) (Figure 1E); the two subgroups were, however, indistinguishable in terms of severity on any of the clinical scales. Similar results were found when testing functional connectivity from the right DCd (Supplemental Table S2) but not when testing functional connectivity from left and right PUT and NAc (Supplemental Table S2, Supplemental Figure S5), revealing the specific relevance of caudate connectivity to cognitive flexibility in OCD patients.

19OCD patients showed impaired goal-directed planning abilities at the hard levels of difficulty as tested with the OTS (Figure 2A) and indexed by the increased number of attempts to obtain the correct response (t83 = −2.427, p =.017; Figure 2B) compared with control subjects. There was no group difference for the easy problems. In the medicated patients, increased self-reported severity of OCD symptoms and anxiety positively correlated with poor goal-directed performance at the hardest level of difficulty (Obsessive-Compulsive Inventory-Revised: r =.6, p < 0.001; State-Trait Anxiety Inventory-State: r =.531, p <.005, both surviving Bonferroni correction) (Figure 3).

20Functional connectivity within a specific frontostriatal circuit predicted patients’ goal-directed planning ability. A higher number of attempts at the most difficult level of the task (6 moves) was associated with reduced functional connectivity between the right PUT and the right dorsolateral PFC (BA 46). In addition, a positive relation was found for the angular gyrus bilaterally (BA 39), whereby better performance was associated with increased functional connectivity from the right PUT (Figure 2C, Supplemental Table S3). Specifically, for the right BA 46, lower connectivity with the right PUT was strongly associated with poor goal-directed planning in the OCD sample (Figure 2D, Supplement) and when including control subjects as well (Supplement). Covariation for age and verbal IQ did not alter the results (p =.007). A mean split according to mean number of attempts revealed significantly reduced functional connectivity between the right PUT and right BA 46 in severely impaired patients (t41 = 3.599, p <.001) (Figure 2E). Those patients also exhibited increased anxiety (t41 = −2.126, p =.04) compared with patients performing better on the task. These results were specific to the PUT and not to DCd or NAc (Supplemental Table S3), revealing the specific relevance of putaminal connectivity to goal-directed executive planning in OCD patients.

21In OCD patients, NAc connectivity was related to clinical scores of anxiety and depression (Supplemental Table S4, Supplement).

22Data-driven network analysis disclosed that, in the OCD sample, nodes corresponding to caudate and putamen as well as the cerebellum were clustered together in a single module (Supplemental Table S5), suggestive of a cohesive functional unit. In contrast, in control subjects, the same striatal and cerebellar regions were distributed among large cortico-subcortical modules (Figure 4A, B). For all the participants, we computed number of connections for each of these nodes; in OCD patients these nodes were more strongly connected to one another (intraconnections) compared with control subjects (t85 = 2.029, p =.046), with no difference for the total number of connections (t85 = 0.338, p =.736) or for number of connections directed toward other nodes (interconnections) (t85 = 0.049, p =.9614) (Figure 4C). Although modularity algorithms are nondeterministic, clustering of these nodes in the same module in OCD and their splitting among different modules in the network of healthy volunteers was robustly observed over multiple runs and several implementations of the analysis (Supplement, Supplemental Figure S6). There were no correlations between network modularity measures and cognitive or clinical measures that survived multiple comparisons (Supplement).

23A double dissociation of cognitive deficits contributing to candidate endophenotypes in OCD of goal-directed behavior and cognitive flexibility was identified for separate frontostriatal circuits. A selective deficit in cognitive flexibility (attentional set-shifting) in OCD was associated with reduced functional connectivity between the ventrolateral PFC and the DCd, but not the PUT. In contrast, impaired goal-directed planning was associated with reduced functional connectivity between the dorsolateral PFC and the PUT, but not the DCd. The latter deficit predicted severity of self-reported OCD symptoms and anxiety state and is consistent with recent hypotheses proposing fundamental goal-directed learning impairments in OCD. Data-driven network analysis provided evidence in OCD patients of denser connectivity within a group of nodes, including the caudate and putamen, as well as the cerebellum.

24Considerable evidence has shown the ventrolateral PFC to be necessary for attentional set-shifting. Excitotoxic lesions of lateral PFC in marmosets produced selective impairments in attentional set-shifting, whereas lesions to the orbitofrontal cortex impaired another form of cognitive flexibility, reversal learning. In a human fMRI paradigm, attentional set-shifting selectively recruited the ventrolateral PFC. Involvement of the striatum in set-shifting however has hitherto been less clear. Healthy volunteers exhibit caudate activation for reversal in rule classification, but not for extradimensional set-shifting. However, set-shifting–related cortical activity was mirrored by activity in the caudate nucleus and dorsal thalamus. Recently, resting-state functional connectivity between the ventral striatum and a priori-selected dorsolateral PFC was reported to be associated with attentional set-shifting in a large sample of healthy volunteers. By contrast, the present study used unbiased whole-brain analysis to demonstrate that functional connectivity between the caudate and ventrolateral PFC was associated with EDs performance in OCD patients (whether or not combined with the control group for this analysis). It is however possible that additional circuitry, including the ventral striatum, may be associated with EDs performance because we did find some associations in the OCD group with this circuit, although these analyses were post hoc and constrained by multiple comparison. It is nevertheless clear that separate circuits relate to cognitive flexibility and goal-directed planning. A limitation of the present study was the insufficient variability in EDs (and also planning) performances in the control subjects to demonstrate the same relation in that group alone.

25Our results provide new evidence that, in OCD patients, the inability to switch attention from a previously relevant dimension to form a new attentional set is intimately related to weakened underlying resting-state connectivity between the DCd and a network of brain regions including the ventrolateral PFC. Consistent with previous data, OCD patients formed attentional sets, as indexed by intact performance on discrimination and intradimensional stages, but showed selective impairment in shifting attention between stimulus dimensions. We augmented previous investigations by showing this effect to be independent of medication, in agreement with evidence in animals and humans that serotoninergic mechanisms are not implicated in EDs performance. The set-shifting deficit was also independent of clinical severity, in agreement with evidence that it may be an endophenotype.

26A separate circuit was relevant for the ability to attain goals via single-contingency, instrumental response sequences. Reduced functional connectivity between the putamen and the dorsolateral PFC was associated with inferior performance in OCD patients alone, as well as in the combined sample including control subjects. In task-related imaging studies, the dorsolateral PFC has been classically implicated in executive planning together with parietal and cingulate cortices. Anatomical data also support our findings; rich reciprocal connections exist between the dorsolateral PFC and the posterior parietal cortex, which project onto overlapping areas of the putamen. In OCD patients, a direct positive association was found such that increased functional connectivity between the PUT and the parietal cortex was predictive of better performance. Goal-directed impairment at the hardest planning levels replicated previous data, with no differences between medicated and unmedicated patients. Goal-directed planning failures were associated with self-reported symptom severity and anxiety only in medicated OCD patients, probably because their underlying symptoms were more severe and mitigated by medication. Further studies of first-degree unaffected relatives will clarify whether goal-directed impairment represents a state or trait marker for OCD.

27Our whole-brain network analysis revealed that, in OCD patients but not in control subjects, nodes belonging to the basal ganglia and cerebellum were more strongly intraconnected, thus corresponding to a unit or “conglomerate.” Results were consistent with previous investigations revealing a higher degree of local connectivity for the putamen in OCD patients and with frequent observations of elevated metabolism of those regions in PET studies during resting state. In other words, the stronger intraconnectivity of basal ganglia plus cerebellar nodes drives their separation into a distinct autonomous unit as a function of diagnosis, analogous to that of high impulsive subjects in a comparable study. Together with the functional connectivity analysis, this evidence parsimoniously suggests that the PFC exerts less top-down control of these subcortical regions. In general, our complementary imaging analyses suggest that increased connectivity within the striatum might coexist with its decreased functional connectivity to frontoparietal cortical regions, being associated with less flexible and impoverished goal-directed forms of behavior. Although the cerebellum has not hitherto been a major focus of interest in OCD, previous and present findings suggest that it merits further study. In line with anatomical evidence of cerebellar frontostriatal circuitry overlapping, we found clustering of striatum and cerebellum in our graph analysis. Much clinical and experimental evidence implicate the cerebellum in executive functions, attentional set-shifting, and motor sequencing, consistent with our evidence that striatal-cerebellar connectivity is implicated in attentional set-shifting. We saw no involvement of the cerebellum in relation to planning, perhaps because planning relates to goal-directed rather than habitual/skilled, or what may become just perfect behavior in OCD. Imbalances in corticostriatal connectivity with increased and decreased functional connectivity within the ventral and dorsal striatal networks, respectively, are consistent with previous investigations. Existing studies of unmedicated patients, however, only provide mixed findings. We directly compared medicated with unmedicated patients, showing that hyperconnectivity of the NAc and PUT to non-PFC cortical areas was more evident in medicated patients. This hyperconnectivity does not necessarily result from medication per se, because the medicated patients might have had more severe underlying symptoms; the hyperconnectivity may therefore reflect OCD symptoms.

28PFC regions found here to be functionally connected to basal ganglia and relevant for specific functions, namely ventrolateral PFC for cognitive flexibility and dorsolateral PFC for goal-directed planning, overlapped with those observed in corresponding task-related fMRI studies. This adds to the burgeoning literature establishing a close relation between resting-state and task-related functional connectivity in a healthy population, and with findings of resting-state functional connectivity related to response inhibition in an OCD population. Despite specific neural changes to be expected during the execution of the task, connectivity at rest appears to represent a functional predisposition that enables task execution. Indeed, our results suggest that the observed behavioral deficits depend on this trait-like predisposition, associated with objective behavioral measures heralding symptom manifestation. Thus, resting-state connectivity might serve as a biomarker predicting individual variability in behavioral performance, as shown here. Resting state is a promising practical tool, considering its relative ease and simplicity of data collection and its potential for providing reliable brain mapping from relatively short durations of scanning. We suggest that coupling of well-standardized behavioral indices and brain measures derived from an easily applicable resting-state sequence is a valuable approach for identifying neurobehavioral markers for severe psychiatric disorders, in line with the recent Research Domain Criteria initiative aimed at identifying the biological bases of mental disorders.

29In conclusion, the present data represent the first identification of discrete striatal-cortical circuits associated with key cognitive endophenotypes for OCD. The data extend the neurobiological model of OCD centered on dysfunction of frontostriatal brain circuits. Our study shows that these alterations have a direct link to cognitive processes severely impaired in OCD and of relevance for its symptoms and diagnosis.