Heterogeneous neuroimaging findings in obsessive-compulsive disorder converge to a common brain network using coordinate network mapping

Abstract

Neuroimaging studies in obsessive-compulsive disorder (OCD) have yielded heterogeneous findings, limiting efforts to localize OCD to specific brain regions or networks. This study applied coordinate network mapping (CNM) to test whether these distributed abnormalities converge onto a common brain network and to explore the biological substrates of this network through associations with neurotransmitter systems. We systematically reviewed and extracted coordinates of significant clusters from 66 Gy matter volume (GMV) and resting-state activity studies involving 3123 patients with OCD and 3116 healthy controls. Using CNM and functional connectome data from 1000 healthy individuals, we identified brain networks functionally connected to these abnormalities. Spatial correlations were examined between identified networks and 19 types of neurotransmitter maps acquired from over 1000 healthy subjects. The GMV and resting-state activity findings in OCD both converged to a common, functionally connected brain network involving the striatum, thalamus, cingulate, supplementary motor area, insula, and inferior frontal and superior temporal gyri. The identified networks derived from the two neuroimaging modalities were significantly correlated with the N-Methyl-D-Aspartate (NMDA) receptor distribution (r = 0.551 and r = 0.541, respectively; p < 0.001). Findings were robust across different methodological parameters and demonstrated good test–retest reliability. This study identifies a robust and reproducible brain network underlying the heterogeneous neuroimaging abnormalities in OCD, helping to interpret previous inconsistencies. Furthermore, the alignment of this network with the cortico-striato-thalamo-cortical circuitry and NMDA receptor distribution underscores its biological and clinical relevance, thereby illuminating potential targets for future neuromodulation and pharmacological interventions.

Introduction

Obsessive-compulsive disorder (OCD) is a chronic and disabling mental health condition characterized by the presence of obsessions, compulsions, or both. Numerous neuroimaging studies have been conducted on patients with OCD; however, the findings are heterogeneous. This heterogeneity may stem from variations in subject characteristics (e.g., age, gender, medication status, OCD subtypes, illness duration, and comorbidity), as well as differences in analytic methods and neuroimaging modalities. Such variability has hindered the ability to consistently localize OCD to specific brain regions or brain networks across individual studies, meta-analyses [[1], [2], [3], [4], [5]], or even mega-analyses [[6], [7], [8]]. Furthermore, there is growing consensus that OCD should be associated with distributed brain networks rather than isolated regions. Recently developed methods, including lesion network mapping (LNM) and its extension—coordinate network mapping (CNM)—have provided novel approaches to overcoming the heterogeneity and to mapping neuropsychiatric disorders to consensual brain networks.

In traditional lesion studies, brain regions associated with clinical symptoms were identified by locating overlapping brain lesion sites among patients with similar symptoms. However, symptoms may result from functional alterations in regions connected to the lesion site, due to the lesion’s remote functional effects [9,10]. To group heterogeneously distributed lesions resulting in the same symptom into a single unifying network, Fox et al. proposed the LNM approach in 2015 [10]. This approach leverages the normative human connectome to identify networks likely to be functionally affected by heterogeneously distributed lesions. In psychiatric disorders where no overt lesions are present, an extended method, termed CNM, has been proposed, where abnormalities in gray matter volume (GMV), cortical thickness, metabolism, perfusion, task-based activity, and resting-state activity, are treated as “lesions”. Applying CNM, heterogeneous neuroimaging findings across studies have been found to be functionally connected with the same brain network for specific disorders or symptoms, including Alzheimer’s disease and delusions in Alzheimer’s disease [11,12], migraine [13], substance use disorders [14], social cognitive and neurocognitive dysfunction in schizophrenia [15], and suicide [16]. These findings suggest that heterogeneous neuroimaging abnormalities associated with a neuropsychiatric condition across different studies are, in fact, functionally connected with a single uniform network. These heterogeneous distributed abnormalities may exert remote functional effects on the same network, ultimately contributing to the development or manifestation of the condition. Therefore, it provides a reasonable and novel explanation for the heterogeneous observations in previous neuroimaging studies on neuropsychiatric conditions.

In this study, we applied CNM to test whether heterogeneously distributed neuroimaging abnormalities in the OCD literature could map to a common brain functional network. If such a brain network exists, it will not only improve our understanding of the neuropathology underlying OCD, but also provide a uniform explanation for the seemingly inconsistent findings in previous neuroimaging studies. Furthermore, to elucidate potential biological substrates for the OCD-associated brain network identified by CNM, we examined whether the identified network spatially overlaps with the distributions of common neurotransmitter receptors or transporters. We hypothesized that heterogeneously distributed neuroimaging abnormalities observed in patients with OCD would be functionally connected to a common brain network, regardless of imaging modality. We further hypothesized that the network identified via the CNM method would exhibit substantial spatial overlap with the distributions of specific neurotransmitter receptors, in particular related to glutamate. Accumulating evidence links glutamatergic dysfunction—particularly involving N-Methyl-D-Aspartate (NMDA) receptor—to the pathophysiology of OCD [17,18]. Abnormal glutamate levels and disrupted NMDA receptor signaling have been observed in key regions implicated in OCD, such as the anterior cingulate cortex, thalamus, and striatum [[19], [20], [21], [22], [23]]. These findings suggest that altered glutamatergic neurotransmission may underlie both structural and functional brain abnormalities observed in OCD and could serve as a promising target for therapeutic interventions. The outcomes of the present study may provide insights into the pathophysiology and potential therapeutic targets of OCD.

Materials and methods

Study search, selection, and data extraction

While the CNM framework allows for the inclusion of various neuroimaging abnormalities—including changes in GMV, cortical thickness, metabolism, perfusion, resting-state activity, and task-based activity—our study focused specifically on GMV and resting-state activity due to data availability and cross-study comparability. Structural magnetic resonance imaging (MRI), resting-state functional MRI (fMRI), and task-based fMRI are the most commonly used modalities in the OCD neuroimaging literature. However, since cortical thickness is not voxel-based and task-based activity studies use different task paradigms and statistical contrasts, they bring difficulty to the consistent extraction of peak coordinates and thereby limiting cross-study comparability. Additionally, studies of metabolism and perfusion are fewer in number. Therefore, we selected GMV and resting-state activity (e.g., amplitude of low-frequency fluctuations [ALFF], fractional amplitude of low-frequency fluctuations [fALFF], or regional homogeneity [ReHo]), which are frequently used and provide a sufficient number of voxel-based, whole-brain studies with standardized coordinate reporting—enabling robust and meaningful CNM analyses.

We previously registered a meta-analysis protocol on neuroimaging findings in OCD in PROSPERO (registration number: CRD42025630391). Two authors (HY and YH) independently identified relevant peer-reviewed articles published in PubMed, Embase, and Web of Science from inception to December 21, 2024. A detailed description of the search strategy, study selection criteria, and data extraction procedures is provided in the Supplement. A flowchart of the article selection process is shown in Fig. S1. For studies reporting more than one contrast, all contrasts comparing patients with OCD to healthy controls on GMV or resting-state activity were included. Coordinates of peak voxels in significant clusters were extracted, regardless of whether they were reported in Montreal Neurological Institute (MNI) or Talairach space. All coordinates were subsequently converted to MNI space using the icbm2tal transformation implemented in GingerALE [24].

Coordinate network mapping for each imaging modality

Following procedures used in previous studies [[11], [12], [13], [14]], we applied CNM to test whether there exists a brain network functionally connected to the heterogeneously distributed neuroimaging abnormalities in OCD (Fig. 1). First, we created 6-mm radius spheres centered on each extracted coordinate to define regions of interest (ROIs). Next, ROI-to-whole-brain functional connectivity (FC) maps were calculated using preprocessed resting-state fMRI data from a normative dataset of 1000 healthy subjects (age: 21.40 ± 2.86 years; male/female: 500/500) derived from the Brain Genomics Superstruct Project (BGSP) [25]. The original BGSP data were acquired on matched Siemens 3T MAGNETOM Tim Trio MRI systems using a 12-channel phased-array head coil. Resting-state blood-oxygen-level-dependent (BOLD) fMRI data were acquired using a gradient-echo echo-planar imaging sequence with a repetition time of 3.0 s, echo time of 30 ms, flip angle of 85°, 47 slices, and 3.0-mm isotropic voxel resolution [25]. The 1000 subjects included in the normative connectome were selected from the larger BGSP dataset based on a combined quality score incorporating normalized DVARS (derivative of variance across registration steady-states), Entropy Focused Criterion, and framewise displacement, with the aim of minimizing the effects of head motion without framewise scrubbing. Anatomical surfaces were reconstructed using FreeSurfer v4.5, and functional preprocessing was performed using a slightly modified version of the Computational Brain Imaging Group (CBIG) fMRI preprocessing pipeline, with modifications and configuration settings designed to approximate those used to generate the original Yeo 1000 connectome [26]. The resulting anatomical images were transformed to MNI152 2-mm isotropic space, and the preprocessed resting-state fMRI data were used to generate the normative functional connectome. Fisher’s z-transformation was applied to each FC map to improve normality. For each ROI, one-sample t-tests were conducted on the 1000 ROI-to-whole-brain FC maps using a family-wise error (FWE)-corrected threshold of p < 0.05 to determine which brain regions were functionally connected to the ROI. Only positive FC was considered, which is in line with the standard procedure of the CNM approach given the ongoing debate regarding the biological interpretation of negative FC [27,28]. The resulting T-map with significant clusters for each ROI was then binarized. The binarized maps derived from all ROIs within the same imaging modality (GMV or resting-state activity) were subsequently compared with each other to generate an “overlap degree map”, where the value of each voxel in the map quantifies the overlap likelihood across all examined ROIs (i.e., the number of binarized maps in which a given voxel is present, divided by the total number of binarized maps derived from all ROIs). The overlap degree maps were then thresholded at 50% to yield the OCD-associated network derived from GMV and the OCD-associated network derived from resting-state activity.

Fig. 1.

Flowchart illustrating the application of coordinate network mapping (CNM) to identify OCD-associated brain networks. The CNM approach was based on previously reported neuroimaging abnormalities in OCD and functional connectome data from 1000 healthy individuals. The resulting networks functionally connected to the heterogeneously distributed neuroimaging abnormalities. OCD = obsessive-compulsive disorder; MNI = Montreal Neurological Institute; ROI = region of interest; FC = functional connectivity; rs-fMRI = resting-state functional magnetic resonance imaging; VBM = voxel-based morphometry.

The 50% threshold was selected based on previous studies [15,29], as well as the results of the random ROI analysis. The random ROI analysis was performed to calculate the likelihood of generating random network overlap at different thresholds (40%, 45%, 50%, and 55%; flowchart shown in Fig. S2). Specifically, for each imaging modality, we randomly selected the same number of ROIs as used above from the 718 nodes in the CAB-NP atlas [30]. The whole CNM analysis was then repeated 10,000 times to generate 10,000 random network overlaps for each threshold. We next calculated, for every brain voxel, the probability of appearing in these random network overlaps; voxels with a value greater than 0 constituted the null network. For each threshold, we subsequently calculated the proportion of the identified OCD-associated network that was present in the null network, computed as the number of overlapping voxels between the null network and the identified OCD-associated network divided by the total number of voxels in the null network. The results indicated that, at the threshold of 50%, the proportion of the identified OCD-associated network present in the null network was below 5% (Fig. S3). Therefore, a 50% threshold was adopted in this study.

Spatial correlation between imaging modalities

We calculated the voxel-wise spatial correlation between the overlap degree maps generated from GMV and resting-state activity. Spatial correlation was assessed using Spearman’s correlation, based on the results of normality tests conducted on all voxels in the two overlap degree maps. Furthermore, a permutation test was conducted to determine whether the observed spatial correlation was stronger than expected by chance. In the permutation test, BrainSMASH was used to generate 10,000 surrogate maps for the voxel-wise overlap degree map derived from one modality, given the potential influence of spatial autocorrelation on the correlation results [31]. Each of the 10,000 surrogate overlap degree maps was then correlated with the overlap degree map from the other modality, generating 10,000 null correlation coefficients. The permutation p-value was calculated as the proportion of permutations that yielded null correlation coefficients higher than the observed value.

Overlap with canonical brain networks

To enhance the interpretability of the generated OCD-associated networks, we calculated the overlap proportion between the identified network derived from GMV or resting-state activity and 8 canonical brain networks, including 7 cortical networks defined by the parcellation of Yeo et al. [26] and 1 subcortical network from the Brainnetome Atlas [32]. The overlap proportion was defined as the number of overlapping voxels between the identified network derived from GMV or resting-state activity and each canonical brain network, divided by the total number of voxels in the respective identified network.

Associations between the overlap degree maps and neurotransmitter maps

In the overlap degree maps, regions with a higher overlap degree are considered to be more likely related to OCD. To test whether the overlap degree of brain regions in OCD is associated with the distributions of neurotransmitter receptors or transporters, we performed spatial correlation analyses between the overlap degree maps and the neurotransmitter maps across the whole brain. Nineteen positron emission tomography (PET)-derived maps of neurotransmitter receptors and transporters—spanning nine neurotransmitter systems, including dopamine, serotonin, norepinephrine, glutamate, gamma-aminobutyric acid (GABA), acetylcholine, histamine, opioid, and cannabinoid—were included in the analysis. These maps were generated from PET data acquired in over 1000 healthy participants, and a group-average tracer map was constructed for each receptor/transporter type. The maps are publicly accessible, and detailed information about the 19 receptor and transporter types is provided in Table S1. Briefly, the overlap degrees and neurotransmitter intensities were extracted from the 116 brain regions defined by the Automated Anatomical Labeling (AAL) atlas. Based on the results of normality tests, Spearman or Pearson correlation analyses were then conducted between the overlap degrees and the receptor/transporter intensities.

Given the potential impact of spatial autocorrelation on the results of the spatial correlation analyses, BrainSMASH was used to generate 1000 surrogate maps for the overlap degree maps derived from GMV and resting-state activity, respectively [31]. Each surrogate map preserved the spatial autocorrelation structure of the corresponding original map. Spearman correlations were then performed between each surrogate map and the receptor/transporter intensity maps, yielding 1000 null correlation coefficients. Therefore, permutation tests (n = 1000) were performed for each of the 38 correlation analyses (2 modalities × 19 receptor and transporter types). To correct for multiple comparisons, the resulting p-values were adjusted using the Bonferroni correction (p < 0.05/38 = 0.001).

Validation and test-retest analyses

To account for the influence of ROI size on the results, we repeated the CNM analyses using spherical ROIs with radii of 2 mm, 4 mm, 8 mm, and 10 mm to calculate FC. Furthermore, to assess the reliability and stability of the findings, we performed the CNM analyses using retest resting-state fMRI data from 747 healthy subjects (age: 21.08 ± 2.64 years; male/female: 338/409), all of whom were part of the original 1000 healthy subjects from the Brain Genomics Superstruct Project [25].

Sensitivity analyses

Two sensitivity analyses were performed to evaluate the influence of sample age composition and individual contrasts on the identified OCD-associated networks. First, we restricted the analysis to studies including adult OCD samples to examine whether the inclusion of pediatric and adolescent samples influenced the identified OCD-associated networks. After identifying the OCD-associated networks in the adult OCD samples, spatial overlap between these networks and those derived from the full set of included OCD studies was quantified using Dice coefficients. The Dice coefficient was defined as 2 × (number of overlapping voxels)/(number of voxels in network A + number of voxels in network B). For descriptive purposes, Dice coefficients of <0.20 were considered to indicate low spatial similarity, 0.20–<0.40 low-to-moderate spatial similarity, 0.40–<0.60 moderate spatial similarity, 0.60–<0.80 high spatial similarity, and 0.80–1.00 very high spatial similarity.

Second, we performed a leave-one-contrast-out sensitivity analysis. Specifically, each contrast was sequentially excluded, and the CNM procedure was repeated using the remaining contrasts. Spatial overlap between the OCD-associated networks derived from each leave-one-contrast-out analysis and those derived from the full set of included OCD studies was quantified using Dice coefficients.

Results

Included studies and extracted coordinates

After selection, 66 studies with 83 contrasts involving 3123 patients with OCD and 3116 healthy controls (HCs) were included. Specifically, for GMV abnormalities, 37 studies with 42 contrasts (1936 patients with OCD and 2016 HCs) were included, from which 178 coordinates of significant clusters were extracted. For resting-state activity abnormalities, 30 studies with 41 contrasts (1227 patients with OCD and 1157 HCs) were included, from which 212 coordinates of significant clusters were extracted. One study included both GMV and resting-state activity analyses [33]. Table S2–S5 and Table S6–S8 provide the subject and imaging characteristics for the GMV and resting-state activity studies, respectively. The coordinates of significant clusters from the GMV and resting-state activity studies are listed in Tables S9 and S10, respectively.

OCD-associated networks derived from GMV and resting-state activity

The OCD-associated network derived from GMV mainly involved the bilateral anterior and middle cingulate, insula, inferior frontal gyrus, superior temporal gyrus, caudate, putamen, pallidum, amygdala, hippocampus, and left thalamus (Fig. 2, part A). The OCD-associated network derived from GMV was primarily distributed across the ventral attention network (30.29%), subcortical network (30.09%), and default mode network (21.42%) (Fig. 3, part A).

Fig. 2.

The identified OCD-associated networks. (A) The OCD-associated network derived from GMV. (B) The OCD-associated network derived from resting-state activity. (C) The OCD-associated networks derived from GMV and resting-state activity overlapped mainly in the bilateral insula, caudate, anterior and middle cingulate, and left thalamus. OCD = obsessive-compulsive disorder; GMV = gray matter volume.

Fig. 3.

Overlap between the identified networks and 8 canonical brain networks. (A) Overlap proportions between the OCD-associated network derived from GMV and 8 canonical brain networks. (B) Overlap proportions between the OCD-associated network derived from resting-state activity and 8 canonical brain networks. The overlap proportion represents the percentage of the identified OCD-associated network that overlaps with each canonical brain network and was calculated as the number of overlapping voxels between the identified network derived from GMV or resting-state activity and each canonical brain network divided by the total number of voxels in the respective identified network. OCD = obsessive-compulsive disorder; GMV = gray matter volume.

The OCD-associated network derived from resting-state activity mainly involved the bilateral anterior and middle cingulate, supplementary motor area, insula, inferior frontal gyrus, superior temporal gyrus, Rolandic operculum, caudate, and thalamus (Fig. 2, part B). The OCD-associated network derived from resting-state activity was primarily distributed across the ventral attention network (49.02%), subcortical network (24.20%), and somatomotor network (9.28%) (Fig. 3, part B).

Furthermore, the OCD-associated networks derived from GMV and resting-state activity overlapped mainly in the bilateral insula, caudate, anterior and middle cingulate, and left thalamus (Fig. 2, part C).

Spatial correlation between overlap degree maps derived from GMV and resting-state activity

Voxel-wise overlap degrees in the overlap degree maps derived from GMV were significantly positively correlated with voxel-wise overlap degrees in the overlap degree maps derived from resting-state activity (rho = 0.816, permutation-based p < 0.001, based on 10,000 permutations, Fig. 4).

Fig. 4.

Association between overlap degree maps derived from GMV and resting-state activity. (A) Significant positive correlation between voxel-wise overlap degrees derived from the GMV studies and those derived from the resting-state activity studies (correlation coefficient rho = 0.816). (B) Permutation tests confirmed that this correlation was stronger than expected by chance (p < 0.001). (C) Overlap degree map derived from the GMV studies. (D) Overlap degree map derived from the resting-state activity studies. GMV = gray matter volume.

Spatial correlations between overlap degree maps and neurotransmitters

The overlap degree maps derived from GMV and resting-state activity were both significantly positively correlated with the distribution of NMDA receptor (r = 0.551, permutation-based p < 0.001, based on 1000 permutations, Fig. 5 [part A]; r = 0.541, permutation-based p < 0.001, based on 1000 permutations, Fig. 5 [part B], respectively). Both associations remained statistically significant after Bonferroni correction for the 38 neurotransmitter-related correlation analyses (corrected p < 0.05/38 = 0.001).

Fig. 5.

Spatial correlations between the overlap degree maps and the distributions of 19 types of neurotransmitter receptors or transporters. (A) Overlap degree map derived from GMV. (B) Overlap degree map derived from resting-state activity. For both parts: The radar plots show the correlation coefficients between receptor/transporter distributions and the overlap degree maps. The scatter plots illustrate significant positive correlations between NMDA receptor distribution and the overlap degree maps. The histograms demonstrate that the observed correlations are not influenced by the potential impact of spatial autocorrelation and are stronger than expected by chance. GMV = gray matter volume; NMDA = N-Methyl-D-Aspartate; 5-HT1A = 5-Hydroxytryptamine Receptor 1A; 5-HT1B = 5-Hydroxytryptamine Receptor 1B; 5-HT2A = 5-Hydroxytryptamine Receptor 2A; 5-HT4 = 5-Hydroxytryptamine Receptor 4; 5-HT6 = 5-Hydroxytryptamine Receptor 6; 5-HTT = 5-Hydroxytryptamine Transporter; α4β2 = Nicotinic Acetylcholine Receptor α4β2 Subtype; CB1 = Cannabinoid Receptor Type 1; D1 = Dopamine Receptor D1; D2 = Dopamine Receptor D2; DAT = Dopamine Transporter; GABAa = Gamma-Aminobutyric Acid Type A Receptor; H3 = Histamine Receptor H3; M1 = Muscarinic Acetylcholine Receptor M1; mGluR 5 = Metabotropic Glutamate Receptor 5; MOR = Mu-Opioid Receptor; NET = Norepinephrine Transporter; VAChT = Vesicular Acetylcholine Transporter.

Validation and test-retest analyses

We conducted several supplementary analyses to support the robustness of the results. Specifically, we repeated the CNM analyses using 2 mm, 4 mm, 8 mm, and 10 mm radius spheres as ROIs to calculate FC. The resulting OCD-associated network derived from GMV or resting-state activity showed high similarity to the network obtained using 6 mm radius spheres as ROIs (Figs. S4 and S5). Furthermore, the generated OCD-associated networks derived from the resting-state fMRI retest data of 747 healthy subjects were nearly identical to those derived from the resting-state fMRI test data of 1000 healthy subjects (Fig. S6).

Sensitivity analyses

After excluding 3 studies reporting functional abnormalities and 8 studies reporting structural abnormalities in samples that included children or adolescents, the OCD-associated networks identified in adult OCD samples (27 functional studies with 38 contrasts and 29 structural studies with 33 contrasts) are shown in Fig. S7. The Dice coefficients between these networks and those derived from the full set of included OCD studies were 0.78 for the functional modality and 0.72 for the structural modality, indicating high spatial similarity between the networks derived from adult samples and those derived from the full set of studies.

In the leave-one-contrast-out sensitivity analysis, the Dice coefficients between the OCD-associated networks derived from each leave-one-contrast-out analysis and those derived from the full set of included OCD studies ranged from 0.72 to 0.97 for the functional modality (Table S11) and from 0.71 to 0.98 for the structural modality (Table S12), corresponding to high to very high spatial similarity across all leave-one-contrast-out analyses.

Discussion

Applying the CNM method and leveraging large-scale human connectome data from 1000 healthy subjects, we identified brain networks that are functionally connected to heterogeneously distributed neuroimaging abnormalities in patients with OCD. These heterogeneously distributed neuroimaging abnormalities were reported in 66 studies with 83 contrasts, involving 3123 patients with OCD and 3116 HCs. The OCD-associated networks derived from GMV and resting-state activity were spatially correlated (rho = 0.816) and mainly involved the striatum, thalamus, cingulate, supplementary motor area, insula, inferior frontal gyrus, and superior temporal gyrus. From the perspective of canonical brain networks, the OCD-associated networks derived from GMV and resting-state activity primarily mapped to the ventral attention network, subcortical network, default mode network, and somatomotor network. These results were robust across different ROI sizes and demonstrated good test-retest reliability. Importantly, the identified networks remained robust when the analysis was restricted to adult OCD samples and when individual contrasts were sequentially excluded. Furthermore, the overlap degree maps derived from GMV and resting-state activity were both significantly positively correlated with the distribution of NMDA receptor (r = 0.551 and r = 0.541, respectively).

Validity of the identified OCD-associated networks

In addition to validating the robustness of the identified OCD-associated networks by repeating CNM analyses using different ROI sizes, retest resting-state fMRI data, adult OCD samples, and leave-one-contrast-out analyses, the following four points further support the validity of the identified OCD-associated networks. First, the identified OCD-associated networks align well with the cortico-striato-thalamo-cortical (CSTC) circuits, which are central to the neurobiological understanding of OCD [34]. These circuits involve interconnected brain regions that form a feedback loop critical for motor control, cognitive processing, and affect regulation. Disruption in these loops is thought to underlie the intrusive thoughts (obsessions) and repetitive behaviors (compulsions) characteristic of OCD. Second, the identified OCD-associated networks mainly involved brain regions that have been the neuromodulation targets with established effectiveness in previous studies, including the supplementary motor area, orbitofrontal cortex, anterior cingulate cortex, and medial prefrontal cortex in transcranial magnetic stimulation [[35], [36], [37]]; the striatum and thalamus in deep brain stimulation [38,39]; and the supplementary motor area and orbitofrontal cortex in transcranial direct current stimulation [40]. The convergence between the identified OCD-associated networks and the targets of previously effective neuromodulation for OCD suggests that the therapeutic effects of neuromodulation in OCD may result from stimulation of the same network whose dysfunction contributes to the development or manifestation of OCD. Third, the consistency in brain regions identified with GMV and resting-state activity suggests highly comparable results across different imaging modalities. Fourth, the overlap degree maps of the identified networks are significantly associated with the distribution of the NMDA receptor in the brain, which is in line with the growing evidence of abnormal glutamatergic signaling in OCD, leading to excitotoxicity or dysregulated synaptic plasticity [41,42]. These associations may provide a biological explanation for the effectiveness of NMDA receptor antagonists—such as ketamine/esketamine, memantine, and amantadine—in treating OCD [41,[43], [44], [45], [46]].

Interpretation of the identified OCD-associated networks

Applying CNM, heterogeneous neuroimaging findings in OCD across previous studies are found to be functionally connected to a single uniform network. The identified OCD-associated network enhances our understanding of its neuropathology. Specifically, the distributed brain abnormalities reported previously may exert functional effects on this common brain network, ultimately contributing to the development or manifestation of OCD [9]. Furthermore, the identified network offers a reasonable and novel explanation for the heterogeneity in previous neuroimaging findings in OCD. It is likely that the inconsistent findings in the literature actually capture different connectivity patterns of this shared brain network.

Recent network-mapping studies in other psychiatric conditions provide further support for this network-level interpretation. Ji et al. used therapeutic network mapping to demonstrate that distinct antidepressant therapies, including pharmacotherapy, electroconvulsive therapy, and magnetic seizure therapy, converged on a common brain network centered on the amygdala [47]. Similarly, Liu et al. showed that heterogeneous neuroimaging changes associated with pharmacotherapy-induced emotional processing recovery across four psychiatric disorders could be reconciled into a common network anchored in the bilateral insula and putamen [48]. Together with the present findings, these studies suggest that heterogeneous regional neuroimaging alterations may reflect different manifestations of alterations within distributed functional networks rather than isolated abnormalities. These convergent findings further support the utility of normative-connectome-based network mapping for identifying distributed network-level substrates of psychiatric disorders and treatment-related brain changes.

Neurotransmitter associations

The significant positive correlations between the distribution of NMDA receptor and the overlap degree maps suggest that brain regions with higher NMDA receptor density may be more likely to be related to the pathological alterations in OCD. These findings are consistent with previous evidence implicating NMDA receptor-related glutamatergic signaling in the pathophysiology of OCD [41,42]. This study therefore identifies an OCD-associated brain network and establishes a potential link among “OCD–neuroimaging correlates–neurotransmitter”, highlighting both the underlying pathophysiology and potential therapeutic targets for OCD.

Implications for future research

This study has important implications for future research on OCD. In particular, the identified network could be used to select neuromodulation targets for the disorder. Prior work in other neuropsychiatric disorders has shown that stimulating different parts of networks identified from the LNM or CNM approach can yield similar treatment effects [[49], [50], [51], [52]]. Furthermore, treatment effects were greater when stimulating areas within the identified networks compared to areas outside the networks [52]. Therefore, brain regions within the OCD-associated networks identified in the present study may all serve as potential targets for neuromodulation interventions, yet their efficacy remains to be tested.

Moreover, the study may also offer implications for medication development, as the identified networks were significantly associated with the distribution of NMDA receptor. This finding aligns with growing evidence implicating glutamatergic dysfunction—particularly involving NMDA receptor—in the pathophysiology of OCD. Abnormal glutamate levels in brain regions such as the anterior cingulate cortex, supplementary motor area, thalamus, and striatum have been consistently linked to compulsive symptoms and altered functional connectivity [[19], [20], [21], [22], [23]]. The NMDA receptor plays a critical role in synaptic plasticity, learning, and neurodevelopment, while its dysregulation may contribute to OCD-related neurobiological changes [53,54]. Meta-analyses of clinical trials have demonstrated the efficacy of glutamatergic agents in alleviating OCD symptoms, with most of these agents targeting NMDA receptor [17]. The significant correlations between the distribution of NMDA receptor and the identified networks may provide a potential biological rationale for the therapeutic effects of NMDA-targeting medications. These include NMDA receptor antagonists, which reduce excessive glutamatergic activity, particularly in hyperactive CSTC circuits [18,55]. Conversely, NMDA receptor co-agonists may enhance NMDA receptor function in underactive or hypo-connected circuits, thereby facilitating synaptic plasticity and fear extinction [18,55]. Together, these studies suggest that both upregulation and downregulation of NMDA receptor activity—depending on the network context—may be beneficial in treating OCD. This network-based pharmacological framework not only supports the development or repurposing of compounds targeting specific glutamatergic subcomponents but also underscores the diverse mechanisms through which NMDA-targeting agents exert their effects. These mechanisms include receptor antagonism, co-agonism, reuptake inhibition, and ion channel modulation—each offering unique opportunities to modulate glutamatergic function within distinct network contexts relevant to OCD.

Limitations

Several limitations should be noted. First, the CNM approach weighted all included studies equally, regardless of sample size and effect size. Future research may improve this method by incorporating these factors as weights. Furthermore, the analysis was restricted to positive FC, and excluding negative FC could omit potentially meaningful anticorrelated networks. In addition, the present study focused on structural and functional alterations in gray matter and did not incorporate white matter functional alterations, largely because relatively few neuroimaging studies have investigated white matter function in OCD. Recent evidence indicates that resting-state BOLD signals in white matter can provide meaningful information about white matter functional organization and dysfunction and may contribute to the characterization of brain disorders [56,57]. Thus, incorporating white matter functional alterations into future CNM studies may provide a more comprehensive characterization of the brain networks associated with OCD. Second, although significant positive correlations were observed between the distribution of NMDA receptor and the identified networks, the NMDA receptor distribution data and the resting-state fMRI data used to identify OCD-associated networks were obtained from two separate groups of healthy subjects. Nevertheless, previous research indicates that the regional distribution of neurotransmitter receptors and transporters is highly consistent across healthy individuals [58]. Third, variations in sample age composition, illness duration, and medication exposure across cohorts may introduce heterogeneity. Although the sensitivity analysis restricted to adult OCD samples yielded networks with substantial spatial overlap with those derived from the full set of included studies, the number of studies involving children or adolescents only was limited, precluding a separate sensitivity analysis in this population. Similarly, the number of studies involving drug-naïve patients with OCD was limited, preventing a separate assessment of the potential influence of medication exposure.

Conclusions

In summary, heterogeneous neuroimaging findings in OCD across previous studies were found to converge on a single, functionally connected brain network. The identified network offers a novel explanation for the heterogeneity observed in previous studies, suggesting that the inconsistent findings in the literature may actually capture different connectivity patterns of this shared brain network. Furthermore, the significant positive correlations between the distribution of NMDA receptor and the identified OCD-associated network suggest that brain regions with higher NMDA receptor density are more likely to be involved in the pathological alterations of OCD. This study, therefore, identifies an OCD-associated brain network and establishes a link among OCD, neuroimaging correlates, and neurotransmitter systems—particularly glutamatergic signaling. These findings highlight both the underlying neurobiological mechanisms and potential therapeutic targets for OCD.

Data availability

The extracted peak coordinates and the identified OCD-associated network (in NIfTI format) are publicly available in the GitHub repository (https://github.com/yanmri/OCD_CNM). The code for generating the overlap degree map is also available in the repository.

Author contributions

HY: Conceptualization, Methodology, Formal analysis, Writing – original draft; YH: Formal analysis, Validation, Writing – original draft; WJ G and WB G: Supervision, Writing – review & editing, Funding acquisition; HC: Conceptualization, Methodology, Supervision, Writing – review & editing, Funding acquisition. All authors read and approved the final manuscript.

Funding

This study was supported by grants from the NIH (Grant No. R01MH138682 and R01MH137501), National Natural Science Foundation of China (Grant No. 82171508), and the Fundamental Research Funds for the Central Universities of Central South University (Grant No. 2024ZZTS0304).

Declaration of competing interest

The authors report no competing interests.

Appendix A. Supplementary data

The following are the Supplementary data to this article.