Functional Connectivity Gradients Reveal Altered Hierarchical Cortical Organization in Functional Neurological Disorder

Abstract

Background:

Neuroimaging studies of functional neurological disorder (FND), a core neuropsychiatric condition, often rely on discrete connections or parcellations that may obscure the brain’s functional network architecture. This study applied a gradient-based approach to examine macroscale cortical organization in FND.

Methods:

We analyzed resting-state functional magnetic resonance imaging (fMRI) data from 64 patients with mixed FND (FND-mixed), 61 age- and sex-matched healthy controls (HCs), and 62 psychiatric controls (PCs) matched on age, sex, depression, anxiety, and post-traumatic stress disorder (PTSD) severity. Functional connectivity gradients were computed to capture dominant axes of cortical organization. Between-group comparisons were conducted for the top three gradients, and associations with symptom severity were investigated. Subtype-specific patterns in functional motor disorder (n=49) and functional seizure (n=24) were also examined. Analyses controlled for age, sex, antidepressant use, and head motion, and were post-hoc adjusted for depression, anxiety, and PTSD-severity, and for childhood maltreatment.

Results:

The FND-mixed group showed alterations across all three gradients relative to HCs and PCs. Gradient 1 revealed increased values in sensorimotor regions, reflecting a shift toward more association-like connectivity. Gradient 2 showed altered differentiation between sensory systems. Gradient 3 exhibited reduced functional separation between representational and modulatory regions, with prominent shifts in the anterior cingulate cortex. Several regions displaying between-group differences also showed correlations with FND and somatic symptom severity. Exploratory analyses revealed overlapping and distinct patterns across subtypes vs. controls.

Conclusions:

We provide novel evidence of atypical hierarchical brain organization in FND, highlighting gradient-based approaches for identifying mechanistically-relevant altered functional brain organization.

INTRODUCTION

Functional neurological disorder (FND) is a condition at the origins of modern-day psychiatry and neurology. Patients with FND experience distressing motor, sensory, and cognitive symptoms not attributable to macroscopic structural brain lesions (1,2). FND is a prevalent and potentially disabling condition, accompanied by significant healthcare costs (3–5). Despite having been neglected throughout the late 20^th century, earning characterizations such as “Psychiatry’s Blind Spot” and “Medicine’s Silent Epidemic”, renewed interest in FND has been catalyzed by the use of a rule-in diagnostic approach (6–8). Although substantial recent progress has been made (9,10), the neurobiological mechanisms underlying FND remain incompletely understood (11).

Resting-state functional magnetic resonance imaging (fMRI) studies of FND have predominantly revealed alterations in somatomotor, salience, and default mode networks (11,12). These investigations often rely on network parcellations or seed-based analyses, which are useful for summarizing connectivity and testing circuit-level hypotheses (13–22). However, such approaches assume discrete network boundaries, which may obscure the fluid and overlapping nature of functional brain organization. Graph theory methods have also been used to probe network architecture alterations by modeling the brain as a set of nodes (regions) and edges (connections) (23–27). These approaches have led to functional brain organization insights in FND (i.e., increased somatomotor network integrated connectivity (27)), yet they do not account for more continuous connectivity patterns.

Gradient mapping offers a complementary perspective for examining macroscale brain organization along multiple continuous dimensions (28–33). These approaches use dimensionality reduction techniques to identify low-dimensional representations of functional connectivity, referred to as gradients. When applied to the cerebral cortex, gradients capture smooth transitions in connectivity and reflect hierarchical dimensions of cortical organization. Gradient-based analyses have been used to study macroscale brain alterations in psychiatric and neurological conditions (34–38). Research has highlighted three gradients reflecting dominant axes of cortical organization (30,39,40): Gradient 1, referred to as the association-sensorimotor gradient (also called a ‘transmodal-unimodal’ gradient), is anchored at one end by the default mode network and at the other by sensory and motor networks and the salience/cingulo-opercular networks; Gradient 2, referred to as the visual-somatomotor gradient, distinguishes between visual and somatomotor sensory systems; and Gradient 3, referred to as the representation-modulation gradient based on functional distinctions drawn from experimental evidence (e.g., 41–47), differentiates the default mode and sensory networks from modulatory/attentional (i.e., frontoparietal, dorsal attention, and salience/cingulo-opercular) networks. This gradient has also been variably referred to as a “multiple demand vs. default mode” (48,49) or “task-positive vs. negative” (50,51) gradient.

Here, we applied a gradient-based framework to investigate alterations in the hierarchical functional organization of the cerebral cortex in patients with FND. We computed connectivity gradients from resting-state fMRI data in a mixed FND cohort (FND-mixed) and compared them to those derived from age- and sex-matched healthy controls (HCs) and psychiatric controls (PCs) matched on age, sex, depression, anxiety, and PTSD symptom severity. Inclusion of PCs allowed us to discern whether observed alterations were specific to FND or reflected features of co-occurring psychiatric conditions or shared risk factors (52,53). We examined between-group differences in the top three gradients, explored subtype-specific effects in patients with functional motor disorder (FND-motor) and functional seizures (FND-seiz), and evaluated associations between gradient values and both core FND symptom severity and non-core somatic symptom burden (54,55). Through these analyses, we aimed to advance understanding of large-scale functional brain organization in FND.

METHODS

Participants

Sixty-four participants comprising a mixed cohort of individuals with FND-motor and/or FND-seiz (FND-mixed; 54 female; mean age [SD]=40.0±13.8 years; average illness duration=4.2±5.3 years, range=0.3–25 years; Table 1) were prospectively recruited from the Massachusetts General Hospital between June 2018 and March 2024 (7,56). FND diagnoses were made based on positive signs and electroencephalography data (FND-seiz only (49)). Recruitment included those with FND-motor and/or FND-seiz, given that many individuals present with mixed symptoms and/or develop distinct FND symptoms longitudinally (58,59). The FND-mixed cohort included 49 individuals with FND-motor (tremor=23; weakness=18; gait=17; speech=14; tics/jerks/spasms=10; dystonia=3 [motor phenotypes were not mutually exclusive]) and 24 participants with FND-seiz (documented=17; clinically established=2; probable=5); nine met criteria for both FND-motor and FND-seiz subtypes (Supplementary Table 1). Data from this cohort have been published in one structural MRI study and one graph-theory fMRI study (27,60).

Table 1.

Demographic and psychometric characteristics of mixed functional neurological disorder (FND-mixed), psychiatric control (PC) and healthy control (HC) samples.

	FND-mixed (N = 64) Mean ± SD or N	PCs (N = 62) Mean ± SD or N (p corrected)	HCs (N = 61) Mean ± SD or N (p corrected)
Age (years)	40.0 ± 13.8	35.6 ± 13.2 (0.2)	36.5 ± 10.9 (0.2)
Sex	F: 54; M: 10	F: 52; M: 10 (0.9)	F: 50; M: 11 (0.7)
SDQ-20	34.7 ± 15.6	22.0 ± 5.2 (<0.001*)	20.3 ± 0.6 (<0.001*)
PHQ-15	13.3 ± 7.3	6.3 ± 3.6 (<0.001*)	2.6 ± 2.2 (<0.001*)
BDI-II	16.8 ± 12.3	14.6 ± 13.1 (0.4)	1.5 ± 2.4 (<0.001*)
STAI-Total	82.0 ± 29.2	79.0 ± 23.6 (0.6)	53.8 ± 9.2 (<0.001*)
PCL-5	28.2 ± 19.6	22.0 ± 18.7 (0.2)	2.6 ± 3.6 (<0.001*)
CTQ-Abuse	31.8 ± 14.8	26.0 ± 12.0 (0.03*)	18.0 ± 3.4 (<0.001*)
CTQ-Neglect	20.8 ± 9.8	19.6 ± 8.8 (0.6)	13.2 ± 4.2 (<0.001*)
SSRI/SNRI	33	29 (0.7)	0 (<0.001*)
Mean FD	0.08 ± 0.05	0.07 ± 0.02 (0.6)	0.05 ± 0.04 (<0.001*)

P values reflect statistical significance between the FND cohort and the control group after False Discovery Rate (FDR) correction for multiple comparisons. Asterisks indicate corrected p-values < 0.05. Statistical comparisons of continuous variables were done using a Mann Whitney U test, while comparisons of discrete variables were done using a Chi-Squared test. Six FND and 1 PC participants had incomplete data. For each variable, counts of missing data were as follows: SDQ-20: 6 FND, 1 PC; PHQ-15: 6 FND; BDI-II: 4 FND; STAI-Total: 4 FND; PCL-5: 5 FND; CTQ: 4 FND. F, Female, M, Male; SDQ-20, Somatoform Dissociation Questionnaire-20; PHQ-15, Patient Health Questionnaire-15; BDI-II, Beck Depression Inventory-II; STAI-Total, Spielberger State-Trait Anxiety Inventory-Total; PCL-5, Post-Traumatic Stress Disorder Checklist for DSM-5; CTQ, Childhood Trauma Questionnaire; SSRI/SNRI, selective serotonin reuptake inhibitor/serotonin norepinephrine reuptake inhibitor use; FD, framewise displacement.

Sixty-two PCs (52 female; mean age [SD]=35.6±13.2 years) and sixty-one HCs (50 female; mean age [SD]=36.5±10.9 years) were recruited from the community. PCs had a lifetime history of depression (n=53), anxiety (n=43), and/or post-traumatic stress disorder (PTSD; n=21) (Forty-two had more than one of these diagnoses; Supplementary Table 2). HCs had no history of psychiatric disorders, and none were on psychotropic medications. Exclusion criteria across the three cohorts are detailed in Supplementary Materials.

Seventeen additional participants (4 FND, 6 PCs, 7 HCs) were enrolled but excluded due to excessive head motion (<120 usable volumes). All individuals signed informed consent and the Mass General Brigham Institutional Review Board approved this study.

Neuropsychiatric Characterization

Participants underwent a Structured Clinical Interview for Diagnostic and Statistical Manual Disorders (see Supplementary Table 1 and Supplementary Table 2 caption for details). Participants also completed psychometric questionnaires, including the Beck Depression Inventory-II (BDI-II), Spielberger State-Trait Anxiety Inventory (STAI), PTSD Checklist-5 (PCL-5), and Childhood Trauma Questionnaire (CTQ). Core FND symptom severity was assessed with the Somatoform Dissociation Questionnaire-20 (SDQ-20), a 20-item measure of the extent to which FND symptoms (e.g., paralysis) were experienced over the past year on a five-point Likert scale (61). Non-core somatic symptom severity was assessed with the Patient Health Questionnaire-15 (PHQ-15), a 15-item measure of how bothersome physical symptoms (e.g., pain, fatigue) were over the past four weeks on a three-point Likert scale (62). Seven participants had incomplete psychometric data (6 FND, 1 PC).

MRI Acquisition and Preprocessing

3T MRI acquisition and preprocessing details are described in Supplementary Materials.

Gradient Analysis

Diffusion Map Embedding.

Functional connectivity gradients were computed using diffusion map embedding, a nonlinear dimensionality reduction technique that can capture the dominant dimensions of spatial variation in high-dimensional functional connectivity data through a set of low-dimensional manifolds (i.e., gradients) (63,64). To do this, resting state functional connectivity (rsFC) matrices were computed for each participant by calculating Pearson correlation coefficients between the time series of every pair of cortical grey matter voxels. Each participant’s matrix was thresholded to retain only the top 10% of connections (including negative connections), with all remaining connections set to zero. We then computed a non-negative, square symmetric affinity matrix for each rsFC matrix using cosine similarity to quantify the similarity between connectivity profiles for each pair of voxels. These affinity matrices were used as inputs to diffusion map embedding, which yielded ten gradients per participant, reflecting the dominant dimensions of rsFC spatial variation. To enable statistical comparisons, individual gradients were aligned to group-level template gradients derived from HCs (see Supplementary Materials for details). Based on a priori hypotheses and prior work, we focused on the three most dominant gradients (30,32,39,40).

Statistical Analysis.

General linear models (GLMs) were used to compute between-group differences in voxel-wise gradient values for the top three gradients across FND and HC/PC cohorts. Comparisons were conducted for FND-mixed vs. PCs and vs. HCs separately, followed by subtype analyses for FND-motor and FND-seiz vs. PCs/HCs. We also conducted exploratory comparisons of non-overlapping, isolated FND-motor (n=40) vs. FND-seiz (n=15) subgroups. All GLMs controlled for age, sex, selective serotonin/norepinephrine reuptake inhibitor (SSRI/SNRI) use (yes/no), and mean framewise displacement (FD). Post-hoc analyses also controlled for: (1) BDI-II, STAI-total, and PCL-5 scores; (2) CTQ-abuse and CTQ-neglect scores. Intersection maps were computed to examine voxels that held across all corrections. To account for multiple comparisons, cluster-wise correction was applied using a Monte Carlo simulation with 10,000 iterations to estimate the probability of false positive clusters at p<0.05.

To further contextualize our findings, we conducted post-hoc seed-to-voxel connectivity analyses using 6mm spherical regions-of-interest (ROIs) around the peak voxel of maximal gradient difference in each group-level gradient comparison (Supplementary Materials). This enabled characterization of low-dimensional gradient findings by clarifying the topography of rsFC differences with specific targets (65,66).

Symptom Correlations

Post-hoc Spearman correlations were computed between mean gradient values extracted from each significant cluster in FND-mixed vs. PCs comparisons and SDQ-20 and PHQ-15 scores. Correlations with the SDQ-20 (reflecting core FND symptoms) were computed for FND-mixed only, while correlations with the PHQ-15 (reflecting non-core somatic symptoms) were computed both within FND-mixed only and across FND-mixed and PCs. Multiple comparison correction using false-discovery rate correction was conducted across clusters for each gradient and symptom scale. For significant correlations, post-hoc partial correlation analyses were performed to test if findings held adjusting for FND subtype (i.e., FND-seiz (yes/no)).

RESULTS

Demographic and Psychometric Comparisons

There were no age or sex differences between FND-mixed, PCs, and HCs. Compared to HCs, FND-mixed had higher scores on all psychometric scales. Compared to PCs, FND-mixed had higher scores on the SDQ-20, PHQ-15, and CTQ-abuse, and did not differ on the BDI-II, STAI-total, PCL-5, CTQ-neglect, or SSRI/SNRI use (Table 1, Supplementary Table 4).

Between-Group Gradient Comparisons

Group-level gradient maps for FND-mixed, HCs, and PCs are shown in Fig. 1. Across all participants, the top three gradients collectively accounted for 46.6% of variance in rsFC organization (variance explained for each gradient shown in Supplementary Fig. 1). These gradients exhibited high spatial similarity to previously published gradients (30,31; Supplementary Table 3), with Gradient 1 (association-sensorimotor) distinguishing default mode and frontoparietal networks from exteroceptive sensory (e.g., somatosensory, visual) and salience/cingulo-opercular networks, Gradient 2 (visual-somatomotor) distinguishing visual from somatosensory/motor networks, and Gradient 3 (representation-modulation) distinguishing default mode and exteroceptive sensory networks from frontoparietal and salience/cingulo-opercular networks.

Figure 1.

Resting-state functional connectivity gradients of the cerebral cortex for patients with FND (n=64; top), healthy controls (HCs; n=61; middle), and psychiatric controls (PCs; n=62; bottom). Group-averaged patterns are shown for the top three gradients explaining dominant axes along which whole-cortex connectivity is organized (left, Gradient 1; middle, Gradient 2; right, Gradient 3). Voxels with similar connectivity patterns are represented by similar colors along the gradient, while voxels at opposite extremes reflect the greatest dissimilarity in connectivity. The sign of each gradient is arbitrary.

Gradient 1 (Association-Sensorimotor).

Compared to HCs, patients in the FND-mixed cohort exhibited increased Gradient 1 values in bilateral precentral and postcentral gyri, slightly extending into the superior and inferior parietal lobules (Fig. 2, top). These regions anchor the sensorimotor end of the gradient, such that increases in gradient values for FND-mixed reflect a shift in similarity of rsFC profiles towards the association end. These findings held correcting post-hoc for BDI-II, STAI-total, and PCL-5 scores (Supplementary Fig. 2), however, when correcting for CTQ-abuse and CTQ-neglect subscales, only left hemisphere findings remained significant. The FND-mixed group also showed decreased Gradient 1 values relative to HCs in bilateral subgenual anterior cingulate and ventromedial prefrontal cortices, and the right rostral middle temporal gyrus — regions that anchor the association end. In this context, decreased values in FND-mixed indicate a shift in similarity of rsFC profiles towards the sensorimotor end of the gradient. These decreases did not remain significant following post-hoc corrections.

Figure 2.

Comparisons of Gradient 1 (association-sensorimotor) for patients with FND-mixed vs. healthy controls (HCs) (top) and FND-mixed vs. psychiatric controls (PCs) (bottom). (A) Results from between-group statistical comparisons. Colors reflect the z-statistic computed from a two-sample general linear model for the primary adjustment for age, sex, SSRI/SNRI use, and mean framewise displacement (z-statistic>1.96; p<0.05 cluster-corrected for multiple comparisons). White outlines reflect regions that also held across all post-hoc corrections for: i) BDI-II, STAI-total and PCL-5 scores; and ii) CTQ-abuse and CTQ-neglect scores. Statistical maps for post-hoc comparisons are visualized in Supplementary Figures 2 and 3. (B) Scatter plots of gradient values. Each dot represents a cortical voxel. Statistically significant voxels are colored (increases/decreases shown in red/blue, respectively), while non-significant (n.s.) voxels are in gray. (C) Density plots of gradient values extracted from statistically significant positive and negative clusters in the primary analysis for FND-mixed (dark red/blue) and respective control groups (light red/blue). BDI-II, Beck Depression Inventory-II; STAI-total, State Trait Anxiety Inventory-Total; PCL-5, PTSD Checklist-5; CTQ, Childhood Trauma Questionnaire.

Compared to PCs, FND-mixed exhibited increased Gradient 1 values in bilateral pre- and postcentral gyri, as well as the right dorsal mid-insula, reflecting a shift from sensorimotor towards the association end of the gradient (Fig. 2, bottom). These findings held when correcting post-hoc for (i) BDI-II, STAI-total, and PCL-5 scores, and (ii) CTQ-abuse and CTQ-neglect subscale scores (Supplementary Fig. 3). No regions exhibited decreased Gradient 1 values.

To further contextualize these findings, we computed seed-to-voxel FC using a ROI centered on the peal voxel of maximal gradient difference for each comparison. Seeds were in regions showing shifts towards the association end of Gradient 1 (vs. HCs: left pre-central gyrus seed; vs. PCs: right dorsal mid-insula/opercular cortex seed) and revealed increased connectivity with association regions in FND relative to both control groups, as well as reduced connectivity with sensorimotor regions in FND relative to HCs (Supplementary Fig. 4).

Gradient 2 (Visual-Somatomotor).

Patients with FND-mixed, compared to HCs, exhibited increased Gradient 2 values in the right middle temporal gyrus and lateral occipital cortex (Fig. 3, top). These differences remained significant across post-hoc corrections (Supplementary Fig. 2). These observed increases suggest a shift in similarity of rsFC profiles towards, and expansion of, the visual end of the gradient. Conversely, decreases in Gradient 2 values for FND-mixed vs. HCs were observed in the right lateral prefrontal and posterior parietal cortices, reflecting a shift in similarity of rsFC profiles from the middle of the gradient towards the somatomotor end. These decreases largely held when adjusting post-hoc for CTQ-abuse and CTQ-neglect but did not hold when adjusting for BDI-II, STAI-total, and PCL-5 scores.

Figure 3.

Comparisons of Gradient 2 (visual-somatomotor) for patients with FND-mixed vs. healthy controls (HCs) (top) and FND-mixed vs. psychiatric controls (PCs) (bottom). (A) Results from between-group statistical comparisons. Colors reflect the z-statistic computed from a two-sample general linear model for the primary adjustment for age, sex, SSRI/SNRI use, and mean framewise displacement (z-statistic>1.96; p<0.05 cluster-corrected for multiple comparisons). White outlines reflect regions that also held across all post-hoc corrections for: i) BDI-II, STAI-total and PCL-5 scores; and ii) CTQ-abuse and CTQ-neglect scores. Statistical maps for post-hoc comparisons are visualized in Supplementary Figures 2 and 3. (B) Scatter plots of gradient values. Each dot represents a cortical voxel. Statistically significant voxels are colored (increases/decreases shown in red/blue, respectively), while non-significant (n.s.) voxels are in gray. (C) Density plots of gradient values extracted from statistically significant positive and negative clusters in the primary analysis for FND-mixed (dark red/blue) and respective control groups (light red/blue). BDI-II, Beck Depression Inventory-II; STAI-total, State Trait Anxiety Inventory-Total; PCL-5, PTSD Checklist-5; CTQ, Childhood Trauma Questionnaire.

Compared to PCs, individuals with FND-mixed showed increased Gradient 2 values in bilateral precentral gyrus/supplementary motor area (SMA), as well as the left mid insula, middle and superior temporal gyri, temporal pole, and parietal operculum (Fig. 3, bottom). These regions reflect a shift in similarity of rsFC profiles towards the visual end of the gradient, with findings largely remaining significant across post-hoc corrections (Supplementary Fig. 3). Decreases were observed in the bilateral lateral prefrontal and occipital cortices and the left precuneus, which remained significant across post-hoc corrections (decreases in the left prefrontal did not hold when adjusting for CTQ-abuse and neglect). These decreases reflect a shift in similarity of rsFC profiles toward the somatomotor end of the gradient.

Seed-to-voxel analyses confirmed that a right lateral occipital/middle temporal gyrus seed exhibiting gradient shifts towards the visual end of Gradient 2 had increased connectivity with visual regions, and reduced connectivity with somatomotor regions, in FND vs. HCs, whereas a right lateral prefrontal seed that shifted towards the somatomotor end showed greater connectivity with somatomotor regions, and reduced connectivity with visual regions, in FND vs. PCs (Supplementary Fig. 4).

Gradient 3 (Representation-Modulation).

Compared to HCs, patients with FND-mixed exhibited increased Gradient 3 values in bilateral precentral gyri, lateral occipital and superior parietal cortices, and the right post-central, supramarginal, and middle frontal gyri (Fig. 4, top). Only the right precentral and middle frontal gyri findings remained significant following both post-hoc corrections (Supplementary Fig. 2). These increases reflect a shift in similarity of rsFC profiles from the modulation end of the gradient towards the representation end. Decreased Gradient 3 values were observed in the bilateral anterior cingulate, superior frontal, and occipital cortices, as well as the right posterior insula and opercular cortex, reflecting a shift in similarity of rsFC profiles from the representation end towards modulation. Decreases in the bilateral anterior cingulate and superior frontal cortices remained significant when correcting for CTQ-abuse and neglect, but no other findings held across post-hoc corrections.

Figure 4.

Comparisons of Gradient 3 (representation-modulation) for patients with FND-mixed vs. healthy controls (HCs) (top) and FND-mixed vs. psychiatric controls (PCs) (bottom). (A) Results from between-group statistical comparisons. Colors reflect the z-statistic computed from a two-sample general linear model for the primary adjustment for age, sex, SSRI/SNRI use, and mean framewise displacement (z-statistic>1.96; p<0.05 cluster-corrected for multiple comparisons). White outlines reflect regions that also held across all post-hoc corrections for: i) BDI-II, STAI-total and PCL-5 scores; and ii) CTQ-abuse and CTQ-neglect scores. Statistical maps for post-hoc comparisons are visualized in Supplementary Figures 2 and 3. (B) Scatter plots of gradient values. Each dot represents a cortical voxel. Statistically significant voxels are colored (increases/decreases shown in red/blue, respectively), while non-significant (n.s.) voxels are in gray. (C) Density plots of gradient values extracted from statistically significant positive and negative clusters in the primary analysis for FND-mixed (dark red/blue) and respective control groups (light red/blue). BDI-II, Beck Depression Inventory-II; STAI-total, State Trait Anxiety Inventory-Total; PCL-5, PTSD Checklist-5; CTQ, Childhood Trauma Questionnaire.

Compared to PCs, increased Gradient 3 values for patients with FND-mixed were observed in bilateral lateral occipital and superior parietal cortices and supramarginal gyri, as well as the left middle and inferior frontal gyri, and inferior temporal gyrus (Fig. 4, bottom). These increases reflect a shift in similarity of rsFC profiles from the modulation end of the gradient towards representation. Most findings remained significant across post-hoc corrections (left frontal findings did not remain significant when correcting for CTQ-abuse and neglect; Supplementary Fig. 3). Decreased values were observed across bilateral cingulate cortex, as well as bilateral precuneus, superior frontal, and occipital cortices, all of which remained significant across post-hoc corrections. These decreases reflect a shift in similarity of rsFC profiles from the representation end of the gradient towards modulation.

Seed-to-voxel analyses showed that seeds shifted towards the representation end of Gradient 3 (vs. HCs: right precentral gyrus seed; vs. PCs: left inferior temporal gyrus seed) had increased connectivity with representational regions, and decreased connectivity with modulatory regions, in FND relative to both HCs and PCs (Supplementary Fig. 4).

FND Subtype Considerations

We next examined between-group comparisons for FND-motor and FND-seiz subtypes vs. HCs and vs. PCs (Fig. 5). For Gradient 1, differences observed for the FND-motor cohort were largely the same as those reported for the full FND-mixed cohort. Findings for FND-seiz were less robust, with increases in the right postcentral gyrus vs. HCs that overlapped with those observed for FND-motor, and no differences observed vs. PCs. For Gradient 2, a similar pattern emerged where the FND-motor findings were largely the same as the pattern reported for FND-mixed. A comparison of FND-seiz vs. HCs revealed no significant differences, and FND-seiz vs. PCs exhibited decreases in the right lateral frontal cortex that overlapped with FND-motor findings. Comparisons for Gradient 3 again revealed a pattern for FND-motor that was largely the same as the pattern reported for the full FND-mixed cohort. Increases in the right postcentral gyrus were not observed in FND-motor vs. HCs, but were observed in FND-seiz vs. HCs. FND-motor and FND-seiz cohorts revealed overlapping decreases in the bilateral anterior cingulate and superior frontal cortices (vs. HCs and vs. PCs), as well as overlapping increases in the left lateral occipital and superior parietal cortices and decreases in the occipital cortex. Direct exploratory comparisons between isolated FND-motor vs. FND-seiz subgroups did not reveal significant differences across the three gradients.

Figure 5.

Comparison of findings for functional motor disorder (FND-motor) and functional seizure (FND-seiz) subtypes compared to controls. Overlays depict voxels that held across all two-sample general-linear models (z-stat>1.96; p<0.05 cluster-corrected for multiple comparisons for Gradient 1 (left), Gradient 2 (middle), and Gradient 3 (right) analyses compared to healthy controls (HCs; top) and psychiatric controls (PCs; bottom). Results for the FND-motor analyses are shown in blue, FND-seiz analyses shown in yellow, and overlapping regions across both subtype analyses vs. controls shown in green.

Symptom Severity Correlations

We observed several significant relationships between self-reported symptoms and gradient values extracted from statistically significant clusters in the FND-mixed vs. PCs analyses. To limit investigations to the most robust findings, only clusters that remained significant for all post-hoc corrections were tested for potential correlations with symptom severity scores. When looking across individuals with FND-mixed and PCs, Gradient 1 values extracted from a cluster in the left pre- and post-central gyri positively correlated with PHQ-15 scores (r_s(115)=0.25, p_corrected=0.01; Fig. 6a). Across both FND-mixed and PCs, Gradient 3 values from a cluster comprised of bilateral anterior cingulate and superior frontal cortex negatively correlated with PHQ-15 scores (r_s(114)=−0.28, p_corrected=0.02; Fig. 6c). Within FND-mixed alone, Gradient 2 values from a cluster comprised of left mid insula and superior temporal gyri/temporal pole positively correlated with SDQ-20 scores (r_s(54)=0.34, p_corrected=0.04; Fig. 6b). All three correlations remained significant in post-hoc partial correlation analyses that additionally controlled for FND subtype (FND-seiz, yes/no).

Figure 6.

Post-hoc correlations between self-reported symptoms and mean gradient values extracted from significant clusters in FND-mixed vs. psychiatric control (PCs) between-group comparisons. (A) The significant positive Spearman correlation across FND-mixed and PCs between mean Gradient 1 values extracted from the cluster shown in red and scores on the Patient Health Questionnaire-15 (PHQ-15). (B) The significant positive Spearman correlation within FND-mixed participants between mean Gradient 2 values extracted from the cluster shown in red and scores on the Somatoform Dissociation Questionnaire-20 (SDQ-20). (C) The significant negative Spearman correlation across FND-mixed and PCs between mean Gradient 3 values extracted from the cluster shown in blue and scores on the PHQ-15. Shaded regions reflect 95% confidence intervals. Outliers were removed prior to computing the correlation coefficients if they had a mean value that exceeded more than 1.5 times the interquartile range above the upper quartile or below the lower quartile. Colored regions on brain maps indicate clusters with significant increases (red) or decreases (blue) in FND-mixed vs. PCs comparisons. For scatterplots, FND-mixed participants are depicted with light green circles and PCs with dark green triangles.

DISCUSSION

Using a gradient-based approach to investigate macroscale functional brain organization, we observed alterations across all three gradients in FND-mixed relative to controls, several of which were associated with individual differences in symptom severity. Notably, Gradient 1 showed connectivity shifts in sensorimotor regions towards more association-like profiles, Gradient 2 revealed altered visual-somatomotor differentiation, and Gradient 3 exhibited reduced separation in rsFC profiles between representational and modulatory regions. We also observed overlapping and distinct alterations in FND-motor and FND-seiz subtypes vs. controls. These findings provide convergent evidence for disrupted hierarchical functional brain organization in FND.

Gradient 1, the association-somatomotor gradient, spans from higher-order association regions (e.g., default mode network) to ‘unimodal’ sensorimotor regions. In FND-mixed, increased Gradient 1 values were observed in the pre- and postcentral gyri, indicating that these sensorimotor regions exhibited more association-like connectivity profiles. These increases were present compared to both HCs and PCs, and largely remained significant after adjustments for psychiatric symptoms and trauma burden, underscoring their specificity to FND. These findings are consistent with recent evidence from our group of increased somatomotor integration (i.e., between-network connectivity) in FND-mixed (27), as well as enhanced link-step connectivity from primary motor cortex to the posterior insula and mid cingulate cortex (23). The current gradient-based approach extends prior findings by revealing not just which connections are altered, but how their functional roles are reorganized within the broader connectivity landscape, offering a macroscale perspective of disrupted organization in FND. Additionally, Gradient 1 values in the left pre- and postcentral gyri were correlated with self-reported non-core somatic symptoms across both patients with FND-mixed and PCs. While these symptoms may reflect comorbid features rather than core FND mechanisms, they are clinically important given that non-motor, somatic symptoms are strongly associated with reductions in health-related quality of life in this population (67), and have been reported to precede onset of core FND symptoms by more than six years (68).

Gradient 2, the visual-somatomotor gradient, is thought to reflect the organization of sensory systems. In the FND-mixed cohort, decreased Gradient 2 values were observed in lateral prefrontal cortices relative to both HCs and PCs, suggesting that attentional regions, positioned towards the middle of the gradient, are shifting towards more somatomotor-like connectivity profiles. The FND-mixed cohort also exhibited increased Gradient 2 values in somatomotor regions compared to PCs, indicating a shift towards more visual-like connectivity that may reflect aberrant processing of sensory signals. Of note, Gradient 2 values in the left mid-insula were positively correlated with core FND symptom severity within FND-mixed. These findings align with the growing literature of a fundamental role for altered sensory processing in FND (69–71).

Gradient 3, the representation-modulation gradient, differentiates regions involved in representing sensory signals (e.g., default mode and sensory networks) from those involved in modulating these signals (e.g., salience/cingulo-opercular and frontoparietal networks). In the FND-mixed cohort compared to both control groups, increased Gradient 3 values were observed in modulatory regions and decreased values in representation regions, suggesting less functional distinction between these processes. Notably, the decrease in Gradient 3 values within the anterior cingulate cortex (perigenual and dorsal regions) was robust relative to PCs and remained significant after adjusting for affective symptoms and trauma burden, indicating specificity to FND beyond shared pathology and risk factors. Moreover, these values were negatively associated with non-core somatic symptom severity across FND-mixed and PCs, suggesting that Gradient 3 alterations may have transdiagnostic relevance for somatic symptom burden.

Analyses of FND subtypes vs. controls revealed shared and distinct alterations. The FND-motor cohort showed patterns largely consistent with those observed in the full FND-mixed cohort, while findings for the FND-seiz group were less robust. For Gradients 1 and 2, there was limited overlap between FND-motor and FND-seiz, with both groups exhibiting Gradient 1 increases in the right postcentral gyrus (vs. HCs) and Gradient 2 decreases in the right lateral frontal pole (vs. PCs). Gradient 3 exhibited the most overlap between subtypes, with convergent decreases in anterior cingulate and occipital cortices (vs. HCs and PCs). These results highlight neurobiological heterogeneity across FND subtypes, alongside shared disruptions in cortical organization. Exploratory comparisons between isolated FND-motor vs. FND-seiz subgroups revealed no significant differences, although analyses were under-powered. This null result suggests that the alterations observed in the FND-mixed cohort are broadly shared across subtypes, while leaving open the possibility that subtle subtype-specific differences may emerge in larger, more balanced samples.

Some alterations also emerged in comparisons of FND-mixed vs. HCs that were not seen vs. PCs. These findings may reflect features that are not specific to FND but instead are shared across psychiatric conditions. For example, decreased Gradient 1 values in FND-mixed vs. HCs in ventromedial prefrontal and subgenual anterior cingulate cortices did not survive post-hoc corrections, and were not present in comparisons against PCs, suggesting that while association-level disruptions may occur, their relation to FND-specific pathophysiology is less clear. Similarly, increases in visual regions for Gradient 2, reflecting a potential expansion of the gradient, appeared only vs. HCs. Gradient 3 increases, while spanning modulatory regions across both control group comparisons, also showed a unique pattern relative to HCs. These findings may reflect more general vulnerabilities or comorbidities, and thus their role in FND’s pathophysiology remains less clear.

Although our findings are derived from fMRI-based rsFC and thus are inherently correlational, we offer an interpretation from a predictive processing, a useful framework for understanding FND (69,72). Functional connectivity gradients have previously been interpreted through a predictive processing framework, which posits that the brain actively constructs predictions about incoming sensory data based on past experiences (30,40). We use this an interpretive rather than mechanistic framework, as our analyses do not directly test predictive processing. Gradient 1 has been proposed to capture a continuum from regions involved in representing abstract, multimodal prediction signals (e.g., default mode network) to regions involved in signaling prediction errors (e.g., sensory and motor networks, salience/cingulo-opercular networks). Accordingly, our findings suggest that regions involved in signaling prediction errors may instead exhibit connectivity profiles resembling prediction-related regions, consistent with accounts of overly strong priors in FND that are not properly updated (73–75). Gradient 2, while requiring more research to aid interpretability, may reflect segregation of exteroceptive sensory systems; alterations here could indicate abnormal attentional modulation of sensory signals, biasing the updating of predictions. Gradient 3 may distinguish regions representing prediction and prediction error signals (e.g., default mode and sensory networks) from those setting their precision (e.g., salience/cingulo-opercular and frontoparietal networks). Alterations along this dimension may therefore reflect atypical predictive and precision-setting processes. Overall, we speculate that our gradient-based findings may reflect disruptions in the hierarchical brain architecture proposed to support the generation, weighting, and updating of predictions.

This study has limitations. While we interpret many of our findings as likely disorder-related, due to their persistence after adjusting for covariates and their presence in comparisons with both control groups, they may also reflect compensatory processes. Additionally, although PCs were matched on key mental health dimensions, diagnostic heterogeneity or unmeasured clinical variables may have influenced results. While all analyses accounted for SSRI/SNRI use, future efforts should consider the influence of other psychotropic medications. Incorporating clinician-rated and objective measures of severity will also strengthen future work (76). Smaller sample sizes in subtype cohorts, particularly FND-seiz, may have limited power to detect subtle effects. In addition, reconciling the shifts in rsFC profiles observed here with structural alterations previously reported in FND cohorts (60,77,78) represents an important next step. Moreover, our analysis focused exclusively on cortical gradients due to methodological limitations of including subcortical structures; future work should explore functional gradients within subcortical areas to provide a more complete picture of macroscale organization in FND. Finally, our analytic choices reflect one of several possible approaches, and future work should examine how alternative methodological decisions (e.g., Procrustes rotation or joint embedding, sparsity of input matrices, preprocessing parameters) may influence the variance explained and overall findings (64).

In conclusion, this study provides evidence that FND is associated with alterations in macroscale cortical organization. We identified shifts across the top three rsFC gradients, reflecting disruptions in hierarchical brain architecture. The inclusion of PCs underscores that these alterations are not solely attributable to concurrent mental health symptoms, supporting their FND specificity. Moreover, associations between gradient values in key regions with FND symptoms and more general somatic symptom severity highlight the mechanistic relevance of altered functional organization. Future work should aim to replicate and extend these findings in larger, clinically diverse samples.

Supplementary Material

Supplement Description:

Supplement Methods, Figures S1-S4, Tables S1-S4

DATA AVAILABILITY

For qualified researchers, analysis code and de-identified data pertaining to study results can be made available following local IRB approval upon reasonable request.