5.0 T MRI reveals characteristic alterations in visual and somatomotor networks associated with migraine chronification: a single-center prospective cohort study

Abstract

Background and objectives

Migraine is increasingly acknowledged as a disorder of large-scale brain network hierarchy rather than a merely focal dysfunction. However, the neural substrates driving the transition from episodic migraine (EM) to chronic migraine (CM)-a key pathological process in migraine chronification, remain poorly delineated.

Methods

Using 5.0 T MRI, we examined 122 participants (30 healthy controls [HC], 66 EM, and 16 CM). Morphometric INverse Divergence (MIND)-based structural similarity networks and resting-state functional connectivity (FC) matrices were constructed from 3D T1 and rs-fMRI data parcellated with the Schaefer-400 atlas. Diffusion map embedding was applied to derive low-dimensional gradients indexing macroscale cortical hierarchy. Group differences in the principal gradient (G1) were assessed using one-way ANOVA with FDR-corrected post hoc tests. Partial Spearman correlations were used to link regional G1 values with attack frequency, disease duration, and anxiety (HAMA), depression (HAMD), and sleep quality (PSQI) scores.

Results

Both MIND- and FC-derived gradients revealed a progressive flattening of the G1 along the HC-EM-CM continuum. These alterations were most prominent in the visual and somatomotor networks. Across the top overlapping regions, lower G1 values, reflecting diminished hierarchical segregation, were consistently associated with higher HAMA and HAMD scores and poorer sleep quality on the PSQI after FDR correction, particularly within visual–somatomotor network.

Conclusions

5.0 T MRI demonstrated a consistent, cross-modal disruption of the cortical hierarchy linked to migraine chronification, emphasizing the reduced hierarchical differentiation of visual-sensorimotor integration and its coupling with affective and sleep dysfunction. Gradient-based metrics may provide promising imaging markers for tracking disease progression.

Introduction

Migraine is one of the leading causes of disability worldwide [1–4] and clinically, it exhibits a strong tendency toward chronification, with approximately 3.1% of episodic migraine (EM) patients progressing to chronic migraine (CM) each yea [5]. However, the neural mechanisms underlying this transition remain largely unclear.

Prior MRI studies have consistently reported abnormalities within the visual and somatomotor systems-regions central to sensory processing and pain modulation [6]. Yet most functional MRI analyses rely on region-wise or graph-based metrics [7–10], which cannot capture the brain’s continuous macroscale organization. These network-level disruptions have been linked to sensory hypersensitivity and impaired pain modulation [11], supporting the view that migraine is a systems-level disorder rather than a focal pathology [12]. Structural MRI investigations have also consistently documented morphometric abnormalities in migraine, including reduced cortical thickness and gray matter volume in the visual [13], somatosensory, and prefrontal cortices [14]. However, conventional graph-theoretical or region-of-interest approaches remain limited in their ability to detect subtle, spatially distributed hierarchical alterations [15].

Recently, the connectome gradient framework has emerged as a powerful method to quantify this large-scale cortical hierarchy by embedding high-dimensional connectivity patterns into low-dimensional manifolds [16]. Cortical gradients provide a low-dimensional representation of the brain’s macroscale hierarchical organization, capturing smooth transitions from unimodal sensory and somatomotor regions to transmodal association cortices. Gradient 1 (G1) captures the principal axis of cortical hierarchy, extending from unimodal visual–somatomotor regions to transmodal association areas. Flattening of G1 reflects reduced differentiation along this sensory-to-association continuum and has been widely used as the primary index of macroscale cortical organization in gradient studies. Using this approach, Lee et al. [7] reported prominent functional network disruptions in EM, emphasizing hierarchical vulnerability. Nonetheless, this study was limited to 3 T data and did not examine structural–functional convergence or hierarchical changes across the full migraine spectrum.

Ultra-high-field 5.0 T MRI provides substantially increased signal-to-noise ratio, enhanced spatial resolution, and improved sensitivity to microstructural variation compared with 3 T. These advantages allow more precise estimation of both morphometric organization and resting-state connectivity. Additionally, the Morphometric INverse Divergence (MIND) framework [17] provides an information-theoretic measure of multivariate morphometric similarity, outperforming correlation-based methods in capturing fine-grained structural coupling.

Here, introducing 5.0 T multimodal MRI, we aimed to comprehensively characterize progressive cortical hierarchical alterations across the migraine spectrum by integrating MIND-based morphometric and functional connectivity (FC)-based connectome gradients. This multimodal gradient approach enables the identification of large-scale hierarchical reorganization bridging structural and functional domains, thereby offering a unified framework for elucidating migraine-related network dysfunction. We hypothesized that migraine chronification is accompanied by a progressive flattening of the principal cortical gradient, which reflects a breakdown of the normal differentiation between unimodal and transmodal systems. Furthermore, we explored whether this gradient disruption-particularly within the visual and somatomotor networks was associated with clinical parameters, including headache frequency, disease duration, and affective or sleep disturbances [18] (Fig. 1).

Fig. 1.

Overview of the study design and analysis pipeline. The research workflow consisted of four major stages. (1) Research population and data modalities: Healthy controls (HC), episodic migraine (EM), and chronic migraine (CM) participants underwent 5.0 T MRI scanning, including high-resolution 3D T1-weighted structural imaging and resting-state fMRI. (2) Data analysis: Morphometric features-gray matter volume (Vol), cortical thickness (CT), surface area (SA), sulcal depth (SD), and mean curvature (MC)-were extracted from FreeSurfer, while regional BOLD time series were derived from fMRIPrep preprocessing. (3) Connectivity matrix construction: Two complementary connectomes were generated based on the Schaefer-400 atlas: a morphometric structural similarity network using the Morphometric INverse Divergence (MIND) framework, and a functional connectivity network derived from pairwise Pearson correlations of regional BOLD signals. (4) Brain gradient analysis: Diffusion map embedding (BrainSpace toolbox) was applied to both connectivity matrices to compute low-dimensional cortical gradients. The first gradient (G1) captured the principal axis of cortical hierarchy from unimodal to transmodal regions, and group-level statistical analyses assessed between-group differences and correlations with clinical measures

Materials and methods

The study was conducted following approval from the local ethics committee of Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University (Ethics Approval No. 2025-0143-01), and adhered to the principles outlined in the Declaration of Helsinki. Written informed consent was obtained from all participants, who retained the right to withdraw from the study at any time.

Study participants

Data were obtained from the Primary Headache Database of Nanjing Drum Tower Hospital. A total of 30 HC, 66 patients with EM, and 16 patients with CM were initially recruited between February and November 2025. All migraine diagnoses conformed to the International Classification of Headache Disorders, 3rd edition (ICHD-3) criteria. Inclusion criteria required participants to be between 12 and 65 years of age, with no history of other neurological or systemic diseases, and capable of independently maintaining a headache diary. Exclusion criteria included secondary headache (e.g., tumor, vascular malformation, infection), psychiatric or behavioral disorders, substance abuse, or contraindications to MRI.

Clinical behavior test

Demographic and clinical information were collected for all participants at the time of MRI scanning. Variables included age, sex, body mass index (BMI), disease duration, monthly attack frequency, headache intensity, and neuropsychological assessments. Headache characteristics were obtained from structured clinical interviews and verified by headache diaries maintained for at least three consecutive months prior to imaging. Headache intensity was assessed using a 10-point visual analogue scale (VAS), and attack frequency was recorded as the number of migraine days per month. Psychological and behavioral comorbidities were evaluated using standardized instruments, including the Hamilton Anxiety Rating Scale (HAMA), Hamilton Depression Rating Scale (HAMD), and the Pittsburgh Sleep Quality Index (PSQI). These scales were administered by trained neurologists or neuropsychologists.

MRI acquisition and analysis

MRI acquisition

MRI data were acquired on a 5.0 T MRI system (uMR Jupiter; United Imaging, Shanghai, China) equipped with a 48-channel phased-array head coil optimized for neuroimaging. High-resolution structural images were obtained using a 3D T1-weighted MPRAGE sequence (TR = 7.5 ms, TE = 2.6 ms, TI = 950 ms, flip angle = 9°, FOV = 256 × 256 mm², matrix = 368 × 368, voxel size = 0.7 × 0.7 × 0.7 mm³). Resting-state functional images were collected using a gradient-echo EPI sequence (TR = 2000 ms, TE = 24 ms, flip angle = 56°, FOV = 216 × 216 mm², matrix = 108 × 108, voxel size = 2 × 2 × 2 mm³), with 240 volumes acquired over approximately 8 min. During scanning, participants were instructed to keep their eyes closed, remain awake and relaxed, minimize head motion, and avoid any deliberate cognitive or emotional activity. All participants were scanned interictally, reducing pain-related motion and attack-related confounds. For all included subjects, head motion was within acceptable thresholds (< 2 mm translation and < 2° rotation).

Preprocessing

Structural preprocessing was performed using FreeSurfer version 7.4.1, with cortical reconstruction and volumetric segmentation executed via the recon-all processing pipeline. The preprocessing steps included skull stripping, intensity non-uniformity correction, surface reconstruction, and the extraction of cortical morphometric metrics—cortical thickness (CT), surface area (SA), gray matter volume (Vol), mean curvature (MC), and sulcal depth (SD)—for each vertex across the cortical mantle.

Functional preprocessing was performed using fMRIPrep [19] version 25.1.3, which integrates tools from FSL, ANTs, and FreeSurfer within a standardized and reproducible BIDS-compliant workflow. The preprocessing pipeline included slice-timing correction, head motion correction, and susceptibility distortion correction using field maps. Functional images were co-registered to each subject’s T1-weighted structural image, then normalized to the MNI152NLin2009cAsym template, and finally spatially smoothed with a 6-mm full width at half maximum Gaussian kernel. To minimize non-neuronal sources of variance, nuisance regression was applied to remove six motion parameters, mean white matter and cerebrospinal fluid signals, and linear trends. In addition, temporal band-pass filtering (0.01–0.1 Hz) was used to retain low-frequency fluctuations relevant to resting-state connectivity. Volumes exhibiting framewise displacement greater than 0.5 were identified as motion outliers and censored from subsequent analyses.

Connectivity matrix construction

The morphometric similarity network was constructed using the MIND framework, which quantifies structural similarity between cortical regions based on multivariate morphometric features. MIND is an information-theoretic approach that computes the inverse divergence between the probability distributions of regional morphological features, thereby capturing the degree of structural resemblance across the cortex. For each subject, five standardized cortical features—CT, SA, Vol, MC, and SD—were extracted from FreeSurfer outputs and entered into the MIND algorithm. This procedure yielded a 400 × 400 morphometric similarity matrices for each subject, based on cortical parcellation defined by the Schaefer-400 atlas [20]. Compared with conventional correlation-based structural covariance methods, MIND provides a more comprehensive and robust characterization of inter-regional morphometric coupling by integrating multiple shape descriptors and minimizing bias from scale differences across features. This framework enables sensitive detection of hierarchical structural reorganization associated with disease progression. In parallel, the FC network was generated by extracting the mean blood oxygen level dependent (BOLD) time series from each Schaefer parcel and calculating pairwise Pearson correlation coefficients, which were subsequently Fisher z-transformed to improve normality, yielding a 400 × 400 FC matrix per subject. Both the structural and functional matrices were then symmetrized and normalized (zero mean, unit variance) prior to gradient decomposition to ensure comparability across modalities.

Gradient computation

Gradient analysis was performed using the BrainSpace toolbox [21]. For each subject, cosine-similarity matrices were derived separately from the MIND-based structural and FC-based functional connectomes. A diffusion map embedding algorithm was then applied to these similarity matrices to identify the principal gradients that capture the macroscale hierarchy of cortical organization. This non-linear dimensionality reduction method projects high-dimensional connectivity data onto a low-dimensional manifold, revealing continuous transitions in connectivity patterns across the cortex. The first gradient (G1) typically reflects the dominant axis of cortical organization, extending from unimodal sensorimotor and visual cortices to transmodal association regions such as the default-mode and frontoparietal control networks, thereby representing the principal hierarchy of cortical integration. Group-level template gradients were generated from the average connectivity matrices, and individual gradients were aligned to the template via Procrustes rotation to ensure cross-subject comparability. The top ten gradients were retained for completeness, although G1 was considered the primary measure of interest in subsequent analyses.

Statistical analysis

For each subject, gradient values were averaged within each cortical region defined by the Schaefer-400 atlas and entered into group-level statistical analyses. A one-way analysis of variance (ANOVA) was conducted to examine differences in G1 among the three groups (HC, EM, and CM). Multiple comparisons across cortical parcels were corrected using the false discovery rate (FDR, P < 0.05). Significant regions from the ANOVA were further examined using post hoc pairwise comparisons to identify the direction and magnitude of gradient alterations. To further assess the distributional changes of cortical hierarchy across groups, the Kolmogorov–Smirnov (KS) test was applied to compare the gradient value distributions. To assess clinical relevance, we performed Spearman’s rank correlations between G1 values in regions showing significant group effects and clinical variables, including monthly attack frequency, disease duration, and neuropsychological scores (HAMA, HAMD, PSQI). In addition, partial Spearman correlation analyses were conducted to account for collinearity among psychological measures, such that correlations involving each scale were evaluated while controlling for the other two (e.g., HAMA controlling for HAMD and PSQI). All correlation results were corrected for multiple comparisons using the Benjamini–Hochberg FDR procedure (q < 0.05). All statistical analyses were conducted using Python (SciPy, Statsmodels) and SPSS version 26.0.

Results

Patients

A total of 125 participants were initially recruited, including 31 HC, 67 patients with EM, and 17 patients with CM. After exclusion due to excessive head motion during scanning or incomplete clinical data, one participant from each group was removed, resulting in a final sample of 105 participants (30 HC, 66 EM, and 16 CM) for the present analysis (Fig. 2). Across all migraine patients, approximately 85% were female.

Fig. 2.

Flowchart of participant recruitment and inclusion. A total of 108 participants were initially recruited, including 31 healthy controls (HC), 67 patients with episodic migraine (EM), and 17 patients with chronic migraine (CM). One participant from each group was excluded due to either incidental findings (HC) or inability to tolerate MRI scanning (EM and CM). The final sample included 30 HC, 66 EM, and 16 CM participants who completed all MRI acquisitions and clinical assessments and were included in the statistical analysis

Behavioral data

In the EM group, patients reported a median migraine duration of 9.0 years (interquartile range [IQR] = 11.0) and an average monthly attack frequency of 2.75 days episodes (IQR = 4.3), while CM patients experienced headaches on average 20 days per month (IQR = 15.0) with a median disease duration of 8.0 years (IQR = 21.0). Across all migraine patients, the mean visual analogue scale (VAS) pain intensity was approximately 7.0, and both anxiety (HAMA) and depression (HAMD) scores were significantly elevated compared with controls (P < 0.05, respectively). Sleep quality, as measured by the PSQI, progressively deteriorated from HC to EM and further to CM, reflecting a worsening trajectory along the course of migraine chronification (Table 1).

Table 1.

Demographic and clinical characteristics of healthy controls (HC), episodic migraine (EM), and chronic migraine (CM) groups

	HC (n = 30)	EM (n = 66)	CM (n = 16)	P value
Age (years)	36.0 (15.0)	34.5 (15.0)	35.5 (41.0)	0.835
Sex (F, female; M, male)	21 F, 9 M	57 F, 9 M	12 F, 4 M	0.153
BMI, kg/m^2	22.6 (4.9)	23.0 (5.5)	22.1 (4.5)	0.863
Disease duration (years)	-	9.0 (11.0)	12.0 (16.0)	0.221
Monthly Attack Frequency (times)	-	2.75 (4.3)	20.0 (15.0)	<0.001
Duration of migraine attack (h)	-	18.0 (16.0)	24.0 (38.8)	0.404
Visual Analogue Scale	-	7.0 (1.0)	7.0 (1.0)	0.019
HAMA	2.0 (4.0)	9.0 (10.0)	10.0 (10.0)	<0.001
HAMD	2.5 (3.0)	7.5 (11.0)	9.0 (10.0)	<0.001
PSQI	4.0 (3.0)	8.0 (6.0)	11.5 (9.0)	<0.001

Values are presented as median (interquartile range, IQR). P values were obtained from Kruskal–Wallis tests for non-parametric data, and from chi-square tests for categorical variables. Significant group differences were observed in monthly attack frequency, Hamilton Anxiety Rating Scale (HAMA), Hamilton Depression Rating Scale (HAMD), and Pittsburgh Sleep Quality Index (PSQI) scores (P< 0.001), indicating progressively greater emotional and sleep disturbances from EM to CM. No significant differences were found in age, sex, BMI, disease duration and attack duration, among groups (P> 0.05)

Neuroimaging data

Averaged group brain gradient and distribution across groups

Group-averaged gradient maps were generated for both the MIND-derived structural similarity network and the FC-derived functional connectivity network across the three groups. All three groups exhibited the canonical macroscale cortical organization represented by the G1, which recapitulated the hierarchical transition from unimodal sensorimotor and visual cortices to transmodal association regions including the frontoparietal control and default-mode networks (Fig. 3). This overall pattern was preserved across both the MIND-derived structural and FC-derived functional modalities, indicating that the fundamental topography of cortical hierarchy remained broadly consistent among groups.

Fig. 3.

Group-averaged cortical gradients derived from structural and functional networks. Panels A-C show the group-averaged cortical gradients derived from the MIND-based structural similarity network for healthy controls (HC), episodic migraine (EM), and chronic migraine (CM), respectively. Panels E-G display corresponding gradients derived from the functional connectivity (FC) network for the same three groups. In both modalities, Gradient 1 (G1) represents the principal axis of cortical hierarchy extending from unimodal sensory-motor and visual regions (blue/purple) to transmodal frontoparietal and default-mode regions (yellow). Scatterplots in panels D (structural) and H (functional) illustrate the two-dimensional gradient space (G1 vs. G2) for each group, demonstrating a progressive flattening and compression of G1 from HC to EM to CM. These findings indicate a consistent topological organization across modalities, with the chronic migraine group exhibiting the most pronounced disruption of the normal cortical hierarchy. Color bars represent gradient loading values

Despite this global preservation, distinct alterations in the spatial configuration of G1 were observed across disease stages. Compared with HC, EM patients already showed a moderate flattening and contraction of G1 (P < 0.01, for both modalities, Fig. 4A and C), particularly in the occipital and central sensorimotor cortices, reflecting early disruption of hierarchical segregation. In contrast, CM patients demonstrated a marked and widespread flattening of G1 (P < 0.01, for both modalities, Fig. 4A and C), with reduced differentiation between unimodal and transmodal territories across both hemispheres, suggesting a profound breakdown of cortical hierarchy with migraine chronification (Fig. 3).

Fig. 4.

Group differences in Gradient 1 (G1) distribution and spatial localization across migraine stages. Panels A–B show results derived from the MIND-based structural similarity network, and panels C–D from the functional connectivity (FC) network. In A and C, kernel density plots illustrate the distribution of G1values for healthy controls (HC), episodic migraine (EM), and chronic migraine (CM). Both modalities demonstrated a progressive flattening of G1 distributions from HC to EM to CM, confirmed by Kolmogorov–Smirnov tests (P < 0.05). Panels B and D display cortical regions showing significant group differences in G1 values based on one-way ANOVA (P < 0.05, FDR-corrected). For the structural gradient (B), significant alterations were mainly localized to the visual and somatomotor cortices, extending into limbic regions. For the functional gradient (D), significant effects were observed in visual, somatomotor, limbic, and default-mode network (DMN) regions. Together, these results indicate a modality-consistent, progressive disruption of cortical hierarchy associated with migraine chronification

Group differences in gradient 1

Significant between-group differences in G1 were identified across multiple large-scale cortical systems (ANOVA, FDR, P < 0.05). Both the MIND-derived structural gradients and FC-derived functional gradients exhibited highly similar patterns, indicating that migraine chronification affects the brain’s macroscale hierarchy in a modality-independent manner.

In the MIND-based gradient, the most prominent alterations were localized to the visual and somatomotor networks, with additional involvement of frontoparietal control, limbic, and default-mode regions. Within the left hemisphere, significant G1 reductions were observed in several visual parcels (LH_Vis_15, LH_Vis_23–25, LH_Vis_29) and somatomotor areas (LH_SomMot_8–9, LH_SomMot_15, LH_SomMot_33), extending into control (LH_Cont_PFCl_13) and posterior default-mode regions (LH_Default_pCunPCC_10). In the right hemisphere, alterations encompassed visual (RH_Vis_21, 25 & 29), somatomotor (RH_SomMot_18, RH_SomMot_38), salience (RH_SalVentAttn_Med_7–8), and orbitofrontal limbic (RH_Limbic_OFC_3 & 6) regions. (Fig. 4B)

Similarly, the FC-based gradient demonstrated significant disruptions within visual, default-mode, limbic, dorsal-attention, and somatomotor networks. Left-hemisphere differences were concentrated in visual areas (LH_Vis_2, LH_Vis_7, LH_Vis_20, LH_Vis_26) and in default-mode regions (LH_Default_PFC_21, LH_Default_Par_3), whereas right-hemisphere effects appeared in visual (RH_Vis_2, RH_Vis_8, RH_Vis_17, RH_Vis_20), dorsal-attention (RH_DorsAttn_Post_6), temporal-pole limbic (RH_Limbic_TempPole_2), and somatomotor (RH_SomMot_2, 8 and 20) territories. (Fig. 4D)

Across both modalities, EM patients already displayed regional gradient flattening relative to HC, most prominently within the visual and somatomotor cortices, while CM patients showed further expansion of these abnormalities into limbic and frontoparietal association regions. The overlapping spatial patterns between the MIND and FC domains underscore a progressive, cross-modal degradation of cortical hierarchy from healthy to episodic and ultimately chronic migraine.

Correlation analysis

To explore the clinical relevance of the observed cortical gradient alterations, partial Spearman correlation analyses were performed to assess the relationship between clinical variables and the G1 values extracted from the top 12 cortical regions with the most prominent between-group differences common to both MIND- and FC-derived gradient maps. These regions were primarily distributed within the visual and somatomotor networks, with smaller clusters in limbic and default-mode areas.

After FDR correction (q < 0.05), both the MIND- and FC-derived gradients exhibited consistent negative associations with HAMA, HAMD, and PSQI scores (Fig. 5A–B). Lower G1 values-reflecting reduced hierarchical differentiation-were associated with higher anxiety, greater depressive symptoms, and poorer sleep quality. Notably, these relationships were most robust within visual and somatomotor regions, indicating that disruption of early sensory hierarchy is closely linked to affective and sleep disturbances in migraine.

Fig. 5.

Partial Spearman correlations between Gradient 1 and clinical variables across migraine stages. Panels A (MIND-based structural gradients) and B (functional connectivity gradients) show partial Spearman correlation matrices between Gradient 1 (G1) values extracted from cortical regions with the most prominent between-group differences and clinical variables, including BMI, attack duration, monthly attack frequency, headache severity, Hamilton Anxiety Scale (HAMA), Hamilton Depression Scale (HAMD), and Pittsburgh Sleep Quality Index (PSQI). Significant associations surviving FDR correction (q < 0.05) are indicated by asterisks. Across both modalities, lower G1 values were significantly associated with higher HAMA, HAMD, and PSQI scores, indicating that reduced hierarchical differentiation is linked to more severe anxiety, depressive symptoms, and poorer sleep quality. In contrast, positive correlations were observed between orbitofrontal G1 values and headache severity scores

Collectively, these findings demonstrate that migraine-related affective and sleep symptoms are tightly coupled to the degree of hierarchical flattening, particularly within visual–somatomotor circuits (Fig. 5).

Discussion

To contextualize our key findings, the present study employed 5.0 T MRI and a multimodal connectomic gradient approach to comprehensively characterize large-scale cortical reorganization during migraine chronification. By integrating MIND-derived structural gradients and FC-derived functional gradients-two complementary metrics capturing structural-functional network topology-we demonstrated a consistent and progressive flattening of the principal cortical hierarchy (G1) from healthy controls to episodic and chronic migraine patients. These results support the notion that migraine is not merely a focal or network-restricted disorder, but rather reflects a systemic breakdown of cortical hierarchical differentiation that evolves in parallel with disease chronification.

Progressive disruption of cortical hierarchy

All groups retained the core canonical hierarchical organization spanning unimodal sensorimotor and visual cortices to transmodal association networks. However, subtle gradient flattening was already evident in patients with episodic migraine, reflecting early-stage impairment in segregation between sensory and association systems. In chronic migraine, this flattening was exacerbated-became more extensive and spatially diffuse-with significant expansion of G1 alterations in the visual, somatomotor [22–25]. This progressive collapse of hierarchical organization is hypothesized to reflects reduced differentiation and increased cross-network coupling among cortical systems that normally maintain hierarchical specialization.

In the present study, the most prominent Gradient 1 alterations were localized to the visual and somatomotor cortices, which together anchor the lower end of the cortical hierarchy and have been repeatedly implicated in sensory hypersensitivity and pain processing in migraine. These results indicate that hierarchical disruption during migraine chronification is primarily driven by alterations within early sensory systems, consistent with prior work showing abnormal excitability, cortical thickness reduction, and altered connectivity in occipital and peri-rolandic regions [26]. Notably, such a loss of network segregation has also been documented in other hierarchical disorders (e.g., major depressive disorder, insomnia, and chronic pain), suggesting it may represent a shared neurobiological signature of maladaptive neuroplasticity in recurrent nociceptive and affective dysregulation conditions [27, 28].

Cross-modal convergence of structural-functional gradient alterations at 5.0 T

A pivotal contribution of the present study is the demonstration that structural and functional connectomic gradients exhibit spatially convergent alterations, reflecting cross-modal convergence of hierarchical disorganization in migraine. High-field 5.0 T MRI afforded enhanced signal-to-noise ratio (SNR) and spatial resolution, enabling reliable quantification of fine-grained cortical gradients and uncovering subtle topological alterations that remain undetectable at conventional field strengths [29]. The parallel gradient flattening across both structural and functional modalities suggests that microstructural remodeling associated with chronic migraine may underpin the observed functional disruptions, pointing to a bidirectional structural-functional interplay in migraine-related cortical reorganization.

Notably, the MIND framework-which integrates five complementary morphometric features- captures complex interregional covariance patterns that extend beyond the limitations of traditional single-feature correlations analyses. This multimodal morphometric integration thereby affords a more comprehensive characterization of anatomical reconfiguration accompanying migraine chronification, overcoming the constraints of unimodal structural metrics that often fail to capture the full complexity of disease-related neuroanatomical changes.

Correlation analysis of sensorimotor-limbic dysregulation and clinical symptom

The most prominent gradient alterations were observed in the visual and somatomotor networks, extending to the limbic and default-mode systems. The visual cortex anchors the lower end of the cortical hierarchy and serves a pivotal role in sensory prediction and integration. Consistent with this, previous structural MRI studies have consistently reported reduced cortical thickness and gray matter volume in the visual cortex of migraine patients-findings that are consistent with the current observation of cortical hierarchical flattening within the visual network [13, 14]. Moreover, sustained aberrant functional activity and hyperexcitability in this region have been widely reported in migraine and may contribute to visual aura and photophobia [30].

Similarly, somatomotor gradient disruption further indicates impaired integration between sensory processing and motor control pathways that are critical for pain perception and regulation [31, 32]. Although small clusters of G1 differences extended into limbic and DMN territories, these effects were modest and spatially limited compared with the widespread sensory-system alterations. Thus, while higher-order association networks may participate secondarily in migraine pathophysiology.

Moreover, the negative correlations between cortical gradient values and scores on HAMA, HAMD, and PSQI scores underscore that hierarchical degradation parallels emotional dysregulation and sleep disturbance-two key comorbidities of migraine chronification across the migraine spectrum. Cortical gradients serve as a reliable proxy for the integrity of hierarchical brain organization, with lower gradient values indicating reduced network segregation and increased functional cross-network coupling among sensory, limbic, and association systems. As migraine progresses from episodic to chronic, this hierarchical degradation becomes more pronounced, paralleling increases in anxiety, depression, and poor sleep quality.

Consistent with clinical observations, numerous studies have demonstrated that chronic migraine patients exhibit significantly higher scores on validated scales assessing anxiety, depression, and sleep dysfunction compared to episodic migraineurs, and these comorbidities intensify with increasing migraine severity and disability [33–35]. It is important to note that anxiety, depression, and sleep disturbances are often strongly interrelated in migraine cohorts. To disentangle independent versus shared contributions, we performed partial correlation analyses. The persistence of significant associations between G1 and psychological measures after adjusting for inter-scale collinearity suggests that gradient flattening reflects both a shared distress component and distinct affective–sleep contributions. This provides a more nuanced interpretation of how cortical hierarchical disruption relates to the clinical phenotype of migraine.

Limitations

Several limitations of the present study should be acknowledged. First, the sample size—particularly for the CM group—was relatively small, which may constrain statistical power to detect subtle effects or interactions. The small CM sample, cross-sectional design, and limited covariate control restrict causal interpretation and render the findings preliminary. Second, subcortical regions were not included, which restricts interpretation of cross-level (cortico–subcortical) hierarchical dynamics. Third, while the Schaefer-400 parcellation offers fine spatial granularity, future studies could explore whether individualized or multimodal brain parcellations strategies yield more reliable gradient alignment and enhance the biological interpretability of findings. Fourth, given the cross-sectional design, the observed gradient flattening reflects an associative pattern rather than a causal mechanism, and the study cannot determine the directionality between hierarchical disruption, migraine chronification, and comorbid symptoms. Thus, longitudinal and interventional studies will be essential to clarify whether normalization of cortical hierarchical organization correlates with symptom remission or therapeutic response.

Conclusion

Using ultra–high-field 5.0 T MRI and multimodal connectome gradients, this study identified a progressive flattening of cortical hierarchy across the migraine spectrum, with the most pronounced disruptions observed in visual and somatomotor systems. These findings highlight the potential relevance of large-scale hierarchical organization to migraine burden; however, they should be interpreted cautiously. Given the modest CM sample size and cross-sectional design, the results reflect associative patterns rather than causal mechanisms, and the directionality between hierarchical disruption, migraine chronification, and comorbid symptoms cannot be inferred. Larger, longitudinal, and fully covariate-adjusted studies are needed to validate these observations, clarify their mechanistic significance, and determine the clinical utility of gradient-based markers in migraine.

Author contributions

JML, HYH, and HYN made substantial contributions to study protocol development, data interpretation, and manuscript drafting. HY, MJS, HYD, YHW, YX, XZ, and QXZ contributed to data interpretation and manuscript drafting. DDK, HYH, HY, MJS, HYD, and YHW were responsible for participant enrollment and data acquisition. JML, HYH, and HYN performed data processing, analysis, and statistical evaluation.

Funding

This work was supported by the Clinical Trials from the Affiliated Drum Tower Hospital, Medical school of Nanjing University (2024-LCYJ-MS-22), The Jiangsu Province “333 High-Level Talent Cultivation Project”, the National Natural Science Foundation of China (82371327) and Jiangsu Provincial Health Commission General Science Program (M2024057).

Data availability

The informed consent obtained from participants did not include permission for public sharing of individual-level raw data. Therefore, these data are not publicly available. De-identified data may be made available to qualified researchers upon reasonable request to the corresponding author and completion of the required data use and confidentiality agreements.

Declarations

Ethics approval and consent to participate

Approval was obtained from Medical Ethics Committee of Nanjing Drum Tower Hospital (Approval No. 2025-0143-01).

Consent for publication

All authors authorize the publication.

Competing interests

The authors declare no competing interests.