Subcortical nuclei and glymphatic system alterations in migraine with aura: a comparative 7T multimodal MRI analysis Authors: Xinyu Wang, Yongqin Xiong, Xiu Liu, Xiaopeng Zong, Caohui Duan, Yuhan Wu, Xiangbing Bian, Song Wang, Jianxin Hu, Luhua Zhang, Haoxuan Lu, Bingbing Zhao, Jiayu Huang, Yan Li, Zhixuan Li, Zhao Dong, Xin Lou Published: 2025 ABSTRACT BackgroundAlthough cortical spreading depression is well-established in migraine with aura (MA), the potential involvement of subcortical nuclei and the glymphatic system (GS) in its pathogenesis—and their utility in distinguishing MA from migraine without aura (MO)—remains poorly investigated. Therefore, we employed multimodal 7T MRI—integrating functional, structural, quantitative susceptibility mapping (QSM), and GS metrics (perivascular space (PVS) morphology, choroid plexus volume, and analysis along the perivascular space (ALPS) index)—to enhance the understanding of MA and to characterize alterations specific to MA relative to MO and healthy controls (HCs).MethodsIn this cross-sectional study, 7T MRI scans were acquired from 35 MA patients, 45 MO patients, and 39 HCs. Group differences were evaluated using general linear models adjusted for age, anxiety, and depression scores (and intracranial volume when evaluating volumetric measures), with false discovery rate (FDR) correction. Partial correlation analyses were performed between abnormal imaging metrics and clinical parameters.ResultsBoth MA and MO showed lower regional homogeneity (ReHo) in the hippocampus(L), thalamus(R), and pallidum(R) compared to HCs (PFDR<0.05). In the volumetric comparisons, MO patients tended to have smaller volumes (especially in the amygdala) than HCs (PFDR<0.05) whereas MA patients showed relatively preserved volumes in these regions. QSM revealed higher iron content in the cortical nuclei of the amygdala in MA compared to MO, while MO had lower iron content relative to HCs (PFDR<0.05). Glymphatic dysfunction was prominent in MA, demonstrated by a lower right ALPS index compared to HCs and enlarged PVS compared to MO. Negative correlations were observed between QSM values of Corticoamygdaloid-transition(R) and headache intensity (r = -0.637, P < 0.001). Although GS metrics did not correlate with clinical parameters, higher amygdalar QSM related to shorter PVS length (cortical nucleus(R): r = − 0.606, P = 0.010), indicating iron–glymphatic coupling.ConclusionsOur findings demonstrate that MA is distinctly characterized by glymphatic dysfunction and aberrant iron deposition in the amygdala, suggesting a novel iron-glymphatic interplay in its pathogenesis. 7T MRI provides compelling evidence that MA and MO are distinct neurobiological entities and establishes robust imaging biomarkers for their differentiation, paving the way for future subtype-specific therapeutics.Supplementary InformationThe online version contains supplementary material available at 10.1186/s10194-025-02258-y. Introduction Migraine with aura (MA) is a primary headache disorder characterized by reversible focal neurological symptoms that precede the headache phase. These aura symptoms typically include visual scintillations or photopsia, visual field deficits, and sensory disturbances such as limb paresthesia or numbness [1]. As a highly prevalent neurological condition, migraine affects approximately 14–17% of the global population, with about one-third of patients experiencing aura [2–4]. Despite its prevalence, the pathophysiology of MA remains incompletely understood. It is well-established that cortical spreading depression (CSD), a propagating wave of cortical depolarization, underlies the aura phase of MA [5]. However, purely cortical and vascular phenomena may not fully account for the complexity of MA. Converging neuroimaging evidence suggests that deeper brain structures, particularly limbic-related regions, are critically involved in the initiation and modulation of migraine attacks. Functional MRI studies have demonstrated disrupted connectivity between the amygdala and hippocampus and key pain-processing regions in patients with migraine [6]. At the same time, structural MRI has revealed volume reductions in these limbic areas that correlate with migraine severity [7]. In addition, hypothalamic, thalamic, and other structural alterations have been observed in migraine, implicating these nuclei in the generation and modulation of attacks [8]. Taken together, these findings support a central role for striato-thalamo-cortical circuits in integrating pain, emotional processing, and autonomic regulation in migraine. Nevertheless, research specifically examining subcortical nuclei in MA is limited and requires further validation [9]. Another persistent question is whether MA and migraine without aura (MO) differ systematically in their neuroimaging characteristics [9–12]. Some studies using 3T MRI have reported differences between these subtypes. For instance, resting-state functional MRI (rs-fMRI) analyses found that MA patients, compared with MO, show altered functional connectivity and network topology in the basal ganglia, frontal gyrus, visual network, default mode network, salience network, and central executive network [13–22]. Similarly, structural MRI studies report that MA patients can have distinct gray and white matter changes—differences in cortical thickness or volume in visual and somatosensory cortices—relative to MO patients [23]. However, most neuroimaging studies of MA to date have only used 3T MRI and single-modality approaches. By focusing mainly on cortical phenomena (e.g., visualizing occipital CSD during aura) [5], these studies provide an incomplete picture of migraine-related brain alterations [24–28]. With advances in magnetic resonance technology, ultra-high-field 7T MRI affords higher spatial resolution and signal-to-noise ratio for migraine research, enabling the detection of subtle microstructural changes in small brain regions [29]. In parallel, a multimodal imaging strategy that combines complementary techniques, such as rs-fMRI, T1-weighted imaging, quantitative susceptibility mapping (QSM) [30, 31], and diffusion tensor imaging (DTI) [32], can yield a multidimensional view of brain changes in MA [33]. Growing evidence has proved dysregulated iron homeostasis [30, 34, 35] and glymphatic system (GS) [36, 37]—an essential cerebrospinal fluid clearance pathway—in migraine pathogenesis. If glymphatic impairment and abnormal iron accumulation contribute to migraine (for example, via impaired clearance of byproducts or neuroinflammation), these factors could be relevant in MA. Nevertheless, this aspect remains poorly investigated using neuroimaging, and no prior study has directly examined glymphatic function or iron deposition in MA, especially in comparison to MO. Based on these considerations, we undertook a multimodal 7 T study across MA, MO, and healthy controls (HCs). We systematically compare these cohorts across key brain regions (including subcortical nuclei, the hypothalamus, thalamus, hippocampus, and amygdaloid subregions) in terms of functional, structural, iron content, and brain-clearance characteristics (perivascular space (PVS) morphology, choroid plexus volume, and the analysis along the perivascular space (ALPS) index). We aimed to: (1) map functional, structural, and QSM alterations within a priori subcortical ROI families (hypothalamic, thalamic, hippocampal, and amygdaloid subregions) regions; (2) quantify GS status; and (3) test for the relationships among multimodal imaging metrics and their correlation with clinical parameters. We hypothesized that (1) a priori subcortical ROI families (hypothalamic, thalamic, hippocampal, and amygdaloid subregions) would show functional, structural, iron, and glymphatic alterations in MA and MO compared with HCs; (2) MA to show greater disturbances in regions potentially affected by repeated CSD and impaired clearance (e.g., glymphatic and iron-related changes) than MO; and (3) these imaging metrics would show meaningful associations with clinical parameters. Materials and methods Participants The study was approved by the Ethics Committee of the Chinese PLA General Hospital (S2023-459-01). All procedures were conducted in accordance with the latest version of the Declaration of Helsinki. Participants were recruited from the Chinese PLA General Hospital and provided written informed consent after receiving a detailed explanation of the study protocol. Participants with migraine were recruited from the International Headache Center of the First Medical Center of the Chinese People’s Liberation Army General Hospital between July 2023 and August 2025. Diagnoses were confirmed by two senior neurologists specializing in headache disorders. Participants were divided into three groups: one group of MA patients, one group of MO patients, and a group of HCs. All migraine patients underwent comprehensive clinical assessments, including documentation of headache characteristics (headache duration, days of headache per month, attack duration), the Visual Analogue Scale (VAS, range: 0–10) for average headache intensity, the Patient Health Questionnaire-9 (PHQ-9, range: 0–27) for depressive symptoms, the Generalized Anxiety Disorder-7 (GAD-7, range: 0–21) for anxiety symptoms, and the Pittsburgh Sleep Quality Index (PSQI, range: 0–21) for sleep quality. Each participant subsequently underwent 7T MRI scanning. To standardize neuroimaging conditions and mitigate pharmacological or stimulant-related artifacts, all participants were required to withhold acute analgesics and caffeinated drinks (e.g., coffee, tea, cola) for at least 12 h prior to the MRI session. All patients underwent MRI examinations during the interictal period, had not taken prophylactic medications within 3 months before the examination, and had not experienced headaches within 72 h after the examination. All participants were right-handed and of Chinese Han ethnicity. Inclusion criteria were as follows: (1) aged between 18 and 65 years; (2) voluntary participation with written informed consent; (3) fulfillment of the International Classification of Headache Disorders, 3rd edition (ICHD-3) diagnostic criteria for MA or MO [38]; (4) no prior use of migraine prophylactic medications; (5) no migraine attacks 24 h preceding MRI scanning. Exclusion criteria included: (1) any neurological or psychiatric disorders; (2) any systemic diseases (e.g., cardiovascular diseases, tumors); (3) other chronic pain conditions (e.g., chronic back pain); (4) pregnancy at the time of scanning; (5) contraindications to MRI; (6) headache occurrence during the MRI procedure. HCs were recruited from primary healthcare facilities or community centers. They were required to be between 18 and 65 years of age, with no personal history of neurological (including headache or chronic pain conditions) or psychiatric illnesses, and no MRI contraindications. This study comprised four analytical modules to investigate differences among patients with MA, MO, and HCs in the following aspects: (1) functional changes in subcortical nuclei and their subregions, assessed via rs-fMRI; (2) volumetric alterations within these structures, as derived from 3D-T1 images; (3) iron content variations measured through QSM; and (4) differences in GS function, evaluated using the Diffusion tensor imaging analysis along the perivascular space (DTI-ALPS) index, choroid plexus volume, and PVS metrics within the basal ganglia. Furthermore, the relationships among multimodal imaging metrics and their correlation with clinical parameters were examined. Image acquisition The sequences employed and features extracted by this research are all shown in Fig. 1. All MRI examinations were conducted on a whole-body 7T scanner (MAGNETOM Terra, Siemens Healthcare, Erlangen, Germany) equipped with an 8-channel transmit/32-channel receive head coil. During the scanning process, participants were positioned supine, and foam padding was applied to restrict head motion. The imaging protocol included the following sequences: rs-fMRI: Acquired using echo-planar imaging (EPI) with the following parameters: voxel size = 1.8 × 1.8 × 1.8 mm3, repetition time (TR) = 2000 msec, echo time (TE) = 24 msec, flip angle (FA) = 90°, field of view (FOV) = 216 × 216 mm, matrix = 120 × 120, slice thickness = 1.8 mm, number of slices = 80, bandwidth = 2314 Hz/Px, and each participant completed one run comprising 240 volumes (total scan duration = 8.0 min); the first 5 volumes were discarded, leaving 235 volumes for analysis. Three-dimensional T1-weighted magnetization-prepared 2 rapid acquisition gradient echo (3D-T1WI-MP2RAGE): voxel size = 0.7 × 0.7 × 0.8 mm3, TR = 6000 msec, TE = 2.21 msec, FA 1 = 4.0°, FA 2 = 5.0°, FOV = 224 × 224 mm, TI1 = 800ms, TI2 = 2700ms, slice thickness = 0.75 mm, 208 slices, bandwidth = 200 Hz/Px or(voxel size = 0.8 × 0.8 × 0.8 mm3, TR = 4300 ms, TE = 2.27 ms, FA 1 = 4.0°, FA 2 = 4.0°, FOV = 240 × 240 mm, slice thickness = 0.75 mm, and slice number = 192, TI1 = 1000ms, TI2 = 3200ms, bandwidth = 200 Hz/Px), Diffusion tensor imaging (DTI): TR = 5120 ms, TE = 59 ms, FOV = 224 × 224 mm, slice thickness = 1.6 mm, voxel size = 1.6 × 1.6 × 1.6 mm3, number of slices = 80, and FA = 180°, bandwidth = 992 Hz/Px, b-values: 0 and 1000 s/mm², 12 non-collinear diffusion-encoding directions at b = 1000 s/mm² plus b0 images, Diffusion tensors were fitted to derive axial diffusivities along the x, y, and z directions, which were then used to compute the DTI-ALPS index. QSM: voxel size = 0.6 × 0.6 × 0.8 mm3, TR = 27 msec, TE = 5.6 msec, FOV = 218 × 218 mm, FA = 12°, bandwidth = 250 Hz/Px, slice thickness = 0.8 mm, 160 sagittal slices. 3D variable flip angle turbo spin echo (TSE): voxel size = 0.5 × 0.5 × 0.6 mm³, TR = 3000 msec, TE = 241 msec, FOV = 210 × 210 mm, FA = 7°, bandwidth = 5000 Hz/Px, slice thickness = 0.55 mm, 208 sagittal slices. Throughout the scan, participants were instructed to remain still, keep their eyes closed, and stay awake. Vigilance was confirmed verbally immediately after the scan to ensure that no participant had fallen asleep during image acquisition. The total acquisition time for the sequences used in the present analyses (rs-fMRI, MP2RAGE, DTI, QSM, and 3D-PVS/T2-TSE) was approximately 40–45 min per participant, not including localizers and optional spectroscopy. Detailed information on the sequences is listed in Supplementary Table 1. Although two 3D T1-weighted sequences were acquired in this study, only one of them was included in the analysis. Fig. 1Overview of the multimodal 7T MRI protocol and extracted imaging metrics. 3D-PVS utilizes a sequence of the 3D variable flip angle turbo spin echo (TSE). 3D-T1 utilizes a sequence of the 3D-T1WI-MP2RAGE; 3D-T1WI-MP2RAGE: Three-dimensional T1-weighted magnetization-prepared rapid acquisition gradient echo. DTI: Diffusion tensor imaging; rs-fmri: resting-state functional MRI; QSM: quantitative susceptibility mapping; PVS: perivascular space; APLS: analysis along the perivascular space; falff: fractional amplitude of low-frequency fluctuation; reho: regional homogeneity Overview of the multimodal 7T MRI protocol and extracted imaging metrics. 3D-PVS utilizes a sequence of the 3D variable flip angle turbo spin echo (TSE). 3D-T1 utilizes a sequence of the 3D-T1WI-MP2RAGE; 3D-T1WI-MP2RAGE: Three-dimensional T1-weighted magnetization-prepared rapid acquisition gradient echo. DTI: Diffusion tensor imaging; rs-fmri: resting-state functional MRI; QSM: quantitative susceptibility mapping; PVS: perivascular space; APLS: analysis along the perivascular space; falff: fractional amplitude of low-frequency fluctuation; reho: regional homogeneity Image analysis FreeSurfer software (7.4.1 [39]) was used for the preprocessing of 3D-T1 structural images and volumetric quantification of subcortical structures. The standard automated processing pipeline was executed [40–53]. Briefly, the following steps were performed: (1) Data preparation: Given the inherently higher noise in 7T MRI, denoising was performed using the DenoiseImage tool to improve signal quality. (2) Reconstruction: The “recon-all” pipeline was run for each subject to perform automated skull stripping, Talairach transformation, volumetric segmentation, and cortical thickness estimation. (3) Quality control: Segmentation results were visually inspected using FreeView to ensure anatomical accuracy. (4) Volumetric extraction: regional volumes of subcortical nuclei, choroid plexus, and subregions. Of note, the native high resolution of the 3D-T1 images was preserved throughout all processing steps. Rs-fMRI data were preprocessed using the DPABI toolbox [54] in MATLAB R2022b. The following standard procedures were applied to ensure data quality and reproducibility: (1) Initial Processing: The first five time points were discarded to allow for magnetic field stabilization. (2) Basic Preprocessing: The remaining 235 time points underwent slice timing correction, motion realignment (reorienting to the anterior commissure–posterior commissure plane), and coregistration with each subject’s high-resolution T1-weighted structural image (which was segmented using the DARTEL algorithm [55]. (3) Spatial Normalization and Smoothing: Functional images were normalized to the Montreal Neurological Institute (MNI) space. For amplitude of low-frequency fluctuation (ALFF) analysis, spatial smoothing was applied with a 6-mm full-width at half maximum (FWHM) Gaussian kernel. Regional homogeneity (ReHo) maps were analyzed without smoothing to preserve local signal characteristics. (4) Metric Calculation: ALFF was computed via fast Fourier transform within the frequency band of 0.01–0.08 Hz and converted to z-scores to obtain mean ALFF (mALFF). ReHo was derived using Kendall’s coefficient of concordance (Kendall’s W) to evaluate the coherence between each voxel and its 27 nearest neighbors, and subsequently normalized to z-scores. (5) Nuisance Regression and Filtering: Nuisance covariates—including signals from white matter and cerebrospinal fluid, as well as head motion parameters based on Friston’s 24-parameter model—were regressed out. The data were then detrended and bandpass filtered (0.01–0.08 Hz). All preprocessed images were visually inspected by two experienced radiologists to confirm quality. (6) Output: Final fractional ALFF (fALFF) and ReHo maps were saved in NIfTI format. Since the standard FreeSurfer atlas does not provide hypothalamic subsegmentations, hypothalamic subregion volumes were extracted using a previously published hypothalamic atlas and H-SynEx methodology [56]. The DTI data were preprocessed and analyzed according to the following comprehensive pipeline: (1)Correction for motion and eddy currents to align temporal sequence; (2) Adjustment of gradient direction tables based on deformation fields from motion correction to enhance tensor estimation accuracy; (3) Fitting of the tensor model using corrected b0 and b1000 images, integrating respective gradient direction and b-value tables for voxel-specific calculations; (4) Generation of diffusion maps in each Cartesian direction (x, y, and z); (5) ROI delineation in projection and association fiber areas of both cerebral brains using fsleyes, with ROIs defined using a 5 mm radius; (6) Extraction of diffusivity values along each direction (x, y, and z) from ROIs delineated in projection and association fiber areas; (7) We computed bilateral ALPS index (left, right, and average of bilateral hemispheres) using the formula [(Dxproj + Dxassoc)/(Dyproj + Dzassoc)]. A higher ALPS index indicates enhanced diffusion and, consequently, improved glymphatic function. QSM processing was performed using the STISuite toolbox. The procedure consisted of the following steps: First, phase and magnitude images were acquired via a 3D gradient-recalled-echo (3D-GRE) sequence and converted into the appropriate format. Second, during data preparation, multiple 3D volumes were concatenated into a single 4D dataset. The phase component of this 4D volume was then unwrapped and rescaled to the range [−π, π]. Third, QSM reconstruction was conducted within STISuite and executing its automated processing pipeline. Fourth, each resultant QSM map was non-linearly registered to the standard MNI-152 template space. The mean susceptibility values in each ROI were then extracted. For the evaluation of PVS features in the basal ganglia, we employed an adapted methodology based on a previously published approach [57]. PVS were automatically segmented on T2-weighted images using a fully convolutional neural network. The basal ganglia—comprising the caudate nucleus, putamen, and globus pallidus—were delineated as the region of interest (ROI) by registering the JHU-Eve atlas to individual native space. Three PVS morphological metrics—path length, volume, and tortuosity—were quantified. After segmenting T1 images with FreeSurfer to obtain the left and right choroid plexus volumes, we calculated the choroid plexus volume by dividing the individual choroid plexus volume by the total TIV to reduce the influence of individual differences [58]. Individual-level T1-weighted images generated from FreeSurfer in native space were non-linearly registered to the MNI-152 standard space using Advanced Normalization Tools (ANTs). This high-precision approach ensured accurate spatial alignment of structural definitions with functional metrics while accounting for inter-subject neuroanatomical variations. Subregions that were too small and did not survive the mask registration process were excluded from subsequent ROI analyses to maintain reliability. The subcortical nuclei examined in this study comprised the bilateral hypothalamus, thalamus, caudate, putamen, pallidum, hippocampus, and amygdala. In addition to global analyses of these structures, we performed subregional analyses of the hypothalamus, thalamus, hippocampus, and amygdala. This focus on these structures was motivated by their roles as key hubs in pain modulation, autonomic regulation, and affective processing. The hippocampus and amygdala are critical limbic regions involved in pain modulation, emotion, and stress responsivity—factors highly relevant to migraine [7]. The thalamus serves as a central relay in the trigeminovascular pain pathway and is well known to be involved in migraine pathophysiology [59]. Clinical features such as cyclic attack timing, sleep disturbances, and autonomic symptoms (e.g., yawning, nausea) strongly suggest hypothalamic involvement in migraine [60]. In summary, we concentrated on the hippocampus, amygdala, thalamus, and hypothalamus because of their documented involvement in migraine’s pain network. Statistical analysis Demographic differences between migraine groups and HCs were assessed as follows: The normality of continuous variables was examined using the Shapiro–Wilk test. Normally distributed variables were summarized as mean ± standard deviation, while non-normally distributed variables were reported as median (interquartile range, IQR). Group differences across the three cohorts were compared using general linear models (GLM), with statistical significance set at P < 0.05. For neuroimaging comparisons among MA, MO, and HC groups, GLMs were applied to evaluate differences in subcortical and subregional values of fALFF, ReHo, iron deposition (QSM), volume, and glymphatic metrics, while covariates were specified by modality: age for fALFF and ReHo; age and PHQ-9 for QSM; and age, PHQ-9, and GAD-7 for glymphatic metrics. The results were considered significant at a false discovery rate (FDR)-corrected P < 0.05, applied separately for rs-fMRI, 3D-T1, and QSM within each predefined ROI set. While comparing volumetric comparisons, we included total intracranial volume (TIV) as an additional covariate. For post hoc pairwise comparisons, we used GLMs restricted to those ROIs that showed a significant group effect and reported them at P < 0.05. For clinical correlation analyses, partial correlation analyses were employed to assess relationships between a small set of imaging metrics that showed robust group differences and clinical parameters (VAS, headache history, days of headache per month, attack duration), controlling for relevant covariates, with statistical significance set at P < 0.05. We did not apply a further formal multiple-comparison correction to these correlations. Results Demographics and clinical data A total of 119 participants were included in this study. Among them, imaging data from 35 patients with MA, 45 patients with MO, and 39 HCs met the quality criteria for analysis in Module 1 and Module 2. For Module 3 (QSM analysis), 25 MA patients, 26 MO patients, and 25 HCs were included. In Module 4 (GS evaluation), 20 MA patients, 24 MO patients, and 26 HCs underwent the relevant MRI protocols. Demographic and clinical characteristics across the four modules are summarized in Table 1. Age was the only variable that showed significant group differences across all four modules. In addition, PHQ-9 scores differed significantly among the groups in Module 3, whereas both PHQ-9 and GAD-7 scores showed significant group differences in Module 4. Table 1Participants’ demographic and clinical traitsCharacteristicMAMOHC P nModule (1/2)/Module 3/Module 435/25/2045/26/2439/25/26NAAgeModule (1/2)27(20, 35.5)36(27, 45)37(29.5, 54)< 0.001Module 337(28.5,52)36(27,45)42(33,50)0.002Module 437(28.75,53)39(30, 47)43(34, 50)0.033Sex(M/F)Module (1/2)13/2410/3515/240.207Module 317/3410/3510/450.358Module 435/1735/1031/100.281Headache history (years)Module (1/2)7(3, 14)10(7, 15)NANAModule 37(3, 12)10(7, 15)NANAModule 410(5.75, 15)10(6.75, 10.25)NANADays of headache per monthModule (1/2)/Module 3/Module 43(1, 5)/3(1, 4.5)/3.25(1, 4.5)4(2, 6)/5(2.25, 6.75)/5(2, 7.25)NANAAttack duration(hours)Module (1/2)/Module 3/Module 46(4.5, 16)/5(4.5, 12)/6.5(4.375, 12)12(8, 36)/12(9, 24)/12(8, 27)NANAHeadache intensity (VAS 1–10)Module (1/2)/Module 3/Module 46(5, 8)/6(5, 8)/6(4.875, 8.25)7(6, 9)/8(6.25, 9)/8(6, 9)NANAPHQ-9Module (1/2)6(3, 10)4(2, 7)4(0, 9)0.747Module 35(3, 10)3.5(2, 7)0(0, 1)<0.001Module 45.5(2.75, 10)3.5(2, 7)1(0, 5)0.011GAD-7Module (1/2)4(0.25, 7)2(1, 7)4(0, 7)0.713Module 34(0,5)2.5(1, 7)1(0, 4)0.202Module 44.5(0.75, 5.5)4(1.75, 7)1.5(0, 4)0.025PSQIModule (1/2)5(3, 9)6(4, 8)5(3, 7.5)0.550Module 35(3, 9)5.5(4, 8.75)4(3, 7)0.439Module 45(4, 9)6(4, 9)4(3, 6.75)0.272The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; VAS: Visual Analogue Scale; GAD-7: Generalized Anxiety Disorder-7; PHQ-9: Patient Health Questionnaire-9; PSQI: Pittsburgh Sleep Quality Index Table 1Participants’ demographic and clinical traitsCharacteristicMAMOHC P nModule (1/2)/Module 3/Module 435/25/2045/26/2439/25/26NAAgeModule (1/2)27(20, 35.5)36(27, 45)37(29.5, 54)< 0.001Module 337(28.5,52)36(27,45)42(33,50)0.002Module 437(28.75,53)39(30, 47)43(34, 50)0.033Sex(M/F)Module (1/2)13/2410/3515/240.207Module 317/3410/3510/450.358Module 435/1735/1031/100.281Headache history (years)Module (1/2)7(3, 14)10(7, 15)NANAModule 37(3, 12)10(7, 15)NANAModule 410(5.75, 15)10(6.75, 10.25)NANADays of headache per monthModule (1/2)/Module 3/Module 43(1, 5)/3(1, 4.5)/3.25(1, 4.5)4(2, 6)/5(2.25, 6.75)/5(2, 7.25)NANAAttack duration(hours)Module (1/2)/Module 3/Module 46(4.5, 16)/5(4.5, 12)/6.5(4.375, 12)12(8, 36)/12(9, 24)/12(8, 27)NANAHeadache intensity (VAS 1–10)Module (1/2)/Module 3/Module 46(5, 8)/6(5, 8)/6(4.875, 8.25)7(6, 9)/8(6.25, 9)/8(6, 9)NANAPHQ-9Module (1/2)6(3, 10)4(2, 7)4(0, 9)0.747Module 35(3, 10)3.5(2, 7)0(0, 1)<0.001Module 45.5(2.75, 10)3.5(2, 7)1(0, 5)0.011GAD-7Module (1/2)4(0.25, 7)2(1, 7)4(0, 7)0.713Module 34(0,5)2.5(1, 7)1(0, 4)0.202Module 44.5(0.75, 5.5)4(1.75, 7)1.5(0, 4)0.025PSQIModule (1/2)5(3, 9)6(4, 8)5(3, 7.5)0.550Module 35(3, 9)5.5(4, 8.75)4(3, 7)0.439Module 45(4, 9)6(4, 9)4(3, 6.75)0.272The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; VAS: Visual Analogue Scale; GAD-7: Generalized Anxiety Disorder-7; PHQ-9: Patient Health Questionnaire-9; PSQI: Pittsburgh Sleep Quality Index Participants’ demographic and clinical traits The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR) Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; VAS: Visual Analogue Scale; GAD-7: Generalized Anxiety Disorder-7; PHQ-9: Patient Health Questionnaire-9; PSQI: Pittsburgh Sleep Quality Index Module 1: functional alterations in subcortical nuclei and subregions After controlling for age as covariates, a GLM was employed to compare fALFF and ReHo values across the three groups in subcortical nuclei and subregions of the hypothalamus, thalamus, hippocampus, and amygdala. Following FDR correction, significant differences in ReHo were observed in the following regions: hippocampus, thalamus, pallidum, (Table 2). Post hoc pairwise comparisons revealed that, compared with HCs, the MA group showed significantly lower ReHo values in hippocampus(L; F=9.23, P=0.003), pallidum(R; F=51.93, P < 0.001), and thalamus(R; F=6.53, P=0.013). Similarly, the MO group exhibited significantly reduced ReHo relative to HCs in hippocampus(L; F=8.65, P=0.040), thalamus(R; F=12.25, P=0.001), pallidum(R; F=96.89, P<0.001) (Fig. 2). Table 2Brain regions displaying significant between-group differences in functional valuesMarkerMAMOHCF P P (FDR corrected) Hippocampus(L)0.45(0.42,0.49)0.45(0.42,0.47)0.47(0.44,0.52)6.940.0010.012Thalamus(R)0.53 ± 0.050.51 ± 0.060.55 ± 0.066.900.0010.012Pallidum(R)0.43(0.39,0.45)0.40(0.36,0.43)0.53(0.46,0.59)57.29<0.0010.007The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; molecular_layer_HP-head: molecular_layer_hippocampus-head Table 2Brain regions displaying significant between-group differences in functional valuesMarkerMAMOHCF P P (FDR corrected) Hippocampus(L)0.45(0.42,0.49)0.45(0.42,0.47)0.47(0.44,0.52)6.940.0010.012Thalamus(R)0.53 ± 0.050.51 ± 0.060.55 ± 0.066.900.0010.012Pallidum(R)0.43(0.39,0.45)0.40(0.36,0.43)0.53(0.46,0.59)57.29<0.0010.007The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; molecular_layer_HP-head: molecular_layer_hippocampus-head Brain regions displaying significant between-group differences in functional values The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR) Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; molecular_layer_HP-head: molecular_layer_hippocampus-head Fig. 2Group differences in ReHo values among MA, MO, and HCs. A, Significant differences in ReHo values were observed in the hippocampus(L) across the three groups. B, Significant differences in ReHo values were observed in the thalamus(R) across the three groups. C, Significant differences in ReHo values were observed in the pallidum(R) across the three groups. *, significant difference from the HC (P < 0.05); **, significant difference from the HC (P < 0.01); ***, significant difference from the HC (P < 0.001) Group differences in ReHo values among MA, MO, and HCs. A, Significant differences in ReHo values were observed in the hippocampus(L) across the three groups. B, Significant differences in ReHo values were observed in the thalamus(R) across the three groups. C, Significant differences in ReHo values were observed in the pallidum(R) across the three groups. *, significant difference from the HC (P < 0.05); **, significant difference from the HC (P < 0.01); ***, significant difference from the HC (P < 0.001) Module 2: volumetric changes in subcortical nuclei and subregions In the volumetric comparisons across groups, TIV was added to the covariates. After applying FDR correction to the volumetric data, significant differences in volumetric values were observed among the three groups in the subregions of amygdala (Table S2). Post hoc analyses suggested that MA patients tended to have smaller volumes than HC in selected amygdalar subregions paralaminar nucleus (FL = 5.90, PL = 0.018; FR = 4.38, PR = 0.040), whereas MO patients showed smaller volumes than HCs in multiple amygdalar nuclei Lateral-nucleus(FL=9.89, PL=0.001; FR=11.17, PR=0.002),Basal-nucleus(FL=10.03, PL<0.001; FR=9.94, PL=0.001),Paralaminar-nucleus(FL=10.24, PL=0.001; FR=8.79, PR=0.002), Whole_amygdala(FL=9.56, PL=0.001; FR=10.32, PR=0.001),Accessory-Basal-nucleus(FR=8.69, PR=0.003),Central-nucleus(FR=10.44, PR=0.002). For full statistical details, please refer to Supplementary Table S2. Module 3: iron content differences assessed by QSM After controlling for age and PHQ-9 scores, and following FDR correction, significant differences in QSM values were observed among the three groups in the Cortical-nucleus and Corticoamygdaloid-transitio (Table 3). Post hoc pairwise comparisons revealed that the MA group exhibited significantly higher QSM values compared to the MO group in the Cortical-nucleus (FL=10.97, PL=0.002; FR=11.70, PR=0.001), and Corticoamygdaloid-transitio (FL=12.78, PL=0.001; FR=10.57, PR=0.002). In addition, the MO group showed significantly lower QSM values than the HC group in the Corticoamygdaloid-transitio (FL=12.87, PL=0.001; FR=6.30, PR=0.015) and Cortical-nucleus (FL=6.90, PL=0.011; FR=7.88, PR=0.007) (Fig. 3). Table 3Brain regions displaying significant between-group differences in QSM valuesMarkerMAMOHCF P P (FDR corrected) Amygdala Cortical-nucleus(L)-0.0003 ± 0.0159-0.0125 ± 0.0180-0.0054 ± 0.01597.670.0010.027Corticoamygdaloid-transitio(L)-0.0101 ± 0.0103-0.0188 ± 0.0104-0.0096 ± 0.01098.92<0.0010.027Cortical-nucleus(R)0.0006 ± 0.0174-0.0122 ± 0.0166-0.0052 ± 0.01427.260.0020.032Corticoamygdaloid-transitio(R)-0.0114 ± 0.0115-0.0196 ± 0.0129-0.0121 ± 0.01176.340.0030.035The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls Table 3Brain regions displaying significant between-group differences in QSM valuesMarkerMAMOHCF P P (FDR corrected) Amygdala Cortical-nucleus(L)-0.0003 ± 0.0159-0.0125 ± 0.0180-0.0054 ± 0.01597.670.0010.027Corticoamygdaloid-transitio(L)-0.0101 ± 0.0103-0.0188 ± 0.0104-0.0096 ± 0.01098.92<0.0010.027Cortical-nucleus(R)0.0006 ± 0.0174-0.0122 ± 0.0166-0.0052 ± 0.01427.260.0020.032Corticoamygdaloid-transitio(R)-0.0114 ± 0.0115-0.0196 ± 0.0129-0.0121 ± 0.01176.340.0030.035The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls Brain regions displaying significant between-group differences in QSM values The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR) Abbreviations: L: left; R: right; MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls Fig. 3Group differences in QSM values among MA, MO, and HCs. A, QSM values differed significantly among the groups in Cortical-nucleus(L). B, QSM values differed significantly between the groups in Corticoamygdaloid-transitio(L). C, QSM values differed significantly between the groups in Cortical-nucleus(R). D, QSM values differed significantly between the groups in Corticoamygdaloid-transitio(R). *, significant difference between two groups (P < 0.05); **, significant difference between two groups (P < 0.01); ***, significant difference between two groups (P < 0.001) Group differences in QSM values among MA, MO, and HCs. A, QSM values differed significantly among the groups in Cortical-nucleus(L). B, QSM values differed significantly between the groups in Corticoamygdaloid-transitio(L). C, QSM values differed significantly between the groups in Cortical-nucleus(R). D, QSM values differed significantly between the groups in Corticoamygdaloid-transitio(R). *, significant difference between two groups (P < 0.05); **, significant difference between two groups (P < 0.01); ***, significant difference between two groups (P < 0.001) Module 4: glymphatic system alterations among the three groups (DTI-ALPS, PVS, choroid plexus volume) GS function was assessed using the DTI-ALPS index, choroid plexus volume, PVS length, PVS volume, and PVS tortuosity. After controlling for age, GAD-7, and PHQ-9 scores, significant differences among the three groups were observed in PVS length, PVS volume, and the right ALPS index (Table 4). Post hoc pairwise comparisons indicated that the MA group exhibited greater PVS length (P=0.022) and PVS volume (P=0.003) compared to the MO group. Additionally, the right ALPS index (P=0.008) and PVS volume (P=0.006) was significantly altered in the MA group than in the HC group (Fig. 4). Table 4Brain regions displaying significant between-group differences in the glymphatic systemMarkerMAMOHCF P PVS length3.14(2.64,4.12)2.77(2.18,2.92)2.79(2.59,3.01)3.140.049PVS volume3.32(1.85,3.73)1.83(1.45,2.44)1.90(1.03,2.52)7.020.002ALPS index(R)1.59 ± 0.171.67 ± 0.281.76 ± 0.223.970.024The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; PVS: perivascular space; ALPS: analysis along the perivascular space; R: right Table 4Brain regions displaying significant between-group differences in the glymphatic systemMarkerMAMOHCF P PVS length3.14(2.64,4.12)2.77(2.18,2.92)2.79(2.59,3.01)3.140.049PVS volume3.32(1.85,3.73)1.83(1.45,2.44)1.90(1.03,2.52)7.020.002ALPS index(R)1.59 ± 0.171.67 ± 0.281.76 ± 0.223.970.024The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR)Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; PVS: perivascular space; ALPS: analysis along the perivascular space; R: right Brain regions displaying significant between-group differences in the glymphatic system The P-values (overall) represent the statistical analysis results among the three groups. Data following a normal distribution were expressed as mean ± SD, whereas data not meeting the normal distribution criteria were expressed as median and interquartile range (IQR) Abbreviations: MA: Migraine with aura; MO: Migraine without aura; HCs: Healthy controls; PVS: perivascular space; ALPS: analysis along the perivascular space; R: right Fig. 4Group differences in GS metrics among MA, MO, and HCs. A, GS metrics differed significantly among MA, MO, and HCs in PVS length. B, GS metrics differed significantly among MA, MO, and HCs in PVS volume. C, GS metrics differed significantly among MA, MO, and HCs in right DTI-ALPS index. *, significant difference between two groups (P < 0.05); **, significant difference between two groups (P < 0.01); ***, significant difference between two groups (P < 0.001) Group differences in GS metrics among MA, MO, and HCs. A, GS metrics differed significantly among MA, MO, and HCs in PVS length. B, GS metrics differed significantly among MA, MO, and HCs in PVS volume. C, GS metrics differed significantly among MA, MO, and HCs in right DTI-ALPS index. *, significant difference between two groups (P < 0.05); **, significant difference between two groups (P < 0.01); ***, significant difference between two groups (P < 0.001) Associations between functional activity, magnetic susceptibility, glymphatic metrics, and clinical parameters Partial correlation analyses were conducted in patients with MA to examine associations between altered brain regions and clinical parameters, adjusting for predefined covariates. The QSM value in the Corticoamygdaloid-transition(R) was negatively correlated with VAS (r = − 0.637, P < 0.001). In contrast, no significant associations were detected between GS metrics and clinical indicators. In a complementary analysis linking imaging alterations to glymphatic function, we only included patients who had completed both QSM and 3D-PVS sequences. The results revealed that higher QSM values in amygdaloid subnuclei were associated with shorter PVS length. Specifically, the Cortical-nucleus(R) (r = − 0.606, P = 0.010) exhibited significant negative correlations with PVS length (Fig. 5). Fig. 5A, The QSM value in the Corticoamygdaloid-transitio (R) correlated inversely with VAS score. B, The QSM values in the Cortical-nucleus(R) correlated inversely with PVS length A, The QSM value in the Corticoamygdaloid-transitio (R) correlated inversely with VAS score. B, The QSM values in the Cortical-nucleus(R) correlated inversely with PVS length Discussion In this 7T multimodal MRI study, we systematically interrogated functional, structural, iron content, and GS alterations in MA. Two principal findings emerged. First, compared to HCs, both MA and MO showed interictal ReHo reductions in limbic–basal ganglia regions, with overlapping decreases in the left hippocampus, right thalamus, and right pallidum. Second, MA and MO showed divergent patterns of iron homeostasis, with MA exhibiting higher magnetic susceptibility than MO in specific amygdalar subnuclei, while MO showed lower susceptibility than HCs in some of these regions. MA, but not MO, was characterized by enlarged basal-ganglia PVS and a reduced right-hemisphere ALPS index compared with controls, indicating more pronounced glymphatic dysfunction in the aura subtype. Together with the QSM results, these findings point to a subtype-specific coupling between glymphatic dysfunction and amygdalar iron content in MA. Furthermore, we observed that iron content in the amygdala was inversely related to PVS length in MA, indicating a potential mechanistic link between iron deposition and glymphatic dysfunction in MA. Our rs-fMRI results showed that both MA and MO patients demonstrate disrupted ReHo in key limbic and basal ganglia regions. In MA, reduced ReHo in the left hippocampus and right pallidum indicates impaired neurolimbic and striato-thalamic network function. The hippocampus, a limbic hub involved in pain processing, stress regulation, and memory, is highly vulnerable to chronic pain and stress [29, 61–64]. Repeated migraine episodes may therefore induce maladaptive hippocampal changes, consistent with reports that frequent headaches correlate with hippocampal atrophy and altered connectivity [65]. Likewise, pallidal hypoactivity in both MA and MO highlights basal ganglia involvement; the pallidum contributes to movement, motivation, and cognitive modulation, and its dysfunction has been linked to migraine frequency [66]. The thalamus is a key relay for trigeminovascular nociceptive signaling, and thalamic dysfunction can exacerbate aberrant sensory processing in migraine [67]. Reduced local coherence in the thalamus may therefore signify impaired integration of nociceptive signals and disrupted thalamo-cortical communication in migraine. Such disturbances likely contribute to both the sensory symptoms (through thalamic and basal ganglia dysfunction) and the affective/cognitive symptoms (through limbic system involvement) associated with migraine. Overall, our findings reinforce that migraine–with or without aura–involves persistent alterations in brain networks that extend beyond the cortex, encompassing regions responsible for the affective-cognitive dimensions of pain. Although volumetric differences of the amygdala were observed in the three-group comparison, this effect was largely driven by a significantly reduced amygdala volume in the MO group compared with HCs. The amygdala, a neurolimbic node governing affective and autonomic responses, plays a central role in pain modulation during attacks [18, 64, 68–70]. Its altered volume suggests greater limbic dysregulation of pain in MO, in line with evidence that migraine involves a “dysfunctional neurolimbic pain network” contributing to impaired pain modulation [7]. These subtype differences imply that while both migraine groups share core limbic-related and basal ganglia disturbances, MO may entail more pronounced deficits in thalamic sensory gating and emotional pain processing circuits. Overall, the results may reinforce that the presence of aura appears to alter migraine’s impact on the brain, leading to structural atrophy in key networks. A novel aspect of our study is the identification of divergent iron deposition patterns in the brains of MA. QSM indicated that MA patients have increased iron content in specific amygdala subnuclei, whereas MO patients tend to have iron levels that are lower than those of HCs in those same regions. Iron in the brain is often a marker of cumulative oxidative stress or microhemorrhages and has been linked to migraine pathophysiology. Repeated migraines can disrupt the blood–brain barrier (BBB) permeability, permitting iron-rich blood proteins to extravasate and deposit in brain tissue [71–77]. Excess iron generates oxidative stress and can damage neurons [76]. Indeed, MRI studies have demonstrated excessive iron in regions like the periaqueductal gray and striatum of migraine patients, correlating with attack frequency [34, 35, 78–81] and a recent QSM study found that chronic migraineurs had higher iron in certain nuclei (e.g., the nucleus accumbens) compared to episodic patients [82]. The amygdala, as part of the limbic pain circuit [83], has been less studied in the past, but our results indicate it is also affected. The unexpectedly low amygdalar QSM in MO could be explained by: (1) amygdala atrophy reduces total iron content (even if iron concentration per tissue is unchanged); (2) increased diamagnetic components (e.g., myelin or glial changes from iron clearance) lower the susceptibility signal [74–77]. In contrast, the elevated QSM in MA denotes greater iron retention, consistent with repeated aura-related neuroinflammation. Notably, within MA, we found that lower QSM in the right corticoamygdaloid-transition was associated with higher headache intensity. One hypothesis is that moderate iron accumulation in these regions represents a neuroadaptive response that dampens pain sensitivity [84, 85]; thus, MA patients with lower QSM would experience more intense pain due to the lack of this adaptation. In summary, our QSM findings point to differential iron homeostasis in amygdalar subregions between migraine subtypes. Along with evidence of neuroinflammation and iron deposition in migraine pain/emotion pathways [83], this suggests that iron dysregulation may distinguish aura vs. non-aura migraine. Longitudinal studies are needed to clarify the relationship between brain iron and migraine frequency/severity, and to test whether modulating iron levels could benefit migraine patients. In addition, MO showed lower magnetic susceptibility than HCs in specific amygdalar subnuclei. Notably, the MO group also exhibited smaller amygdalar volume than HCs. Given the non-specificity of QSM, and considering the modest sample size together with the reduced amygdalar volume in MO, this pattern should be confirmed in larger, independent cohorts. Our results provide preliminary support for the involvement of the GS in MA. Unlike the original approach of Taoka et al., who quantified the ALPS index only in the left hemisphere [32], we calculated the ALPS index in both hemispheres, given that glymphatic function is generally considered a global brain phenomenon. Measuring ALPS in both hemispheres and using the mean value provides a more robust estimate of overall glymphatic activity, while also allowing detection of any potential lateralized effects. This bilateral quantification strategy has also been adopted in several recent studies [37, 86]. We found that MA patients had significantly greater PVS volume and length than MO, and a lower right-hemisphere DTI-ALPS compared to controls. This indicates that MA is associated with diminished glymphatic clearance. The GS is a brain-wide waste clearance pathway that operates mainly during sleep, driven by arterial pulsation and astroglial aquaporin-4 channels [36, 87, 88]. CSD, the event underlying aura, is thought to transiently close perivascular channels and impede glymphatic flow [36]. Our findings support that repeated aura episodes could cumulatively impair waste clearance, with compensatory PVS enlargement and a lowered ALPS index [25]. Notably, only the MA group – not MO – showed a significant ALPS reduction compared to controls, implying that aura predisposes patients to greater glymphatic dysfunction. This may explain why MA patients have been reported to exhibit higher cerebrospinal Fluid inflammatory marker levels (e.g., IL-1β, calcitonin gene-related peptide (CGRP)) [36], as impaired clearance could lead to their accumulation. The MO group did not significantly differ from controls on PVS metrics, but glymphatic involvement in MO cannot be ruled out. In our sample, MA showed much more pronounced PVS changes (likely due to CSD events), whereas MO changes were minimal. Our PVS analysis was limited to basal ganglia regions, so it is possible that glymphatic alterations in other brain areas were not captured. We also found no significant correlations between glymphatic indices and clinical features in MA, suggesting that glymphatic dysfunction is more an inherent trait of the aura subtype or cumulative disease burden rather than directly tied to immediate headache frequency. Larger studies are needed to fully elucidate glymphatic changes in migraine and their clinical significance. Importantly, we discovered a link between iron deposition and PVS changes in MA, suggesting that the perivascular channels that surround these subnuclei may be fewer/shorter or exhibit reduced patency, consistent with impaired lymphoid clearance efficiency. This is in agreement with findings from aging research, which show that individuals with reduced glymphatic function tend to have greater brain iron retention [89]. The GS likely contributes to the clearance of excess iron or iron-bearing compounds from the brain [89]; therefore, when glymphatic transport is compromised, iron may progressively accumulate in certain regions. Conversely, iron overload can exacerbate oxidative damage to small vessels and perivascular tissues, potentially worsening glymphatic flow [89]. We speculate that in MA, a vicious cycle may occur: recurrent CSD leads to glymphatic impairment, which allows iron and other metabolites to build up, and this iron deposition in turn further impairs the perivascular clearance pathways. Over time, this feedback loop could contribute to chronic inflammation and structural changes. Although our cross-sectional data cannot prove causality, the iron–glymphatic coupling we identified opens a new perspective on migraine pathophysiology. It underscores the possibility that improving glymphatic function might also help normalize brain metabolic homeostasis in migraine. Several limitations of this study should be acknowledged. First, the sample size, especially for the QSM and glymphatic analyses, was modest and may have limited our power to detect smaller effects or subgroup differences. Our findings will need replication in larger cohorts. Second, the cross-sectional design cannot establish causal relationships or the temporal dynamics of the observed changes. Longitudinal studies, e.g., following patients before and after migraine chronification or before and after aura development, would be valuable to determine whether the iron and glymphatic alterations are consequences or contributors to migraine progression. Third, while we controlled for age, depression, and anxiety in our analyses, the MO group had higher baseline anxiety/depression scores, which could independently influence brain function and structure. The extent to which the observed differences are due to migraine subtype per se versus these comorbidities is difficult to fully disentangle. Fourth, our multimodal MRI protocol, while comprehensive, has inherent technical challenges at 7T and requires combining data from slightly different sequence parameters in some cases; we mitigated this with careful preprocessing, but residual variance could remain. Fifth, we focused on specific ROI, which provides depth in those areas but does not capture whole-brain network interactions. Finally, while FDR correction was applied to the primary between-group comparisons, post hoc pairwise tests and clinical association analyses were reported at an uncorrected threshold and not further adjusted for multiple comparisons, which needs caution and warrant replication in larger samples. Conclusions In conclusion, while both MA and MO exhibit dysfunction in limbic and deep gray matter regions, they diverge in structural, iron, and glymphatic changes. MO showed broader reductions in ReHo and mild atrophy (especially in the amygdala), along with higher anxiety/depression levels, indicating that chronic pain stress negatively affects emotion-processing centers [85]. MA, in contrast, shared some functional abnormalities with MO but did not show the same degree of limbic atrophy; instead, MA patients demonstrated elevated amygdalar magnetic susceptibility and greater glymphatic impairment. This suggests that recurrent CSD events in aura may forge a pathophysiological link between migraine headache mechanisms, disrupted iron metabolism, and impaired clearance of neurotoxic byproducts, which is less prominent in MO [36]. These findings highlight neurobiological differences between migraine subtypes. Clinically, our results imply that the management of migraine could be improved by addressing these central disturbances. Interventions targeting limbic dysfunction (e.g., treating comorbid anxiety/depression or neuromodulation of emotional circuits) and enhancing glymphatic clearance (e.g., optimizing sleep or other methods to promote cerebrospinal fluid flow) may complement conventional pain therapies and provide more comprehensive relief for patients across different migraine subtypes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1