Impaired glymphatic function and advanced brain age gap in chronic migraine: a multimodal MRI study

Abstract

Objectives

This study investigated whether the coupling between cortical blood-oxygen-level-dependent (BOLD) signals and cerebrospinal fluid (CSF) flow, a physiologically motivated but indirect proxy related to glymphatic-associated dynamics, moderates the relationship between migraine disability (MIDAS) and brain age gap (BAG).

Methods

Patients with episodic migraine (EM), chronic migraine (CM), and healthy controls were recruited. BAG was estimated using a deep learning-based framework, and glymphatic-related hemodynamic-CSF dynamics were assessed using BOLD-CSF coupling. A moderation model, adjusted for sleep quality (PSQI) and biological covariates using HC3 robust standard errors, evaluated the statistical moderating role of BOLD-CSF coupling on the MIDAS-BAG association.

Results

A network-specific pattern of altered BOLD-CSF coupling was identified, with higher-order network abnormalities observed specifically in CM (p = 0.037). CM patients exhibited a significantly greater BAG than EM patients and healthy controls (p = 0.007). BOLD-CSF coupling significantly moderated the association between MIDAS and BAG (interaction β = -0.104, p = 0.035; Wald test p = 0.036). In individuals with relatively stronger coupling, MIDAS was more strongly associated with BAG (β = 0.066, p < 0.001), whereas this association was attenuated in those with weaker coupling (β = 0.035, p = 0.005).

Conclusions

Altered higher-order network BOLD-CSF coupling may represent a relevant imaging feature of migraine chronification. BOLD-CSF coupling was associated with interindividual differences in the relationship between migraine-related disability and BAG, independent of sleep quality. These findings support the potential relevance of glymphatic-related physiological processes in chronic migraine.

Supplementary Information

The online version contains supplementary material available at 10.1186/s10194-026-02384-1.

Introduction

Migraine chronification represents a profound neurobiological transition beyond an increase in headache frequency, involving extensive structural and functional remodeling [1]. While the cumulative burden of chronic migraine (CM) is associated with persistent alterations in brain morphology [2], the physiological mechanisms driving these “structural scars” remain poorly understood. Traditional morphometric analyses often focus on localized changes [3]; however, the concept of Brain Age offers a holistic measure of brain health by utilizing deep learning to predict biological age from MRI data [4]. The resulting brain age gap (BAG) has emerged as a sensitive biomarker for accelerated aging in CM, with recent evidence demonstrating an advanced BAG of 4.16 years in CM patients that is notably absent in episodic migraineurs (EM) [5]. Crucially, the lack of a direct correlation between BAG and headache frequency in existing reports suggests that the relationship between pain burden and structural aging may be non-linear and may be influenced by additional physiological processes.

A candidate process relevant to this relationship is the glymphatic system, which has been implicated in brain fluid transport and metabolite clearance. Current migraine research has primarily relied on the Diffusion Tensor Image Analysis along the Perivascular Space (DTI-ALPS) index to assess glymphatic-related alterations; however, the biological specificity of this metric remains under debate, as it may be influenced by white matter geometry rather than directly reflecting fluid dynamics [6]. As an alternative, the coupling between cortical blood-oxygen-level-dependent (BOLD) signals and ventricular cerebrospinal fluid (CSF) flow (BOLD-CSF coupling) has emerged as a physiologically motivated imaging marker associated with coordinated hemodynamic-CSF dynamics [7]. Prior work in sleep and neurodegenerative disorders suggests that this measure may be relevant to glymphatic-related physiology, although it does not directly quantify glymphatic clearance [8]. Physiologically, BOLD-CSF coupling is thought to reflect the temporal coordination between large-scale neural/hemodynamic fluctuations and compensatory CSF movement. Prior studies have shown that, particularly during non-rapid eye movement sleep, slow neural activity is followed by hemodynamic oscillations that are subsequently coupled with CSF inflow, a physiological state in which CSF-mediated waste clearance is believed to be enhanced. One plausible basis for this phenomenon is that changes in cerebral blood volume within the fixed intracranial space are accompanied by reciprocal CSF displacement, thereby linking hemodynamic oscillations to CSF dynamics. Because glymphatic transport depends on CSF movement and exchange with interstitial fluid, BOLD-CSF coupling may capture a physiological process relevant to glymphatic function, albeit indirectly [7].

Although BOLD-CSF coupling has been investigated in sleep physiology and neurodegenerative disorders, its application to migraine remains largely unexplored. Migraine provides a distinct disease context because it is characterized by recurrent neurovascular perturbation, altered sensory processing, and substantial clinical burden, particularly in chronic migraine. Applying BOLD-CSF coupling to migraine may therefore extend prior work by evaluating whether coordinated hemodynamic-CSF dynamics differ across migraine subtypes and are associated with structural brain aging. In this study, we examined whether network-level BOLD-CSF coupling alterations were differentially distributed across migraine subtypes and whether such alterations were associated with BAG.

We tested three related hypotheses: (1) functionally, BOLD-CSF coupling would show differential alterations across lower-order and higher-order networks, with higher-order network changes being more apparent in chronic migraine; (2) structurally, chronic migraine would be associated with an advanced BAG; and (3) statistically, BOLD-CSF coupling would moderate the association between migraine-related disability and BAG. Given the cross-sectional design and the indirect nature of the imaging marker, these hypotheses were intended to address associations rather than causal mechanisms.

Methods

Participants

This case-control neuroimaging study recruited patients with migraine from the Headache Clinic of the Second Affiliated Hospital and Yuying Children’s Hospital, Wenzhou Medical University, from April 2024 and is scheduled to continue until October 2025. Age- and sex-matched healthy controls with no personal or first-degree family history of primary headache disorders were recruited from the local community after screening by medical history interview and neurological examination. The protocol was approved by the Ethics Committee of the Second Affiliated Hospital of Wenzhou Medical University (2024-K-137-01) and complied with the Declaration of Helsinki, with written informed consent obtained from all participants. Inclusion criteria for migraine patients were: age between 16 and 65 years; right-handedness to minimize hemispheric variability in neuroimaging analysis; absence of prophylactic headache treatment for ≥ 3 months prior to enrollment to minimize pharmacological confounding; and the ability to undergo MRI scanning and clinical assessments; exclusion criteria for both groups included history of other neurological disorders or severe psychiatric illnesses, severe systemic diseases, contraindications to MRI, and headache attacks occurring within 48 h before or after the scheduled scan. Diagnoses of episodic or chronic migraine were established by experienced headache specialists (C.Y.G., F.F.W., L.Y.) according to the International Classification of Headache Disorders, 3rd edition [9]. To minimize pharmacological influences, only patients not receiving preventive migraine medication were enrolled, and all participants were required to discontinue analgesics for at least 48 h before scanning; if intolerable headache attacks occurred during this period and required analgesics, the participant was withdrawn from the study and provided appropriate symptomatic treatment. All subjects underwent comprehensive neurological examination and completed standardized headache-related questionnaires. To control for circadian regulation of glymphatic clearance, all MRI scans were acquired between 19:00 and 23:00. To ensure image acquisition during the interictal phase, absence of headache was confirmed on the scan day via direct interview and reconfirmed 48 h post-scan by telephone follow-up; data were excluded if headache occurred within this periscan window.

A total of 341 participants were initially enrolled. After eligibility assessment and imaging quality control, 38 participants were excluded from the final analysis: 8 due to white matter hyperintensities with a Fazekas grade > 1, 10 due to imaging quality issues, including motion artifacts or poor registration quality identified by visual inspection after T1 registration, and 20 due to migraine attacks occurring within 48 h after MRI scanning. Consequently, the final dataset comprised 303 participants, including 65 healthy controls, 200 patients with episodic migraine, and 38 patients with chronic migraine (Fig. S1).

Beyond the standard neurological examination, all participants completed a battery of validated instruments to capture the multidimensional burden of migraine, including the Migraine Disability Assessment (MIDAS) for quantifying lost productive time and the Headache Impact Test-6 (HIT-6) as a global measure of functional impact. Psychological symptom severity was evaluated using the Patient Health Questionnaire-9 (PHQ-9) for depression and the Generalized Anxiety Disorder-7 (GAD-7) for clinical anxiety levels. Furthermore, given the established role of sleep in glymphatic function and its circadian regulation, subjective sleep quality and disturbances were assessed using the Pittsburgh Sleep Quality Index (PSQI), while pain intensity during the interictal phase and the typical severity of attacks were quantified via a Visual Analogue Scale (VAS) ranging from 0 to 10.

MRI data acquisition

Imaging was performed on a 3.0T scanner (Ingenia Elition X, Philips Medical, Best, the Netherlands) equipped with a 32-channel head coil. Structural MRI: High-resolution T1-weighted images were acquired using a 3D turbo field echo (TFE) sequence (TR/TE = 8.1/3.7 ms; flip angle = 8°; FOV = 240 × 240 mm²; matrix = 240 × 240; 188 sagittal slices; voxel size = 1 × 1 × 1 mm³ [isotropic]; slice thickness = 1 mm). Resting-state fMRI: Functional images were obtained using a BOLD-sensitive EPI sequence (TR/TE = 2000/30 ms; flip angle = 90°; FOV = 256 × 256 mm²; matrix = 80 × 80; 56 axial slices; voxel size = 3.2 × 3.2 × 3.2 mm³; slice thickness = 3.2 mm; gap = 0 mm; multiband factor = 2). The resting-state scan lasted approximately 400 s.

Imaging preprocessing and BOLD-CSF coupling quantification

Registration and tissue segmentation

Structural and functional images were integrated using FSL (https://fsl.fmrib.ox.ac.uk/fsl/docs/index.html) and ANTs (https://github.com/ANTsX/ANTs). T1-weighted images were skull-stripped using FSL BET (f = 0.5), followed by ANTs SyN nonlinear registration to align structural data to native functional space. FSL FAST was then used for tissue segmentation, and individualized gray matter (GM) masks were generated with a probability threshold of 0.3. All cases underwent visual inspection to verify the quality of skull stripping, tissue segmentation, and structural-to-functional registration before signal extraction. To ensure signal extraction precision, the Schaefer 400-parcel atlas was inverse-mapped from MNI space to the native functional space and further constrained by the GM mask [10]. This refinement ensured that subsequent BOLD signals were derived exclusively from anatomically verified gray matter voxels.

The spatial alignment of the transformed gray matter masks with the preprocessed functional images was verified through visual inspection. The voxel-wise BOLD (vBOLD) signal was extracted from each voxel within the gray matter mask, while the global BOLD (gBOLD) signal was calculated by averaging the time series of all voxels within the mask [11].

CSF masks were positioned on the bottom slices of the Resting-state functional MRI (rs-fMRI) images in individual functional space (Fig. 1). Visual inspection confirmed that these bottom slices consistently included the cerebellum for all subjects, thereby enabling the capture of the CSF cross-slice inflow effect as previously described [7, 10]. The CSF regions were manually selected for each subject.

Fig. 1.

Methodological framework for BOLD and CSF signal extraction. (Top) Spatial localization of the gray matter (GM) and CSF regions of interest. The individualized GM mask (purple overlay) was segmented from T1-weighted images and aligned to the native functional space. The CSF mask (white) was manually positioned on the most inferior slices of the rs-fMRI acquisition to capture the cross-slice inflow effect, as these slices consistently included the cerebellum and surrounding cisterns. (Bottom) Representative time series from a single participant showing the cortical gray matter BOLD signal (blue line) and the average CSF signal (orange line). Both signals have been z-score normalized to illustrate their corresponding amplitude fluctuations and temporal synchrony

Quantifying the coupling between BOLD signal and CSF inflow

The coupling between cerebrospinal fluid (CSF) flow and hemodynamic signals (gBOLD and vBOLD) was evaluated using cross-correlation analysis [11]. For gBOLD-CSF coupling, we calculated the cross-correlation function between the gBOLD signal and the CSF signal across various time lags (ranging from − 10 to 10 s for each subject) during the rs-fMRI scans. The negative correlation coefficient peaked at a lag of + 2 s, indicating that the interaction between hemodynamic signals and CSF dynamics was most significant at this time point at the group level. This relationship was further validated using the negative derivative of the BOLD signal [7]. In the vBOLD-CSF analysis, we compared coupling differences between lower-order networks, including the sensorimotor network (SMN) and visual network (VN), and higher-order networks, including the default mode network (DMN) and frontoparietal network (FPN) [12] across different disease states.

Brain age estimation and feature extraction

Structural MRI preprocessing

Structural T1-weighted images were processed using the automated recon-all pipeline in FreeSurfer (version 8.0.0) (http://surfer.nmr.mgh.harvard.edu/). This standard procedure includes motion correction, intensity normalization, brain extraction, and automated tissue segmentation. Each processed scan underwent rigorous visual inspection to ensure the accuracy of cortical surface reconstruction and subcortical segmentation.

We extracted a total of 150 morphometric features to characterize individual brain morphology. Cortical parcellation was performed according to the Desikan-Killiany (DK) atlas [13], from which cortical thickness (n = 68) and cortical surface area (n = 68) were derived across both hemispheres. Additionally, the volumes of 14 subcortical structures were obtained based on the Aseg atlas [14].

Brain age prediction model

Individual biological brain age was estimated using a previously validated and publicly available pre-trained brain-age model developed by Yu et al. [4]. This model was trained on a large lifespan sample of 35,683 healthy individuals aged 5–90 years and further evaluated in independent datasets for cross-dataset generalizability and longitudinal consistency. In the present study, we applied this externally trained model directly to our cohort without retraining. Brain-age prediction was based on 150 structural morphometric features derived from T1-weighted MRI, including 68 cortical thickness measures, 68 cortical surface area measures, and 14 subcortical volume measures. Our cohort age range (16–65 years) fell within the training range of the original model, supporting the applicability of the model to the current sample. For each participant, the BAG was calculated as the predicted brain age minus chronological age, with positive values indicating an older-appearing brain and negative values indicating a younger-appearing brain.

To reduce age-related bias in brain-age estimation, we applied a linear regression-based bias correction procedure [15]. Because the model was trained externally and transferred to a single-center clinical cohort, residual domain effects related to cohort composition, scanner platform, and acquisition protocol cannot be fully excluded. Therefore, BAG was interpreted as a global structural aging marker derived from a validated external model.

Moderation of gBOLD-CSF coupling on brain age gap

We performed a moderation analysis to investigate whether the relationship between migraine burden (measured by the MIDAS) and BAG was contingent upon the strength of gBOLD-CSF coupling. All statistical analyses were conducted in Python using the statsmodels and scipy libraries.

Model specification

To ensure the robustness of our findings, we included Sex (as a categorical factor), total intracranial volume (TIV), and PSQI as covariates in the model. Sex was treated as a dummy-coded categorical variable, while all continuous variables (MIDAS, BOLD-CSF, TIV, and PSQI) were mean-centered to reduce multicollinearity and improve the interpretability of the interaction term. The regression equation was defined as follows:

Robust inference and diagnostics

Given the observed non-normality of residuals (Shapiro-Wilk p < 0.05) and potential heteroscedasticity, we employed HC3 robust standard errors (Heteroscedasticity-Consistent standard errors) for all parameter estimations. This approach provides more conservative and reliable p-values and confidence intervals when OLS assumptions are partially violated.

To evaluate the overall contribution of the moderation effect, we utilized a Wald test to compare the interaction model against a restricted model. Multicollinearity was assessed using the Variance Inflation Factor (VIF), with a threshold of < 5 indicating no significant collinearity issues.

Simple slope analysis

To further interpret the interaction, simple slope analyses were performed at the 25th and 75th percentiles of the moderator. Because BOLD-CSF coupling was quantified as a negative cross-correlation coefficient, with more negative values indicating stronger coupling, the 25th and 75th percentiles were defined as relatively stronger and weaker coupling, respectively.

Results

Demographic and clinical characteristics

The demographic and clinical characteristics of the participants are summarized in Table 1. A total of 303 individuals were included in the final analysis, including 38 patients with CM, 200 patients with EM, and 65 healthy controls (HC). No significant group differences were observed in age (p = 0.058) or sex distribution (p = 0.091), suggesting that these factors were unlikely to confound the subsequent neuroimaging analyses. As expected, headache frequency was significantly higher in the CM group than in the EM group (p < 0.001). Similarly, CM patients had significantly higher HIT-6 and MIDAS scores than EM patients (both p < 0.001), indicating greater disease burden and disability. No significant between-group difference was found in VAS score (p = 0.208). Significant group differences were observed across all psychological and sleep-related measures (all p < 0.001). CM patients reported the highest levels of depressive symptoms (PHQ-9) and sleep disturbance (PSQI). Both migraine groups had higher anxiety scores (GAD-7) than healthy controls, with the greatest burden observed in the CM group.

Table 1.

Demographic and clinical characteristics of participants

Age	CM(n = 38)	EM(n = 200)	HC(n = 65)	Statistic	p
	38.15(10.89)	33.70(9.71)	36.56(11.09)	5.69	0.058
Sex(Female)	33 (86.84)	170 (85.00)	48 (73.85)	4.78	0.091
VAS	7.00(1.75)	6.00(3.00)		1.59	0.208
Frequency	20.00(15.00)	3.00(2.00)		92.89	< 0.001
PHQ-9	7.00(4.00)	5.00(6.00)	1.00(4.00)	43.81	< 0.001
GAD-7	4.50(7.50)	5.00(6.00)	1.00(3.00)	40.69	< 0.001
PSQI	9.00(4.75)	7.00(5.00)	5.00(4.00)	30.37	< 0.001
HIT-6	66.00(6.75)	53.00(7.00)		81.00	< 0.001
MIDAS	54.00(51.75)	9.00(9.00)		67.43	< 0.001

Data are presented as median (IQR)

CM: Chronic Migraine; EM: Episodic Migraine; HC: Healthy Controls.VAS: Visual Analogue Scale; PHQ-9: Patient Health Questionnaire-9; GAD-7: Generalized Anxiety Disorder-7; PSQI: Pittsburgh Sleep Quality Index; HIT-6: Headache Impact Test-6; MIDAS: Migraine Disability Assessment.P-value: Calculated by Kruskal-Wallis test for comparison among the three groups. p < 0.05 denotes statistical significance

Extraction of CSF inflow and BOLD signals

The aggregate gBOLD-CSF cross-correlation function demonstrated a robust negative peak at a +2s lag across all subjects (Fig. 2A). This timing was consistently observed at the subgroup level (Fig. 2C). Validation using the negative derivative of the gBOLD signal further confirmed the physiological stability of this coupling, showing a zero-crossing at + 2 s (Fig. 2B). Consequently, the cross-correlation strength at + 2.0 s was utilized as the quantitative metric for subsequent analyses.

Fig. 2.

Characterization and comparison of global BOLD-CSF coupling. (A) Mean cross-correlation function between the gBOLD signal and CSF inflow across various time lags (− 10 to 10 s) for all participants. The peak negative correlation occurs at a lag of + 2 s (red dot), representing the maximal physiological coupling between hemodynamic and CSF dynamics. (B) Validation of the coupling using the negative derivative of the gBOLD signal (-d/dt gBOLD) and CSF signal. The cross-correlation curve and the zero-crossing at + 2 s further support the BOLD-CSF coupling mechanism. (C) Group-level cross-correlation curves for healthy controls (HC, blue line), episodic migraine (EM, green dashed line), and chronic migraine (CM, pink dashed line). Shaded areas represent the standard error of the mean. (D) Quantitative comparison of gBOLD-CSF coupling strength at the + 2 s lag. Both EM and CM patients exhibit significantly attenuated coupling compared to HC (p = 0.0141). Statistical analyses were performed using a general linear model adjusted for age and sex. Asterisks (*) indicate significant group differences at p < 0.05

Altered BOLD-CSF coupling across migraine subtypes

A General Linear Model (GLM), adjusted for age and sex, revealed significant inter-group differences in gBOLD-CSF coupling (p = 0.0141). Post-hoc analyses showed that HCs exhibited significantly stronger coupling than both EM and CM patients (FDR-p < 0.05).

Regarding the comparison of average network coupling strength, a significant main effect of group was observed in low-order networks (p = 0.0258; Fig. 3A). Post-hoc comparisons revealed that both EM and CM patients exhibited significantly attenuated vBOLD-CSF coupling, manifesting as a shift toward less negative values relative to healthy controls (both p < 0.05). In high-order networks, a significant overall group effect was also observed (p = 0.0371; Fig. 3B). However, post hoc analyses showed that only CM patients had significantly reduced coupling magnitude relative to healthy controls (p < 0.05), whereas the EM group did not differ significantly from healthy controls.

Fig. 3.

Network-level comparison of BOLD-CSF coupling across migraine subtypes. (A) BOLD-CSF coupling strength within low-order networks across HC, EM, and CM groups. A significant overall group difference is observed (Overall p = 0.0258, adjusted for age and sex), with both EM and CM patients exhibiting significantly attenuated coupling compared to HC. (B) BOLD-CSF coupling strength within high-order networks across the three groups. Statistical analysis reveals significant group-level alterations (Overall p = 0.0371, adjusted for age and sex). This spatial pattern suggests differential involvement of lower-order and higher-order networks, with higher-order network BOLD-CSF coupling alterations being more apparent in the chronic state. For all plots, box boundaries represent the interquartile range (IQR), and the horizontal line indicates the median. Asterisks (*) denote significant pairwise differences (p < 0.05) relative to the HC group

Accelerated brain aging in chronic migraine

After controlling for total intracranial volume (TIV) and sex, a General Linear Model (GLM) was employed to assess inter-group differences. Significant differences were observed in the BAG across the three groups (p = 0.00732; Fig. 4). Specifically, post-hoc analyses revealed that patients with CM exhibited a significantly elevated BAG relative to both HCs (p < 0.05) and patients with EM (p < 0.01), whereas no significant disparity was found between the EM and HC cohorts.

Fig. 4.

Accelerated structural brain aging in chronic migraine. Comparison of the BAG among HC (blue), EM (green), and CM (pink). Results indicate that the CM group manifests a significantly advanced BAG compared to both the HC group (p < 0.05) and the EM group (p < 0.01). Notably, no significant difference (ns) was observed between the HC and EM groups. Statistical analyses were performed using a general linear model adjusted for total intracranial volume (TIV) and sex (Overall p = 0.00732). Box boundaries denote the interquartile range (IQR), the horizontal bar represents the median, and individual data points are overlaid to show the distribution

Moderation of BOLD-CSF coupling on the association between MIDAS and BAG

Moderation analysis revealed a significant interaction between MIDAS and BOLD-CSF coupling in their association with BAG (interaction β = -0.104, p = 0.035). A Wald test further indicated that inclusion of the interaction term significantly improved model fit (F = 4.432, p = 0.036), after adjustment for sex, TIV, and PSQI.

Simple slope analysis showed that the association between MIDAS and BAG differed according to the level of BOLD-CSF coupling (Fig. 5). Specifically, in individuals with relatively stronger coupling, higher MIDAS scores were more strongly associated with increased BAG (β = 0.066, p < 0.001). In those with weaker coupling, this relationship remained significant but was less pronounced (β = 0.035, p = 0.005). These findings suggest that BOLD-CSF coupling was associated with differences in the strength of the relationship between migraine-related disability and BAG.

Fig. 5.

Moderating effect of BOLD-CSF coupling on the relationship between MIDAS and BAG. Regression lines illustrate the simple slopes of MIDAS predicting BAG at different levels of BOLD-CSF coupling: strong coupling (blue, 25th percentile) and weak coupling (red, 75th percentile). Individual data points are color-coded by their respective BOLD-CSF coupling levels (blue for values below the median, red for values above the median). Shaded regions indicate the 95% confidence intervals estimated using HC3 robust standard errors. The regression model is adjusted for TIV, sex, and PSQI. Statistical results for simple slopes, including regression coefficients (β values) and p-values, are provided in the legend

Discussion

This study identified significant associations between BOLD-CSF coupling, an indirect imaging marker related to coordinated hemodynamic-CSF dynamics, and structural brain aging in migraine. Three main findings emerged. First, BOLD-CSF coupling showed a network-specific pattern of alteration, with lower-order sensory networks affected across migraine subtypes and higher-order network abnormalities more evident in chronic migraine. Second, patients with chronic migraine exhibited a significantly advanced BAG relative to episodic migraine and healthy controls. Third, BOLD-CSF coupling moderated the association between migraine-related disability and BAG, such that the strength of this association differed across levels of coupling. Taken together, these findings support a relationship between BOLD-CSF coupling abnormalities and chronic migraine, while remaining associative in nature given the cross-sectional design.

A strength of the present study is the use of BOLD-CSF coupling as a dynamic imaging metric that may capture physiologically relevant hemodynamic-CSF coordination, offering a complementary perspective to the more commonly used DTI-ALPS index [16]. While the DTI-ALPS index is widely utilized due to its technical accessibility, its biological specificity as a direct marker of glymphatic clearance remains a subject of ongoing debate [17–19]. Emerging evidence suggests that the ALPS index may be sensitive to age-related variations in white matter microstructure [20], functioning as a composite metric that reflects both potential glymphatic orientation and underlying structural integrity. Furthermore, as a static diffusion-based measure, DTI-ALPS cannot capture the temporal dynamics of hemodynamic and CSF oscillations that may be relevant to metabolite clearance.

In contrast, BOLD-CSF coupling evaluates the functional synchronization between neural-driven hemodynamic shifts and reciprocal CSF oscillations [7, 21]. In migraine, a disorder associated with neurovascular and metabolic disturbances, this dynamic measure may provide additional information beyond static diffusion-based indices. The present study extends prior BOLD-CSF coupling research by applying this dynamic metric to migraine and by showing that BOLD-CSF coupling alterations were not uniform across migraine subtypes, but instead showed a network-specific pattern that was more evident in chronic migraine. Moreover, by linking BOLD-CSF coupling to BAG and testing its statistical moderating role in the MIDAS-BAG association, this study moves beyond disease-control comparisons and provides an association-based framework for understanding interindividual differences in the relationship between migraine-related disability and structural brain aging [11]. However, BOLD-CSF coupling should still be interpreted as an indirect imaging marker rather than a direct measure of glymphatic clearance [22].

The spatial distribution of BOLD-CSF coupling alterations suggests differential network involvement across migraine subtypes [23, 24]. The presence of lower-order sensory network abnormalities in both EM and CM may reflect broader sensitivity of sensory systems to repeated nociceptive input. In contrast, the more selective involvement of higher-order networks in CM may indicate that chronic migraine is associated with more widespread disruption extending beyond sensory systems. This interpretation is also consistent with previous reports of large-scale functional reorganization in chronic migraine [25].

One possible explanation for this pattern is that higher-order networks such as the DMN and FPN are metabolically demanding and play central roles in large-scale brain integration [26–28]. As a result, altered coupling in these networks may be relevant to the clinical and physiological characteristics of CM [29, 30]. Their preferential involvement in CM may also be associated with the advanced BAG observed in this group, although this interpretation remains speculative and should be tested in longitudinal studies.

The observation of a significantly advanced BAG in the CM group suggests that chronic migraine is associated with a more pronounced global structural aging phenotype than episodic migraine [2, 5, 31]. In contrast, the EM group showed a BAG comparable to that of healthy controls. These cross-sectional findings indicate subtype-related differences in structural brain aging, but they do not establish whether these differences precede, follow, or co-occur with chronic migraine.

Beyond the observed spatial patterns, an important finding of this study was that BOLD-CSF coupling statistically moderated the association between MIDAS and BAG. This effect remained significant after adjustment for PSQI, suggesting that the interaction was not fully explained by differences in subjective sleep quality. Specifically, the association between migraine-related disability and BAG was stronger in individuals with relatively stronger coupling (β = 0.066, p < 0.001), whereas this association was attenuated but remained significant in individuals with relatively weaker coupling (β = 0.035, p = 0.005). These findings suggest that the relationship between clinical burden and structural brain aging may differ across levels of hemodynamic-CSF coupling [8, 32]. The observed moderation effect may provide a useful framework for future studies examining interindividual differences in structural brain aging in migraine. If replicated in longitudinal cohorts, this pattern may help determine whether BOLD-CSF coupling has prognostic or stratification value. In contrast, the attenuated MIDAS-BAG association observed in the weaker-coupling group suggests that the clinical-structural relationship may not be uniform across individuals. This finding should not be interpreted as evidence that symptom-focused treatment loses value, but rather as an indication that the relationship between migraine-related disability and structural brain aging is more complex than a simple linear model. Future longitudinal studies with direct physiological monitoring are needed to determine the temporal and biological relevance of BOLD-CSF coupling in migraine [28].

Several limitations should be acknowledged. First, the cross-sectional design precludes causal inference; therefore, the observed associations cannot determine whether altered BOLD-CSF coupling precedes, follows, or co-occurs with structural brain aging in migraine. Second, although BOLD-CSF coupling is a physiologically motivated imaging marker, it remains an indirect proxy and does not directly quantify glymphatic clearance. Its interpretation may also be influenced by multiple physiological factors, including respiration, cardiac pulsatility, vascular dynamics, CO₂ fluctuations, and arousal state. Third, the resting-state fMRI sequence used in this study had a TR of 2 s, which limits the ability to resolve faster physiological components related to glymphatic dynamics, including cardiac- and respiratory-driven CSF pulsations described by Kiviniemi et al. [33]. Fourth, although the CM-to-EM ratio reflects clinical prevalence, the relatively modest CM sample size may have limited statistical power, particularly for interaction analyses. Fifth, the brain age model was externally trained and then applied to our cohort; although this approach improves methodological consistency, residual cohort- or scanner-related domain effects cannot be entirely excluded. Finally, while BAG is a useful global marker of structural brain aging, it may not fully capture localized migraine-related morphological abnormalities. Future longitudinal, multi-center studies with optimized physiological monitoring are needed to validate and extend these findings.

Conclusion

In summary, this study identified a network-level pattern of BOLD-CSF coupling alterations in migraine, with higher-order network abnormalities being particularly evident in chronic migraine and accompanying an advanced brain age gap. In addition, BOLD-CSF coupling significantly moderated the association between migraine-related disability and BAG, suggesting that the relationship between cumulative clinical burden and structural brain aging may vary across physiological states. Together, these findings support the potential relevance of glymphatic-related hemodynamic-CSF dynamics to chronic migraine and structural brain aging. Given the indirect nature of BOLD-CSF coupling and the cross-sectional design, these results should be interpreted cautiously and regarded as a basis for future longitudinal and mechanistic studies.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CM

Chronic migraine

EM

Episodic migraine

HC

Healthy controls

BAG

Brain Age Gap

MRI

Magnetic Resonance Imaging

fMRI

Functional MRI

rs-fMRI

Resting-state functional MRI

EPI

Echo-Planar Imaging

T1w

T1-weighted

TFE

Turbo Field Echo

TR

Repetition Time

TE

Echo Time

FOV

Field Of View

MNI

Montreal Neurological Institute

FSL

FMRIB Software Library

ANTs

Advanced Normalization Tools

BET

Brain Extraction Tool

FAST

FMRIB’s Automated Segmentation Tool

GM

Gray matter

DK

Desikan-Killiany

Aseg

Atlas of Subcortical Segmentation

BOLD

Blood-oxygen-level-dependent

CSF

Cerebrospinal fluid

vBOLD

Voxel-wise BOLD

gBOLD

Global BOLD

DTI-ALPS

Diffusion Tensor Imaging Analysis Along The Perivascular Space

SMN

Sensorimotor network

VN

Visual network

DMN

Default mode network

FPN

Frontoparietal network

GLM

General Linear Model

TIV

Total intracranial volume

FDR

False Discovery Rate

IQR

Interquartile range

VAS

Visual Analogue Scale

HIT-6

Headache Impact Test-6

MIDAS

Migraine Disability Assessment

PHQ-9

Patient Health Questionnaire-9

GAD-7

Generalized Anxiety Disorder-7

PSQI

Pittsburgh Sleep Quality Index

Author contributions

Y.L and X.L: Conceptualization, supervision, funding acquisition, and project administration. C.D. and F.F.: Methodology, data curation, formal analysis, and writing-original draft. X. L., Z.T., X.Z., Y.G. and X.L.: Investigation, validation, and software. C.L., M.Y, M.H., Y.H., and K.L: Resources, visualization, and data collection. Y.C., K.C, and B.H.: Writing-review & editing, and technical support.

Funding

This study was funded by the Health Commission of Zhejiang Province (2024KY1275 to YL).

Data availability

For reasonable requests, please contact the corresponding author to obtain all data related to this study.

Declarations

Ethics approval and consent to participate

Our study was approved by the Ethics Committee of the Second Affiliated Hospital of Wenzhou Medical University (ethics approval No. 2024-K-137-01). Written informed consent was obtained from all participants prior to their inclusion in the study.

Consent for publication

All authors consent to the publication.

Competing interests

The authors declare no competing interests.