A potential neuroimaging biomarker for depression in migraine patients: imbalance in excitation and inhibition in the medial superior frontal cortex

Abstract

Background: Depression is a common psychological and behavioural symptom experienced by people with migraines. However, the underlying neural mechanisms of depressive symptoms in this population are unclear.

Methods: In this study, we investigated the neural mechanisms underlying excitation–inhibition imbalance in patients with migraine accompanied by depressive symptoms (MD) using the Hurst exponent. We used Analysis of covariance to compare differences in the Hurst exponent between patients with MD, patients with migraine accompanied by non-depressive symptoms (MnD), and healthy controls (HC). Spearman's correlation analysis was used to calculate the correlation between the Hurst exponent of different brain regions and clinical variables.

Results: Compared with patients with migraine accompanied by MnD, patients with MD showed an increased Hurst exponent in the left lingual gyrus and medial superior frontal cortex. Compared with HC, MD showed an increased Hurst exponent in the right cerebellum crus 1 and right lingual gyrus. Compared with HC, MnD showed an increased Hurst exponent in the left cerebellum crus 2 and medial superior frontal cortex. The Hurst exponent of the medial superior frontal cortex in patients with migraine was significantly positively correlated with depression scores (r = 0.3501, p = 0.0001).

Conclusions: Our study has identified imaging features of excitatory–inhibitory imbalance in MD. Imbalances in excitation and inhibition in the medial superior frontal cortex may serve as a promising candidate imaging marker for diagnosing and intervening in depressive symptoms in migraine.

Introduction

Migraine is a complex neurological disorder typically characterized by recurrent episodes of moderate-to-severe throbbing headache. Beyond pain, migraines are associated with adverse psychological outcomes, including an increased risk of depression.¹ Individuals with migraine are two to four times more likely to develop depression than those without migraine; this bidirectional relationship complicates clinical management and is associated with poorer prognosis in comorbid populations.²

An imbalance between excitation and inhibition (E/I) is considered a key pathophysiological feature of depression.³ Elevated spontaneous neuronal activity in the medial prefrontal cortex (mPFC) of women with depression has been linked to increased glutamate concentrations.⁴ Furthermore, ketamine treatment has been shown to reduce functional connectivity between the cerebellum and both the default mode network and the frontoparietal network during response inhibition tasks.⁵ Collectively, these findings suggest that disrupted inhibitory control between the mPFC and the cerebellum contributes to the pathophysiology of depression.

Neurotransmitters play a central role in maintaining the homeostasis of excitatory and inhibitory synaptic transmission.⁶ Glutamate promotes excitatory neuronal firing, whereas γ-aminobutyric acid (GABA) mediates inhibitory signaling, together sustaining a dynamic balance. Dysregulation of this system may result in excessive glutamatergic activity or insufficient GABAergic inhibition, thereby disrupting E/I balance.⁶ The midbrain–cortical system provides dopaminergic projections to the mPFC, a region involved in higher-order cognitive processes such as motor planning, attention, and behavioral inhibition.⁷ Dopamine exerts its primary inhibitory effects in the mPFC indirectly through the activation of local GABAergic interneurons.⁸

Magnetic resonance spectroscopy and electroencephalography are commonly used to assess E/I balance; however, findings across these modalities are often inconsistent, raising concerns regarding their reliability.⁹ The Hurst exponent has emerged as a novel metric for characterizing E/I balance. Both computational simulations of resting-state functional magnetic resonance imaging (rs-fMRI) signals and animal studies have demonstrated a robust relationship between E/I dynamics and the Hurst exponent.^10,11 Derived from the temporal autocorrelation properties of rs-fMRI time series, the Hurst exponent quantifies fractal characteristics of neural activity. Notably, it varies inversely with the E/I ratio: higher values indicate reduced excitability, whereas lower values reflect increased excitability.¹⁰

In the present study, we employed the Hurst exponent to quantify E/I balance in patients with migraine without depressive symptoms (MnD), patients with migraine with depressive symptoms (MD), and healthy controls (HC). We further examined correlations between regional Hurst exponent alterations and clinical variables. Finally, receiver operating characteristic (ROC) curve analysis was conducted to evaluate the utility of the Hurst exponent in distinguishing MD from MnD.

Experimental procedures

Patients

Between January 2025 and January 2026, the study was carried out. It comprised 77 HC, 68 patients with migraine and non-depressive symptoms (MnD), and 54 patients with migraine and depressive symptoms (Fig. 1). The MD group included 42 patients with migraine without aura, 6 patients with migraine with aura, and 6 patients with chronic migraine. The MnD group included 55 patients with migraine without aura, 8 patients with migraine with aura, and 5 patients with chronic migraine. Experienced neurologists screened the migraine patients, who were selected from a tertiary headache center. The International Classification of Headache Disorders, Third Edition served as the foundation for the diagnostic criteria.¹² The following requirements had to be met by the patients: (a) they had to be right-handed; (b) they had to be at least 18 years old; (c) they had to have migraine symptoms for more than 6 months; (d) they had not taken any psychiatric medications; and (e) they had to avoid headache episodes or acute migraine medication for at least 3 days before the fMRI scan.

Fig. 1.

Flowchart of participant enrollment.

The Visual Analogue Scale (VAS) was used to measure the intensity of headaches. Depression and anxiety were assessed using the Patient Health Questionnaire-9 (PHQ-9) and Generalized Anxiety Disorder-7 (GAD-7), respectively. Headache impact test-6 (HIT) was used to measure the impact of headaches on patients’ quality of life. The MD group consisted of participants who scored 10 or above on the PHQ-9, and the MnD group consisted of those who scored 4 or below.¹³ We selected PHQ-9 scores of ≥10 and ≤4 to maximize clinical contrast and avoid heterogeneity in subclinical symptoms. Our university's Institutional Review Board approved this study. Each subject provided written informed permission.

MRI scanning

A 3.0 Tesla magnetic resonance imaging scanner (Ingenia Elition, Philips Medical, Best, the Netherlands) was used to acquire the MRI images. Coronal T2-weighted imaging, 3D T1-weighted imaging, and rsfMRI were all part of the MRI procedure. Brain structural abnormalities were assessed by T2-weighed imaging. Functional images were obtained axially using a gradient-echo planar imaging sequence: TR = 2000 ms, TE = 30 ms, Multiband factor = 2, parallel imaging factor = 1.8, slice = 35, thickness = 3.2 mm, gap = 0 mm, field of view (FOV) = 256 mm × 256 mm, acquisition matrix = 80 × 80, FA = 90°, and total 200 volumes. 3D T1-weighted imaging was obtained using a three-dimensional turbo fast echo procedure. The parameters were as follows: thickness = 1 mm; gap = 0 mm; slices = 170; FA = 8°; TR/TE = 8.1/3.7 ms; slices = 170; FA = 8°; thickness = 1 mm; gap = 0 mm; acquisition matrix = 256 × 256; FOV = 256 mm × 256 mm. Participants were told to close their eyes and stay still throughout the scanning.

fMRI processing

The DPABI toolkit version 7.2 (rfmri.org) was used to preprocess fMRI scans. The first 10 time points are removed to reach magnetic field balance. The remaining data were subjected to slice timing and motion correction. Head movement parameters: translation less than 1.5 mm, rotation less than 1.5^o. Volumes with a mean framewise displacement greater than 0.3 mm were removed. The fMRI images were then co-registered to each person's high resolution T1 anatomical scan and normalized to the MNI152 template. The normalized image was smeared using a 6 mm full-width at half-maximum Gaussian kernel. Linear detrending were applied to the smoothed fMRI images. Finally, each voxel's time series was regressed to remove Friston-24 head motion artifacts, cerebrospinal fluid signals, and white matter signals.

Hurst exponent

The final preprocessed data were used to estimate the Hurst exponent. This was computed as a fractional integral process utilizing a time series model, a discrete wavelet transform, and maximum likelihood estimation. In particular, the Fractal Toolbox's bfn_mfin_ml.m function was utilized,^10,11 with the “filter” argument set to “haar” and the “lb” and “ub” parameters set to [1.5, 10] and [−0.5, 0], respectively.¹¹

Spatial correlation with neurotransmitter profiles

The aim was to investigate whether spatial variations in the Hurst exponent are associated with specific neurotransmitter profiles. The JuSpace Toolbox was used to analyse the spatial correlation between differences in SC-FC coupling values and the neurotransmitter atlas. We selected neurotransmitters associated with excitatory and inhibitory functions: metabotropic glutamate receptor 5 (mGluR5) and GABA, as well as dopamine D1, dopamine D2, and the dopamine transporter.¹⁴ To determine whether the observed patient-specific correlations deviated from a zero distribution, exact p-values were calculated based on 1000 permutations for each analysis. These p-values were then false discovery rate (FDR)-corrected based on the number of tests in each comparison.¹⁴

Statistical analysis

Statistical analysis was performed using MATLAB 2019b. The Kruskal–Wallis test was used to assess differences in age and duration among the three groups, while the χ² test was used to assess differences in gender distribution. The Mann–Whitney test was used to compare clinical scales across the three groups.

We used the Lilliefors test to verify the normality of the Hurst exponent. We employed ANCOVA to compare differences in the Hurst exponent and FC maps among the three groups. Post hoc analysis for comparisons between two groups used the least significant difference test. For statistical comparisons, we regressed covariates for gender, age, education level, and head movement. We used the Gaussian random field approach to perform multiple comparison correction on the z-maps derived from the ANCOVA, setting voxel-wise p-values below 0.005, and cluster-wise p-values below 0.05. We calculated the correlations between clinical variables in the MD and MnD groups. Spearman partial correlation analysis was performed to assess the relationships between the Hurst exponents of the abnoram brain regions and both demographic variables and biomarker assessments, with gender, age, education level, and head motion included as covariates (p < 0.05, FDR-corrected). We performed ROC curve analysis to evaluate the utility of the study results for clinical diagnosis, using leave-one-out cross-validation.

Ethical approval and consent to participate

This study was conducted in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration. The institutional review board at our university gave its approval to this project (approval number: 2024-K-137-01). All participants gave their informed, written consent.

Results

Neuropsychological results

Demographic and medical data are presented in Table 1. No significant differences were observed in terms of age (Kruskal–Wallis test: c² = 1.234, p = 0.166), gender (χ² test: c² = 1.910, p = 0.166), education (Kruskal–Wallis test: c² = 0.587, p = 0.745), or mean framewise displacement (Kruskal–Wallis test: c² = 0.831, p = 0.659) between the three groups. However, significant differences were observed between the MD and MnD groups in terms of PHQ-9 (z = 9.50, p < 0.001), GAD-7 (z = 6.08, p < 0.001), and HIT (z = 5.00, p < 0.001) scores. There are significant positive correlations between the PHQ-9 scores and GAD-7 (r = 0.4235, p = 0.0014) and HIT scores (r = 0.3353, p = 0.0132) in MD. There are significant positive correlation between the PHQ-9 scores and GAD-7 scores (r = 0.3431, p = 0.0042) and between the VAS scores and HIT scores (r = 0.8012, p < 0.0001) (Fig. 2) in MnD.

Fig. 2.

Correlation analysis of clinical scales. (A) There is a significant positive correlation between the PHQ-9 scores of MD and GAD-7 (r = 0.4235, p = 0.0014) and HIT scores (r = 0.3353, p = 0.0132). (B) There are significant positive correlation between the PHQ-9 scores of MnD and GAD-7 scores (r = 0.3431, p = 0.0042) and between the VAS scores of MnD patients and HIT scores (r = 0.8012, p < 0.0001).

Table 1.

Demographics and neuropsychological data.

	MD	MnD	HC	z/c2	p
Gender, n (M/F)	54 (19/35)	68 (25/43)	77 (28/49)	1.91	0.166
Age (years)	31.7 ± 8.0	33.6 ± 8.9	34.7 ± 9.3	1.23	0.166
Education (years)	16.3 ± 3.6	16.6 ± 3.7	16.7 ± 3.7	0.58	0.757
Duration (years)	9.1 ± 8.10	10.1 ± 9.1	–	1.05	0.106
Frequency (d/m)	8.1 ± 4.1	7.0 ± 3.1	–	2.40	0.082
VAS	6.8 ± 1.7	6.4 ± 1.6		−1.06	0.068
PHQ-9	13.8 ± 3.2	2.3 ± 1.5	–	9.50	<0.001
GAD-7	9.6 ± 4.8	2.4 ± 2.3	–	6.08	<0.001
HIT	58.4 ± 5.3	51.1 ± 5.0	–	6.50	<0.001
Mean FD	0.2 ± 0.1	0.1 ± 0.0	0.1 ± 0.0	0.83	0.659

Note: Data represent mean ± SD. MD: migraine accompanied by depressive symptoms; MnD: migraine accompanied by non-depressive symptoms; HC: healthy controls; d/m: day per month; VAS: visual analogue scale; PHQ-9: Patient Health Questionnaire-9; GAD-7: General Anxiety Disorder-7; HIT: Headache Impact Test-6; FD: framewise displacement.

Abnormal Hurst exponent in MD

The HC has a higher Hurst exponent in the parietal and frontal lobes, indicating relatively lower excitability. In the cerebellum and temporal lobes, the Hurst exponent is lower, indicating higher excitability (Fig. 3A; Table 2). The mean Hurst exponent in MnD indicates lower excitability within a narrower range in the frontal lobe, as well as higher excitability in the temporal lobe and cerebellum (Fig. 3A; Table 2). The mean Hurst exponent in MD is associated with more widespread reduced excitability in the frontal lobe, as well as increased excitability in the temporal lobe and cerebellum (Fig. 3A; Table 2). Compared with MnD, MD showed increased Hurst exponent in the left lingual gyrus and medial superior frontal cortex (Fig. 1B); compared with HC, MD showed increased Hurst exponent in the right cerebellum crus1 and right lingual gyrus. Compared with HC, MnD showed increased Hurst exponent in the left cerebellum crus2 and medial superior frontal cortex (Fig. 3B; Table 2). In patients with migraine, the Hurst exponent of the medial superior frontal cortex showed a significant positive correlation with PHQ-9 scores (r = 0.3501, p = 0.0001) (Fig. 3C). Weak but statistically significant positive correlations were also observed with GAD-7 (r = 0.2016, p = 0.0260) (Fig. 3D) and HIT (r = 0.1821, p = 0.0447) (Fig. 3E) scores.

Fig. 3.

Statistical analysis results of Hurst exponent in the MD, MnD, and HC groups. (A) The HC has a higher Hurst exponent in the parietal and frontal lobes, indicating relatively lower excitability. In the cerebellum and temporal lobes, the Hurst exponent is lower, indicating higher excitability. The mean Hurst exponent in MnD indicates lower excitability within a narrower range in the frontal lobe, as well as higher excitability in the temporal lobe and cerebellum. The mean Hurst exponent in MD is associated with more widespread reduced excitability in the frontal lobe, as well as increased excitability in the temporal lobe and cerebellum. (B) Compared with MnD, MD showed increased Hurst exponent in the left lingual gyrus and medial superior frontal cortex. Compared with HC, MD showed increased Hurst exponent in the right cerebellum crus1 and right lingual gyrus; compared with HC, MnD showed increased Hurst exponent in the left cerebellum crus2 and medial superior frontal cortex. (C) The Hurst exponent of the medial superior frontal cortex in patients with migraine was significantly positively correlated with PHQ-9 (r = 0.3501, p = 0.0001). (D) The Hurst exponent of the medial superior frontal cortex in patients with migraine was weak but significantly positively correlated with GAD-7 (r = 0.2016, p = 0.0260). (E) The Hurst exponent of the medial superior frontal cortex in patients with migraine was weak but significantly positively correlated with HIT (r = 0.1821, p = 0.0447).

Table 2.

Brain regions with significantly different Hurst exponent in the MD group compared with the MnD and HC groups.

			MNI coordinates
Brain regions	Voxels	BA	x	y	z	Z value
MD vs. MnD
Left lingual gyrus	45	19	−18	−54	−12	3.9225
Medial superior frontal cortex	35	9	3	42	48	4.0661
MD vs. HC
Right cerebellum crus1	37	5	42	−51	−27	4.2394
Right lingual gyrus	40	19	15	−72	−3	3.4311
MnD vs. HC
Left cerebellum crus2	36	5	−42	−57	−45	−4.2030
Medial superior frontal cortex	37	9	3	39	48	−4.0644

Note: MNI, Montreal Neurological Institute; BA, Brodmann area.

Correlation with neurotransmitter profiles

After spatial correlation analysis between Hurst exponent differences and neurotransmitter profiles, the Hurst exponent differences between MD and MnD shown the significantly correlation with D1 (r = 0.3081, p = 0.0150) (Fig. 4A). The Hurst exponent differences between MD and MnD shown the significantly correlation with D1 (r = −0.2617, p = 0.0159) (Fig. 4B) and DAT (r = −0.2780, p = 0.0159) (Fig. 4C).

Fig. 4.

Correlation with neurotransmitter profiles. (A) After spatial correlation analysis between Hurst exponent differences and neurotransmitter profiles, the Hurst exponent differences between MD and MnD shown the significantly correlation with D1 (r = 0.3081, p = 0.0150). (B) The Hurst exponent differences between MnD and HC shown the significantly correlation with D1 (r = −0.2617, p = 0.0159). (C) The Hurst exponent differences between MnD and HC shown the significantly correlation with DAT (r = −0.2780, p = 0.0159).

Receiver operating characteristic curve

The ROC curve shows that the Hurst exponent of the medial superior frontal cortex has an accuracy of 70% (95% CI: 69%–72%), a sensitivity of 74% (95% CI: 72%–75%), and a specificity of 62% (95% CI: 61%–64%) in distinguishing between MD and MnD (Fig. 5).

Fig. 5.

Receiver operating characteristic curve. The ROC curve shows that the Hurst exponent of the medial superior frontal cortex has an accuracy of 70%, a sensitivity of 74%, and a specificity of 62% in distinguishing between MD and MnD.

Discussion

To our knowledge, this is the first study to investigate the microstructural characteristics of excitation–inhibition (E/I) imbalance in patients with migraine accompanied by depressive symptoms. The brain regions showing E/I imbalance were primarily located in the cerebellum, medial superior frontal cortex, and lingual gyrus. Moreover, the E/I ratio in the medial prefrontal cortex was able to distinguish between MD and MnD.

Compared with healthy controls, MnD exhibited increased excitability in the medial superior frontal cortex and cerebellum, whereas MD showed reduced excitability in the medial superior frontal cortex. Previous studies have reported increased whole-brain functional connectivity in patients with migraine relative to healthy controls.¹⁵ Investigations of cortical reactivity also frequently indicate heightened cortical excitability in migraine.¹⁶ In one transcranial magnetic stimulation study, cortical excitability was assessed using magnetic suppression of the perceptual accuracy curve, and migraine patients demonstrated cortical hyperexcitability. Similarly, positron emission tomography revealed increased metabolism in the brainstem and reduced metabolism in the medial frontal, parietal, and somatosensory cortices.¹⁷ Dysregulation of activation and inhibition in these cortical regions may reflect dysfunction in inhibitory pathways, leading to cortical hyperexcitability and ultimately contributing to migraine onset.¹⁶

In depression, molecular and cellular alterations—including impaired NMDA receptor function, disrupted GABAergic interneuron signaling, epigenetic repression of genes involved in synaptic plasticity, and chronic neuroimmune activation—disturb synaptic gain and E/I balance, particularly within the prefrontal–hippocampal circuit.^18,19 The mPFC in depression is characterized by reduced volume, neuronal atrophy, and impaired excitatory synaptic density and function. Ketamine exerts rapid antidepressant effects by blocking NMDA receptors on GABAergic interneurons, thereby rapidly disinhibiting excitatory pyramidal neurons and promoting neuronal plasticity in the mPFC.²⁰

Migraine is associated with cortical hyperexcitability followed by exhaustion: previous studies have reported that repeated migraine attacks lead to metabolic exhaustion in prefrontal regions, manifesting as reduced metabolism on PET and altered functional connectivity.^16,17 The E/I imbalance in the mPFC is an intrinsic neurobiological feature of migraine, reflecting the long-term erosion of prefrontal inhibitory control caused by recurrent attacks.¹⁷ The finding of increased excitability in MnD aligns with this “exhaustion” model. Depressive symptoms lead to changes in MD neurotransmitters, promoting neuronal plasticity in the mPFC.²⁰ This alters the pre-existing E/I imbalance in the prefrontal cortex.

D1 and D2 dopamine receptors increase interneuronal excitability and thereby influence cortical network activity.²¹ Dopamine modulates reward processing and cognitive function through circuits involving the ventral tegmental area, prefrontal cortex, and nucleus accumbens. Abnormal receptor function within these circuits can disrupt the excitability of prefrontal pyramidal neurons.²² D2 antagonists can reverse glutamatergic overexcitation and reduced GABAergic inhibition, thereby modulating exercise-induced cortical excitability.²³

Significant correlations were observed between depression and anxiety scores in both the MD and MnD groups. People with migraines often experience symptoms of depression or anxiety, and these two conditions frequently coexist.²⁴ Depression scores on the MD were significantly correlated with headache-related disability scores, while pain scores on the MnD were also significantly correlated with these scores. Patients with traumatic brain injury-related headaches who exhibit depressive symptoms tend to have higher HIT scores than those without depression. Furthermore, HIT scores show a positive trend in relation to the severity of depression.²⁵ Following acupuncture treatment, migraine patients showed a significant reduction in VAS and HIT scores.²⁶ The MnD questionnaire primarily refers to the impact of pain on daily life, whereas the MD questionnaire is characterized by depressive symptoms taking the place of pain as the primary factor affecting daily life.

The Hurst exponent in the medial superior frontal cortex achieved an Area Under Curve (AUC) of 0.70 (sensitivity 74%, specificity 62%) for discriminating MD from MnD. A specificity of 62% yields a false-positive rate of 38%, potentially leading to unnecessary psychiatric referrals, patient anxiety, and inappropriate treatment. Given this limitation, the clinical translational value of this biomarker lies not in standalone use but as a prescreening or risk-stratification tool within a sequential testing paradigm.²⁷ Patients flagged by the Hurst exponent could undergo confirmatory clinical assessment (e.g., PHQ-9 interview or structured diagnostic interview), thereby mitigating false-positive burden while preserving objective neurobiological information. Multimodal integration strategies can be used to improve diagnostic performance.^28,29

This study has several limitations. First, this study defined MD (migraine + depression) solely by a PHQ-9 ≥ 10, while the PHQ-9 ≥ 10 threshold indicated moderate depressive symptoms; our findings may not fully generalize to patients with a formal diagnosis of major depressive disorder. Second, longitudinal studies are required to determine whether changes in E/I balance are consistent with the present results over time. In this study, we did not collect physiological noise, which may have a potential impact on the Hurst exponent. In future studies, physiological noise should be collected to assess its effect on the Hurst exponent. Third, our rsfMRI scan duration of 6 min 40 s (200 time points), while within the range validated for wavelet-based Hurst estimation, is shorter than ideal for fractal analysis. Longer scan durations (≥8–10 min) would further enhance the reliability of Hurst exponent estimates. Future studies should validate our findings with extended acquisition protocols. Finally, the neurotransmitter atlas used in this study was derived from healthy individuals rather than patients, which may introduce bias.

Conclusions

This study demonstrates that alterations in excitation–inhibition balance, indexed by the Hurst exponent, are present in migraine patients with depressive symptoms, particularly within the medial superior frontal cortex. These findings suggest that prefrontal E/I dysregulation may contribute to the neurobiological link between migraine and depression. While the discriminative performance is moderate, the Hurst exponent may represent a promising candidate imaging marker that warrants further validation in larger and longitudinal cohorts.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Author contributions

Conceptualization: XL, YL

Data curation: XL, FF, YC, KC, JC, YL

Formal analysis: FF, YC, KC, JC, YL

Investigation: XL, FF, YC, KC

Methodology: JC, YL

Project administration: XL, YL

Resources: FF, YC, KC, YL

Software: XL, YC

Supervision: XL, YL

Validation: XL, FF, YC, KC, JC, YL

Visualization: XL, FF, YC

Writing – original draft: XL

Writing – review & editing: XL, JC, YL

Funding information

The Key Project of Medical Science Research of Hebei Province (20200002) (to JM Cheng).