Patients with EM and CM aged 18 to 60 years admitted to the headache clinic of the Department of Neurology, Brain Hospital of Nanjing Medical University (Nanjing, Jiangsu, China) were included for this study. The patients conformed to the International Classification of Headache Disorders, Third Edition, Migraine Diagnostic Guidelines [16]. The patients were classified according to the frequency of headaches. EM was classified as people having fewer than 15 headache days monthly. In contrast, CM was classified as having more than 15 days of headaches per month, lasting more than 3 months, and having migraine characteristics at least 8 days per month [17]. The selected patients were right-handed, and none received migraine prophylaxis. Patients with MwoA and no history of medication overuse were chosen for this study. Patients with psychiatric symptoms, or those taking medications that affect the central nervous system, were unable to undergo MEG examinations, and those who were uncooperative were excluded. The control group consisted of age-range-matched healthy individuals without headaches or a family history of headaches. This study was approved by the Institutional Review Board of Nanjing Brain Hospital (Batch number: 2016-KY023). Informed consent forms were signed by all participants.

In this study, information on migraine was collected from the study population by trained neurology professionals, including the time of migraine onset, monthly headache attack frequency, headache duration, headache severity, and medication use of the patients. Scores for the Hamilton Anxiety Rating Scale (HAMA) and the Hamilton Depression Rating Scale (HAMD) were obtained for all included individuals. Visual Analog Scale (VAS) scoring was performed for CM and EM patients.

Patients with MwoA underwent resting state MEG recording during the interictal period, with the requirement of the subject being migraine attack-free 48 hours before and 24 hours after MEG data collection. If this was not the case, a new MEG examination was required.

Data acquisition was performed using a whole scalp CTF 275-channel MEG system (VSM MedTech Systems, Inc., Coquitlam, BC, Canada) in a magnetically shielded room in the MEG room of Nanjing Brain Hospital. Before data collection, each subject was checked thoroughly to ensure no metal objects were on their body. We placed a coil at the base of the nose and in front of both ears as an anatomical marker for accurate integration with the cranial magnetic resonance imaging (MRI). First, the 3-minute empty chamber recording was performed, followed by MEG data recording. This would facilitate better capturing of the sensor and background noise and be used to determine the noise covariance for source analysis. The sampling frequency of MEG data was 6000 Hz, and each group of MEG data acquisition time was 120 s. A total of six groups were used for data acquisition, and at the same time, data collection underwent noise reduction. Subjects were required to remain supine, relaxed, awake, and with their eyes closed during MEG data collection. The subject’s head movement distance was not to exceed 5 mm before and after data collection, which would lead to the rejection of the collected data. If the patient fell asleep, the data was rejected, requiring data re-collection.

All patients enrolled were subjected to a 3.0 T head MRI scan (Siemens, Munich, Germany). Three-dimensional T1-weighted anatomical images were acquired using a fast gradient echo sequence (repetition time/echo time (TR/TE) = 1900/2.48 ms). The parameters were as follows: field of view, 250 $\times$ 250 mm; flip angle, 9°; voxel size, 0.48 $\times$ 0.48 $\times$ 1 mm${}^3$; and matrix dimensions, 512 $\times$ 512. Three localization coils placed during MEG data acquisition were used to match with the cranial MRI in order to determine the anatomical location of the MRI presentation.

We performed the following operations to eliminate environmental artifacts and non-brain activity signals from the MEG data: (1) manual removal of artifacts due to head position shifts or noise interference; (2) filtering of signals at 50 Hz and its harmonics to eliminate power line contamination; and (3) to capture sensor and ambient noise, MEG recording started with a 3-minute empty chamber recording, leading to noise covariance calculation for offline source analysis to adjust for the residual and stationary instrument, ambient, and sensor noise components. A cortical source analysis model was obtained using the FreeSurfer image (Version 7.11, Athinoula A. Martinos Center for Biomedical Imaging, Boston, MA, USA) analysis package to reconstruct T1-weighted structural body images in the surface model automatically. The program performs reconstruction work of the scalp, brain white matter, and gray matter and performs stereoscopic image analysis of the cerebral surface to assess the borders of the white and gray matter. A successive period of 60 s was chosen for the study to avoid peak discharge interferences on the magnetoencephalographic signal. Six frequency bands were selected for MEG analysis in this study: delta (2–4 Hz), theta (5–7 Hz), alpha (8–12 Hz), beta (15–29 Hz), gamma 1 (30–59 Hz), and gamma 2 (60–90 Hz).

Depth-weighted minimum parametric estimation was used for assessing source-level-based cortical activation in the MEG data. This method has proved relatively stable in previous studies [18]. A forward model for the minimum norm estimation (MNE) analysis was developed using the overlapping spheres method, describing each cortical vertex as a current dipole with approximately 15,000 vertices. Then, the distribution of the current sources was estimated using the following inverse operators: (1) the direction of current sources was restricted to be positive to the cortical surface; (2) a depth-weighting algorithm compensated for the inhomogeneous sensitivity of the current direction and depth; and (3) to minimize numerical instability, we used the regularization parameter $\lambda$${}^2$ = 0.33, which reduces the sensitivity of the MNE to noise and obtains the spatially smoothed solution, the inverse of the signal-to-noise ratio of the MEG record. Brainstorm software (http://neuroimage.usc.edu/brainstorm) was used to analyze the depth-weighted MNE, downloaded freely using the GNU (https://www.gnu.org/gnu/about-gnu.en.html#f1) General Public License.

Twelve brain regions associated with DMN were selected, including the bilateral IPC, MPFC, MTC, PCU, PCC, and LTC. We used the spm_maff8 function, which is derived from SPM12’s Brainstorm (The University of Southern California, Los Angeles, Los Angeles, CA, USA; The Hôpital de la Salpêtrière, Paris, France; Los Alamos National Laboratory, Los Alamos, NM, USA), to register each subject’s T1 MRI to the Montreal Neurological Institute (MNI) coordinate system, computed the 4 $\times$ 4 affine transform automatically, and in the T1 template body defined regions of interest (ROIs) using the Desikan-Killiany cortical atlas. The spectral power of the sources was estimated by calculating the average current density at each vertex within the ROI. The power spectral density (PSD) on every ROI derived using the Welch technique (window duration of 5 s with 50% overlap). The PSD value is the ratio of spectral power of every ROI chosen to the total power of entire spectrum over every frequency bin: Relative PSD (f) = PSD (f)/i [total PSD (fi)], where fi refers to the single frequency from the absolute PSD. This procedure normalizes PSD values across subjects’ brain regions [19].

Corrected amplitude envelope correlation (AEC-c) analysis was used to assess the choppy functional connectivity of the ROI. Several studies have shown that AEC-c analysis is highly stable and reproducible in FC network studies [20, 21]. Based on the approach reported in previous studies, the signal pairs were orthogonalized before envelope calculation to remove pseudo-connections resulting from field diffusion and volume conduction effects [22]. The amplitude envelope is the absolute value of the Hilbert transform of a given cortex, which reflects the amplitude fluctuation with time and can be obtained from the bandpass-filtered cortex source activity for each frequency band [21]. If s(t) is a random time segment, the Hilbert transform is defined as:

$${\mathrm{S}}_H(t)=\frac{1}{\pi }\underset{-\mathrm{\infty }}{\overset{+\mathrm{\infty }}{\int }}\frac{\mathrm{s}(\tau )}{(t-\tau )}d\tau$$

The Hilbert envelope was then split into equal time units, and the mean values of the envelopes within every time window were obtained. The FC metric was determined based on the Pearson correlation between the mean values and indicated by AEC-c values. The amplitude envelope of cortical oscillatory activity that correlates the two ROIs is measured as the AEC-c value. If the AEC-c values are high, the amplitude envelope synchronization fluctuations are strong. The AEC-c values were calculated and the 12 $\times$ 12 adjacency matrix was estimated.

Independent samples t-tests, one-way analysis of variances, or chi-squared tests were used to compare between-group differences in demographic and clinical data. We used the U test to compare the spectral power and AEC-c values of ROIs in six frequency bands between the MwoA and HC groups. Comparisons between the spectral power and AEC-c values of each ROI in every frequency band in all three groups were performed using the Kruskal-Wallis test. Statistical significance was set at *p 0.05 after Bonferroni correction.

As there were three groups of subjects and 12 ROIs at the spectral power level for CM, EM, and HC, the p-value was corrected 3 $\times$ 12 = 36 times. For MwoA and HC, there were two groups and 12 ROIs, and the p-value in the spectral power analysis was corrected 12 times. Twelve ROIs were investigated in matrix analysis at the FC level. Thus, the p-values in the FC analysis of the two groups were corrected 66 times, and 66 $\times$ 3 = 198 times for the three groups.

In addition, correlation analysis and logistic regression analysis was used to investigate the correlation between the clinical features and the spectral power or FC differences between EM and CM. Receiver operator characteristic (ROC) curve analysis was used to evaluate clinical diagnostic efficacy. All statistical analyses were performed using SPSS 24.0 (SPSS Inc., Chicago, IL, USA).

After rigorous screening by headache specialists in the Department of Neurology at the Brain Hospital, 33 MwoA patients were included in this study, including 17 CM and 16 EM patients. An additional cohort of 22 HC participants was also included. There were no statistical differences between CM, EM, and HC regarding sex and age. Headache frequency, VAS score, and number of painkillers taken per month were higher in the CM group than in the EM group. HAMA scores and HAMD scores were higher in the CM and EM groups than in the HC group, while no significant differences were observed in anxiety and depression scores between the CM and EM groups. Detailed comparisons are shown in Table 1.

Compared with HC, the spectral power of MwoA were different, mainly in the delta, theta, and alpha bands. In contrast, no significant differences were observed in the beta, gamma 1, and gamma 2 bands. The significant differences in FC were mainly in the theta, alpha, and beta bands, with no significant differences in the other bands.

In the MwoA group within the bilateral PCU (Left, L p = 0.024, Right, R p=0.024), PCC-R (p = 0.048), and MTC-L (p = 0.012), the spectral power of the delta band was significantly higher than that of the HC. In the theta band, significant differences were mainly concentrated in the bilateral cortices of the MTC (L p = 0.024, R p = 0.0004), PCU-L (p = 0.036), and PCC-R (p = 0.012), with higher spectral power in the MwoA group than in HC. In the alpha band, the spectral power was significantly lower within the bilateral MTC (L p = 0.0006, R p = 0.0004), PCU (L p = 0.003, R p = 0.048), IPC-R (p = 0.036), PCC-R (p = 0.001), and LTC-L (p = 0.048) in the MwoA group in comparison with HC (Fig. 1).

Fig. 1.

Differences in the spectral power plots between the migraine without aura (MwoA) and HC groups. (A) Shows the spectral power plot of the MwoA group; (B) Shows the spectral power plot of the HC group; (C) Shows the spectral power plot with the differences between the MwoA and HC groups. In (C), ✩ denotes that the spectral power of the MwoA group is higher than that of the HC group; $△$ denotes that the spectral power of the MwoA group is lower than that of the HC group. IPC, inferior parietal cortex; MPFC, medial prefrontal cortex; MTC, medial temporal cortex; PCU, precuneus; PCC, posterior cingulate cortex; LTC, lateral temporal cortex; MwoA, migraine without aura; HC, healthy controls. p 0.05 is statistically significant after Bonferroni correction for multiple comparisons.

For FC, the AEC-c values in the theta band in the MPFC-L and PCC-R (p = 0.019) were significantly lower in the MwoA group than in the HC group. The AEC-c values in the alpha band within the MPFC-L and MTC-L (p = 0.0003), MPFC-L and PCC-R (p = 0.032), MPFC-R and MTC-R (p = 0.006), and MTC-L and PCC-R (p = 0.028) were significantly lower in the MwoA group than in HC group. The AEC-c values within the beta band in the MPFC-L and MTC-L (p = 0.003) were significantly lower in the MwoA group compared with the HC group (Fig. 2).

Fig. 2.

Comparison of functional connections by group. (A) Shows functional connectivity plots of the MwoA with HC groups, showing the functional connectivity of two groups of regions of interest (ROIs) with significant differences in theta, alpha, and beta bands. (B) Shows the functional connectivity plots of the CM, EM, and HC groups, showing the functional connectivity of three groups of ROIs with significant differences in delta, alpha, and beta bands. PCC, posterior cingulate cortex; MPFC, ventral medial prefrontal cortex; MTC, medial temporal cortex.

The spectral power differences between the CM, EM, and HC groups were concentrated in the delta, theta, and alpha bands. The significant differences in FC were mainly in the delta, alpha, and beta bands, with no significant differences in other bands.

The spectral power in the theta band was significantly higher in the EM group within the MTC-R (p = 0.036) compared with the HC group (Fig. 3). The FC in the delta band was significantly lower in the EM group within the MPFC-R and MTC-R (p = 0.027) compared with the HC group. The FC in the beta band was significantly lower in the MPFC-L and MTC-L (p = 0.030) in the EM group compared with the HC group (Fig. 2).

Fig. 3.

Comparison of the spectral power of the CM, EM, and HC groups. Alpha spectral power with * indicates a significant difference between the spectral power of the EM and CM groups (the MPFC-L (p = 0.002) and LTC-R (p = 0.036)). *p 0.05 after Bonferroni correction for multiple comparisons. The values corresponding to the bars in Fig. 3 are the mean $\pm$ SD.

The CM and HC groups differed in spectral power and FC in more cortical layers. The spectral power in the delta band within the IPC-R (p = 0.004), PCC-R (p = 0.004), and bilateral PCU (L p = 0.010, R p = 0.005) was significantly higher in the CM group than in the HC group. The spectral power in the theta band within the MTC-R (p = 0.014) and PCC-R (p = 0.010) was significantly higher in the CM group than in the HC group. The spectral power in the alpha band within the bilateral IPC (L p = 0.016, R p = 0.001), bilateral MTC (L p = 0.002, R p = 0.002), bilateral PCU (L p = 0.001, R p = 0.007), bilateral LTC (L p = 0.036, R p = 0.004), MPFC-L (p = 0.008), and PCC-R (p = 0.0005) was significantly lower in the CM group than in the HC group (Fig. 3). The ACE-c values in the alpha band within the MPFC-L and MTC-L (p = 0.0006), MPFC-L and PCC-R (p = 0.014), and MTC-L and PCC-R (p = 0.033) were significantly lower in the CM group than in the HC group (Fig. 2).

The spectral power in the alpha band within the MPFC-L (p = 0.002) and LTC-R (p = 0.036) cortex was significantly lower in the CM group than in the EM group (Fig. 3), while no significant differences were found in the FC between the two groups.

The brain regions that differed in spectral power between the EM and CM groups were MPFC-L and LTC-R. Correlation analysis of the spectral power of MPFC-L and LTC-R with the number of days of headache in the EM and CM groups showed that the spectral power of MPFC-L in the alpha band was negatively correlated with the number of headache days per month (r = –0.90, p 0.001) (Fig. 4). However, there was no correlation between LTC-R and the number of headache days (p $>$ 0.05). Binary logistic regression analysis of the difference in MPFC-L spectral power between the CM and EM groups was statistically significant (odds ratio (OR) = 0.559, 95% confidence interval (CI) (0.378–0.828), p = 0.0004). The ROC curve of MPFC-L spectral power in the alpha band is shown in Fig. 5. The area under the curve, sensitivity, and specificity of the ROC are shown in Table 2.

Fig. 4.

Correlation between MPFC-L spectral power in the alpha band and number of headache days per month. MPFCL, left medial prefrontal cortex.

Fig. 5.

ROC curve of the MPFC-L for differentiating episodic migraine from chronic migraine.