2.1 Design and study duration.
This is the primary analysis of the reported data using a cross-sectional design. Other imaging data (MR spectroscopy (MRS) and Arterial Spin Labeling (ASL) MR imaging) have been collected for all participants and results are presented elsewhere (33). No statistical power calculation was conducted prior to the study. The sample size was based on the available data (during the study interval) and was similar to a recent ASL study in episodic migraine patients (34). All data was collected between December 2013 and July 2015.
2.2 Participants
19 right-handed HC, 17 right-handed patients with EM and 12 patients with CM were included for the study. The detailed demographic data are listed in table 1. All patients fulfilled the modified ICHD-III-beta diagnostic criteria for EM or CM (35). None of the HC demonstrated signs of EM or CM according to these criteria (family history of migraine was allowed). During the enrollment process, we excluded all patients, which suffered from comorbid tension-type headache. Six of 12 CM also fulfilled the criteria for medication overuse headache (MOH), which is line to the literature (36, 37). MOH is defined as headache that develops or significantly worsens during overuse of acute pain medication (35). For all participants, exclusion criteria were severe psychiatric disorders, cardiac problems (e.g. severe hypertension), other headache disorders or other neurologic disorders such as epilepsy, stroke, traumatic brain injury, neck injury or cerebrovascular disease. All participants completed prospective headache diaries, the Migraine Disability Assessment (MIDAS) (38) and Hamilton Anxiety (HADS-A) and Depression (HADS-D) Score (39) questionnaires. Acute and prophylactic medication was recorded prior to the study interval for each patient (see table 2). We assessed the attacks/month based on the MIDAS questionnaire. Here, the label “headache attacks/month rate” (table 1) reflects the average number of migraine headache days in the last three months prior to the MRI (i.e., an attack frequency of 4.0 in EM means that the average number of headache days was four per month across in this group). We recorded aura occurrence in all patients electronically in a table. Patients were free from migraine attacks at least 48 hours before and after the scan. The study was approved by the ethics committee of canton Zurich (KEK number E-37/2007), Switzerland. All subjects provided written informed consent prior to study enrolment. Both groups received 50 Swiss Francs reimbursement for their study participation. Patients were recruited by advertisement (Intranet of the Hospital and mailing lists) and word-of-mouth.
Table 1. Demographic details of the subjects in the study. * We found no significant difference (p > 0.05) in age and sex between the groups: EM - HC, CM - HC and EM – CM (unpaired t-tests and Chi-Square test, respectively). MIDAS values in days. EM: episodic migraine, CM: chronic migraine, HC: healthy controls, F: female, M: male. * The label “headache attacks/month rate” represents the average number of migraine headache days in the last three months prior to the MRI (i.e., an attack frequency of 4.0 in EM means that the average number of headache days was four per month across in this group).
Group
|
N
|
Age (years)*
|
Sex*
|
MIDAS
|
HADS - A
|
HADS - D
|
Headache attacks / month*
|
EM
|
17
|
32.7 ± 9.9
|
F = 13,
M = 4
|
19.65 ± 20.61
|
5.3 ± 3.9
|
3.4 ± 2.6
|
4.0 ± 3.8
|
CM
|
12
|
38.19 ± 16.15
|
F = 8,
M = 4
|
55.50 ± 12.76
|
5 ± 3.46
|
4.67 ± 2.96
|
18.50 ± 4.25
|
HC
|
19
|
31.7 ± 9.2
|
F = 10,
M = 9
|
N/A
|
3.4 ± 2.3
|
1.3 ± 1.2
|
N/A
|
EM/CM with aura
|
12EM/5CM
|
35.68 ± 13.15
|
F = 12,
M = 5
|
30.47 ± 28.32
|
5.11 ± 3.55
|
3.88 ± 2.69
|
8.64 ± 8.71
|
Table 2. List of the preventive (prophylaxis and acute medication) therapy for each patient. Abbreviations: EM - Episodic migraine, CM - Chronic migraine, SA - simple analgesics, SA* - simple analgesics (not for every attack), B2 - Riboflavin, Mg - Magnesium, Q10 - coenzyme Q10, MOH - medication overuse headache. MOH is defined as headache that develops or significantly worsens during overuse of acute pain medication.
EM
|
Acute
|
Prophylactic
|
MOH
|
CM
|
Acute
|
Prophylactic
|
MOH
|
Subj. 1
|
SA
|
|
|
Subj. 1
|
SA
|
B2, Mg
|
no
|
Subj. 2
|
SA
|
|
|
Subj. 2
|
Triptans
|
|
yes
|
Subj. 3
|
Triptan
|
|
|
Subj. 3
|
Triptans, SA
|
|
yes
|
Subj. 4
|
Triptan, SA
|
|
|
Subj. 4
|
Triptans
|
|
no
|
Subj. 5
|
SA
|
|
|
Subj. 5
|
Triptans
|
|
no
|
Subj. 6
|
Triptans
|
|
|
Subj. 6
|
Triptans
|
|
yes
|
Subj. 7
|
Triptans
|
|
|
Subj. 7
|
Triptans
|
Betablocker
|
yes
|
Subj. 8
|
SA, opiates
|
|
|
Subj. 8
|
Triptans, SA
|
|
yes
|
Subj. 9
|
SA
|
|
|
Subj. 9
|
SA
|
|
no
|
Subj. 10
|
SA*
|
|
|
Subj. 10
|
Triptans
|
Riboflavin
|
yes
|
Subj. 11
|
SA
|
|
|
Subj. 11
|
SA
|
|
no
|
Subj. 12
|
SA
|
|
|
Subj. 12
|
SA
|
|
no
|
Subj. 13
|
SA*
|
|
|
|
Subj. 14
|
Triptans, SA
|
|
|
Subj. 15
|
SA*
|
|
|
Subj. 16
|
SA
|
|
|
Subj. 17
|
Triptans, SA
|
B2, Mg, Q10
|
|
2.3 Data acquisition
Whole-brain magnetic resonance imaging (MRI) was performed on a 3T scanner (Philips Ingenia, Netherlands) with a 32-channel receive-only head coil at the Neuroimaging Center of the University Hospital Zurich. 3D T1-weighted magnetization prepared rapid gradient echo (MPRAGE) sequence was acquired for each subject with TE/TI/TR = 2.52/900/1900 ms, flip angle = x°, field of view (FOV) = 256 × 256 mm2, matrix size = 256 × 256, slab thickness = 192 mm, voxel size=1 × 1 × 1 mm3. Subjects’ scans were examined for any major anatomical abnormalities by an experienced neuroradiologist.
2.4 Data analysis
2.4.1 Cortical and subcortical morphometric analysis
Data from all subjects were analyzed using FreeSurfer version 5.3.0 (http://surfer.nmr.mgh.harvard.edu). This automated anatomic parcellation procedure enables one to extract reliable estimates of various cortical and subcortical measures including thickness, volume, area, curvature etc. (40). The procedure includes several steps: intensity normalization, skull stripping, Talairach transformation, and atlas-based assignment of neuro-anatomical labels, which are described in detail in previous studies (41, 42). To describe here briefly, all subjects were run using the “recon-all” processing stream with default parameters to create a cortical surface model. This process includes above-mentioned procedures along with motion correction, averaging of multiple T1 volumes, removal of non-brain tissue and grey matter white matter boundary tessellation to create the surface model. This obtained model is then further used with its intensity and continuity information from the entire 3D volume in segmentation and deformation procedures to generate cortical thickness, calculated as the closest distance from the gray/white boundary to the gray/CSF boundary at each vertex on the tessellated surface (41).This process of obtained morphometric measures have been validated using histological (43) and manual measurements (44) and have been demonstrated to have very high reliability across different scanner and field strengths (45). These cortical and subcortical morphometric measures are very effective in depicting the regional alterations. However, the effect of these regional changes to other associated regions could have a significant impact in overall information transfer leading to various functional modifications. Hence, to observe these network level differences between the groups the computed cortical thickness (CT) and subcortical volumes (SCV) from all the subjects were further processed using brain network analysis.
2.4.2 Brain network analysis
Graph theoretical measures of network modularity, distance, and local information transfer was computed using the CT and subcortical volumes obtained from FreeSurfer using Brain Connectivity Toolbox (46) (https://sites.google.com/site/bctnet/). The group level correlations between the cortical regions and subcortical volumes and the differences between them were then computed in different network densities for observing the steady topological changes (47, 48). The details of the analysis have been explained elsewhere (49, 50).
Among different measures computed in the study, below is the brief overview of those relevant for the study, with simplistic illustration in figure 1.
a) Modularity is a measure of the degree, to which the network is subdivided into densely interconnected nodes (modules) with sparse connections to other network or modules. Louvain algorithm was used for computing the modularity which is a hierarchical clustering algorithm, that recursively merges communities into a single node and executes the modularity (51).
b) Transitivity is the ratio between the number of triangles and the number of triplets in the graph.
c) Assortativity is correlation coefficient between the degrees of all nodes on two opposite ends of a link, higher (positive) assortativity indicating the nodes tend to link to other nodes with the same or similar degree.
d) Clustering coefficient is the fraction of triangles around a node representing the node’s neighbors that are also neighbors of each other.
e) Betweenness centrality of a node is the fraction of all shortest paths in the network that contain a given node. A node with higher edge-betweenness centrality participates in a large number of shortest paths.
In addition, network-based statistic (NBS) was used to assess differences in the inter-regional connectivity between the groups. NBS analysis performs the mass-univariate testing at every connection comprising the graph controlling for multiple comparisons through evaluating the null hypothesis at the level of interconnected subnetworks rather than individual connections (52). Here, the connectivity matrices obtained from the association of CT and SCV between the regions across a range of network densities were subjected to NBS analysis. The analysis primary goal was to identify the sub-network with the regions shown to have significant difference in various network properties using graph theoretical measures. Further details regarding the procedure are mentioned elsewhere (53, 54).
2.4.3 Statistical analysis
For the graph theoretical framework analysis, we used the CT values of a network comprising 34 cortical regions in each hemisphere based on Desikan atlas (55). Subsequently, we used the sub-cortical volumes from nine regions in each hemisphere and brain stem for the sub-cortical network. For assessing the statistical significance of graph metrics between patients and HC, a nonparametric permutation tests with 5000 iterations were applied (56, 57). Given that, CT is sensitive to age and sex, they were further used as covariates for the analysis. In each repetition, the regional data for each subject were randomly reassigned to one of the two groups and an association matrix was obtained. The network measures were then calculated for all the networks at each density. Here, density represents cost of the network computed by fraction of present connections to all possible connections. Hence, the network measures derived at each density would specify the alterations in network behavior at different levels of fragmentation (from full, partial to discontinuous connectivity). This method of thresholding ensures that all the regions (nodes) of the network are connected while discarding spurious connections (edges) (47, 58). The actual between-group difference in network measures was then placed in the corresponding permutation distribution and a two-tailed p-value (at 5% significance level, false discovery rate (FDR) corrected) was calculated based on its percentile position (59).
To assess the statistical significance for CT correlations with different clinical parameters, QDEC – a FreeSurfer statistical toolbox was used. Here, surface maps depicting regions with significant differences in the correlation with CT at each vertex were determined with general linear models (GLMs) using p < 0.001 as the threshold for a significant cluster. In addition, we further performed the GLM analysis to observe the association between the CT change and different clinical scores including HADS-A, HADS-D, hours of sleep and attacks per month.
To validate the significance of these network measures, we further applied support vector machine analysis to predict the clinical scores used in the diagnostic criteria for migraineurs. Here, we performed a support vector regressor (SVR) analysis – representing a machine‐learning‐based multiple regression method - that could associate the observed and trained values and present the regression coefficient for the accuracy of the prediction (60). The regression coefficient of 0.5 obtained after 10-fold cross validation is considered borderline significant result.