Cerebral gray matter volume in patients with chronic migraine: correlations with clinical features

Among the patients who were consecutively admitted to our headache clinics, 20 patients (Table 1) provided informed consent to participate in the present study. Per the International Classification of Headache Disorders, 3rd edition, beta (ICHD-3 beta) criteria [9], the 20 patients were diagnosed as having de novo CM during their first visit, i.e. they did not have a previous history of medication overuse. As a confirmation, we ensured that patients’ mean monthly tablet intake (2.8 ± 3.1 tablets/month; Table 1) was below the lower limit set by the International Classification Committee for medication overuse [9]. All patients had an established history of episodic migraine without aura (ICHD-3 beta code 1.1), and used nonsteroidal anti-inflammatory drugs as acute medication. With the exception of four patients who had mild headaches (mean visual analogue scale score = 2.5) without migrainous features, all of the patients with CM underwent the MRI scans during a headache-free state. Inclusion criteria were as follows: no history of other neurological diseases, systemic hypertension, diabetes or other metabolic disorders, connective or autoimmune diseases, medically treated depression, and/or any other type of primary or secondary headache. Patients did not always experience the headaches on the same side. To avoid the bias of pharmacologic treatment, no prophylactic treatments were allowed during the previous 3 months. For comparison, we enrolled 20 HCs of comparable age and sex distribution, who were recruited from among medical school students and healthcare professionals. The HCs had no personal or familial history (1st- or 2nd-degree relatives) of migraine or any detectable medical conditions and were not on any regular medications. The HCs were randomly scanned between patients. To avoid variability owing to hormonal changes, female participants underwent MRI outside of their pre-menstrual or menstrual periods. All scanning sessions were performed in the afternoon (16:00–19:00). For both HCs and patients, additional exclusion criteria were abnormal structural MR images of the brain and/or abnormal pathological findings, including WM lesions. All participants received a complete description of the study and granted written informed consent. The ethical review board of the Faculty of Medicine, University of Rome, Italy, approved the project.

A Siemens Magnetom Verio 3-Tesla scanner was used to acquire all images. Structural scans of the brain were acquired for each participant using a T1-weighted three-dimensional sagittal magnetisation-prepared rapid gradient echo sequence with the following parameters: 176 slices, repetition time = 1900 ms, echo time = 2.93 ms, slice thickness = 1 mm, and an in-plane resolution of 0.508 × 0.508 mm. The raw and preprocessed images were manually inspected for artefacts and image quality. Moreover, the ‘check sample homogeneity’ function in CAT12 (http://​www.​neuro.​uni-jena.​de) was used to identify images with poor quality and incorrect preprocessing. None of the acquired and preprocessed image series showed abnormalities.

Image data processing was conducted using SPM12 (www.​fil.​ion.​ucl.​ac.​uk), and the CAT12 toolbox in the MatLab environment (www.​mathworks.​com) was used to perform the VBM analysis [10]. The images acquired for each participant were reoriented to have the same point of origin (anterior commissure) and spatial orientation. A non-linear deformation field was estimated that best overlaid the tissue probability maps on the individual subjects’ images. Three tissue components, namely the GM, WM, and cerebral spinal fluid (CSF), were obtained to calculate the overall tissue volume (GM, WM, and CSF volume) and total intracranial volume in the native space. Afterwards, all of the native-space tissue segments were registered to the standard Montreal Neurological Institute template (the standard included in SPM12) using the affine registration algorithm. The diffeomorphic anatomical registration through the exponentiated lie algebra (DARTEL) toolbox was applied to all participants’ GM and WM to refine the inter-subject registration. In the last step of DARTEL, the GM tissues are modulated using a non-linear deformation approach to compare the relative GM volume adjusted for individual brain size. Furthermore, the voxel values in the tissue maps are modulated by the Jacobian determinant that was calculated during spatial normalization [11]. Once the preprocessing pipeline was completed, a quality check was performed using a CAT12 toolbox to assess the homogeneity of the GM tissues. Lastly, each participant’s modulated and normalised GM tissue segments were smoothed with an 8-mm full width at half maximum Gaussian filter.

We used CAT12 for all of the statistical analyses. First, a two-sample t-test was performed to compare the GM volume between patients and HCs. For all analyses, we included age, sex, and total intracranial volume as covariates (Additional file 1: Figure S1). The patients’ relative GM volume changes were initially assessed at a threshold of p < 0.05 (corrected for multiple comparisons [family-wise error]). However, the small sample size and previous VBM evidence showing the involvement of specific brain structures in the process of migraine recurrence [8, 12] prompted us to also perform an exploratory analysis using less-conservative uncorrected thresholds of p < 0.001 and p < 0.005 throughout the whole brain. Thereafter, to identify whether the regional GM volume changes were correlated with patients’ clinical features, we performed multiple univariate regression analyses using the CAT12 model design tool that included the severity of the headache attacks (0–10), disease duration (years), number of days per month with headaches (n), attack duration (h), number of tablets taken per month (n), and duration of the chronic headache phase (months) as independent variables. These inferences have been performed at a level of p < 0.001 uncorrected.

The so-called mesocorticolimbic reward circuit consists of a complex network of cortical and subcortical regions that are responsible for the effects of positive and negative reinforcement (reward and aversion) [14]. Together, the regions in this network, including the OFC, pallidum, amygdala, hippocampus, and temporal pole, integrate information related to reward processing, emotion, and memory to modulate striatal activity [15]. In brief, dense OFC fibres converge in the central and lateral parts of the ventral striatum (activated by reward-related behavioural paradigms). The amygdala and temporal pole play key roles in the emotional coding and recalling of salient stimuli [16]. The amygdala also projects to the ventral striatum, which in turn sends efferent projections to the pallidum [17]. The reward circuit is a critical component of the brain disease model of addiction [18], of which CM due to medication overuse is thought to belong [19]. In particular, volume reductions in the amygdala and hippocampus have been previously reported in adults affected by substance abuse disorders and in their apparently healthy offspring, indicating a possible neurodevelopmental component [14, 18]. Other studies have demonstrated that the temporal pole [20], amygdala [3, 6], and pallidum [21] are involved, with a variable level of statistical significance, in the process of migraine chronification. In interictal CM, the amygdala was previously found to be atypically connected with regions in the superior frontal cortex and occipital cortex [22]. However, to the best of our knowledge, no other study has specifically reported a true reduction of GM volume in the OFC of patients with CM without a history of medication overuse. Neuroimaging studies of patients with CM with a history of medication overuse identified abnormal blood-oxygen-level dependent activity [23], GM volume [5, 12, 24], and metabolism [25] in the OFC region. In a VBM study, Riederer et al. [24] observed that patients with MOH had significantly less GM in the OFC and that the reduced GM volume in this area was associated with the treatment response. In the ¹⁸F-fluorodeoxyglucose positron emission tomography study by Fumal et al. [25], patients with MOH who underwent successful withdrawal from acute medications had greater metabolism reductions in the OFC after withdrawal than they did before withdrawal, leading the authors to conclude that the latter could predispose a subgroup of migraineurs to overuse analgesics. Interestingly, in episodic migraineurs, new-onset medication overuse was associated with baseline poor performance in tasks related to orbitofrontal function [26]. Considering these findings, we propose that patients with slightly reduced GM volume in regions that are part of the mesocorticolimbic reward circuit may be at risk of decreasing the threshold for the tendency to consume analgesics.

In animals and humans, the deep cerebellar nuclei process noxious stimuli [27‐29] and participate in pain perception and inhibition through their connections with the brainstem nuclei and thalamus [30, 31]. In HCs, when the cerebellar activity is forcibly enhanced, e.g. through neuromodulatory techniques, then the pain threshold is increased, i.e. the antinociceptive effects of the cerebellum are heightened [32]. Here, we found that the GM volume within cerebellar lobule VIIIa and Crus II was slightly reduced in patients with CM compared to HCs. Cerebellar lobule VIIIa, which represents part of the face within the cerebellum [33], has been shown to play a role in trigeminal nociception [31], while Crus II seems to be more active during non-noxious emotional processing [29, 34] and cognitive associative learning [35]. Our exploratory volumetric MRI data partially agree with those of Bilgiç and co-workers [4] who found reduced right, as well as left, cerebellum volume in patients with CM with a high rate of medication overuse and under migraine prophylaxis in comparison with HCs. Interestingly, previous neuroimaging studies of patients with CM with excessive acute medication intake identified elevations in the cerebellar metabolism [25] and volume [12]. We think that these observations are still in line with our exploratory results, as our correlation analysis revealed that higher acute medication intake was linked to higher cerebellar GM volume. Overall, we reason that abnormal macrostructural patterns in the cerebellum may be a predisposing factor that may lead to MOH development. Moreover, we found that the longer the history of migraine, the lower was the neuronal volume of the cerebellum, an observation that could be interpreted as indicating that time-dependent plastic changes in the cerebral microstructure are correlated with the chronic perpetuation of migraine attacks, or that the cerebellum is anatomically susceptible to the emotional/cognitive consequences of a chronic disorder. Our findings in patients with CM coincide with those showing that morpho-functional abnormalities in the periaqueductal gray area, which is interconnected with the cerebellum [36], are positively correlated with the disease duration in patients with CM [37], i.e. the longer the disease duration the higher the dysfunctional cerebellar antinociceptive effects.

Located on the lateral surface of the temporal pole, the MTG is involved in several cognitive functions. Reduced GM volume and changes in the functional connectivity of the left MTG have been linked to the severity of the clinical symptoms associated with social anxiety or phobia, of which a core feature is anticipatory anxiety, i.e. a state of continuous alertness for an imminent or likely threatening event [38], such as a headache. It was postulated that subjects affected by anticipatory anxiety are more prone to engage in avoidant behaviours from potential threatening events [39]. Social phobia is a disorder that is frequently diagnosed in individuals with juvenile [40] and adult [39, 41] CM, and, very likely, even in animal models of CM [42]. Some researchers have suggested that being affected by both CM and social phobia configures a state of phobic avoidance that is associated with the fear of a migraine attack, which may explain why some patients take analgesics at the smallest indication of a headache and why such patients may be at risk of developing medication overuse or of decreasing the threshold for analgesic consumption [39, 43]. However, since we did not assess patients’ psychiatric profiles, and considering the exploratory nature of our uncorrected findings, we cannot draw a definitive conclusion about the link between reduced GM volume in the MTG and social phobia in terms of its ability to promote migraine chronification and medication intake. To clarify this, future studies investigating the cerebral microstructure and connectivity in patients with CM with/without medication overuse should include assessments of patients’ psychiatric profiles.

In the present study, we discovered that patients with CM had slightly reduced GM volume in visual area 17 and visual association area 18 compared to HCs. It is worth noting that pain can be related to vision. Studies of cortical function show that tonic pain induces marked spontaneous [44] or evoked [45] electroencephalographic and functional imaging [46] changes in the occipital regions. Recently, in a group of patients with episodic migraines, we found reductions in the functional connectivity between the visuo-spatial system and the so-called default mode network between attacks [47], while during attacks, the connectivity was reduced between the executive and dorso-ventral visual attention networks [48], stressing that occipital areas could be involved in the attentional processes to pain and in some aspects of pain representation [49]. Notably, the visual presentation of affective pictures modulates occipital functional activation and, at the same time, pain perception differently in patients with CM than it does in HCs [50], perhaps through direct occipital-to-brainstem trigeminal nuclei connections [51]. Therefore, our results in patients with CM tend to show the morphological correlates of aberrant attentional processes to head pain and of anomalous representations of pain.

Certain limitations of the present study should be acknowledged. First, the sample size was small, thus GM volume changes in patients versus HCs were apparent only when data was assessed with very liberal uncorrected thresholds. Additionally, we did not analyse the psychiatric profiles of the patients, although we think that patients’ social anxiety symptomology may contribute to their clinico-morphological status.