Headache-related circuits and high frequencies evaluated by EEG, MRI, PET as potential biomarkers to differentiate chronic and episodic migraine: Evidence from a systematic review

We used the advanced search functionality of the Web of Science (WoS), Scopus, and PubMed electronic databases to conduct the initial literature screening. On the one hand, WoS and Scopus have a high coverage in Life Sciences (with Web of Science having highly selective journal coverage) [12]. On the other hand, PubMed has the most exhaustive journal coverage of the three [12]. For this reason, the selection of these three databases is optimal for a biomedical systematic review. The query strings included within the searching boxes for each of the databases are shown in Table 1. The searches were conducted by two independent researchers (JG-P and VM-C), who also conducted subsequent screening, as well as the evaluation and review of the studies found using the electronic databases. After selecting the appropriate searching terms, these were searched in the title, abstract, and keywords. The search was performed without start date until the date of the database search (9^th May 2022).

The electronic database search was supplemented with manual searches for published, unpublished and ongoing randomized clinical trials (RCTs) in ClinicalTrials.gov by means of the selecting “Migraine” as “Condition or disease” and “episodic, chronic” as “Other terms”. This search was performed on 20^th May 2022. Then, JG-P and VM-C conducted a manual screening by selecting only those RCTs using EEG, MEG, fMRI, PET, or fNIRS, and including both CM and EM groups. The resulting RCTs were analysed and the related articles were incorporated to the list obtained from the electronic database search.

The search was limited to articles published in English. We also reviewed the reference lists of the selected articles and reviews to identify studies that were missed in the search process. The studies were assessed for duplicates, while verifying that the eligibility criteria were met (i.e., inclusion and exclusion criteria, listed in Table 2). A first screening of the articles was conducted by reading the abstracts of the searching results. All eligible studies were then screened in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines [13]. This screening was performed independently by JG-P and VM-C. A final selection to be included in the systematic review was independently proposed by JG-P and VM-G. Discordances were resolved by consensus.

The digital database search resulted in a total of 106 records (Fig. 1a). Two additional articles were identified through a manual search of the reference lists of articles and reviews. These 108 articles were assessed for duplicates, retrieving 52 of them for the subsequent screening process. The articles were then screened by reading the title, abstract and keywords (when available). Twenty-one articles do not meet the eligibility criteria (Table 2) after this first initial screening, thus, 31 documents including original studies and review articles were selected for an in-deep evaluation (i.e., an in-depth reading of the full text). Concurrently, only 4 studies from the 80 identified RCTs were eligible for evaluation. After a second screening process, a comprehensive review of the 35 documents according to PRISMA reporting guidelines and following the eligibility criteria were conducted. Thus, 24 studies were finally included in this systematic review since 11 did not meet all requirements.

No studies were found that meet inclusion criteria until 2011 (see Fig. 1b). More than half of the studies were published in the last 5 years. Table 3 summarizes the main data obtained from the 24 selected studies. Of them, 22 were original articles, while 2 were review articles [14, 15]. Results of the original articles were based on EEG (3 studies) [16‐18], MEG (6 studies) [19‐24], fMRI (10 studies) [25‐34], and PET (3 studies) [35‐37] (see Fig. 1c). Noteworthy, none of the 24 included studies come from RCTs or used fNIRS recordings.

Nine original papers and two review articles [14‐24] investigated, directly or indirectly, the differences in neural patterns between CM and EM from an electrophysiological or magnetophysiological perspective. The statistical power of the studies was diverse, with studies using populations ranging from 25 subjects (15 CM and 10 remitted EM) [22] to over 300 subjects (24 healthy controls, 48 CM and 232 EM) [18].

Two of these studies were review articles [14, 15]. In the most recent of them [14], authors investigated resting-state activity as a potential brain signature for migraine patients (both for CM and EM). Although the search for differences between CM and EM was not the objective of the study, the authors described and summarized very interesting findings on characteristic patterns in migraine both from the purely oscillatory point of view and from the more advanced perspective of the analysis of functional connectivity (i.e., non-directional as opposed to effective connectivity) as well as its characterization using parameters derived from graph theory. In addition, they highlighted a previous study [19] (we further analyse it latter) in which the node degree (sum of the connectivity of a certain node with all the others in the network) showed significant differences between CM and EM in primary and secondary somatosensory cortices, insula, anterior cingulate cortex (ACC) and medial frontal cortex.

In the other review article [15], the authors shed light on the underlying mechanisms that associate sleep disturbances and chronic headaches. Interestingly, a relationship between an increase in slow-wave sleep accompanied by a reduction in beta activity during migraine attacks was described. According to the authors, this could suggest that the observed changes in sleep dysregulation reflect the process of headache chronification (i.e., the transition from EM to CM), rather than simply reflecting differences between the ictal and interictal states of migraine.

In the most recent of the original articles on EEG, Gomez-Pilar and colleagues [16] performed a spectral analysis to find spectral bands of interest in 39 controls, 42 CM and 45 EM. Using bootstrap and other robust statistical techniques, the authors showed a specific frequency band around high beta in which the CM group statistically differed from the EM group during resting state. Although the differences are widespread on the scalp, they seem to be more concentrated in the left hemisphere.

The other two EEG studies used visual evoked potentials (VEPs) [17] or steady-state visual evoked potentials (SSVEPs) [18] to compare CM and EM groups. In both cases, significant differences were found in high frequency bands: in the beta band in the SSVEPs study (occipital region, photic driving power response at 20 Hz) and in the gamma band in the VEPs study. It should be noted, however, that the differences in the gamma band seem to be due to differences of the power line artifact around 60 Hz, which should be removed in spectral studies [38‐40]. As the authors indicated in the limitations of their study, this removal was not performed, so the results should be interpreted with caution.

Only one of the 6 MEG studies used recordings during resting state [20]. As mentioned in the review study [14], the connectivity in different areas was analysed by node degree. The connectivity was calculated using the imaginary part of the coherence, which reduces volume conduction effects [41]. The beta band showed significant differences between CM and EM in various brain regions after source analysis, but ACC was the region that showed the greatest between-group differences.

The 5 remaining MEG studies showed heterogeneous results. Two of them did not report significant differences between the two groups using time–frequency analysis of emotional stimulation responses [21] o temporal analysis of visual evoked fields (VEF) [23]. Hsiao and colleagues [20] reported a similar somatosensory gating response associated with the treatment outcomes both in CM and EM groups. However, in this case, authors did report between-group differences in the amplitude of the somatosensory evoked field (characterized by a peak around 50 ms) after 3-month treatment. Finally, two studies from Chen and colleagues [22, 24] found differences in the temporal analysis of VEFs. Using a well-designed block-based procedure, the authors studied patients’ habituation to stimuli by measuring the percent change in P100m amplitude between the first block and subsequent blocks. Statistically significant differences were found between the interictal EM group and the CM patients. Since the percentage change was greater in EM, this indicated a habituation of the CM group compared to a potentiation in EM patients.

Together, these M/EEG studies account for between-group differences frequently based on the early response in evoked potentials (i.e., fast frequency responses) or on alterations in beta band in resting state studies.

fMRI is the brain activity acquisition modality with the largest number of studies for the comparison between CM and EM (see Fig. 1c). Although the temporal resolution of this technique is several orders of magnitude lower than EEG or MEG, its spatial resolution allows a fine inspection of the specific activity of different brain regions. Given the somewhat obvious relationship one might expect between migraine and pain-related circuits [42, 43], or the less obvious relationship with emotion regulation [44, 45], fMRI facilitates researchers to directly study these networks. Thereby, after the screening, ten studies using the blood oxygenation level dependent (BOLD) activity from fMRI recordings to distinguish between CM and EM were selected for the systematic review.

Three studies [27, 28, 31] showed differences between CM and EM in functional connectivity in the hypothalamus, this being the most reproducible result. Chen and colleagues [27] analysed structural and functional connectivity in a conventional resting state design. Being its anatomical results the most significant, in particular the proposal of the volume of the hypothalamus as a marker of CM, they also showed an interesting finding on the functional level. Specifically, authors reported statistically significant differences between CM and EM in functional connectivity between the hypothalamus and the right medial orbital gyrus (MorG). According to the authors, the increased connectivity in CM may reveal the role of the anterior hypothalamus in altered sleep responses or emotional and execution dysfunction in CM. The results reported by Lerebours et al. [28] agreed with these findings. They found a significantly increased connectivity between the anterior hypothalamus and the spinal trigeminal nucleus in CM in comparison with EM. This highlights the major role of the anterior hypothalamus in migraine, particularly its relationship with medication overuse. In the third study whose findings involved the hypothalamus (the first published of the three) [31], the authors used four different stimuli in a pseudorandomized order. During the administration of gaseous ammonia (as a painful stimulus), activity within the right anterior hypothalamus was significantly higher in CM group than in EM group during ictal stage. Together, these studies speak for the importance of the anterior hypothalamus in attack generation and migraine chronification (mainly via medication overuse).

The findings of the other studies might seem, at first glance, heterogeneous. However, all of them involved, in a direct or indirect way, neural pathways related to pain circuits and/or emotion processing. A clear example is the study from Chen and colleagues [34], in which the authors studied the functional connectivity of the marginal division of neostriatum, involved in the modulation of pain. A decreased connectivity in CM and CM with medication-overuse headache was found in this region as compared to EM group. Also, in the study of Imai et al. [30], an increased functional connectivity between ACC and the right occipital gyrus was reported in a set of 31 CM patients as compared to 31 EM patients. It is noteworthy that the ACC is probably the cortical area that has been most frequently linked to pain [46]; specifically, it appears to be involved in the emotional reaction to pain, rather than to the perception of pain itself [47]. As ACC, amygdala is associated with the emotional-affective dimension of pain [48]. Interestingly, CM patients show increased functional connectivity between amygdala and inferior temporal gyrus (ITG) and orbitofrontal gyrus (OFG) compared to EM, as reported by [32], shedding light on the role of the amygdala in the neurolimbic pain-modulating in the migraine.

Another study [33] from Hubbard and colleagues showed decreased functional connectivity in CM between primary somatosensory cortex (S1) and both lateral occipital cortex (LOC) and dorsomedial prefrontal cortex (DMPFC). As the authors stated, S1 has a relevant role in processing the sensory-discriminative components of pain. In line with these findings, Chen and colleagues [26] used the regional homology analysis method (ReHo) to analyse the BOLD fluctuations. Although, unfortunately, no comparison was reported between infrequent EM and CM, a large variety of brain areas showed significant differences between frequent EM and CM (see Table 6 in [26] for details). The areas exhibiting the higher statistically significant differences were the left and right precentral gyrus, i.e., the S1, supporting the results of Hubbard and colleagues [33].

In the study of Dai and colleagues [25], an increased functional connectivity between habenula and salience network was exhibited in CM relative to EM group. Habenula brings input from the hippocampus and basal ganglia structures, among others [49], while salience network (primarily composed of the anterior insula and dorsal ACC) collaborates in the integration of emotional and cognitive information [50]. Finally, in the study of Bogdanov et al. [29], the analysis of the salience network also shows interesting results. Along showing differences in the motor cortex and superior temporal sulcus, authors reported significant differences between CM and interictal EM in salience network regions, such as the insula, the thalamus, the ACC and the S1.

All together, these studies support the involvement of neural circuits and brain networks that process, directly or indirectly, the stimuli and responses related to pain and emotion. The differences found between CM and EM using fMRI seem to be robust, converging across conditions (resting state vs. stimuli processing), migraine stage (ictal, interictal) in a variety of designs and analysis (functional connectivity, neural activation or ReHo).

Only three studies [35‐37] analysed the differences in metabolic activity between EM and CM via PET neuroimaging. Metabolic differences between both migraine subgroups were only found in terms of µ-opioid (µOR) availability [36], but not when measuring 5-HT [37] or fluorodeoxyglucose [35] levels.

Jassar et al. [36] used [¹¹C]carfentanil to measure µ-opioid (µOR) availability in 7 CM patients, 8 EM patients, and 7 healthy controls (HC). CM showed significantly lower µOR non-displaceable binding potentials than HC in thalamus and left caudate. This ictal µOR dysfunction of CM extended to the limbic system, i.e., right parahippocampal region and right amygdala, in CM relative to EM. Additional analyses suggested that the increased µOR receptor-mediated neurotransmission in limbic system of CM is highly modulated by the attack frequency, pain severity and sensitivity. These results are in line with the evidence from fMRI studies, since this µOR dysfunction is involved in pain networks.

On the contrary, negative results were found by Deen and colleagues [37] in the evaluation of brain serotonin 5-HT levels after injection of [¹¹C]SB207145 (a specific 5-HT₄ receptor radioligand) in 16 CM patients, 15 EM patients, and 16 HC subjects. Although CM group exhibited significantly higher 5-HT levels than HC group, no significant differences between CM and EM levels and no association between this metric and number of monthly migraine days were found. Authors concluded that high brain 5-HT levels may be a trait marker of the migraine brain rather than a risk factor for conversion from EM to CM.

In line with the previous study, Torres-Ferrus and colleagues [35] used [¹⁸F]FDG radiotracer to perform interictal PET and MRI scans to 7 MC patients, 8 EM patients, and 11 HC subjects. The authors, however, did not find statistically significant differences between CM and EM groups. CM showed significant frontotemporal hypometabolism than HC, while EM presented intermediate values. Only the bilateral temporal lobe in EM yielded significant differences as compared to HC. However, as mentioned, no significant differences between CM and EM were found in terms of cerebral metabolism. Interestingly, no significant differences were found when compared both migraine groups as a whole (i.e., CM and EM together) versus HC group.

The nine original M/EEG articles showed a rather heterogeneous methodological design. Only two studies analysed the EEG [16] or the MEG [19] at rest from different perspectives: spectral and nonlinear analysis versus connectivity analysis. Despite their differences, both found differences between CM and EM in the beta band. Five other studies focused on evoked responses related to visual activity. Although with heterogeneous results, most found differences between groups in early evoked potentials, which are related to high-frequency responses. In particular, the studies by Chen et al. [22, 24] showed differences in the amplitude in potentials that appear at 50 ms and 100 ms, that is, related to the alpha (1/100 ms = 10 Hz) and beta (1/50 ms = 20 Hz) bands.

Albeit tentative, the results indicate that the differences appear recurrently in the fast frequency bands (beta) and, to a lesser degree, in alpha band. These findings seem independent of the type of analysis used, involving either spectral, connectivity, or temporal analysis evaluating the amplitude/latency of evoked potentials. Therefore, the potential biomarkers seem to be focused on fast frequencies, regardless of the type of analysis used. This, however, requires further study before being confirmed.

Once evaluated the studies that give information about the differences in brain dynamics between CM and EM, we were interested in the particular brain circuits involved in this differentiation. fMRI was then assessed revealing consistent findings in differentiating between CM and EM. These differences lie in changes in the neural pathways associated with pain. These differences seem to be due to the pain chronification process [26, 30, 31], often linked to medication overuse [10, 51‐53]. Differences in the hypothalamus were the most replicated, which is known to be involved in homeostatic functions and pain control [54]. Differences were also observed in other regions related to pain, such as the marginal division of neostriatum [34], the ACC [30], the amygdala [32], and the S1 [33, 55], among others. These findings converge across brain activity acquisition modality, ictal stage, analysis design, and recording condition, evidencing the robustness of this pattern.

Among the PET studies, only the work of Jassar and colleagues [36] reported statistically significant differences between CM and EM. Particularly, they found intergroup differences in terms of µOR availability in the limbic system. The ictal µOR dysfunction in right parahippocampal region and right amygdala of CM compared to EM is also in line with the evidence obtained from fMRI studies, as pain-related neural pathways are suggested to play a key role in migraine chronification.

Unlike other neurological or psychiatric diseases that show delocalized alterations and/or generalized changes in neural dynamics, migraine and in particular the differentiation between CM and EM subgroups seems to lie in specific brain regions and concrete spectral or connectomic changes. Throughout this comprehensive systematic review, we found functional metrics with great potential to become true biomarkers in the near future. However, these biomarkers are not yet of sufficient reliability and accuracy, and more effort should be made to validate and extend the main findings. Therefore, we would like to encourage further research in these directions to increase statistical power and drive migraine diagnosis toward an objective examination.

Most discriminative differences between CM and EM were found in beta bands using EEG and MEG and brain circuits related to pain (i.e., thalamus, amygdala, salience network, etc.) by means of fMRI and PET. We recommend further investigations in these directions, as beta abnormalities (e.g., [16, 19, 22, 24]) and µOR disfunction [36], as well as functional connectivity alterations (e.g., [27, 28, 30, 31]) in pain networks which could play a key role in the chronification of migraine. We also want to draw attention to the fact that most of the previous work evaluated multiple brain regions without a specific anatomical or topographical a priori hypothesis. In future studies, specific designs should be considered that seek confirmation of previous findings, but with an adequate calculation of the appropriate sample size, considering the effect size reported by these previous studies.

Regarding the recording modality, a simultaneous acquisition of EEG and fMRI would ideally be required in order to simultaneously undertake a spatial and frequency analysis with sufficient resolution. These resources, however, are rarely accessible and affordable by the majority of the health centres. Actually, a biomarker based on PET scans, fMRIs or MEG would not usually be accessible in large population settings where migraine burden is high, and resources are restricted. An alternative is to carry out an adequate source analysis from EEG recordings (high-density recordings when possible). Therefore, future research might want to connect key brain substrates to peripheral markers for future diagnostic and prognostic purposes.

It is not clear whether future studies should focus on recordings at rest or during the performance of a task. Although most studies focus on recordings acquired under task condition (frequently visual), several studies have shown evidence that migraine has a strong impact on neural activity at rest (see [14] for an excellent review on this topic). Regardless of the condition (rest or task), different study designs and specific new methodologies can be carried out. Particularly important would be those techniques that allow studying the temporal evolution of brain states, i.e., neural dynamics, in previously determined regions. Since there seem to be differences between groups in specific frequency bands, the transition speed between states could shed light on potential biomarkers to distinguish CM and EM. Very recently, these techniques have begun to be applied in the context of migraine [56]. However, they have not yet been used for the specific distinction of migraine subgroups. Future studies should consider investigating these techniques applied to the search for biomarkers of migraine subgroups.

Finally, although the studies reviewed here did not distinguish between low-frequency (8 or fewer migraine days per month) and high-frequency (between 8 and 15 migraine days per month) EM, a growing and robust body of evidence suggests that these two categories could have a well-differentiated brain substrate [57]. In addition, accordingly with the reviewed literature, differences between CM and EM are much more likely to be found when subjects with EM are restricted to low-frequency EM. Together, this suggest that CM and EM represent a gradual difference that becomes more pronounced as headache frequency increases. In this context, the classification between CM and EM may not be purely binary, meaning that high-frequency EM may be in many patients more similar to CM rather than to low-frequency EM, in terms of disability, treatment response and biomarkers. Therefore, the current binary classification criteria based on the self-reported number of headache days per month could be deemed as an arbitrary distinguisher. Although strengthening this preliminary evidence is still necessary, these findings show that high-frequency EM could have a clinical and biological behaviour similar to chronic migraine.