Introduction

Pathophysiology of migraine is complex and, so far, no biomarker for any of the phases of this cyclic disease exists. During the last decade, advanced neuroimaging modalities are increasingly used to understand migraine pathophysiology and disease mechanisms in the search for imaging markers of migraine. An often-used imaging technique is the resting-state or the so-called functional connectivity (FC) magnetic resonance imaging (fMRI), which has been applied in increasing number of migraine studies, since the first paper was published in 2011 [1]. Ideally, resting-state FC studies may be used to unveil migraine mechanisms.

The migraine resting-state literature is often analyzed and presented in several different ways, which makes it hard to compare results across studies, and findings are at times difficult to understand and are rarely reproduced. Thus, definitive imaging biomarkers for migraine have still not been identified limiting the usefulness and applicability of FC data.

Still, several well-performed resting-state FC studies and reviews [2] are available but a systematic review of the consistency of findings is missing. In the present review, we wish to provide an overview of all published conventional resting-state FC studies and discuss what we have learned so far based on FC findings.

Methods

Literature search

Two authors (JMH and FMA) performed search on the PubMed.com website to identify all original articles with resting-state FC data in migraine patients. The literature search was finalized on Pubmed.com September 20th, 2018. We used the following search terms: #1 resting state fMRI and migraine, #2 functional connectivity and migraine, and #3 functional connectivity fMRI and migraine. The search was restricted to human studies published in English language within 10 years, up to September 20th, 2018. Reviews, pediatric studies, case-reports, all other headache diagnoses and letters were excluded. We also assessed reference lists of the found articles for additional relevant studies. Moreover, we excluded all studies that did not use conventional resting-state analysis but other modalities, e.g. functional connectivity density, Granger causality, amplitude of low-frequency fluctuations, and regional homogeneity. Articles, in which the method was not properly described or if data on the comparison to a non-headache control group was not available were also excluded (expect if migraine attacks were compared to an interictal phase). Finally, studies testing treatment effect were also excluded. These exclusion criteria were chosen to include comparable studies in this review.

Data extraction

To screen for inclusion and exclusion criteria, the senior authors (JMH and FMA) assessed all abstracts found in the initial search. The selected studies were then sent to the co-authors (KS, WSvH, DD, AP, AS, BMI, EB, IS, LDA, and LF) who then read the text and extracted further information, i.e. origin of study, study population, method and main findings.

Resting-state functional connectivity MRI

The imaging method is based on blood-oxygen-level dependent (BOLD) recordings of the resting brain (i.e. the person lying in the MRI scanner is relaxing with closed eyes, but not sleeping). Every voxel in the obtained image of the brain emits a signal with a specific frequency. The higher the degree of synchronization of signal frequency between two different voxels, the more functional connected are these voxels, and vice versa. Brain areas displaying a particular level of similarity represent a functional connectivity network. Thus, all areas in the brain are more or less functionally connected to each other. The use of this method depends on the change in the functional connectivity between areas in a network, when measured in two different conditions or population samples.

Results

Our search strategy was finalized September 20th, 2018 and resulted in a total of 219 results, including 94 unique results, from which following were excluded: 15 reviews, 12 stimulation studies, nine non-conventional FC modalities, six examining effect of treatment (acupuncture), five non-migraine studies, five non-FC studies, four non-original articles, one pediatric study, and one study was retracted. Further eight studies were excluded because the method was not properly described or lack of a non-headache control group. One study was subsequently included from the reference lists. We ended up with a total of 28 studies, including 25 during the interictal phase (Table 1) and three during the ictal phase (Table 2) of migraine (Fig. 1). The studies were published between 2011 and 2017 and originated from five different countries, including China = 11; USA = 6; Italy = 6; Denmark = 4; Taiwan = 1.

Table 1 Functional connectivity MRI during the interictal phase of migraine compared with non-migraine controls
Full size table
Table 2 Functional connectivity MRI during and outside of the ictal phase of migraine
Full size table
Fig. 1
figure 1

Flow chart of the literature search on functional connectivity (FC) studies in migraine

Full size image

Interictal migraine versus non-headache controls

Twenty-five published studies reported data comparing interictal migraine with non-migraine non-headache controls. In 12 studies a migraine without aura (MO) population was examined, while pure migraine with aura (MA) was only investigated in a single study. In four studies, data for both MA and MO groups were reported separately, whereas mixed results were reported in the remaining eight studies.

When comparing migraine patient to controls, the functional connectivity was changed within or with a number of different networks or seed areas: periaqueductal gray network [1, 23], left [3, 7] dorsal [5] and right [3, 25] anterior cingulate cortex, fronto-parietal-network [4], right occipital lobe [5], left medial [5] and bilateral [7] prefrontal cortex, right cerebellum [5], brainstem [5], bilateral central executive network [6, 20], left [16] salience network [6, 20], default mode network [6, 8, 14, 15, 20, 21], right thalamus [7], right [7] and anterior [9] insula, amygdala [9, 10, 24], bilateral caudate [11], right nucleus accumbens [11], hypothalamus [12], right executive control network [13], left dorsal attention network [16], right cuneus [16], visual network [17], marginal division of neostriatum [18], primary visual cortex [19], primary auditory cortex [19] and bilateral primary somatosensory cortex [26]. All areas with abnormal connectivity to the above-mentioned networks are shown in Table 1 and Additional file 1 and Fig. 2.

Fig. 2
figure 2

Overview of areas which have affected functional connectivity to 20 different networks reported in a total of 25 studies of interictal migraine patients compared with healthy controls

Full size image

Ictal migraine versus non-headache controls

Three conventional resting-state FC studies (one MA and two MO) have been published during compared to outside of migraine attacks. Following networks or areas showed altered connectivity during the attack versus outside of the attack: salience network [27], somatosensory network [27], default mode network [27], left pons [28] and right thalamus [29]. All areas with abnormal connectivity to the above-mentioned networks and areas are shown in Table 2.

Discussion

Based on this first systematic review of isolated conventional FC studies in migraine, we report that several areas and networks throughout the brain, brainstem and cerebellum showed altered connectivity in interictal and ictal migraine studies.

The findings are very diverse, with change in FC in many area thought to relevant for migraine as well as several other areas. The fact that almost all published studies report changes to some degree in all areas studied makes it difficult to gather the results into a coherent model, of specific activation patterns of activation in migraine.

All included studies (Tables 1 and 2) shared many characteristics; they used a 3 T MRI scanner, same type of patients (either MA or MO according to the International Classification of Headache Disorders criteria) and controls and in addition analyzed data using almost similar approaches (ICA or seed-based) in either the FSL or MATLAB-based software packages. Seed-based analysis can be affected by the chosen seed. Alterations in the default mode network (DMN) is most frequently reported. However, selection of different seed coordinates for DMN could potentially be the reason why FC changes in the DMN are different across studies. The strength of ICA is that it is independent of seed selection and more reproducible findings should be expected. The ICA-approach has been used in 10 studies and even in these studies different findings were reported.

Migraine is a heterogeneous disorder (with different disease duration, attack frequency, co-morbidity, effect of treatment, presence of aura), which might cause variation in results between studies. We did, however, only include studies where headache was diagnosed according to strict and uniform International Classification of Headache Disorders criteria.

In recent resting-state fMRI studies supplementary analyses like the Granger causality [30,31,32] have been introduced to investigate if FC changes can be linked to migraine phenotypes in the examined populations, but even here findings cannot be reproduced. As it is clear from Additional file 1 the findings are scattered and show very little overlap (Additional file 1). Moreover, none of the reported FC changes may be specific for migraine as other studies reported similar or exact same network changes in several other conditions, including fibromyalgia [33], Parkinsonian syndromes [34, 35] altered consciousness states [36], systemic lupus [37] and chronic hepatitis C virus infection [38]. Thus, it can be suspected that this FC method is at all not reproducible, which may be due to lack of sensitivity and specificity. Furthermore, to the best of our knowledge no sample size or power calculation guidelines are available for resting-state FC, with the consequence that a meaningful sample size for a resting-state FC study remains unknown. To avoid spurious findings, it would be useful to consider either sharing of data or joining patients in multicenter studies to allow for better and more reproducible studies.

As is already the norm for clinical trials, FC studies should be based on publically available protocols. It is also noteworthy that since very few studies report “negative results” or no changes in FC, primary endpoints should be chosen before initiating studies, as is already the case for randomized clinical trials (RCT). The fact that few (if any) results are reproducible, strongly suggest that stricter methodological guidelines for FC studies are warranted.

Almost half of the presented studies included only MO patients which gives a total sum of 348 MO patients, where 120 MA patients can be calculated in our tables. The FC method may be useful for the study of specific sub-types of migraine if these are clearly selected beforehand, preferable based on a calculation of the necessary number of patients, and with a clear hypothesis to be tested.

The FC method is very versatile and may potentially help improve our understanding of underlying disease mechanisms and even define biomarkers or migraine. Based on this systematic review, we suggest that the current lack of uniform study design, a priori hypothesis and diverse analyses and terminology makes it difficult to apply the available data for a coherent understanding of migraine.

Conclusions

Imaging, including FC studies could potentially help improve our understanding of underlying disease mechanisms, but so far no reproducible biomarkers of migraine have been identified. Future FC studies should either pool existing data to extract information about sub-phenotypes of migraine patients or follow guidelines similar to RCT guidelines in case of design of new FC studies.