Increased intrinsic brain connectivity between pons and somatosensory cortex during attacks of migraine with aura

Abstract

The neurological disturbances of migraine aura are caused by transient cortical dysfunction due to waves of spreading depolarization that disrupt neuronal signaling. The effects of these cortical events on intrinsic brain connectivity during attacks of migraine aura have not previously been investigated. Studies of spontaneous migraine attacks are notoriously challenging due to their unpredictable nature and patient discomfort. We investigated 16 migraine patients with visual aura during attacks and in the attack‐free state using resting state fMRI. We applied a hypothesis‐driven seed‐based approach focusing on cortical visual areas and areas involved in migraine pain, and a data‐driven independent component analysis approach to detect changes in intrinsic brain signaling during attacks. In addition, we performed the analyses after mirroring the MRI data according to the side of perceived aura symptoms. We found a marked increase in connectivity during attacks between the left pons and the left primary somatosensory cortex including the head and face somatotopic areas (peak voxel: P = 0.0096, (x, y, z) = (−54, −32, 32), corresponding well with the majority of patients reporting right‐sided pain. For aura‐side normalized data, we found increased connectivity during attacks between visual area V5 and the lower middle frontal gyrus in the symptomatic hemisphere (peak voxel: P = 0.0194, (x, y, z) = (40, 40, 12). The present study provides evidence of altered intrinsic brain connectivity during attacks of migraine with aura, which may reflect consequences of cortical spreading depression, suggesting a link between aura and headache mechanisms. Hum Brain Mapp 38:2635–2642, 2017. © 2017 Wiley Periodicals, Inc.

INTRODUCTION

The migraine aura is one the most striking features of clinical neurology. Migraine aura symptoms, typically in the form of transient visual, sensory or language disturbances, directly reflect cerebral cortical dysfunction in awake and conscious patients [Headache Classification Committee of the International Headache Society (IHS), 2013]. These symptoms are most likely caused by the electrophysiological phenomenon of cortical spreading depression (CSD) [Leao, 1944], though CSD has only been indirectly observed in humans during aura [Charles and Baca, 2013]. Migraine aura is firmly linked to the subsequent disabling headache phase of the migraine attack but the mechanisms that provide this link are not yet known. Uncovering these early, attack‐initiating processes is crucial for the understanding of migraine pathophysiology and for the targeting of novel prophylactic treatments [Tfelt‐Hansen and Olesen, 2012].

A prevailing theory of migraine attack pathogenesis holds that intrinsic brain events, such as CSD, may activate the trigeminovascular pathway via descending pain modulatory circuits. In animals, CSD has been shown to activate neurons in the trigeminal ganglion [Zhang et al., 2010] and brainstem nuclei via stimulation of trigeminal afferents [Bolay et al., 2002; Karatas et al., 2013; Zhang et al., 2011]. Altered intrinsic brain dynamics due to CSD may thus contribute to the development of the headache phase of a migraine attack.

Resting state functional MRI (rs‐fMRI) is a noninvasive tool for the study of intrinsic brain signaling. We previously investigated brain connectivity using rs‐fMRI in 40 patients with typical aura outside of attacks compared with individually matched healthy controls and found no abnormalities [Hougaard et al., 2015a]. However, abnormal signaling may develop during attacks of migraine with aura. Particularly, the affected cortical areas (e.g., visual areas during visual aura) could establish functional connections to other brain areas known to be involved in migraine, thereby linking aura and pain. Imaging studies during migraine attacks are extremely challenging due to the unpredictable nature of the attacks and especially because patients are severely incapacitated during attacks. In the present study, we investigated migraine patients with typical visual aura in the course of spontaneous attacks using rs‐fMRI. We hypothesized that intrinsic connectivity between visual areas and areas involved in migraine attack‐initiation would increase during attacks.

MATERIALS AND METHODS

Study Design and Participants

Patients were eligible for inclusion if they were aged between 18 and 65 years, and had a verified diagnosis of migraine with aura in accordance with the International Headache Society criteria [Headache Classification Committee of the International Headache Society (IHS), 2013].

Exclusion criteria were as follows: a history of any other primary headache (except episodic tension‐type headache for <5 days per month), pregnant or breast‐feeding women, contraindications for MRI (i.e., metal in the body or claustrophobia), cardiovascular or cerebrovascular disease, or psychiatric disease or drug abuse. We recruited patients via announcement on a Danish website for recruitment of volunteers to health research (http://www.forsoegsperson.dk) and on the intranet for hospital workers in the Capital Region of Denmark. Enrolment was done at Rigshospitalet Glostrup from Nov 28, 2012, to Jan 14, 2014. The Ethical Committee of the Capital Region of Denmark (H‐3‐2012‐073) approved the study. All patients gave written consent after receiving detailed oral and written information and the study was done in accordance with the Helsinki II Declaration of 1964, with later revisions. We applied a paired longitudinal design in which patients served as their own controls (i.e., investigated during attack vs. outside of attack), and thus no control subjects were included.

General Procedures

All patients were asked to telephone a member of the research team (AH or FMA) when they experienced an attack of migraine with aura. Patients were instructed to come to the hospital by taxi (fare reimbursed) to undergo an MRI investigation following attack onset. The patients were also instructed to write down details of their aura symptoms during attacks and they were interviewed about their symptoms on arrival to the hospital before MRI scanning. MRI was performed as described below. Additionally, all participants who were scanned during attacks were subsequently scanned on an attack‐free day where patients had to be headache (48 h) and migraine (72 h) free. Analgesics or triptans were not allowed 48 h before scanning on both experimental days. Headache intensity was evaluated on a verbal rating scale (VRS) from 0 to 10 (0: no pain, 10: worst pain imaginable).

MRI Procedures

MRI was performed on a 3T MR unit (Philips Achieva) using a 32‐element phased‐array head coil. To minimize head movement foam pads were placed in the head coil in both temple regions. Anatomical images were acquired using a 3D T1‐weighted MP‐RAGE (magnetization‐prepared rapid gradient‐echo) sequence with repetition time (TR) of 6.9 ms, echo time (TE) of 2.78 ms, field‐of‐view 263 × 281 × 150 mm³ and matrix size 256 × 256 mm³, 137 sagittal slices, 1.1 × 1.1 × 1.1 mm³ voxels).

Functional imaging used a gradient‐echo echo planar imaging sequence [32 slices of 4.0 mm thickness; slice gap 0.1 mm; field of view 230 × 230 mm; in‐plane resolution 2.9 × 2.9 mm; repetition time 3.0 s; echo time 35 ms; flip angle 90°; and SENSE (SENSitivity Encoding factor 2)]. Prior to acquisition, two volumes were omitted from the analysis to ensure steady‐state longitudinal magnetization. The lighting conditions in the scanner room were identical during each scan. Patients were instructed to keep their eyes closed, but to stay awake during the scan. They performed no additional task to avoid task‐related brain activations. A complete scan comprised 200 volumes and, thus, lasted 600 s.

Data Analysis

We used two complementary approaches to investigate changes in resting state intrinsic connectivity of spontaneous fMRI signals: (1) Seed‐based connectivity and (2) Independent Component Analysis (ICA).

All analyses were carried out with and without the addition of time from attack onset until scanning, and pain intensity as covariates. In order to normalize the data, images of patients with right‐sided visual aura symptoms were mirrored in the left–right direction so that the right hemisphere in standard space represented the symptomatic (i.e., contralateral to the perceived aura symptoms) side of all patients. All analyses were carried out on this laterality‐normalized dataset as well as data in the original orientation.

Data were initially processed using FMRIB's Software Library (FSL) including motion correction and spatial smoothing with a 5 mm full width at half maximal Gaussian kernel. In order to remove motion‐related, scanner‐related and physiological noise, we applied FSL's ICA‐based Automatic Removal of Motion Artifacts (ICA‐AROMA) [Pruim et al., 2015b]. We chose this method of data denoising because it is particularly suited for identifying motion artifacts [Pruim et al., 2015a] and since migraine patients could be expected to exhibit more movement during attacks, due to pain and discomfort, which could theoretically influence the results. Following this general denoising, white matter and cerebrospinal fluid was delineated by automated segmentation using FSL's FAST [Zhang et al., 2001]. Signals specifically from white matter and cerebrospinal fluid segmentations were extracted and removed from the data by including them as nuisance regressors in a general linear model (GLM). Nonlinear registration using FMRIB's Nonlinear Image Registration Tool (FNIRT) was applied between the subject's structural image and the Montreal Neurological Institute (MNI) standard space.

Seed‐Based Approach

We performed seed‐based resting state correlation analyses on seeds located in cortical visual areas (primary visual cortex, secondary visual cortex, V3, V4, and V5) and the lateral geniculate nucleus (LGN) of each hemisphere as defined by the FSL version of the Jülich Histological Atlas [Eickhoff et al., 2006]. Additional seeds were right and left side of the pons [MNI coordinates: (x, y, z) = (±8, −24, −32), 6 mm sphere], the hypothalamus [(x, y, z) = (0, 2, −6), 6 mm sphere], and the periaqueductal grey, as reported in a previous study [Mainero et al., 2011], [left: (x, y, z) = (−2, −28, −6) and right (4, −28, −6), 3 mm sphere]. Thus, a total of 17 seed locations were investigated.

Average time courses were extracted from these seeds and fitted with a voxel‐wise GLM consisting of an explanatory variable representing the demeaned extracted time courses. Statistical testing comparing the networks produced by the seeds in the ictal versus the interictal phase was performed using paired statistical testing by permutation‐based non‐parametric testing [Nichols and Holmes, 2002] (5,000 permutations) with threshold‐free cluster enhancement [Smith and Nichols, 2009], family‐wise error (FWE) corrected for multiple voxel comparisons. A two‐sided FWE‐corrected P < 0.05 was considered statistically significant. Results were not corrected for testing of multiple seed regions. These analyses were carried out for each seed with and without the addition of a set of covariates: time from attack onset until scanning and pain intensity during scanning.

ICA‐Based Approach

To compare networks obtained by ICA we used an entirely data‐driven regression approach (dual regression) that allows for voxel‐wise comparisons of resting state functional connectivity data. The details of this method have been described elsewhere [Filippini et al., 2009]. In short, the concatenated multiple fMRI data sets of all scans (ictal and interictal) were decomposed into common resting state networks using FSL MELODIC. The resulting networks were then tested in a voxel‐wise manner for differences between the ictal and interictal state using nonparametric permutation testing as described above. ICA analyses were carried out (a) using automatic estimation of number of components, and (b) with the total number of components restricted to thirty. Correction for multiple voxel comparisons was carried out as described above. A two‐sided FWE‐corrected P < 0.05 was considered statistically significant. Analyses were carried out with and without the addition of the nuisance variables described above.

RESULTS

We recruited 58 patients of whom 16 (nine women and seven men) completed functional and structural imaging on both scanning days and were included in the analyses (see Fig. 1). Characteristics of migraine attacks are shown for each patient in Table 1. The mean age at the time of scanning was 35.1 years (range: 22.3–58.8 years, SD 12.1 years). The median migraine with aura attack frequency was 12 attacks per year (range 1–24 attacks/year, SD 6.8 attacks/year) and the median disease duration was 16.2 years (range 5.2–32.8 years, SD 7.5 years). Three patients reported concomitant attacks of migraine without aura. On the attack day, patients gave a detailed description of their aura symptoms. All patients experienced visual symptoms initially during the attack. In some patients, visual aura was followed by sensory (n = 4) and aphasic (n = 1) symptoms. The mean time from aura onset to start of MRI was 8.2 h (range 1–22 h, SD 6.2 h). The mean headache intensity during scanning was 4.4 (range 2–8, SD 1.7). Patients did not exhibit more head motion during the ictal scan compared with the interictal scan (ictally: 0.659 mm vs. interictally: 0.655 mm absolute displacement, 95% CI of difference: −0.157 to 0.148 mm, P = 0.95, paired T‐test).

Figure 1.

Flowchart of the patient inclusion process.

Table 1.

Characteristics of migraine attacks in 19 patients during the MRI scanning sessions

Patient no.	Aura side	Aura symptoms	Aura duration (min)	Time to scan (h)	Headache side	Headache intensity	Associated symptoms (nausea/photophobia/phonophobia)
1	Left	V	20	2.5	Bilateral	5	Yes/yes/no
2	Right	V	30	13.4	Left	4	Yes/no/no
3	Left	V	18	12.3	Right	6	Yes/yes/yes
4	Right	V	20	56.9	Right	3	No/no/no
5	Left	V	60	21.6	Bilateral	2	No/no/no
6	Left	V	30	9.6	Right	3	Yes/no/no
7	Left	V	40	8.5	Right	5	No/yes/yes
8	Left	V	20	13.7	Bilateral	2	No/yes/no
9	Left	V	25	16.7	Right	6	Yes/yes/yes
10	Right	V	35	12.8	Left	3	Yes/yes/yes
11	Left	V + S	30	46.5	Right	2	Yes/yes/yes
12	Right	V + S	30	44.8	Left	6	Yes/yes/yes
13	Right	V	40	25.0	Left	5	No/yes/yes
14	Left	V + S	40	10.6	Right	5	No/yes/yes
15	Bilateral	V+ S + A	40	16.0	Left	5	No/yes/yes
16	Left	V	30	19.0	Right	8	No/yes/yes

Seed‐Based Analyses

For data in the original orientation (not flipped according to visual aura lateralization) we found a marked increase in connectivity during attacks between the left pons and the left primary somatosensory cortex corresponding to the head and face somatotopic areas [peak voxel: P= 0.0096, (x, y, z) = (−54, −32, 32), see Fig. 2]. Connectivity between the left pons and left superior parietal lobule also increased significantly during the attack phase (Fig. 2). For aura‐side normalized data, we found increased connectivity during attacks between visual area V5 and the lower middle frontal gyrus, both in the symptomatic hemisphere, that is, the hemisphere contralateral to the reported visual aura symptoms, [peak voxel: P = 0.0194, (x, y, z) = (40, 40, 12), see Fig. 3].

Figure 2.

Intrinsic connectivity for seed placed in the left pons, data in the original orientation (not flipped according to visual aura lateralization). A. The red sphere marks the seed location. Green: Areas functionally connected to the seed during the interictal phase. Blue: Areas functionally connected to the seed during spontaneous attack of migraine with aura. R marks the right hemisphere side; x, y, and z gives MNI coordinates for the slices. B. Areas of significantly increased connectivity during attacks compared with the attack‐free state. Connectivity did not decrease in any areas during attacks. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 3.

Intrinsic connectivity for seed placed in area V5 (MT) in the symptomatic hemisphere, i.e. the hemisphere contralateral to the perceived aura symptoms (aura‐side normalized data). A. The red area marks the seed location. Green: Areas functionally connected to the seed during the interictal phase. Blue: Areas functionally connected to the seed during spontaneous attack of migraine with aura. S marks the symptomatic hemisphere side; x, y, and z gives MNI coordinates for the slices. B. Areas of significantly increased connectivity during attacks compared with the attack‐free state. Connectivity did not decrease in any areas during attacks. [Color figure can be viewed at http://wileyonlinelibrary.com]

We found no changes in connectivity for any other seed region for data in the original orientation or aura‐side normalized data.

Analyses with added covariates did not significantly alter these results and showed no relation between the connectivity changes and time from attack onset to scanning or pain intensity.

ICA‐Based Analyses

Using the ICA‐based approach and automated dimensionality estimation, the denoised fMRI signals were decomposed into 56 (original orientation) and 58 (aura‐side normalized) individual intrinsic networks. No changes from the interictal to the ictal state were found for any of these networks.

DISCUSSION

The important novel discovery of this study was an intrinsic connection between the dorsal pons and an ipsilateral hemispheric network including the head and face areas of the somatosensory cortex during spontaneous attacks of migraine with aura. In addition, connectivity increased between the visual cortical area V5 (also known as MT, middle temporal) and the ipsilateral frontal cortex specifically in the aura‐affected hemispheres.

Increased Connectivity Between Pons and Parietal Cortex

We selected the pons as a seed region for this study since it is generally believed to be a key structure in the pathophysiology of migraine attacks. Increased brainstem activity specifically in the dorsal lateral pons has previously been reported in PET studies of spontaneous migraine attacks primarily without aura [Afridi et al., 2005; Weiller et al., 1995] and in attacks of migraine triggered by glyceryl trinitrate [Afridi, 2005] leading to a theory of the brainstem as a “migraine generator.” Whether this activation is migraine‐specific or if it is a general feature of ascending/descending pain transmission is not known.

In the present study, we observed a functional connection between the left side of the pons and the left somatosensory cortex during attacks. It is not obvious how this connection is established. The majority of patients reported aura symptoms in the left visual field and right‐sided pain (Table 1). This corresponds well with previous reports of patients generally experiencing headache contralateral to aura symptoms in more than 90% of cases [Russell and Iversen, 1994]. Thus, in most patients in the present study, the pathophysiological process would start in the right hemisphere while the subsequent painful state would be associated with left hemisphere activation. In animals, CSD causes activation of ipsilateral first‐order neurons [Zhang et al., 2010] and of neurons in the ipsilateral trigeminal nucleus caudalis at the level of spinal segments C1–C2 [Bolay et al., 2002; Zhang et al., 2011] and from there the nociceptive pathway further projects to the contralateral thalamus and somatosensory cortex. In addition, a recent functional MRI study of brainstem responses to trigeminal pain in healthy subjects reported pain‐related activation of the contralateral pons [Schulte et al., 2015]. The authors suggested that this area of increased activation corresponded to the locus coeruleus, which is the principal site for brain synthesis of norepinephrine and is responsible for mediating sympathetic responses during acute stress [Benarroch, 2009]. Thus, our data likely reflects synchronous input to the contralateral pons and somatosensory cortex due to activation of trigeminovascular nociceptors during the pain phase of the migraine attack. We did not find similar connectivity changes from the seed in the right pons. However, we speculate that a similar process could take place in patients with right‐sided aura and left‐sided head pain. Interestingly, the pontine intrinsic networks were not restricted to one side of the brainstem but rather indicated bilateral connections (Fig. 2). It should be noted that due to the relatively low spatial image resolution of this study it is difficult to draw firm conclusions about the lateralization of small structures such as pontine nuclei.

In addition, functional MRI of the brainstem poses challenges especially in terms of physiological noise and spatial normalization. We applied robust ICA‐based denoising and removal of residual noise by means of white matter and CSF regression. Co‐registration was carried out in a two‐step approach with initial linear registration, which was subsequently refined by non‐linear registration. In spite of this, physiological noise and normalization issues may have influenced the results. We were not able to detect increased connectivity involving the pons when flipping the images according to perceived aura lateralization. This is possibly due to the fact that resting state networks are not symmetrically distributed in the brain [Tian et al., 2011], and therefore there is a risk that the flipping procedure will “blur” the images in some regions, thus decreasing instead of increasing sensitivity. This effect seems particularly likely in the brainstem where even small displacements of seeds potentially will have consequences for the resulting connectivity.

Increased Connectivity Between V5 and Frontal Cortex

The extrastriate areas, that is, cortical visual areas other than V1, likely play an important role in visual migraine aura. CSD‐like BOLD‐signal changes were reported to initiate from area V3A during an attack of visual aura [Hadjikhani et al., 2001], and cortical structural abnormalities have been reported in V5 (also known as MT) and V3A of migraine patients [Granziera et al., 2006]. Area V5 is primarily involved in visual motion processing and, accordingly, deficits of motion perception have been demonstrated in MA patients interictally [McKendrick and Badcock, 2004].

It thus seems plausible that the observed abnormal connectivity pattern of the symptomatic hemisphere is directly associated with the visual aura. Interestingly, the frontal cortical area of increased connectivity corresponds to a component of a functional network showing hyperresponsivity to visual stimulation in the symptomatic hemispheres of patients with visual aura outside of attacks [Hougaard et al., 2015b]. While the exact role of the frontal cortex is not clear, the observed abnormal ictal connectivity indicates signaling between areas that are affected during aura and cortical areas involved in other processes, possibly the initiation of subsequent events in the migraine attack. Thus the results support aura‐induced alteration of intrinsic brain signaling reaching beyond brain areas affected by CSD in the aura phase.

Similar studies of intrinsic brain connectivity in migraine aura patients compared with controls, performed outside of attacks, have reported reduced connectivity between the anterior insula and extrastriate cortical visual areas, including area V3 [Niddam et al., 2015] and reduced connectivity within the so‐called “executive control network” in the middle frontal gyrus and the anterior cingulate gyrus [Tessitore et al., 2015]. We recently investigated a large group (N = 40) of interictal migraine patients with typical aura compared with individually age‐ and sex‐matched healthy controls using this technique and found no differences in intrinsic brain connectivity [Hougaard et al., 2015a]. The observed discrepancy between studies during interictal periods could be due to the cyclic nature of migraine, intrinsic networks could change in the pre‐ictal [Amin et al., 2016] or post‐ictal states, even when patients appear asymptomatic.

The ICA‐based approach showed no significant differences between the ictal and interictal states.

The strength of the seed‐based approach is that specific hypotheses involving certain anatomical regions are tested. The ICA‐based approach will ideally test all available functional networks and is considered to be more robust but also less sensitive than the seed‐based approach. Importantly, the two approaches are conceptually different: the same seed region may be part of several ICA derived networks [Joel et al., 2011].

In the present study, we did not observe abnormal connections directly between affected (i.e., visual) cortical regions and areas of the brain believed to be involved in migraine pain mechanism. This may suggest that CSD is not able to activate pain‐related structures, indicating that migraine pain is not a cause of CSD, but rather that pain and aura are both caused by unknown, parallel mechanisms [Charles, 2012]. Another explanation could be that the CSD‐induced signaling to pain‐related structures only occurs briefly during aura symptoms, since patients in the present study were scanned during the headache phase, when the aura was over. However, scanning during or immediately after spontaneous aura is not feasible due to the short‐lasting and unpredictable nature of the aura. An interesting approach would be to provoke migraine aura using hypoxia as demonstrated in a recent study [Arngrim et al., 2016] in order to perform rs‐fMRI in the aura phase.

In conclusion, rs‐fMRI in 16 migraine patients during and outside of spontaneous migraine attacks with typical aura revealed increased intrinsic connectivity between the pons and cortical areas involved in head and face pain processing increased during attacks as well as between extrastriate visual and frontal cortex. These observations further underline the important role of the pons in migraine pathophysiology and provide novel evidence of intrinsic brain connectivity alterations related to migraine aura.

DISCLOSURES

M. Ashina is a consultant or scientific advisor for Allergan, Amgen, Alder, ATI and Eli Lilly, primary investigator for an Amgen 20120178 (Phase 2), 20120295 (Phase 2), 20130255 (OLE), 20120297 (Phase 3) and GM‐11 gamma‐Core‐R trials, and reports grants from Lundbeck Foundation (R155‐2014‐171) University of Copenhagen, Research Foundation of the Capital Region of Copenhagen, Danish Council for Independent Research‐Medical Sciences and Novo Nordisk Foundation (NNF11OC101433) during the conduct of the study. A. Hougaard, F.M. Amin, H.B.W. Larsson, and E. Rostrup report no disclosures relevant to the manuscript. Funding sources played no role in study design, data collection, analysis, interpretation, manuscript preparation, or submission. The corresponding author had full access to and the final responsibility to submit all study data for publication.