Background
Migraine is a disabling and common neurological disorder, typically characterized by unilateral, throbbing, pulsating headache [
1]. It affects about 12 % of general population and causes substantial personal and social burden [
2]. During headache attacks, more than 60 % of migraineurs develop cutaneous allodynia (CA) [
3,
4], which is defined as a painful perception of an innocuous stimulation of the skin [
5]. CA is associated with high frequency of attacks, and has been proposed as a risk factor for progression to chronic migraine [
6,
7]. In addition, it has been also proposed as a predictor of poor response to triptan therapy [
8]. Therefore, probing the mechanism of CA in migraine may have significant implications for our understanding of the pathophysiology of migraine and its prognosis.
The hypothetical theory regarding the pathophysiological mechanism of CA is central sensitization of trigeminal neurons that take inputs from the intracranial and extracranial structures [
9‐
11]. Central sensitization is a condition in which neurons have lower activation thresholds, increased responsiveness to afferent inputs, increased spontaneous activity, and enlarged receptive fields [
12]. It is hypothesized that sensitization of the second-order trigeminovascular neurons in the spinal trigeminal nucleus propels the development of cephalic allodynia, and sensitization of the third-order trigeminovascular neurons in the posterior thalamic nuclei mediates the development of extracephalic allodynia [
13,
14]. Functional neuroimaging studies in humans and animals have shown activation of the trigeminal neurons in the posterior/dorsal thalamic nuclei in migraine [
9,
15,
16]. In addition, activation of the trigeminal system has been found to be associated with the ascending nociceptive transmission via the trigemino-thalamo-cortical pathway [
17]. As we know that thalamus is a relay region subserving both sensory and motor processing, with axonal fibers not only projecting out to the cerebral cortex in all directions but also receiving feedback informations from those multiple cortical areas [
18]. For pain perception, thalamus is believed to modulate ascending nociceptive information before transmitting nociceptive inputs to cortical structures [
19,
20]. Furthermore, thalamus is also a part of the descending pain inhibition pathways that modulate nociceptive inputs from spinal trigeminal nucleus [
21]. However, how the posterior thalamus affects the brain systems and whether it is affected by those brain regions in migraineurs with CA are poorly understood.
Effective connectivity and functional connectivity are two effective techniques to address the integration of functionally specialized areas in human brain [
22]. They both allow the inference of regarding communication between spatially remote brain regions, whereas effective connectivity, utilizing information about time-lagged relationships between brain regions, allows additional inferring about the directionality of information transfer within functionally connected networks [
23]. Previous resting-state functional magnetic resonance imaging (rs-fMRI) studies of migraine have found abnormal functional connectivity between thalamus and other pain matrix regions, such as periaqueductal grey (PAG), nucleus cuneiformis (NCF), or the affective regions (including insula, anterior cingulate cortex, and amygdala) [
24‐
26]. However, these studies are neither directly investigating the CA in migraine, nor are applying the thalamic-level effective connectivity analysis. Therefore, we hypothesize that studying the direct influence (i.e., effective connectivity) of thalamus with other structures can strengthen our understanding of the neural circuitry of CA in migraine.
In the present study, we aim to investigate the effective connectivity patterns of bilateral posterior thalamus with the rest of the brain in migraineurs with CA. Granger causality analysis (GCA) is a special approach to explore such effective connectivity among regions, and can be used to obtain the deductive network based on certain hypothetical seed region without requiring prior knowledge [
27]. So by employing the GCA method on rs-fMRI data, we can reconstruct an effective connectivity network associated with thalamus, which would provide us with fresh insights into the descending or ascending pain pathways at the thalamic-level in migraine with CA. To the best of our knowledge, this is the first study to examine the causal interactions between the posterior thalamus and the rest brain regions in migraineurs with CA using rs-fMRI.
Methods
Subjects
Thirty-four (
N = 34) right-handed migraineurs (aged 19–44 years) without aura as well as 25 age-, sex-, and handness-matched healthy controls were recruited for this study. The diagnosis of migraine without aura was made according to the Second Edition of the International Classification of Headache Disorder (ICHD- II) [
28]. The protocol of this study was approved by the Medical Ethics Committee of West China Hospital, Sichuan University, and all participants provided written informed consent. To avoid any possible pharmacological interference, all subjects had to be weaned off analgesic drugs 1 week or longer, not in preventive treatment and had not used any other drugs for at least 1 month prior to study participation. Subjects did not having a migraine attack at least 72 h prior to scanning. In addition, patients would be excluded if they had a migraine precipitated during the 2-day following-up. Potential subjects were excluded if they had any contraindication to MRI, had a prior brain injury, had a neurologic disorder other than migraine, had a psychiatric disorder other than anxiety or depression, or if they had any acute or chronic pain disorder other than migraine.
Clinical parameters
The Allodynia Symptom Checklist-12 (ASC-12) scale was administrated to divide all enrolled migraineurs into two groups: migraineurs with or without cutaneous allodynia (MWCA or MWoCA). ASC-12 is a 12-item questionnaire presented interictally to migraine patients, estimating the prevalence and severity of CA in the migraine population. The yielding score of each subject was placed into one of 4 categories: 0–2 = no allodynia; 3–5 = mild allodynia; 6–8 = moderate allodynia; 9 or more = severe allodynia. Then all participants were interviewed regarding demographic features (e.g., age, sex, and education). Additional information including the migraine history (e.g., onset age, frequency and duration of attacks, headache characteristic, and the pain intensity (evaluated with the visual analogue scale (VAS))) and the impact of headache (evaluated with the Migraine Disability Assessment Scale (MIDAS) [
29] and Headache Impact Test (HIT)-6 [
30]) was collected in those ones with migraine. For all subjects, psychiatric assessment, including 24-Hamilton Depression Scale (24- HAMD) [
31] and 14-Hamilton Anxiety Scale (14-HAMA) [
32] was also administered to assess the depression and anxiety state.
Statistical analysis of clinical data
Demographic and clinical features of MWCA compared to MWoCA and HC respectively were conducted using an independent sample t-test or chi square test, as appropriate.
Data acquisition
The experiment was performed on a 3.0 Tesla scanner (Trio Tim, Siemens, Erlangen, Germany) with a 16-channel birdcage head coil, and tightly padded clamps were used to minimize head motion. A routine T1 weighted imaging was first performed, and then the resting-state fMRIs were obtained by using an echo-planar imaging sequence with the following parameters: voxel size 3.75 × 3.75 × 5 mm3, TR 2000 ms, TE 30 ms, FOV 240 × 240 mm2, matrix 64 × 64, and slice thickness 5 mm with no gap, number of slices 30, number of time points 180. Throughout the scanning, subjects were instructed to lay in the scanner supine, relax, close their eyes but be awake.
Data spatial processing
The data were pre-processed using SPM8 (The Wellcome Department of Cognitive Neurology, London, UK,
www.fil.ion.ucl.ac.uk/spm/software/spm8). The first five time points of the resting-state data were discarded due to instability of the initial MRI signal, leaving 175 time points remaining for further processing. Then the functional images were slice-timing corrected and realigned to the first volume using a six-parameter rigid body transformation. Subjects with maximum head translation exceeded 2 mm or maximum rotation exceeded 2° were excluded. The mean image generated was then spatially normalized into standard stereotactic space, using the Montreal Neurological Institute (MNI) echo planar imaging (EPI) template. Computed transformation parameters were applied to all functional images, interpolated to isotropic voxels of 2 mm
3 and the resulting images were smoothed using an 4-mm full-width half-maximum (FWHM), isotropic Gaussian kernel. Then, using the Data Processing Assistant for resting-state fMRI (DPARSF) package [
33], linear drift was removed. A band-pass frequency filter (0.01 < f < 0.08 Hz) was then applied to reduce physiological high frequency noise [
34]. To further reduce the effects of confounding factors unlikely to be involved in specific regional correlation, we also removed several sources of spurious variance by linear regression, including six head motion parameters, and average signals from cerebrospinal fluid, white matter according to previous fMRI studies [
35,
36].
Granger causality analysis and statistical analysis
We used Granger causality to describe the effective connectivity between the reference time series of the seed regions (left and right posterior thalamus (PTH)) and the time series of each voxel within the whole brain. The coordinates (peak MNI coordinates: the right PTH = 3, −14, 6; the left PTH = −6, −21, −3) were obtained based on previous studies that showed structural or functional alterations in PTH in migraineurs [
37,
38]. Two 6 mm-radius sphere seeds based on the peak coordinates were then designed for GCA. Bivariate first-order coefficient-based voxelwise GCA was performed using the REST-GCA in the REST toolbox (
http://www.restfmri.net) [
39]. Granger causality estimates the causal effect of the seed region on every other voxel in the brain (X to Y effect), as well as the Y to X effect, the causal effect of every voxel in the brain (Y) on the seed region (X). A positive coefficient from X to Y indicates that activity in region X exerts a causal influence on the activity region Y in the same direction (i.e., positive influence). Similarly, a negative coefficient from X to Y suggests that the activity of region X exerts an opposing directional influence on the activity of region Y (i.e., negative influence). Using this approach, we are able to build a Granger-causal model based on the temporal elements of regional BOLD activity. The generated voxel-wise GCA maps were than transformed to z scores to improve the normality [
40]. Two-tailed two-sample
t-tests were conducted on the causal effects to compare MWCA to MWoCA and HC respectively with a Gaussian random field (GRF)-corrected significance (voxel-level of
P < 0.01 and joint cluster-level of
P < 0.05). Age, sex ratio, and education were applied as covariates in the two-sample
t- tests.
Clinical correlation analysis
To investigate the association between allodynia severity and abnormal effective connectivity, the regions showing significantly different (increased or decreased) Granger influences between the MWCA and MWoCA, or MWCA and HC comparisons were extracted as regions of interest (ROIs). Mean Granger causality values within these ROIs were correlated against the ASC scores of MWCA using Spearman correlation analysis. Statistical analyses were performed in SPSS 17.0 (SPSS Inc., Chicago, Illinois, USA) and threshold was set at P < 0.05.
Discussion
The present rs-fMRI study with Granger causality analysis demonstrated effective connectivity alterations from and to the bilateral PTH in MWCA comparing to MWoCA and HC respectively. Specifically, the abnormal effective connectivity pathways included: 1) one thalamo-cortico-thalamic circuit and one ipsilateral feedback inflow to the PTH in MWCA compared with MWoCA, i.e., bidirectional connectivity between the bilateral PTH and the mPFC (including the ventral and dorsal part), and inflow from the left limbic regions (including the amygdala, hippocampus, and BG) to the ipsilateral PTH; 2) one bidirectional thalamic-cortical circuit and another two ipsilateral feedback inflows to the PTH in MWCA compared with HC, i.e., bidirectional connectivity between the right PTH and the bilateral temporoparietal regions (including the STG and IPL), inflows from the left parietooccipital regions (including the precuneus, cuneus, and PCC) to the ipsilateral PTH and from right DLPFC (including the IFG and MFG) to the ipsilateral PTH. Moreover, the disturbed effective connectivities between PTH and cuneus, as well as PTH and MFG were associated with the severity of allodynia. These results confirmed the previous notion that central sensitisation of the third-order trigeminal neurons in the thalamus was implicated in the mechanism of CA in migraine [
14,
41]. Additionally, some regions in those abnormal effective connectivity pathways were involved in affective, cognitive, and sensory discriminative dimensions of pain, which suggested that CA implicated a multi-dimensional pain processing dysfunction.
One remarkable finding of the present study was the abnormally causal connectivity between the bilateral PTH and the mPFC (including the dorsal and ventral mPFC) in both MWCA vs. MWoCA and MWCA vs. HC. Results of the comparison between the two migraine groups suggested that CA may be associated with the decreased inflow from the left dmPFC to the ipsilateral PTH and increased inflow from the right vmPFC to the ipsilateral PTH. mPFC has widespread anatomical and functional connections with thalamus [
42], and is thought to play a critical role in emotional-cognitive processing [
43]. Furthermore, the two subregions of mPFC (i.e., vmPFC and dmPFC) have functional differentces,: the dmPFC is involved in evaluation of the perceived pain sensation [
44], while the vmPFC has an important role in regulation of sensory and affective pain [
45]. Therefore, the present results suggested that patients with CA had an impaired ability of evaluation of the perceived pain sensation. This was consistent with the definition of CA which is defined as pain (an exaggeration of the sensory inputs) resulting from the application of a non-noxious stimulus such as heat, cold or pressure to normal skin. Results also suggested that MWCA showed increased descending pain modulation which might be considered a response to the exaggerated pain information. Previous study has suggested that the medial thalamus and the mesial cortex (including the mPFC and ACC) involve the affective-motivational component of pain [
46]. Other resting-state fMRI studies also found abnormal mPFC activity to be implicated in the abnormal affective pain perception, cognition, and modulation in migraineurs [
25,
36,
47]. However, whether the dysfunctional effective connectivities between PTH and mPFC contribute to CA or are just cumulative effective results of CA still merits further study.
Another abnormal effective connectivity associated with CA was the decreased influence from the amygdala, hippocampus, putamen, and pallidum to the left PTH. All these subcortical regions have dense connections to thalamus [
18,
48], and have been implicated in the cognitive/affective aspect of pain processing in migraine [
49‐
52]. The involvement of these subcortical regions may correlate with the strongly relevant individual concern, like fears and anxiety [
53] resulting from recurring and repetitive attacks of migraine [
54]. Furthermore, the result is supported by a previous fMRI study which found that migraine patients with very high frequency of migraine attacks per month in response to the thermal stimuli demonstrated significantly lower activation throughout the caudate, putamen, and pallidum nuclei of the BG than those with low frequency of migraine attacks [
51]. Therefore, based on these findings, we speculate that present abnormal cognitive and emotional pain processes may result from the accumulating damages from recurring and repetitive attacks, which may involve the cognitive-emotional impairment in migraineurs with CA.
Reduced influence from the left parietooccipital area (including cuneus, precuneus, and PCC) to the ipsilateral PTH was also found in MWCA comparing with HC, and was negatively correlated with the ASC (allodynia assessment index), which suggested the abnormal feedback flow through the parietooccipital-thalamic circuit to be implicated in the underlying mechanism of CA. Most of these regions are important parts of the default mode network (DMN), which is the most consistently reported intrinsic connectivity networks during resting-state and has been implicated in sensation integration, self-relevant, and internally cognitive-attentional dimensions of pain in migraineurs [
55,
56]. Decreased connectivity between the parietooccipital regions and PTH reported herein may suggest an abnormal feedback from DMN regions to PTH, and certainly affects the following pain processing as relevant to the DMN function in migraine. In addition, the results also found increased influence from the right DLPFC (including IFG and MFG) to ipsilateral PTH in MWCA vs. HC, as well as its positive relationship with ASC. Concerning pain perception, the prefrontal regions including IFG have an important role in modulating pain processing during attentional manipulation paradigms [
57]. DLPFC, along with medial thalamus, has also been reported to engage in the pain processing during heat allodynia [
58]. Thus we speculate that the DLPFC exerts abnormal active control on sensory integration involving in the CA through the cortico-thalamus pathway.
The result revealed an abnormal bidirectional interaction between the right PTH and the bilateral temporoparietal areas (including the STG and IPL) in MWCA vs. HC. Specifically, there were increased inflow from the bilateral temporoparietal area to the right PTH and decreased outflow from the right PTH to the bilateral temporoparietal areas. Previous morphological and functional studies in migraine showed that migraineurs have reduced gray matter in STG and IPL [
59,
60], as well as less pain-induced activation of STG than HC in the interictal state [
61]. STG is an essential structure involved in auditory processing and social cognition [
62], and is generally associated with the IPL to serve the so-called ventral stream of the visual pathway [
63], which acts as a link between auditory and visual processing, perception and memory [
64]. When taken together with the functional properties of these brain regions, the present findings suggested that migraineurs suffered from a condition of global dysfunction of sensory integration and memory processes, as well as abnormal descending modulation through the cortico-thalamic pathway.
Study limitations
There are several limitations in the present study. First, the sample size of patients was small. The severity of allodynia symptoms, the length of disease course, the history of medication use and other characteristics were possible confounding variables, but we could not make any subgroup comparisons due to the few participants. Second, the current study did not present the large results of the MWoCA vs. HC, because we exclusively focused on the results that was directly associated with CA comparing MWCA with either MWoCA or HC. Finally, we have not performed a long-term evaluation or rescanned the subjects yet, so whether the functional alteration is a predictor or a consequence of central sensitization cannot be determined by the present study.
Competing interest
The authors declare that they have no competing interests.
Authors’ contribution
TW: experimental design, image data analysis, results interpretation, manuscript drafting; WZ: manuscript revision, Grammatical error correction; NC: data acquisition, results interpretation, manuscript drafting; JL: results interpretation, manuscript drafting; JPZ, QL, HH, and LH: manuscript revision; JRZ and QYG: conception and design of the study, results interpretation, manuscript revision; All authors read and approved the final manuscript.