Eighteen patients suffering from medication overuse headache and migraine (with and without aura) were recruited in the headache outpatient department of the University Clinic of Hamburg Eppendorf and scanned at two time points: before and after medication withdrawal. The time points of scanning had to differ for a minimum duration of 8 weeks to allow the efficacy of drug withdrawal. In the patient group, we assessed the number of headache days in the month before the first scan and continuously until the second scan. The number of headache days in the month before each scan were taken as measure of withdrawal success. We also assessed which and how often analgesia were used and whether and when a preventative medication was started. Two patients already used preventative medication (1 Flunarizin, 1 Venlaflaxin in combination with Valproic Acid) at the first measurement, while 11 patients used preventative drugs at the second time point after drug withdrawal (4 Metropolol, 4 Flunarizin, 1 Amitriptylin, 1 Amitriptylin in combination with Flunarizin, and 1 Venlaflaxin in combination with Valproic acid) leaving 7 without preventative medication. Patients further completed German versions of the Becks Depression Inventory (DBS-II) (12) and a symptom checklist (SCL-90-R) (13) to assess changes in depression and anxiety scores. Anxiety was measured as the average score in 10 anxiety related questions (questions: 2, 17, 23, 33, 39, 57, 72, 78, 80, 86). Additionally, 18 age- and gender matched healthy controls without a history of headache were invited to participate and were also scanned twice at the same interval as the patients to control for time and session effects. Controls were free of any neurological or psychiatric diseases, did not suffer from headache and had no 1st grade relatives suffering from primary headache. Details regarding clinical data can be found in Table 1.

The experiment followed a well-established paradigm inducing trigeminal nociception which has been published before in detail (14–17). In short, volunteers were exposed to 15 standardized trigeminal nociceptive stimuli (gaseous ammonia) and several control stimuli through an air-proofed Teflon tube which was, on one end, connected to an olfactometer outside of the scanning room. Control conditions were either an olfactory (rose odor), an odorless (air puffs) stimulation (both used 15 times for 1 s each), or a visual stimulus (flickering checkerboard, shown 16 times for 4 s each). All conditions were pseudo-randomly presented with a jittered time lag and after each stimulus the participants rated the stimulus for intensity and unpleasantness on a visual analog scales reaching from 0 to 100. As we focused on nociceptive processing, we only analyzed ratings of the painful stimulations and used two-tailed two sample t-test for both time points individually to assess differences between healthy volunteers and patients at these two time points. The protocol was approved by the Ärztekammer Hamburg (PV2814). All subjects gave written informed consent in accordance with the Declaration of Helsinki.

Both sessions were identical. All subject were instructed how to perform the experiment and went into the 3 Tesla MR Scanner (TIM TRIO, Siemens, Germany). Here, high resolution MPRAGE (1 mm³) structural images and functional images (echo planar imaging sequence, repetition time 2.62 s, echo time 30 ms, flip angle 80°, field of view 220 × 220 mm², 2 × 2 mm² in-plane resolution, 40 axial slices per volume with 2 mm slice thickness, and 1 mm gap) during the actual experiment were acquired in approximately 55 min. Further details on the MR parameters can be found elsewhere (18–20).

Structural scans from both sessions were analyzed with the Computational Anatomy Toolbox (CAT12 toolbox by Christian Gaser, http://www.neuro.uni-jena.de/cat/) for statistical parametric mapping (SPM12, http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) to assess gray matter volume alterations by means of voxel based morphometry (VBM) (21, 22). For this approach the structural image from both sessions of each subject were co-registered to their mean, segmented and normalized to MNI space and modulated to preserve gray matter volume. Resulting images had an isotropic resolution of 1.5 mm³ and were then smoothed by an 8 mm³ isotropic Gaussian kernel and entered the statistical group comparison. Total intracranial volume was extracted from the CAT12 results and used as a covariate of no interest for all following statistical analyses. Statistical analyses included (1) two sample t-tests for the first as well as second session to compare the group means, flexible factorial analysis of variance (ANOVA) with the factors subject, group and time to statistically assess (2) differences between time points for both groups individually as well as (3) the interaction of the factors group and time (group × time). Results are presented at a statistical threshold of p < 0.05 family wise error (FWE) small volume corrected (using a sphere with 12 mm radius as search volume and an initial voxel-level threshold of p < 0.001) with a cluster extent of more than 30 voxel (101.25 mm³ ~ 0.10 mL). Region of interest were hypothesized from the current available literature on medical overuse headache and used as coordinates for the small volume correction.

Regions of interest resulting from the VBM analysis were further correlated with the reduction of headache days per month (absolute and percentage) by means of Spearman's correlations to assess their predictability of the clinical outcome.

The functional images of both sessions were denoised, using spatially adaptive non-local means (23) denoising (implemented in the aforementioned CAT12 toolbox), realigned and slice time corrected. The structural image from the first session was co-registered to the mean image of both session, segmented and normalized to MNI space. Normalization parameters of the structural images were applied to all functional images, warping also them to MNI space. Afterwards functional images had an isotropic resolution of 2.0 mm³ and were smoothed with an 8 mm³ isotropic Gaussian kernel and entered a general linear model (GLM) analysis. Here 5 conditions (ammonia, air puffs, rose odor, visual stimulation, and all button presses) were modeled by convolving a hemodynamic response function (implemented in SPM12) with the onsets of the stimuli and fitting them to the actual data.

Contrast images from the condition pain and air puffs (differences of their beta-images which result as fitting parameters from the GLM analysis) signifying strength of neuronal activity during nociception entered the statistical group comparison, modeled similar to the one in the VBM analyses, including (1) two sample t-tests for the first as well as second session to compare the group means, flexible factorial ANOVA with the factors subject, group and time to statistically assess (2) differences between time points for both groups individually as well as (3) the interaction of the factors group and time (group × time). As for the VBM analyses, results for the functional imaging are presented at a statistical threshold of p < 0.05 FWE small volume corrected (using a sphere with 12 mm radius as search volume and an initial voxel-level threshold of p < 0.001) with a minimum cluster extent 30 voxel (240 mm³ =∧ 0.24 mL). Region of interest were hypothesized from the current available literature on medical overuse headache and used as coordinates for the small volume correction.

Alterations of functional connectivity of the OFC were analyzed using Psychophysiological Interaction (PPI) analysis as implemented in SPM12 (24) and statistically tested using aforementioned ANOVA. Here we used the coordinates from the original work of Fumal (5) and colleagues with a sphere of 12 mm radius as seed region and the contrast between the nociceptive (ammonia) and control (air puffs) condition as psychological variable. The same aforementioned statistical thresholds were used. The coordinates for the seed of the PPI was not taken from the VBM results of this data-set to prevent circular reasoning (25).

To investigate coherences between structural and functional features we calculated an exploratory regression analysis using baseline GMV of the OFC cluster as a covariate for the contrast nociception at time point 1 > time point 2.

Age and time interval between scans of patients and controls were not significantly different at an alpha level of α = 0.05. As we were specifically interested in withdrawal, we only included patients into this analysis who were successful in withdrawing medication. Patients who were unsuccessful in medication withdrawal (n = 2 out of 20 patients) withdrew consent to be scanned a second time and were not available for statistical comparison. Between scans, the patients showed a reduced headache frequency after withdrawal of 12 ± 5 days per month on average which equaled 54 ± 20 % (mean ± standard deviation) relative headache reduction (RD) calculated as RD = (T1–T2)/T1^*100 where T1 signifies the number of headache days per month at time point 1 and T2 at time point 2, respectively (Table 1). The reduction corresponds to a change from a chronic to an episodic migraine. Pain ratings were significantly higher in the patient than in the control group only at the first time point (p = 0.029 for intensity and p = 0.047 for unpleasantness, respectively) at an alpha level of α = 0.05. However, there were no differences within the groups between the first and second time point. Pain intensity as well as unpleasantness ratings were available in all but one patient. Neither anxiety (SCL-90-R) nor depression (DBS-II) scores were significantly different between the two scanning times for the patient group.

Decreased gray matter volume in patients as compared to controls in the first time point (before drug withdrawal) were found in the hippocampus, precuneus, inferior frontal gyrus (IFG), and bilateral orbitofrontal cortex (OFC), more specifically the medial orbital gyrus (MOG). The comparison between patients and controls at time point 2 (after a minimum of 8 weeks of drug withdrawal) revealed similar results namely decreased gray matter volume in OFC, hippocampus, IFG and additionally superior frontal and middle temporal gyrus and the cerebellum. However, an interaction analysis between the factors group and time did not reach significance. See Figure 1 and Table 2 for details.

The comparison of gray matter volume before and after drug withdrawal revealed an additional decrease of gray matter volume in 4 clusters including cuneus, superior temporal gyrus, putamen, and cerebellum (Table 2). The healthy participants showed no significant changes in gray matter between the first and second time point.

The gray matter volume of the right MOG before withdrawal showed a positive relation with the clinical outcome in reduction of headache days per month (absolute) (r = 0.560, p = 0.016) and relative reduction (RD)(r = 0.515, p = 0.029) (Figure 2).

We found a significantly increased activation in the left spinal trigeminal nucleus (sTN), right operculum (lowering threshold bilateral), and left posterior insula after withdrawal compared to before (Figure 3A and Table 3). No change of BOLD signal was found in the control group between time points. The Interaction analysis of nociceptive stimulation of group x time showed no significant effect at the chosen threshold similar to the comparison between healthy controls and patients at both time points.

Following medication withdrawal, e.g., at time point 2 the right MOG showed a significantly increased connectivity to the right STN, bilateral cerebellum in the patient group only. See Table 3 and Figure 3B.

Taking the average value of the baseline GMV of the individuals' left OFC cluster (coordinates −12, 38, −24) as a covariate for the contrast (ammonia-air)_post-(ammonia-air)_pre did not show any significant results in the F-contrast using a voxel-wise FWE-corrected threshold of p < 0.05. Lowering the threshold to an uncorrected threshold of p < 0.001 and a minimum cluster size of 30 voxel in the negative T-contrast (i.e., lower GMV of the OFC at baseline explains variance of increase in nociceptive activation) reveals 6 clusters, namely: left planum polare, left anterior insula, right cerebellum and STN, right operculum, right fusiformis gyrus, and left posterior insula. Patients and controls were analyzed together. Supplementary Image 1 shows these results superimposed on an MRI).

The interpretability of the results of this study needs to take the relatively small number of subjects into consideration, although the longitudinal approach allows taking each patient (and healthy participant) as their own controls and such reducing the variance. A possible limitation is that we investigated patients with and without a preventative medication. Ideally, such a study would only include patients who are medication free. However, as for ethical reasons the imaging study had to follow clinical decisions, some patients were on preventative medication to help decreasing migraine frequency. Given that preventative medications did not show functional changes in the areas we found in MOH patients but rather changes in the thalamus (18) and hypothalamus (19), we do not think it plausible that preventatives alone explain our findings. On the whole there are few imaging studies in MOH. Our data add to these findings that the OFC shows a decreased functional connectivity between the OFC and top-down and bottom-up pain processing regions namely the STN and cerebellum. Future studies should try to investigate patients who do not take preventatives.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.