Central modulation in cluster headache patients treated with occipital nerve stimulation: an FDG-PET study

In the 1^st group, patients (n = 4), underwent a PET session after 24 to 30 months of ONS. All patients but one belonging to the 2^nd group (n = 6) were scanned before implantation (baseline). In both groups, for each session after implantation, patients were scanned with the stimulator switched on (ON) and off for 3 days (OFF). This period was arbitrarily fixed in line with our previous observation of headache recurrence within 2 days on average after switching off the stimulator [8]. They underwent 2 more pairs of scans after 1 and 6 months of ONS. None of the patients had a cluster attack during PET, nor within the 12 hours preceding or following the procedure.

Data collected in patients were compared to a pool of 39 drug-free healthy volunteers (HV) (18 males, 21 females, mean age 45 ± 16 years SD) without headache history.

PET data were obtained on a Siemens CTI 951 16/32^® scanner (Siemens, Erlangen). Data were corrected for attenuation and background activity. Resting cerebral metabolism was studied after intravenous injection of 5-10 mCi (185-370 MBq) [18F]fluorodeoxyglucose (FDG). Subjects were scanned in a dark room, with minimal environmental noise.

Because we selected CH patients with side-locked unilateral attacks and as there was no side shift during the observational period, we flipped the PET scans of the patients with right-sided symptoms in the axial plane so that we could analyze all subjects together (all "left-sided"). We used a binary classification of patients responding or not to ONS with an arbitrary, but clinically relevant, cut-off point for responders set at 50% decrease in attack frequency. Given the data from previous functional imaging studies, we conducted our analysis with an a priori hypothesis towards regions known to be involved in CH and other TACs [5], areas which are modulated by ONS in chronic migraine [13] and areas belonging to the pain matrix.

Data were analysed using statistical parametric mapping (SPM8 version; Wellcome Department of Cognitive Neurology Institute of Neurology, London, UK; http://​www.​fil.​ion.​ucl.​ac.​uk/​spm) implemented in MATLAB (version 7.1, MathworksInc., Sherborn, MA). Images were spatially normalized into a standard stereotactic space using a symmetrical MNI (Montreal Neurological Institute) PET template [15] and smoothed using a 14 mm full-width-half-maximum (FWHM) isotropic kernel [16]. T-test was used with a significance level set at p < 0.001 uncorrected (p < 0.05 FDR).

The first design matrix included the scans of the 39 HV, of the 4 patients of the 1^st group performed 24 to 30 months post-ONS and of the 6 patients of the 2^nd group, scanned before implantation (baseline), 1 month and 6 months post-ONS. Our first analysis identified brain regions with a significant hyper- or hypometabolism in drCCH patients as compared to HV independent on time of scanning or ONS stimulation. Then, we looked for brain regions showing short term ONS-induced (ON versus OFF) increase or decrease in metabolism independent on delay since ONS implantation, in the early post-ONS phase (scans obtained after 1 month) and in the late post-ONS phase (scans obtained after 6 and 24-30 months). Finally, we compared metabolic activity measured during baseline, early phase (1 month) and late phase (≥ 6 months) post-ONS (independent of ONS stimulator settings) searching for progressive increases and decreases in metabolism over time.

In a second design matrix we searched for differences between the subgroups of 7 responders and 3 non-responders. Here, we included only the PET data obtained in the late phase (≥ 6 months; both with stimulator ON and OFF) and the HV scans, searching for regions with metabolic differences between the two groups (i.e., increased metabolism - as compared to HV - present in responders but not in non-responders).

For all analyses, the resulting set of voxel values for each contrast, constituting an SPM of the t-statistics (SPM{t}), was transformed to the unit normal distribution (SPM{Z}) and thresholded at p = 0.001. All group results were thresholded at false discovery-corrected p < 0.05, corrected for the whole brain volume. For differences between responder and non-responder subgroups, results were corrected for multiple comparisons within the regions of interest identified during the previous whole group analyses by employing a small volume correction (10 mm radius sphere).

The enhanced FDG uptake found ipsilaterally in hypothalamus of drCCH patients respective to HV is in line with the reports showing increased activity in this area with H2¹⁵O-PET or fMRI during attacks [18‐20]. However in our study all patients were scanned between attacks, i.e. at a time point when hypothalamic activation has not been reported yet. Another FDG-PET study of episodic CH comparing patients during and between bouts revealed no change in hypothalamic glucose uptake [21].

Areas belonging to the pain matrix like the cingulate gyrus or midbrain (periaqueductal grey - PAG) are classically activated in various pain states including headaches [5]. We also found activation in the cerebellum, in line with previous imaging studies showing consistent cerebellar activation across the spectrum of pain from visceral to somatic, acute to chronic [22]. Activation of cerebellar vermis and anterior lobe in drCCH may be particularly strong due to dense somatotopically arranged trigemino-cerebellar connexions [22], but also because there are direct connections between the ventro-posterior hypothalamic area and the cerebellum as shown by MRI tractography in a drCCH patient treated with hypothalamic DBS [23].

The perigenual ACC (PACC) is of particular interest. Contrary to our results, it was found hypometabolic compared to HV in episodic CH [21]. However, when the authors compared the same patients during and between bouts, the PACC was hypermetabolic during the bout despite the absence of an attack at the time of scanning.

Lower pons activation has been described during attacks of hemicrania continua, but not of CH [5]. In hemicrania continua, dorsal pontine activation is ipsilateral, like in our study, while posterior hypothalamic activation is observed on the opposite side, unlike in CH.

The pulvinar where we found ipsilateral activation in drCCH patients has not been a region of interest in CH before. Pulvinotomy and electrical stimulation of the pulvinar have been used successfully in the treatment of chronic pain in humans [24]. In functional neuroimaging studies, pulvinar hypermetabolism is either associated with pain state [25] or with pain relief after various procedures [24], including ONS in chronic migraine [13].

We have no straightforward explanation for the increased metabolism of the ipsilateral visual cortex in our patients. Such activation has not been reported hitherto. Photophobia ipsilateral to the pain is a frequent attack-associated symptom in various TACs, in particular CH [26]. Like in migraineurs, the visual cortex of CH patients might thus be more sensitive to light stimuli. This hypothesis seems unlikely, however, as all subjects were scanned in a dark room.

To summarize, the interictal FDG-PET hyperactive pattern of drCCH patients comprises areas reported to be hypermetabolic mainly during TAC attacks (ipsilateral hypothalamus and pons), but also during a bout of the disorder outside of an attack (perigenual ACC).

We found no significant differences between PET recordings performed with stimulator ON or OFF within a 72 hour delay. This finding is similar to Matharu et al.'s [13] observations in chronic migraine, except that they were not able to scan the patients OFF and pain-free because of an almost immediate recurrence of pain after interrupting the stimulation. They concluded that central structures were not modulated in chronic migraine by ONS beyond the stimulation period.

Here, lack of short-term metabolic modification favours a slow neuromodulatory effect of ONS, as we suspected before [8]. Interestingly, in CCH patients treated with hypothalamic DBS, May et al. [27] found rapid metabolic changes with H₂ ¹⁵O-PET in various brain structures involved in cluster headache and more generally in the pain matrix within only 10 minutes of switching the stimulator ON or OFF, but there were no clinical correlates suggesting that the therapeutic effect of DBS is also due to slow CNS changes.

Our subanalysis with a smaller significance level revealed a change in left lower brainstem metabolism. It might indicate a short term ONS effect in the trigemino-cervical complex. As expected from the neuro-anatomical connexions, the trigemino-cervical complex could relay stimulation to more rostral structures allowing for neuroplastic modulation of their activity. In comparison, a recent study of transcutaneous electrical nerve stimulation (TENS) in arthritic rats suggests that its antinociceptive effect is mediated by ascending activation of the opioid system originating in the ventrolateral PAG and projecting via the rostral ventro-medial medulla to the spinal cord [28].

Hypermetabolism of most overactive areas in drCCH patients compared to HV decreased after ONS. This was particularly obvious for the anterior cingulate cortex, left pulvinar, left visual cortex, left lower pons, cerebellum and midbrain. Conversely, baseline hypometabolism of sensori-motor cortices increased after ONS. The noteworthy exception to these post-ONS metabolic changes is the ipsilateral hypothalamus. This is precisely the region activated during CH attacks on the side of the pain [18] and where increased gray matter density is found on voxel-based MRI between attacks [29]. Our findings in drCCH contrast with those by Sprenger et al [21] in episodic CH where no significant metabolic change was found in the hypothalamus, either outside or during the bout. If replicated, our data suggest that persistent hypothalamic activation is a hallmark of chronic CH. They may explain why attacks are non-remittent but also why some ONS-treated, pain-free patients still have autonomic attacks [8]. A similar conclusion was drawn for the dorsal rostral pons in chronic migraine following the finding of persistent activation in this area despite ONS-induced pain relief [13]. The persistence of an ipsilateral hypothalamic activation despite reduced attack frequency also confirms that ONS is no more than a symptomatic therapy, as already suggested by the recurrence of attacks after interruption of the stimulation [8].

The metabolic changes observed after ONS could be due to the stimulation itself or to the reduction of attack frequency. Our protocol did not allow to favour either mechanism. However, despite the small number of non-responders in our study, some insight might be gained from the comparison of patients who clearly responded to ONS and those who did not.

Comparison of ONS responders and non-responders revealed increased FDG uptake in the PACC of the former. This area is of interest for several reasons. First, PACC plays a major role in central opioidergic pain control system. It is selectively activated during analgesia induced by the μ-receptor agonists fentanyl and remifentanyl compared with placebo [30], providing evidence that opioidergic analgesia is mediated by activation of descending antinociceptive pathways. Second, PACC was found hypometabolic respective to HV in episodic CH, but its activity increased significantly during the bout [21]. Knowing the pivotal role of PACC in descending pain control, these authors hypothesized that deficient endogenous antinociceptive mechanisms between bouts might predispose CH patients to the disorder and to its recurrence. Concordantly, Sprenger et al [31], using PET with the opioidergic ligand [11C]diprenorphine, demonstrated an inverse linear relationship between the duration of CH and opioid receptor availability in the rostral ACC (and ipsilateral hypothalamus). A recent case report by the same group of a drCCH patient in whom low dose levomethadone induced complete remission of attacks favours this hypothesis [32].

Compared to TENS that was shown to induce analgesia through activation of a PAG-RVM-spinal cord pathway [28], ONS could activate this descending pain-control pathway even further up-stream at the level of PACC. Its therapeutic effect in drCCH patients could thus be due to progressive restoration of activity in deficient opioidergic antinociceptive pathways.

We are well aware of some methodological flaws that may limit the strength of our findings.

First, the number of patients included is rather small. This is the case in most similar studies as drCCH patients are rare and ONS is an emerging treatment modality for which only 38 cases, including ours, have been published. In a much commoner condition like chronic migraine, metabolic imaging studies have been limited to less than 10 patients [13].

A second shortcoming is the dichotomy of the PET design. PET studies were not planned in our initial pilot study of ONS in 5 patients as the outcome was uncertain and the sample considered too small. As clinical efficacy was encouraging, we were allowed to recruit 6 additional patients, for whom imaging studies were planned prospectively. For greater sample size, patients from the 1^st group were also proposed to undergo PET. This explains why the latter had neither baseline nor 1 month scans and why long-term treatment periods vary between 6 and 30 months. We know since that there is no further clinical improvement after 6 months of ONS and that in most patients attacks recur after stimulation interruption whatever the duration of ONS [8]. This is why we decide to merge scans obtained between 6 and 30 months of ONS, albeit statistically questionable.

Finally, one may argue that the prophylactic drugs taken by the patients may have influenced the PET results. This cannot be ruled out, as HV did not take any medication. However, pharmacotherapy remained stable in all patients of the 2^nd group except one and was similar in ONS responders and non-responders.