in brain of Fibromyalgia patients inflammantory glail cells seen to be overly activated in study
Brain glial activation in fibromyalgia – a multi-site positron emission tomography investigation
Abstract
Fibromyalgia (FM) is a poorly understood chronic condition characterized by widespread musculoskeletal pain, fatigue, and cognitive difficulties. While mounting evidence suggests a role for neuroinflammation, no study has directly provided evidence of brain glial activation in FM. In this study, we conducted a Positron Emission Tomography (PET) study using [11C]PBR28, which binds to the translocator protein (TSPO), a protein upregulated in activated microglia and astrocytes. To enhance statistical power and generalizability, we combined datasets collected independently at two separate institutions (Massachusetts General Hospital [MGH] and Karolinska Institutet [KI]). In an attempt to disentangle the contributions of different glial cell types to FM, a smaller sample was scanned at KI with [11C]-L-deprenyl-D2 PET, thought to primarily reflect astrocytic (but not microglial) signal.
Thirty-one FM patients and 27 healthy controls (HC) were examined using [11C]PBR28 PET. 11 FM patients and 11 HC were scanned using [11C]-L-deprenyl-D2 PET. Standardized uptake values normalized by occipital cortex signal (SUVR) and distribution volume (VT) were computed from the [11C]PBR28 data. [11C]-L-deprenyl-D2 was quantified using λk3. PET imaging metrics were compared across groups, and when differing across groups, against clinical variables.
Compared to HC, FM patients demonstrated widespread cortical elevations, and no decreases, in [11C]PBR28 VT and SUVR, most pronounced in the medial and lateral walls of the frontal and parietal lobes. No regions showed significant group differences in [11C]-L-deprenyl-D2 signal, including those demonstrating elevated [11C]PBR28 signal in patients (p’s≥0.53, uncorrected). The elevations in [11C]PBR28 VT and SUVR were correlated both spatially (i.e., were observed in overlapping regions) and, in several areas, also in terms of magnitude. In exploratory, uncorrected analyses, higher subjective ratings of fatigue in FM patients were associated with higher [11C]PBR28 SUVR in the anterior and posterior middle cingulate cortices (p’s<0.03). SUVR was not significantly associated with any other clinical variable.
Our work provides the first in vivo evidence supporting a role for glial activation in FM pathophysiology. Given that the elevations in [11C]PBR28 signal were not also accompanied by increased [11C]-L-deprenyl-D2 signal, our data suggests that microglia, but not astrocytes, may be driving the TSPO elevation in these regions. Although [11C]-L-deprenyl-D2 signal was not found to be increased in FM patients, larger studies are needed to further assess the role of possible astrocytic contributions in FM. Overall, our data support glial modulation as a potential therapeutic strategy for FM.
1. INTRODUCTION
Fibromyalgia (FM) is a poorly understood chronic condition characterized by widespread musculoskeletal pain, fatigue, unrefreshing sleep, memory and attention difficulties, among other symptoms (Clauw, 2014). While the etiology of FM is unknown, central mechanisms are strongly implicated, including evidence of abnormalities in structure, function, and molecular chemistry of the central nervous system (CNS) (Albrecht et al., 2016; Clauw, 2014; Dehghan et al., 2016; Flodin et al., 2014; Gracely et al., 2002; Harris et al., 2007; Jensen et al., 2009; Jensen et al., 2010; Jensen et al., 2013; Kuchinad et al., 2007; Loggia et al., 2014; Loggia et al., 2015a; Napadow and Harris, 2014; Schreiber et al., 2017; Schroeder et al., 2016; Wood, 2008), though some evidence points to peripheral alterations as well (Oaklander et al., 2013; Uceyler and Sommer, 2013).
Dysregulation of neuroimmune activation is one potential mechanism contributing to previously reported central aberrations and central sensitization in FM. For instance, FM patients demonstrate elevated levels of fractalkine and interleukin-8 (IL-8) in cerebrospinal fluid (CSF) (Backryd et al., 2017; Ka setoff et al., 2012; Kosek et al., 2015). Both chemokines are implicated in neuron-glial communication (Montague and Malcangio, 2017; Puma et al., 2001), and have been associated with central sensitization and pain (Kosek et al., 2015; Montague and Malcangio, 2017). However, no study to date has clearly demonstrated that glial activation occurs in the brain of FM patients. Acknowledging a role for neuroimmune dysfunction in FM would open the exploration of glial modulation as a therapeutic option for this condition.
In the human CNS, glial activation can be studied in vivo using positron emission tomography (PET) and radioligands that bind to the 18-kDa translocator protein (TSPO), such as [11C]PBR28, which displays nanomolar affinity to this protein (~0.5nM; Imaizumi et al., 2008). Located mainly on the outer mitochondrial membrane, TSPO expression is low in healthy CNS tissue, but is widely upregulated in microglia and astrocytes under inflammatory conditions (Lavisse et al., 2012; Rupprecht et al., 2010). Our group has recently used TSPO PET imaging to document neuroimmune activation in the central and peripheral nervous system of patients with chronic low back pain (CLBP) (Albrecht et al., 2018; Loggia et al., 2015b). While no TSPO PET studies of FM patients have been published yet, a possible link between TSPO and FM pathophysiology is provided by an association between the Ala147Thr polymorphism (rs6971) in the TSPO gene and FM symptom severity and cerebral pain processing (Kosek et al., 2016).
The aims of the current study were to evaluate the hypothesis that brain TSPO binding in FM patients, as assessed using the [11C]PBR28 PET ligand, is 1) elevated compared to healthy controls (indicating the presence of glial activation) and 2) correlated with specific symptoms attributable to FM pathophysiology. An additional third aim was to tease out the most likely cellular sources of increased TSPO binding because, while TSPO upregulation in neuroinflammatory responses consistently colocalizes with microglia, an accompanying astrocytic component has been observed in some (Liu et al., 2016; Rupprecht et al., 2010; Toth et al., 2016; Wei et al., 2013), but not all cases (Abourbeh et al., 2012; Mirzaei et al., 2016). To this end, a smaller sample of FM patients received a PET scan with [11C]-L-deprenyl-D2, which binds to monoamine oxidase B (MAO-B) with high specificity in postmortem tissue (~30 fmol/mg tissue; Gulyas et al., 2011). Because the expression of MAO-B in glial cells is thought to be predominantly, if not exclusively, within astrocytes (Ekblom et al., 1994), we reasoned that the presence of elevations in [11C]-L-deprenyl-D2 signal within regions also demonstrating TSPO elevations would support the presence of an astrocytic contribution to the TSPO signal. Conversely, the absence of [11C]-L-deprenyl-D2 in regions showing elevated TSPO signal would suggest a predominantly microglial cellular source.
2. MATERIALS AND METHODS
2.1. Study Design
This collaborative project combines data from two independent research centers to amass a [11C]PBR28 PET imaging cohort of FM patients and matched healthy controls (HC): Massachusetts General Hospital (MGH) in Boston, MA, United States, and Karolinska Institutet (KI) in Stockholm, Sweden. Both studies were approved by ethical commitees (MGH – Partners Human Research Committee; KI – Regional Ethical Review Board in Stockholm), and all subjects provided informed consent. The initial study design and data collection at each site were completed independently, and the decision to aggregate the data into a common analysis was made after completion of data collection. Potential confounds attributable to site-specific differences were taken into account in the statistical models of the combined dataset. Group differences identified in the primary analysis of the combined dataset were also assessed within each site separately. Because arterial plasma data were collected for all subjects scanned at Karolinska, we decided to perform two levels of analyses. The first analysis took advantage of the arterial blood data collected at Karolinska, in order to obtain quantitative distribution volume (VT) metrics using kinetic modeling. The second analysis employed a blood-free ratio approach previously validated in separate cohorts (Albrecht et al., 2017), and utilized the enhanced statistical power of the combined sample (31 patients vs. 27 controls). An evaluation of the agreement between VT computed in the smaller (KI only) sample and SUVR in the larger, combined sample (KI+MGH) was perfomed by assessing the extent of spatial overlap of the group differences, as well as the regional correlation of these metrics.
2.2. Subjects
In total, 31 FM patients (29 female, 50.7±11 y/o) and 27 controls (25 female, 49.4±11 y/o) received a [11C]PBR28 PET brain scan. A total of 11 FM patients (11 female, 51.5±8.2 y/o), and 11 controls (11 female, 51.0±7.0 y/o) received a [11C]-L-deprenyl-D2 brain scan. All FM patients had received a diagnosis of FM from a physician, and met the 2011 modifications of the American College of Rheumatology classification criteria for fibromyalgia (Wolfe et al., 2011). FM patients were excluded for the presence of any pain conditions other than FM. Exclusionary criteria for all subjects at either site included: history of major psychiatric illness, neurological illness, cardiovascular disease, inability to communicate in English (MGH) or Swedish (KI), and contraindication for PET/MR scanning (e.g., pacemaker, metallic implants, pregnancy, etc). Addtionally, at MGH, the use of benzodiazepine medications was exclusionary, except for clonazepam, alprazolam, and lorazepam, that show negligible binding to TSPO in vitro, even at large clinical doses (Canat et al., 1993; Clow et al., 1985; Gehlert et al., 1985; Kalk et al., 2013; Wamsley et al., 1993). At KI, benzodiazepines were not exclusionary, however no patients reported benzodiazepine use (see Results). Please see Supplementary Table 1 for a detailed list of exclusion criteria by site.
2.2.1. Karolinska Institutet.
Eleven patients diagnosed with FM (11 female, 51.8±8,6 y/o) and 11 HC (11 female, 51.5±9.0 y/o) were matched according to age, sex and genotype for the Ala147Thr TSPO polymorphism (rs6971) which affects binding of TSPO radioligands, including [11C]PBR28, both in vitro and in vivo (Collste et al., 2016; Kreisl et al., 2013a; Owen et al., 2010; Owen et al., 2012). Sixteen subjects were Ala/Ala (i.e., high-affinity binders;;HABs: FM n=8; HC n=8) and six were Ala/Thr (i.e., mixed-affinity binders; MABs: FM n=3; HC n=3). No Thr/Thr (i.e., low-affinity binders; LABs) were included. Eleven FM patients and 11 HC received a brain [11C]PBR28 PET scan. Moreover, six of the 11 FM patients that completed the [11C]PBR28 scan and agreed to participate in an additional PET scan, as well as five additional FM patients and 11 sex and age-matched healthy controls not previously examined, received a [11C]-L-deprenyl-D2 PET scan. In addition to meeting 2011 ACR diagnostic criteria, all patients in the KI cohort also met the ACR 1990 criteria (Wolfe et al., 1990).
2.2.2. Massachusetts General Hospital.
Twenty patients diagnosed with FM (18 female, 48.0±1.2 y/o) were group matched with 16 HC subjects (14 female, 50.2±13 y/o) according to age, sex, and TSPO polymorphism. Twenty-four subjects were HABs (FM n=14; HC n=10), and 12 were MABs (FM n=6; HC n=6). No LABs were included in the study. At the MGH site, one patient also had chronic hepatitis C, one had idiopathic CD8 lymphocytopenia, and one had Meniere’s Disease and diabetes. These subjects were not outliers on any of the PET imaging measures, and results were unaffected if these subjects were excluded.
2.3. Clinical Assessment
All FM patients completed the following clinical questionnaires: 2011 American College of Rheumatology self-report survey for the assessment of FM (ACR; Wolfe et al., 2011), FM Impact Questionnaire (FIQ; Bennett et al., 2009), Beck Depression Inventory (BDI; Beck et al., 1961), and Pain Catastrophizing Scale (PCS; Sullivan et al., 1995). All items were completed on the day of the scan, with the exception of the ACR survey for all KI patients and two MGH patients, and the BDI for MGH patients, which were completed only during the screening visit. Because 17 MGH patients completed the ACR survey on both the scan day and during the screening visit, we were able to confirm that this questionnaire has good temporal stability by correlating the scores across visits. All subscales of the ACR survey showed high to moderate intercorrelations (total: r=0.785, p<0.001; symptom severity: r=0.788, p<0.001; widespread pain index: r=0.728, p=0.001; fatigue: r=0.534, p=0.027; trouble thinking: r=0.581, p=0.015; waking up tired: r=0.810, p<0.001), supporting the temporal stability of these measures, and therefore the appropriateness of evaluating these variables in relation to imaging metrics collected at different timepoints.
Prior to tracer injection on the scan day, all patients also rated their pain on a visual analog scale, anchored by 0 (“No pain at all”) and 100 (“Most intense pain tolerable”).
2.4. Positron Emission Tomography (PET) and Magnestic Resonance (MR) Imaging
2.4.1. Karolinska Institutet.
PET imaging was performed using the High-Resolution Research Tomograph (Siemens Molecular Imaging, Knoxville, TN, USA) at the PET centre at Karolinska Institutet, Stockholm, Sweden. Structural MR images were collected using a 1.5-T Siemens Avanto scanner at Medicinsk Röntgen at Odenplan prior to the first PET scan (TR=1790 or 1800ms, TE=3.53 or 2.8ms, flip angle=15° or 8°, voxel size=1mm isotropic). Prior to PET scanning, subjects received a cubital vein catheter for intravenous radioligand administration and a radial artery catheter in the contralateral arm for arterial blood sampling. To minimize motion during the PET data acquisition, each participant wore an individually-designed helmet, placed in a frame holder.
Preparation, injection and PET data acquisition for [11C]PBR28 have been described previously (Collste et al., 2016; Kanegawa et al., 2016). Average administered radioactivity of [11C]PBR28 (MBq) was – HC: 416±40 (mean±SD), FM: 385±71; average specific radioactivity (GBq/μmol) – HC: 306±195, FM: 239±74; average injected mass (μg) – HC: 0.60±0.34, FM: 0.61±0.21. Average administered radioactivity of [11C]-L-deprenyl-D2 was – HC: 364±46, FM: 366±45; average specific radioactivity – HC: 193±105, FM: 213±82 ; average injected mass – HC: 0.48±0.30, FM: 0.36±0.13. PET data were acquired for 63 minutes both for [11C]PBR28 and [11C]-L-deprenyl-D2. Manual samples were drawn at 2, 4, 6, 8, 10, 15, 20, 25, 30, 45, and 60 minutes for [11C]PBR28 and 1, 2, 4, 6, 8, 10, 16, 20, 30, 40, and 60 minutes for [11C]-L-deprenyl-D2. For one control subject in the [11C]-L-deprenyl-D2 dataset, there was a technical issue with the arterial blood collection, and thus this subject was excluded from the analysis. Arterial blood data pre-processing was performed using Kaleidagraph 4.1 software (Synergy Software) as described previously (Collste et al., 2016). Radioligand metabolism correction was performed using the parent fraction in PMOD v3.3 (pixel-wise modelling software; PMOD Technologies Ltd., Zurich, Switzerland) where individual parent fraction data was fit with a 3-exponential model.
2.4.2. Massachusetts General Hospital.
Imaging was performed at the MGH/HST Athinoula A. Martinos Center for Biomedical Imaging in Charlestown, MA. [11C]PBR28 was produced in-house using a procedure modified from the literature (Imaizumi et al., 2007). [11C]PBR28 scans were performed for 90 minutes with an integrated PET/MR scanner consisting of a dedicated brain avalanche photodiode-based PET scanner in the bore of a Siemens 3T Tim Trio MRI (Kolb et al., 2012). A multi-echo MPRAGE volume was acquired prior to tracer injection (TR/TE1/TE2/TE3/TE4=2530/1.64/3.5/5.36/7.22 ms, flip angle=7°, voxel size=1mm isotropic) for the purpose of anatomical localization, spatial normalization of the imaging data, as well as generation of attenuation correction maps (Izquierdo-Garcia et al., 2014). Average administered radioactivity of [11C]PBR28 (MBq) was – HC: 457±57 MBq, FM: 505±40; average specific activity (GBq/μmol) – HC: 77.4±30, FM: 71.8±26; average injected mass (μg) – HC: 2.32±0.8, FM: 2.78±1.1.
2.5. PET Data Analysis and Quantification
2.5.1. Kinetic analysis.
Estimation of [11C]PBR28 distribution volume (VT) was performed using Logan graphical analysis with a metabolite corrected plasma input function (Logan et al., 1990), based on five frames from 33 to 63 minutes. For each PET scan, a parametric VT image was generated using the stationary wavelet aided parametric imaging (WAPI) approach (Cselenyi et al., 2002). WAPI analysis of TSPO binding has been previously shown to be sensitive to within-subject changes in VT (Forsberg et al., 2017; Jucaite et al., 2015), and has shown both high correlation with VT estimated with the two-tissue compartment model (2TCM), and good reliability for 63 minutes of data (Collste et al., 2016).
Quantification of [11C]-L-deprenyl-D2 data was performed as described previously, utilizing the 2TCM with three rate constants (K1, k2, k3) with PMOD 3.3 (Sturm et al., 2017). The outcome measure λk3 was calculated as (K1/k2)*k3 which has been shown to reflect the regional enzyme concentration more accurately than k3 alone (Fowler et al., 1995; Logan et al., 2000). There is presently no validated method to produce λk3 parametric images with the WAPI methodology we used for the [11C]PBR28 analysis. For this reason, [11C]-L-deprenyl-D2 data was analyzed only using a region-of-interest (ROI) approach. Because an aim of this project was to assess the presence of a possible astrocytic component to TSPO signal in FM, the regions identified as statistically different across groups in the [11C]PBR28 voxel-wise SUVR analyses (see below) were selected as ROIs. Addtionally, λk3 values were computed for 23 anatomically definted ROIs from the AAL atlas, whole brain, and whole gray matter for exploratory analyses.
2.5.2. SUVR analysis.
Static [11C]PBR28 PET images were reconstructed from 33–63 minutes post-injection PET data, the latest 30-minute period available at both sites. Standardized uptake value (SUV) images were calculated by normalizing images by injected dose/body weight. SUV ratio images (SUVR) were obtained via normalization by PET signal from a pseudo-reference region (i.e., occipital cortex, identified using the occipital cortex label from the AAL atlas available in PMOD (Tzourio-Mazoyer et al., 2002)). We have previously utilized this approach for quantification of [11C]PBR28 PET data in both chronic low back pain patients and in patients with ALS (Albrecht et al., 2017), showing that group differences in SUVR (in the thalamus and motor cortex, in pain and ALS patients, respectively) can be similarly observed using VT (or VT ratio; DVR) estimated with 2TCM, and that SUVR and DVR are strongly correlated. However, since some medial portions of the occipital cortex exhibited group differences in the VT analysis (see Results), these were excluded from the occipital pseudo-reference region. A general linear model (GLM) analysis with genotype and injected dose as regressors of no interest, revealed that the mean SUV extracted from the occipital region defined above did not show any significant effects of Group (F1,52=4.38×10−6, p=0.99), Site (F1,52=0.006, p=0.72), or a Group*Site interaction (F1,52=0.08, p=0.20). These results support the appropriateness of using this region as a pseudo-reference in this particular study.
2.5.3. Image Post-processing.
For both SUVR and VT images, FSL and Freesurfer tools were used for image processing. PET images were co-registered to individual structural T1 images, normalized to MNI standard space, and spatially smoothed with an 8mm FWHM Gaussian kernel, as in Loggia et al., (2015b) and Albrecht et al., (2017).
2.6. Statistical analysis
Differences in continuous variables were assessed by performing a GLM analysis with Group and Site as fixed factors, and a Group*Site interaction term. Significant interaction terms were decomposed with post-hoc planned comparisons of least squares means. Differences in the distribution of categorical variables were assessed with Chi-Square tests. Between-site differences in FM questionnaire scores were assessed with two-sample t-tests.
As the primary analysis, voxelwise group comparisons of [11C]PBR28 VT and SUVR maps were performed with FSL’s FEAT GLM tool (www.fmrib.ox.ac.uk/fsl, version 5.0.7), using a voxelwise cluster-forming threshold of z>2.3 and a (corrected) cluster significance threshold of p<0.05 to correct for multiple comparisons. TSPO polymorphism (Ala/Ala, Ala/Thr), was included as a regressor of no interest in both analyses. Study site (MGH, KI) was included as an additional regressor of no interest in SUVR analysis, as this was perfomed on the combined sample from both institutions. Because there was a statistically significant effect of Site (F1,54=37.9, p<0.001) and a significant Group*Site interaction for injected dose (F1,54=4.23, p=0.044), this variable was also added as a covariate of no interest in all group analyses using the combined dataset (KI+MGH). There were also significant effects of Site for specific activity and injected mass, but no significant Group*Site interactions. Therefore, we performed supplementary voxelwise analyses including these variables as regressors of no interest, in order to ensure that including them in the statistical model had negligible effects on the outcomes. Because no subcortical effects were detected, and for ease of visualization, imaging results were visualized on a surface (FreeSurfer’s fsaverage).
For follow up analyses and illustration purposes, clusters of significant group differences from the SUVR or VT analysis (Figs 1–3) were parcellated into separate anatomically-constrained subregions, using the labels from the Harvard-Oxford probabilistic atlas (using an arbitrary threshold of 30). Average SUVR and/or VT values were extracted from these regions, for the purposes of visualizing data, comparing group effects within each site independently, and assessing relationships between SUVR and clinical variables and between outcome metrics (SUVR and VT). Although anterior middle cingulate (aMCC) did not exhibit significant group differences in the voxelwise VT analysis, this region was significant in the SUVR analysis (see Results); therefore, average VT was extracted from aMCC to test for potential VT differences not detected in the voxelwise analysis, particularly because this region is highly relevant to pain processing (Kragel et al., 2018; Shackman et al., 2011). Partial correlation analyses were used to assess correlations between PET signal extracted from regions exhibiting significant group differences in the voxelwise analyses and continuous clinical variables (i.e. ACR [total score, symptom severity score, widespread pain index], FIQR, BDI, PCS, and current VAS pain), correcting for TSPO polymorphism. These analyses were performed using SUVR, as these values were available for all patients at both sites. To assess the relationship between extracted PET signal and ordinal clinical variables (individual ACR symptom severity items: “fatigue”, “trouble thinking or remembering”, “waking up tired”), we performed GLM analyses with TSPO PET signal as the dependent variable, clinical score as a fixed factor (“slight or mild problem”, “moderate problem”, or “severe problem”), and TSPO polymorphism and study site as regressors of no interest. Post-hoc planned comparisons of least squares means were performed to decompose significant main effects. Additionally, partial correlation analysis was used to evaluate the association between [11C]PBR28 VT and SUVR in participants for whom both measures were available, correcting for TSPO polymorphism.
Figure 1. Voxelwise group differences in [11C]PBR28 VT.
A: Surface projection maps displaying areas with significantly elevated [11C]PBR28 VT in FM patients compared to controls (FM – n=11; HC – n=11) in voxelwise analyses (KI-only sample). B: average ± standard deviation VT extracted from several regions. The S1/M1, dLPFC and precuneus data were extracted from the clusters identified as statistically significant in the voxelwise VT analysis. For these regions, the plots are displayed for illustrative purposes only, and the level of statistical significance noted for each plot reflects that of the voxelwise analyses. For the aMCC, the data was extracted from a region, independently identified based on the results of the SUVR voxelwise analysis (see Fig. 2). The level of statistical significance noted for this region reflects the result of a region-of-interest analysis.
SPL – superior parietal lobule, S1 – primary somatosensory cortex, M1 – primary motor cortex, SMG – supramarginal gyrus, dlPFC – dorsolateral prefrontal cortex, SMA – supplementary motor area, PCC – posterior cingulate cortex, dmPFC – dorsomedial prefrontal cortex. The barplots for S1/M1, dlPFC and Precuneus are for illustrative purposes. The barplot for aMCC illustrates an ROI analysis (p=0.071)
Reed more
Reference<https://pmc.ncbi.nlm.nih.gov/articles/PMC6541932/
