Large-scale momentary brain co-activation patterns are associated with hyperalgesia and mediate focal neurochemistry and cross-network functional connectivity in fibromyalgia.

Patients with fibromyalgia who met the 2011 American College of Rheumatology criteria and pain-free healthy controls were recruited for the study. Full details of inclusion and exclusion criteria for patients with fibromyalgia have been published previously. 43 However, briefly, right-handed female patients with fibromyalgia aged 18 to 65 who reported at least 4 of 10 on a visual analogue scale based on recall over 7 days with presence of pain over 50% of days were included in the study. For HC, inclusion criteria were (1) age 18 to 65 years; (2) female; (3) right handed; (4) pain equaled 0 on a 0 to 10 visual analogue scale based on 7-day recall; (5) do not have fibromyalgia or an associated pain disorder, including migraine, temporomandibular joint disorder, chronic pelvic pain, or chronic fatigue syndrome or myalgic encephalomyelitis; (6) willingness to complete all study procedures; and (7) capable of giving written informed consent. The exclusion criteria for patients with fibromyalgia and pain-free healthy controls were (1) presence of autoimmune or inflammatory disease that causes pain; (2) history of head injury or loss of consciousness; (3) peripheral neuropathy; (4) routine daily use of narcotic analgesics, marijuana, or history of substance abuse; (5) use of stimulant medications (eg, amphetamine/dextroamphetamine, methylphenidate, dextroamphetamine, etc); (6) pregnant or nursing; (7) severe psychiatric illnesses; (8) MRI contraindications; (9) vascular surgery in the lower limbs; and (10) use of pro re nata narcotic or over-the-counter pain medication before MRI session. The University of Michigan Institutional Review board reviewed and approved study protocols, and written informed consent was obtained from participants in accordance with the Declaration of Helsinki. The current study used baseline cross-sectional data from a longitudinal neuroimaging trial of acupuncture, published previously. 31 , 43 In short, after phone screening, both participants with fibromyalgia and healthy control participants were invited to complete a baseline behavioral and MRI assessment. The behavioral session consisted of cuff pain calibration and clinical or behavioral questionnaires. The MRI session consisted of the following runs relevant to this study: resting-state functional MRI (REST), 1 H-MRS measuring levels of GABA+ and Glx (glutamate+glutamine), and sustained evoked pressure pain functional MRI (PAIN; Fig. 1 A). Experimental setup, pain sensitivity results, and brain co-activation pattern (CAP) generation procedures. (A) Experimental setup of sustained evoked pressure pain on the left leg through a rapid cuff inflator in the MRI scanner, along with sequence of relevant functional MRI (fMRI) and proton magnetic resonance spectroscopy ( 1 H-MRS) scans performed during the session. (B) Patients with fibromyalgia require a lower cuff pressure to achieve target pain relative to healthy controls, that is, patients with fibromyalgia show greater experimental pain sensitivity (hyperalgesia) compared with healthy controls. *** P < 0.001. In addition, cuff pressure is negatively correlated with clinical pain interference in fibromyalgia such that greater pain interference is associated with lower cuff pressure (ie, greater pressure pain sensitivity). (C) k -means clustering was conducted on concatenated frames of REST across all subjects to define the CAPs. The resultant CAP templates were then applied in a supervised fashion to each frame in PAIN. Occurrence rates for each CAP per subject were computed as the number of frames with the assigned CAP label divided by the number of frames for that scan run, REST or PAIN. aINS, anterior insula; Glx, glutamate+glutamine; GABA+, γ-aminobutyric acid. During MR imaging, painful pressure stimulation was applied continuously for 6 minutes to the left gastrocnemius muscle (referred to as PAIN; Fig. 1 A) using an electronically regulated inflation system with a velcro cuff (Hokanson Inc, Bellevue, WA). Pressure intensity was held steady for the entire 6 minutes at a level individually calibrated for each participant at an earlier behavioral testing session to evoke a pain rating of approximately 40 of 100 numerical units (ie, P40). After cuff deflation at the end of the PAIN run, participants retrospectively reported ratings of perceived pain intensity in the initial 2 minutes, middle 2 minutes, and final 2 minutes of the cuff being inflated (0 = “no pain,” 100 = “worst pain imaginable”). The calibration procedure at the behavioral testing session involved administering an ascending series of discrete, 10-second cuff pressures starting at 40 mm Hg with 20 mm Hg increments. Participants verbally reported the perceived pain intensity of each pressure on a numerical rating scale (0 = “no pain”, 100 = “worst pain imaginable”) after cuff deflation. The procedure was stopped when a pain rating ≥70 of 100 was obtained. A rough approximation of the stimulus–response curve was obtained by plotting the pressures and the corresponding ratings. This curve was interpolated to infer a corresponding pressure intensity to the P40 rating, which was then used at the MRI session. This method of quantitative sensory testing safely targets deep-tissue nociceptors without ischemia or tissue injury, has been used to study predictors of clinical phenotypes, 44 and has been easily adapted to the MR environment to study functional brain characteristics under sustained pain. 34 The REST and PAIN runs were conducted for 6 minutes each in an awake, eyes-open state. Whole-brain blood oxygen–level-dependent (BOLD) images at both REST and PAIN were acquired on a 15-channel head coil in a 3.0T scanner (Philips Ingenia, Best, Netherlands) using a T2*-weighted echo-planar sequence (repetition time [TR]/echo time [TE] = 2000/30 milliseconds, flip angle = 90°, matrix = 80 × 80, slice number = 38, slice thickness = 3.5 mm, size = 2.75 × 2.75 mm 2 , number of volumes = 180, and total scan time = 360 seconds). Pneumatic belt was used to collect respiratory traces and these traces were fed into an in-house version of RETROICOR 18 for physiologic correction of the BOLD signal. For functional co-registration and 1 H-MRS voxel placement (see the following section), high-resolution T1-weighted images (Axial 3D-MP-RAGE, 0.9 mm 3 isotropic, TR/TE = 8.2/3.7 milliseconds, matrix = 240 × 240, slice number = 154) were also collected. The fMRIprep 1.1.8 13 minimal preprocessing pipeline (reliant on Nipype 1.1.3 19 ) was implemented on the T1w, REST, and PAIN images. The N4BiasFieldCorrection (ANTs 2.2.0) 53 algorithm for intensity correction and the antsBrainExtraction.sh (ANTs 2.2.0) algorithm for skull-stripping were performed on the T1w image. Surface reconstruction on the T1w image was performed using recon-all (FreeSurfer 6.0.1), 9 , 36 followed by nonlinear registration and spatial normalization to the 2009c ICBM152 template through antsRegistration (ANTs 2.2.0). 2 The REST and PAIN functional images were linearly co-registered using 9 degrees-of-freedom to the T1w (bbregister, FreeSurfer). 21 Motion correction of functional images was performed using Mcflirt (FSL 5.0.9), 30 followed by nonlinear warping to the MNI152NLin2009cAsym space. Anatomical CompCor (aCompCor) was performed using principal components analysis on combined signals extracted from the cerebrospinal fluid (CSF) and white matter (masks constructed in native space) 4 to generate 6 regressors for each run. Skull-stripping on the functional images was performed through fslmaths using a dilated mask in Montreal Neurological Institute (MNI) space. The mcflirt parameters, aforementioned aCompCor regressors, censoring (based on a framewise displacement threshold of 0.4 mm), and high-pass temporal filtering (0.008 Hz) were applied to the preprocessed image (3dTproject, AFNI). 37 Only 2.14% of frames were censored and replaced with a 0 index in preparation for CAP analyses (see “Generation of co-activation patterns” section below). Finally, an iterative smoothing algorithm (3DBlurToFWHM, AFNI) with a 6 mm FWHM was applied. Any scan runs with greater than 30% discarded frames were excluded, and subjects with both REST and PAIN pairs were included for the final analysis (see section “Sample used for final analyses”). Proton magnetic resonance spectroscopy spectra were acquired from the right aINS with a 3 × 2 × 3-cm 3 voxel in 2 separate scan runs, through native T1w-guided voxel placement during the scan session. The right aINS was studied as our previous studies showed aberrations in neurotransmitter levels in this region in patients with fibromyalgia. 1 , 15 , 22 Single-voxel point-resolved spectroscopy (PRESS; TR/TE = 2000/35 milliseconds, 32 averages, VAPOR water suppression) was collected to quantify basal levels of Glx. Data obtained from PRESS were analyzed with LCModel, 48 including a CSF correction performed using the SPM12 (Wellcome Trust Centre for Neuroimaging, London, United Kingdom) software. A successful metabolite estimation was indexed by Cramer–Rao Lower Bounds <20%. A separate GABA+-edited Mescher–Garwood PRESS (MEGA-PRESS) scan ((TE = 70 milliseconds, TE1 = 13.4 milliseconds, TE2 = 56.6 milliseconds); TR = 2000 milliseconds; 320 transients; 2048 datapoints; spectral width = 2 kHz; frequency selective editing pulses (14 milliseconds) applied at 1.9 ppm (ON) and 7.46 ppm (OFF)) was performed to assess GABA+ estimates 45 and was processed in the MATLAB-based toolbox Gannet 3.1.5. 11 The processing in Gannet included spectral registration for frequency-and-phase correction, weighted averaging, zero-filling to 32,768 datapoints, and 3 Hz exponential line broadening. A 3-Gaussian function was used to model the GABA+ (at 3 ppm) and the Glx (3.75 ppm) resonances in a 2.79 to 4.2-ppm range of the difference spectra (fitted using nonlinear least squares). The “Coregister” and “New Segment” functions of SPM12 were used by Gannet to estimate gray matter, white matter, and CSF within the MRS volume. The resulting GABA+ estimates are in pseudoabsolute molal units (GABA+[i.u.] as moles of GABA+ per kilogram of solute water) and corrected for partial volume effects. 17 The acquisition and preprocessing details of PRESS and MEGA-PRESS have been published previously. 43 Short-Form Brief Pain Inventory (BPI) was the primary clinical outcome, which is broken into the severity (ie, intensity dimension) and interference (ie, broader impact on functioning) subscales. 8 Brief Pain Inventory-Pain Severity measures worst pain in 24 hours, least pain in 24 hours, pain on average, and pain right now. Brief Pain Inventory-Pain Interference measures the impact of pain in the past 24 hours on general activity, mood, walking ability, household and outside work, interpersonal relationships, sleep, and life enjoyment. Furthermore, secondary clinical measures such as American College of Rheumatology 2011 Fibromyalgia survey criteria, 56 pain catastrophizing, 49 and PROMIS 5 (depression, anxiety, physical function, fatigue, sleep) were also collected. A k -means clustering algorithm (unsupervised machine learning approach) was used to obtain CAPs. Analysis was run on MATLAB R2018b using the k-means function. The algorithm groups similar topographic patterns in single fMRI frames across all subjects, which results in clusters known as CAPs. The number of clusters is determined by the value of k . To ease computational complexity of the clustering algorithm, z -normalized voxels in each frame were resampled to 6 ×6 × 6 mm and a gray matter mask was applied to obtain 6033 voxels per frame. All resampled frames of subjects with fibromyalgia and healthy control subjects were concatenated during REST (14,580 frames after censoring) resulting in an input matrix of 14,580 frames × 6033 voxels. Once processed, the algorithm output a 14,580 frames × 1 vector containing indices or labels of each CAP. Frames that were censored during preprocessing were labelled with a 0. Frames corresponding to each CAP index were averaged and normalized to obtain z -statistic CAP maps. Co-activation patterns span multiple brain subsystems; hence, CAPs were named after maximal resemblance to specific major canonical networks, 3 , 57 verified through voxelwise spatial correlation (fslcc). Next, the centroids from each CAP obtained from the k -means algorithm were applied to each frame for every subject in PAIN using maximum spatial similarity (supervised machine learning method) and this procedure gave rise to CAP labels or indices for PAIN. Note that the REST and PAIN images were not concatenated together in the initial unsupervised k -means clustering procedure for 2 reasons: (1) we sought to define CAPs based on spontaneous brain activity representing major intrinsic functional networks that are not confounded by stimulus-related effects 28 and (2) the effects of stimulus-related modulation of CAPs during PAIN may otherwise get diluted when REST and PAIN images are concatenated. The k -means procedure was repeated in increments of 2 starting from k = 4 until k = 14. The procedure for optimal k is described below. For each scan run (REST or PAIN) for each subject, the occurrence rate of each CAP was computed by the ratio of the number of frames that were labeled as the corresponding CAP to the total number of frames for that scan. For example, an occurrence rate value of 0.2 at REST for subject 1 means that the CAP is present 20% of the time during the REST run for this subject. This metric is simple to understand and allows for between-subject and cross-state or task comparisons. The ideal number of CAPs in the brain is unknown and is an active area of research 39 —higher CAP numbers may explain greater variance but at the cost of reduced parsimony and low generalizability (and vice versa); therefore, for the relevant research question of CAPs in fibromyalgia, we used maximum discriminability of the CAPs between patients with fibromyalgia and healthy controls as the marker for a suitable number of CAPs because this would best reflect the underlying brain processes in pain. Maximum discriminability was implemented through a multivariate algorithm called random forest classification (implemented through scikit-learn), where the occurrence rates at PAIN relative to REST (ie, PAIN—REST) were inputs, along with corresponding fibromyalgia and healthy control subject labels. A leave-one-out method was used for cross validation. A spherical (6 mm) seed in the right aINS 29 (MNI coordinates: 32, 16, 6) was used to extract fMRI time series, and seed-to-voxel correlation analysis was used to evaluate whole-brain connectivity maps for the right aINS. Time series from the right aINS seed (fslmeants) were used as a GLM regressor (fsl_glm) to obtain whole-brain parameter estimates and associated variances, for each participant. These parameter estimates and variances were then passed on to group-level analysis, conducted on FMRIB's Local Analysis of Mixed Effects (FLAME 1 + 2) to improve mixed-effects variance estimation. As age influences neuroimaging outcomes, it was included as a regressor of no-interest in all analyses. Multiple comparisons family-wise error correction was conducted using Gaussian random field cluster threshold ( Z > 2.3) and significance at corrected P < 0.05. For calculation of Glx to GABA+ ratio as a proxy for basal excitatory over inhibitory neurotransmitter levels in the aINS, tissue concentration values obtained from the PRESS scan for Glx was divided by the values obtained from the MEGA-PRESS scan for GABA+. The resultant ratio was z -normalized. Higher values on this ratio indicate greater basal excitatory over inhibitory neurotransmitter levels and lower values on this ratio indicate lower basal excitatory over inhibitory neurotransmitter levels. For this study, a subset of patients with fibromyalgia with both REST and PAIN scan runs were used (detailed flow of the larger set of patients from this study has been published previously 43 ). Of the 78 subjects with fibromyalgia and 23 healthy control subjects who were recruited for MRI scanning (based on inclusion and exclusion criteria mentioned in the “Participants” section above), 63 patients with fibromyalgia and 19 healthy control subjects were used in CAP and functional connectivity analyses, meeting the following criteria: (1) both usable REST and PAIN scans per subject, as determined by quality checks; (2) had maximum framewise displacement 0.4 mm excluded with no more than 30% of total frames excluded per scan; and (4) complete and usable clinical and demographic variables, cuff pressure levels, and corresponding pain ratings. In patients with fibromyalgia, the mean FD was 0.099 ± 0.068 and the spatial standard deviation of successive difference images (DVARS) was 1.151 ± 0.065. In healthy controls, the FD was 0.081 ± 0.043 and the spatial standard deviation of successive difference images was 1.161 ± 0.071. In addition, a smaller subset of 47 patients with fibromyalgia and 11 healthy controls had both PRESS (for Glx) and MEGA-PRESS (for GABA+) collected and/or met usability criteria (see “ 1 H-MRS acquisition and preprocessing” section above for the usability criteria). Besides image-based statistics, regular statistical analyses such as between-group and paired t tests and Pearson r correlations were performed in GraphPad Prism version 9.0.0 for macOS (San Diego, CA, www.graphpad.com ). To determine whether relationships assessed with Pearson r were directionally different in patients with fibromyalgia relative to healthy controls, the single-tailed Fisher z cocor algorithm 10 was used. For mediation analyses, bias-corrected bootstrapped (10,000×) mediation was conducted using the Process Macro 25 on IBM SPSS Statistics 26 (IBM, Armonk, NY), and estimates of indirect effects were computed at the 95% confidence level. All charts were created on GraphPad Prism version 9.0.0 for macOS (San Diego, CA, www.graphpad.com ). Figures 4 D, 5 D were created with BioRender ( www.biorender.com ). Figure 1 A was created with Mind the Graph ( www.mindthegraph.com ).

Descriptive statistics for each demographic and clinical measure for the 63 subjects with fibromyalgia and 19 healthy control subjects are reported in Table 1 . Medication usage has been published previously. 43 Mean ± SD is listed for each clinical variable in the table below. P value has been obtained from between-group t tests. PROMIS scores have been reported as normative T scores ( https://www.healthmeasures.net/score-and-interpret/interpret-scores/promis ). Descriptive statistics for each demographic, clinical, and experimental pain outcome in patients with fibromyalgia and healthy controls. For the experimental pain outcomes, patients with fibromyalgia required lower cuff pressure (135.00 ± 57.65 mm Hg) to reach target experimental pain relative to healthy controls (196.70 ± 72.66 mm Hg; P < 0.001; Fig. 1 B), suggesting hyperalgesia in fibromyalgia. Within patients with fibromyalgia, the level of cuff pressure was associated with BPI severity ( r = −0.31, P = 0.012, 95% CI = [−0.518 to −0.067]), BPI interference ( r = −0.32, P = 0.011, 95% CI = [−0.526 to −0.078]; Fig. 1 B), and PROMIS fatigue ( r = −0.30, P = 0.016, 95% CI = [−0.510 to −0.056]), suggesting that hyperalgesia was associated with clinical metrics. For cuff pain ratings, a 2 (group: fibromyalgia, healthy controls) by 3 (time: initial, middle, last) mixed ANOVA was conducted, showing a significant effect of time ( F = 25.71, P < 0.001) and effects approaching significance for group ( F = 3.53, P = 0.064) and group × time ( F = 2.33, P = 0.101). Mean ± SD of cuff pain ratings during the PAIN scan run is shown in Table 1 . The procedure for k -means clustering is summarized in Figure 1 C. The outcomes of k -means clustering resulted in varied spatial patterns ranging from k = 4 to k = 14 in increments of 2 (shown in Supplementary Fig. 1, available at http://links.lww.com/PAIN/B866 ). Multivariate classification on occurrence rates showed that patients with fibromyalgia could be distinguished from healthy controls with the highest accuracy of 74.4% for k = 8 and k = 10 (accuracy levels for all other values of k is shown in Supplementary Table 1, available at http://links.lww.com/PAIN/B866 ). We prioritized parsimony, therefore k = 8 was chosen for all subsequent analyses. The 8 patterns were paired into mirror motifs (ie, pairs with overlapping spatial structure, but opposite signs/valence). 28 As shown in Figure 2 A, the CAPs consist of the following mirror motifs: (1) Two CAPs encompassing the frontoparietal and salience or ventral attention networks, henceforth abbreviated as FPN+ (SLN−) and SLN+ (DMN−), respectively. (2) Two CAPs encompassing the attentional systems (salience/ventral and dorsal attention) and the default mode network, henceforth abbreviated as SLN+/DAT+ (DMN−) and DMN+ (SLN−/DAT−), respectively. (3) Two CAPs encompassing the sensorimotor and limbic networks, henceforth abbreviated as SMN+ (Lim−) and Lim+ (SMN−), respectively. (4) Finally, 2 CAPs encompassing the visual and frontoparietal networks, henceforth abbreviated as VIS+ (FPN−) and FPN+ (VIS−), respectively. (1) Two CAPs encompassing the frontoparietal and salience or ventral attention networks, henceforth abbreviated as FPN+ (SLN−) and SLN+ (DMN−), respectively. (2) Two CAPs encompassing the attentional systems (salience/ventral and dorsal attention) and the default mode network, henceforth abbreviated as SLN+/DAT+ (DMN−) and DMN+ (SLN−/DAT−), respectively. (3) Two CAPs encompassing the sensorimotor and limbic networks, henceforth abbreviated as SMN+ (Lim−) and Lim+ (SMN−), respectively. (4) Finally, 2 CAPs encompassing the visual and frontoparietal networks, henceforth abbreviated as VIS+ (FPN−) and FPN+ (VIS−), respectively. Co-activation patterns (CAPs) and occurrence rates. (A) Spatial maps of the CAPs produced through k -means clustering at optimal k = 8, organized in pairs with opposite valence (mirror motifs). The CAPs are, namely, FPN+ (SLN−), SLN+/DAT+ (DMN−), SMN+ (Lim−), VIS+ (FPN−), and their corresponding mirror motifs. (B) Occurrence rates for each CAP at REST and PAIN, plotted separately for patients with fibromyalgia and healthy controls. Paired t tests were conducted for REST–PAIN pairs at each CAP and significance testing results are indicated with the following symbols: *** P < 0.001, ** P < 0.01, * P < 0.05, # P < 0.1. FPN, frontoparietal network; SLN, salience network (also known as the ventral attention network); DAT, dorsal attention network; DMN, default mode network; SMN, sensorimotor network; Lim, limbic network; VIS, visual network. Using the labels of CAPs from clustering, occurrence rates (ie, percent of frames labelled with a given CAP relative to all frames for a scan run per subject) were calculated. To explore whether occurrence rates of CAPs were modulated because of state changes (ie, from REST to PAIN), paired t tests were conducted, separately for patients with fibromyalgia and for healthy controls. For fibromyalgia, the SLN+/DAT+ (DMN−) and DMN+ (SLN−/DAT−) CAPs occurred less frequently during PAIN relative to REST, whereas the SMN+ (Lim−), Lim+ (SMN−), VIS+ (FPN−), and FPN+ (VIS−) CAPs occurred more frequently during PAIN relative to REST. For healthy controls, the FPN+ (SLN−) CAP occurred less frequently during PAIN relative to REST, whereas the VIS+ (FPN−) and FPN+ (VIS−) occurred more frequently during pain (means, distribution of individual datapoints for each subject, and P values for each comparison have been highlighted in Fig. 2 B and Supplementary Fig. 2, available at http://links.lww.com/PAIN/B866 ). To tackle the multiple comparison problem across many paired t tests, we also ran multivariate classification (see previous section, “Clustering reveals 8 optimal CAPs”). Because CAPs have been hypothesized to be the precursor (ie, building blocks) of functional connectivity and previous studies have found heightened connectivity between the insula and default mode network in fibromyalgia, 35 , 40 , 46 we explored how occurrence of CAPs that encompass the aINS and DMN ultimately lead to heightened aINS–DMN functional connectivity. As shown in Figure 3 A, the right aINS was weighted heavily in the SLN+/DAT+ (DMN−) CAP (aINS happens to be a key hub region 32 of the SLN 54 ); hence, individual time series were extracted from the right aINS seed and correlated voxelwise to compute whole-brain functional connectivity. Whole-brain regressions of SLN+/DAT+ (DMN−) occurrence rates with right aINS connectivity in fibromyalgia revealed clusters in the posterior cingulate cortex (PCC) or precuneus (cluster peak z -stat = 4.04; peak coordinate: x = −2, y = −44, z = 30) and the medial prefrontal cortex (mPFC; cluster peak z -stat = 4.22; peak coordinate: x = 4, y = 56, z = 14; Fig. 3 B). The associations were such that as the SLN+/DAT+ (DMN−) occurred less frequently during PAIN relative to REST, the functional connectivity of the right aINS and DMN subregions increased during PAIN relative to REST (R aINS–PCC/precuneus connectivity, r = −0.51, 95% CI = [−0.67 to −0.30]; R aINS–mPFC connectivity, r = −0.50, 95% CI = [−0.67 to −0.29]; scatter plots are shown in Fig. 3 C). In healthy controls, the CAP–functional connectivity associations were not significant for the PCC or precuneus ( r = −0.07, P = 0.772, 95% CI = [−0.51 to 0.40]) or for the mPFC ( r = −0.26, P = 0.284, 95% CI = [−0.64 to 0.22]). The correlation values for CAP–functional connectivity relationships were significantly different between fibromyalgia and healthy controls for the PCC or precuneus (Fisher z = −1.75, P = 0.04) but not for the mPFC (Fisher z = −1.01, P = 0.16). Lower SLN+/DAT+ (DMN−) occurrence rates are related to increased insular to DMN subregion functional connectivity during sustained evoked pain in fibromyalgia. (A) Right anterior insula (aINS) seed overlayed on the SLN+/DAT+ (DMN−) co-activation pattern (CAP). (B) Voxelwise maps of whole-brain regression of SLN+/DAT+ (DMN−) occurrence rates (PAIN–REST) with right aINS connectivity (PAIN–REST), showing clusters encompassing the medial prefrontal cortex (mPFC) and the posterior cingulate cortex (PCC) or precuneus in fibromyalgia (display threshold z > 2.3). (C) SLN+/DAT+ (DMN−) occurrence rates (PAIN—REST) are negatively correlated with right aINS to mPFC and right aINS to PCC connectivity (PAIN–REST) in fibromyalgia, such that as occurrence rates are lower, the connectivity between the right aINS and DMN subregions are higher during PAIN. SLN, salience network (also known as the ventral attention network); DAT, dorsal attention network; DMN, default mode network. To explore how neurotransmitters in hub regions such as the aINS give rise to brain-wide CAPs such as the SLN+/DAT+ (DMN−) CAP, we computed the ratio of Glx to GABA+ in the right aINS. Figure 4 A shows mean spectra, model fit, and corresponding peaks. In fibromyalgia, the ratio of Glx to GABA+ in the right aINS was associated with the change in SLN+/DAT+ (DMN−) occurrence rate (PAIN relative to REST; r = −0.35, P = 0.01, 95% CI = [−0.58 to −0.07]). The magnitude and direction of this correlation was similar with those in healthy controls, but the correlation was not statistically significant ( r = −0.32, P = 0.35, 95% CI = [−0.77 to 0.35]; note that healthy controls had a low sample size of Glx/GABA+ values, N = 11). In addition, the correlation values were not statistically different between patients with fibromyalgia and healthy controls (Fisher z = −0.09, P = 0.47). Pooled analysis across both patients with fibromyalgia and healthy controls also showed a statistically significant correlation ( r = −0.34, P = 0.009, 95% CI = [−0.55 to −0.09]). Together, the association was such that the greater the basal excitatory over inhibitory neurotransmitter levels in the right aINS (ie, higher Glx/GABA+), the less frequent was the occurrence of SLN+/DAT+ (DMN−) during PAIN relative to REST (refer to Fig. 4 B for patients with fibromyalgia only and Supplementary Fig. 3 for pooled correlation across both patients with fibromyalgia and healthy controls, available at http://links.lww.com/PAIN/B866 ). Greater basal Glx to GABA+ ratio in the aINS is associated with decreased SLN+/DAT+ (DMN−) occurrence rates in fibromyalgia. (A) Proton magnetic resonance spectroscopy ( 1 H-MRS) conducted in the right aINS. Mean spectra and model obtained from the 1 H-MRS scan sequences PRESS and MEGA-PRESS, showing the respective glutamate+glutamine (Glx) and γ-aminobutyric acid (GABA+) peaks. (B) Ratio of Glx to GABA+ in the right aINS was negatively associated with the SLN+/DAT+ (DMN−) occurrence rates in fibromyalgia, such that greater excitatory over inhibitory neurotransmitter levels was associated with lower occurrence of SLN+/DAT+ (DMN−) during PAIN relative to REST. (C) Mediation analysis showed that in fibromyalgia, the Glx to GABA+ ratio in the right aINS and connectivity between the right aINS and PCC were mediated by the occurrence rate of SLN+/DAT+ (DMN−). (D) Mechanistic model proposing that the basal excitatory over inhibitory neurotransmitter levels in the right aINS orchestrates momentary occurrence of SLN+/DAT+ (DMN−) during PAIN. Lesser momentary occurrence of the SLN+/DAT+ (DMN−) during PAIN indicates greater enmeshment of the right aINS (a key salience network node) with the default mode network and vice versa. Cho, choline; Cr, Creatine; NAA, N-acetylaspartate; ppm, parts per million; BootSE, bootstrap standard error; BootCI, bootstrap confidence interval; SLN, salience network (also known as the ventral attention network); DAT, dorsal attention network; DMN, default mode network; PRESS, point-resolved spectroscopy; PCC, posterior cingulate cortex. It is known that neurotransmitters in circuits drive hemodynamic activity 41 and possibly cross-network functional connectivity. We sought to understand how CAPs may be a mechanistic intermediary between Glx to GABA+ ratio in the aINS and aINS–DMN connectivity in fibromyalgia through simple bootstrapped mediation analysis. Results showed that in fibromyalgia, a greater Glx to GABA+ ratio in the aINS was associated with greater functional connectivity of the right aINS to the PCC during PAIN (relative to REST) indirectly through less frequent occurrence of the SLN+/DAT+ (DMN−) CAP during PAIN (relative to REST; β = 0.464, BootSE = 0.275, BootLLCI = 0.032, BootULCI = 1.093, Figs. 4 C,D). The direct effect of Glx to GABA+ ratio in the right aINS on a PAIN-induced increase in right aINS–PCC connectivity was not significant (β = 0.104, SE = 0.418, LLCI = −0.738, ULCI = 0.947), suggesting that the effect is transmitted through modulation of the SLN+/DAT+ (DMN−) CAP in fibromyalgia. Because of insufficient sample size of Glx to GABA+ ratio in healthy controls (N = 11), we were unable to perform this mediation analysis in healthy controls alone. However, we performed an additional pooled analysis of patients with fibromyalgia and healthy controls under 2 considerations: (1) patients with fibromyalgia and pain-free healthy controls may be along a spectrum of possible brain configurations and (2) there might be a general principle of basal neurotransmitter levels and functional connectivity during evoked pain. The pooled mediation analysis showed a similar association of greater Glx to GABA+ ratio in the right aINS with greater right aINS–PCC connectivity (PAIN relative to REST) indirectly through less frequent occurrence of the SLN+/DAT+ (DMN−) CAP (PAIN relative to REST; β = 0.451, BootSE = 0.242, BootLLCI = 0.062, BootULCI = 0.999). The direct effect was not significant (β = 0.045, SE = 0.367, LLCI = −0.690, ULCI = 0.781) in the pooled analysis either (see Supplementary Fig. 4, available at http://links.lww.com/PAIN/B866 ). We next assessed whether CAPs were associated with cuff pressure levels (ie, degree of hyperalgesia) and clinical pain interference in fibromyalgia. The occurrence of SLN+/DAT+ (DMN−) CAP was negatively associated with clinical pain interference ( r = −0.29, P = 0.02, 95% CI = [0.046-0.5]), such that less frequent occurrence of SLN+/DAT+ (DMN−) during PAIN relative to REST was associated with greater clinical pain interference in fibromyalgia (Fig. 5 A). Clinical variables in fibromyalgia are associated with SMN+ (Lim−) and SLN+/DAT+ (DMN−) co-activation pattern (CAP) occurrence rates. (A) Occurrence rate of SLN+/DAT+ (DMN−) is negatively correlated with pain interference, such that lower occurrence of SLN+/DAT+ (DMN−) during PAIN is associated with higher clinical pain interference in fibromyalgia. (B) Occurrence rate of SMN+ (Lim−) is positively correlated with pain interference and negatively correlated with cuff pressure, such that greater occurrence of SMN+ (Lim−) during PAIN is associated with higher clinical pain interference and greater experimental pain sensitivity in fibromyalgia, respectively. (C) Relationship between cuff pressure and clinical pain interference in fibromyalgia is mediated by the occurrence of SMN+ (Lim−) during PAIN. (D) Proposed mechanistic model of how momentary CAPs are involved in the relationship between experimental pain sensitivity and clinical pain interference in fibromyalgia. Greater experimental pain sensitivity brings about higher momentary occurrence of SMN+ (Lim−), which explains higher clinical pain interference in fibromyalgia, and vice versa. BootSE, bootstrap standard error; BootCI, bootstrap confidence interval; SMN, sensorimotor network; SLN, salience network (also known as the ventral attention network); DAT, dorsal attention network; DMN, default mode network; Lim, limbic network. In addition, the occurrence of SMN+ (Lim−) CAP was positively associated with clinical pain interference ( r = 0.36, P = 0.004, 95% CI = [0.12-0.56]) and negatively associated with cuff pressure ( r = −0.36, P = 0.004, 95% CI = [−0.56 to −0.12]) in fibromyalgia, such that more frequent occurrence of SMN+ (Lim−) during PAIN relative to REST was associated with greater clinical pain interference and less cuff pressure (ie, more hyperalgesia) in fibromyalgia (Fig. 5 B). The SMN+ (Lim−) CAP–cuff pressure association was not significant in healthy controls ( r = −0.09, P = 0.72, 95% CI = [−0.52 to 0.38]) and the correlation values for patients with fibromyalgia and healthy controls were not significantly different from each other (Fisher z = −1.02, P = 0.154). We next assessed whether PAIN-induced increases in SMN+ (Lim−) CAP was concurrent with PAIN-induced decreases in SLN+/DAT+ (DMN−) and found a significant relationship in fibromyalgia ( r = −0.31, P = 0.014, 95% CI = [−0.52 to −0.067]; Supplementary Fig. 5, available at http://links.lww.com/PAIN/B866 ) and in healthy controls ( r = −0.55, P = 0.015, 95% CI = [−0.80 to −0.13]). The correlation values were not significantly different between patients with fibromyalgia and healthy controls (Fisher z = 1.06, P = 0.145). Next, the difference between PAIN-induced SMN+ (Lim−) and SLN+/DAT+ (DMN−) occurrence rates was used as a measure of temporal distance between 2 CAPs. In fibromyalgia, the difference between PAIN-induced SMN+ (Lim−) and SLN+/DAT+ (DMN−) occurrence rates was positively associated with clinical pain interference ( r = 0.39, P = 0.002, 95% CI = [0.16, 0.58]) and negatively associated with cuff pressure ( r = −0.26, P = 0.036, 95% CI = [−0.48 to −0.013]). The relationship was such that as the brain spent more frequent time in SMN+ (Lim−) and less frequent time in SLN+/DAT+ (DMN−) during PAIN relative to REST, the greater clinical pain interference and the lesser the cuff pressure required to achieve target experimental pain (ie, more hyperalgesia; Supplementary Fig. 6, available at http://links.lww.com/PAIN/B866 ). The difference between PAIN-induced SMN+ (Lim−) and SLN+/DAT+ (DMN−) occurrence rates was not related to cuff pressure in healthy controls ( r = −0.076, P = 0.764, 95% CI = [−0.51 to 0.39]) and the correlation values were not different between patients with fibromyalgia and healthy controls (Fisher z = −0.68, P = 0.250). The brain is a mediator of psychophysical outcomes and clinical processes—here we sought to understand the relationship between hyperalgesia (assessed through cuff pressure) and clinical pain interference through bootstrapped simple mediation analysis. Results showed that lower cuff pressure was associated with greater clinical pain interference in fibromyalgia indirectly through more frequent occurrence of the SMN+ (Lim−) CAP during PAIN (relative to REST; β = −0.0039, BootSE = 0.0021, BootLLCI = −0.0090, BootULCI = −0.0007, Fig. 5 C). The effect of hyperalgesia on clinical pain can be explained through the occurrence rate of SMN+ (Lim−) during sustained pain (Fig. 5 D). A separate mediation showed that lower cuff pressure was associated with greater clinical pain interference in fibromyalgia indirectly through more frequent occurrence of the SMN+ (Lim−) CAP relative to less frequent occurrence of SLN+/DAT+ (DMN−) during PAIN (relative to REST; β = −0.0030, BootSE = 0.002, BootLLCI = −0.0070, BootULCI = −0.0010, Supplementary Fig. 7, available at http://links.lww.com/PAIN/B866 ). Thus, the effect of hyperalgesia on clinical pain can additionally be explained through the relative amount of time spent by the brain in SMN+ (Lim−) and SLN+/DAT+ (DMN−) CAPs during sustained pain (Supplementary Fig. 8, available at http://links.lww.com/PAIN/B866 ).

In this cross-sectional study, we evaluated potential brain mechanisms underlying experimental pain hypersensitivity in fibromyalgia. In particular, we used a clustering technique to understand the topographic nature and pain-related modulation of single instantaneous timepoints of the BOLD signal (ie, CAPs). To briefly summarize our results, we found evidence for hyperalgesia in the cuff pain paradigm, which was related to clinical metrics. During the deep-tissue sustained cuff pain stimulation, several CAPs in the brain were modulated which distinguished patients with fibromyalgia and pain-free healthy controls in multivariate classification with 74.4% accuracy. Notably, 1 CAP encompassing the salience or ventral and dorsal attentional systems (SLN+/DAT+ [DMN−]) occurred less and a separate CAP encompassing the sensorimotor systems (SMN+ [Lim−]) occurred more frequently during sustained pain stimulation relative to the resting state. We also found that the SLN+/DAT+ (DMN−) CAP was associated with the Glx to GABA+ ratio and functional connectivity of a key node of the SLN, the aINS. Specifically in individuals with fibromyalgia, the effect of higher excitatory over inhibitory neurotransmitter levels in the aINS on heightened functional connectivity between aINS and DMN subregions was transmitted through decreased occurrence of the SLN+/DAT+ (DMN−) CAP during sustained evoked pain. Moreover, we found that higher occurrence of the SMN+ (Lim−) CAP during sustained evoked pain was associated with greater hyperalgesia (ie, lower cuff pressure) and clinical pain interference in fibromyalgia. This relationship was also present when higher SMN+ (Lim−) CAP occurrence was accounted for by the lower SLN+/DAT+ (DMN−) CAP occurrence during sustained evoked pain. Finally, we found that higher occurrence of the SMN+ (Lim−) CAP (and higher occurrence of the SMN+ [Lim−] CAP relative to lower occurrence of the SLN+/DAT+ [DMN−] CAP) mediated the relationship between greater hyperalgesia and greater clinical pain interference. Until recently, the most common method of studying brain networks has been through time series correlation (ie, functional connectivity). Through CAP analysis, patterns of activity at instantaneous timepoints can be examined, affording more nuanced spatiotemporal inferences. It is believed that CAPs are shaped by intermittent and transient neuronal events 38 , 39 and a task-related modulation of time spent in a CAP reflects changes in underlying task-related neuronal activity. Our sustained evoked pain paradigm increased time spent in a CAP encompassing the sensorimotor cortex (both primary somatosensory and primary motor; S1 and M1), likely due to spinothalamic afference from the leg causing tonic firing in S1 to encode the pain stimulus. This finding corroborates previous research of S1 leg connectivity being increased with a sustained pain stimulus. 34 The sustained evoked pain paradigm also modulated the SLN+/DAT+ (DMN−) CAP encompassing the attentional systems (salience/ventral and dorsal), which consists of the aINS and the cingulate. The degree of modulation of this CAP was contingent on the basal neurotransmitter levels in the aINS, which indicates that neurotransmitters in hub regions may determine state-dependent expression of the corresponding CAP. The aINS is a hub region that drives overall SLN dynamics 54 and interfaces with multiple subsystems 47 —depending on the state or task, membership of the aINS gets emmeshed or distributed across multiple networks. 46 During sustained pain, we found decreased time spent by the SLN+/DAT+ (DMN−) CAP relative to the resting state, which is likely due to enmeshment of the attentional systems with the DMN, resulting in SLN+/DAT+ (DMN−) CAP remaining in its home network less frequently. In healthy individuals, the attentional systems (salience/ventral and dorsal) and DMN represent anticorrelated networks. 16 Accumulated evidence from more than a decade of research has shown enmeshment of the SLN and DMN (ie, reduced anticorrelation or even positive correlation) in fibromyalgia and other chronic pain conditions. 26 , 35 , 40 , 46 Pathological communication between networks or the inability to disengage the brain's self-referential systems and salience detection system (which integrates external and internal signals 54 ) may represent the propensity of the central nervous system to amplify innocuous or aversive signals in conditions such as fibromyalgia. 7 Despite the wealth of evidence of heightened SLN–DMN connectivity, the elements that constitute the building blocks of this connectivity remain largely understudied because of a lack of multimodal tools and analytic techniques. Our previous multimodal study in patients with endometriosis showed that increased excitatory neurotransmitter levels in the aINS were associated with increased aINS–mPFC connectivity, 1 positing that excitatory signaling in hub regions such as the aINS may orchestrate cross-network connectivity patterns. 33 The current study adds to our understanding of multiscale brain organization—because CAPs may be a precursor to functional connectivity, 39 the CAP that encompasses the aINS (and broader salience/ventral attention network) may be a mechanistic intermediary to heightened cross-network aINS–DMN connectivity produced by local excitatory and inhibitory neurotransmitter levels. Fibromyalgia may be conceptualized as a disorder of consciousness, where a floodgate of sensory information amplified in the central nervous system results in pain (the other end of the spectrum being diminished information processing in anesthetized states 42 ). This information floodgate may explain symptoms such as sensitivity in nonsomatosensory domains such as auditory and visual hypersensitivity 24 in patients with fibromyalgia. Recent research (Huang et al.) has found evidence that the aINS plays a key role in the gating of conscious access. 27 Importantly, the study by Huang et al. identified the aINS as a key component in facilitating alternating activity in the DMN and the DAN. Activation of the aINS initiated this network shift by activating the DAN and suppressing the DMN. It remains unknown how gates of conscious access and associated network states lead to multisensory augmentation in fibromyalgia. The current study suggests that heightened excitatory over inhibitory neurotransmitter levels in the aINS may orchestrate more promiscuous task-positive attentional systems (both dorsal and salience/ventral) with lower suppression of the DMN, resulting in a weaker gating mechanism that allows a flood of conscious information. Validation of our findings is needed to determine whether this mechanism is involved in fibromyalgia symptomatology. To the best of our knowledge, this is the first study using CAP analytic methods in fibromyalgia. Alternative methods that investigate instantaneous or shorter timescales also exist. One notable method is dynamic functional connectivity, which aims to understand how pairwise relationships between regions covary with time. Co-activation pattern analysis differs vastly from dynamic functional connectivity 39 ; however, the approaches are complementary and inferences from each method may corroborate each other. A recent study (Cheng et al. 6 ) used the same sustained evoked pain paradigm in a different cohort of patients with fibromyalgia and found that patients with increased temporal summation of pain showed sustained dynamic connectivity between nodes of the sensorimotor and salience or ventral attention networks. Our finding that patients with greater experimental pain sensitivity spent more time in the SMN+ (Lim−) CAP during the pain task is consistent with their results. Interestingly, those patients who did not show temporal summation of pain had “gaps” in the dynamic connectivity time series, that is, periods with lack of sensorimotor–salience or ventral attention communication. Although CAPs were not evaluated in the study by Cheng et al., it can be posited that the “gaps” in dynamic functional connectivity might be marked by less modulated SLN+/DAT+ (DMN−) and SMN+ (Lim−) CAPs. Taken together, the study by Cheng et al. and the current study each provide a unique perspective on brain dynamics occurring at shortened timescales in response to evoked pain in fibromyalgia. Our study is not without limitations. We did not have sufficient sample size of both Glx and GABA+ in healthy controls (N = 11) to make reasonable inferences about group differences. For this reason, future validation with larger studies is needed to discover if an underlying neurochemical imbalance is driving network enmeshment uniquely in FM. In addition, the Glx to GABA+ ratio does not allow us to make inferences about postsynaptic receptor interactions in the aINS (ie, whether the net effect on the aINS is excitation or inhibition). In addition, we remain cautious using the Glx to GABA+ ratio to make claims about excitatory or inhibitory balance as the term has been derived from electrophysiological evidence and may not always corroborate with findings from proton spectroscopy. Furthermore, unlike previous studies, 28 we did not find global CAPs that are typically linked to arousal states. Our analysis pipeline did not use global signal regression, so we attribute lack of global CAPs to careful physiological correction (using both peripheral physiologic measurements and aCompCor 13 ). The CAPs did not correlate with pain intensity; however, they did relate to the pain interference subscale. It might be that the time a specific brain network is maintained (ie, a CAP occurrence rate) may relate more to the degree that an individual with fibromyalgia has difficulty functioning (ie, pain interference) and less so with the intensity of the pain. Finally, the multivariate classification between patients and controls can be influenced by differences in reported experimental pain intensity such as during the last 2 minutes of pressure pain. In summary, spatiotemporal patterns at instantaneous timepoints act as the bridge between local neurotransmitter levels and functional connectivity during sustained pain. The amount of time spent by certain spatiotemporal patterns in the brain may be responsible for heightened pain sensitivity in fibromyalgia. CAPs may be used in conjunction with other biological measurements for diagnostic, predictive, and therapeutic developments in fibromyalgia. 52

The brain is a complex system with distinct macroscale topographic organization. 12 Functional neuroimaging over the past few decades has provided insight into this topographic organization through the discovery of canonical spatial networks. 57 Although most inferences in functional neuroimaging have been made in the spatial domain, fewer inferences have been made along the temporal axis. Because the brain is a dynamic entity with the spatial properties evolving over time, understanding temporal properties affords critical insight into various behavioral states, including pain. The underlying architecture of this macroscale organization along space and time is diverse, composed of many cell types, neurotransmitters, and circuits. 12 This multiscale organization is hierarchical and is characterized by several highly connected regions known as hubs, which orchestrate global communication and subserve coupling and switching between networks. 47 However, few studies have studied how local neurotransmitter levels in hubs influence global networks and consequently pain. In particular, glutamatergic (excitatory) and γ-aminobutyric acid (GABA)-ergic (inhibitory) neurotransmission has been shown to be responsible for neural coding and information propagation, 58 but its influence on global functional networks remains unknown. Pain serves as an evolutionarily beneficial threat detection mechanism, but it can become maladaptive in states such as fibromyalgia (also described as centralized 7 or nociplastic pain 14 ). A wealth of functional neuroimaging research has shown that fibromyalgia is associated with altered central nervous system function, 23 which emerges from an imbalance of pronociceptive and antinociceptive activity and aberrant communication between the salience or ventral attention, default mode, and sensorimotor networks. 7 , 15 , 20 , 22 , 34 , 46 , 50 Furthermore, proton magnetic resonance spectroscopy ( 1 H-MRS) studies have shown heightened levels of local excitatory neurotransmitters 22 and diminished levels of local inhibitory neurotransmitters 15 in fibromyalgia, which has been causally corroborated in reverse translational animal studies. 55 In this study, we used multimodal data to further our understanding of the brain's role in fibromyalgia, along both spatial and temporal axes. We acquired 1 H-MRS in the anterior insula (aINS), a key hub of the salience network, to assess glutamate (along with glutamine) and GABA concentrations. In addition, we used co-activation pattern (CAP) analysis during an evoked pain task to understand macroscale organization. Co-activation patterns are spatial patterns of brain activity found in single instantaneous timepoints of the functional MRI time series. 38 , 39 Co-activation patterns repeat over time and they are believed to constitute the building blocks of overall functional connectivity between brain regions. 38 , 39 The occurrence rate of a CAP is the percentage of time that the brain displays that pattern of activity. Co-activation patterns may be produced by avalanches of neuronal activity initiated by specific regions (ie, brief bursts of local activity from hub regions that have a network-wide cascading effect). 51 Thus, CAPs may be the mechanistic bridge between hub neurochemistry and large-scale network communication. Here, we examined CAPs (and their occurrence rates) during a sustained evoked pain task, and we hypothesized that CAPs are a mechanistic bridge between hub neurochemistry and cross-network communication in fibromyalgia. Furthermore, we hypothesized that CAPs mediate the relationship between pain sensitivity and clinical measures in fibromyalgia.

Supplemental digital content associated with this article can be found online at http://links.lww.com/PAIN/B866 .

The authors have no conflict of interest to declare.