A picture is worth a thousand words: linking fibromyalgia pain widespreadness from digital pain drawings with pain catastrophizing and brain cross-network connectivity

Introduction

Pain catastrophizing is a pain-targeted psychosocial construct comprised of helplessness, pessimism, and magnification of pain-related symptoms and complaints [50]. Catastrophizing is prominent in fibromyalgia [20], and has been proposed to contribute to widespread pain at multiple body sites [15,19,43,53]. In fact, fibromyalgia patients frequently report pain in numerous body regions, and even whole-body pain [57]. Widespread multisite pain has been associated with a wide array of adverse demographic, lifestyle, and general health factors [29], chronic work disability [44,51], and even increased risk of cancer and cardiovascular-related mortality [39]. The specific bodily patterns and sensory qualities of clinical pain is often highly heterogeneous and may reflect important information about individual symptomatology, which is lost in one-dimensional pain ratings [48]. However, fine-grained assessment of pain widespreadness using modern digital pain drawing techniques is still lacking, and may be an important tool for understanding how maladaptive brain plasticity contributes to both catastrophizing and pain widespreadness in chronic pain patients.

Several studies have linked catastrophizing to chronic pain widespreadness. In patients with knee osteoarthritis, high pain catastrophizing is associated with an increased number of non-knee painful sites [19]. Similarly, in chronic Low Back Pain (cLBP), higher levels of catastrophizing is associated with more widespread pain and higher disability [45]. Another study found that widespread pain (pain in >7 body sites) after a motor vehicle collision was predicted by pre-collision depressive symptoms and catastrophizing, but not factors such as road speed limit, extent of vehicle damage, or airbag deployment [12]. Thus, psychological factors often contribute to pain widespreadness (though this may not extend to all pain disorders [22]), which may contribute to the development of chronic pain through a negatively reinforcing cycle of increasing pain-related worrying and spread of pain to previously non-painful body locations [53].

Brain imaging studies using functional MRI (fMRI) have identified several abnormalities in resting functional network connectivity for chronic pain, including fibromyalgia, likely facilitated by pain catastrophizing. In particular, recent studies indicate that cross-network resting state connectivity between Salience (SLN) and Default Mode (DMN) Networks encodes clinical pain severity [23,37,40,41]. For example, in cLBP patients, DMN/insula connectivity is associated with clinical pain intensity, especially in patients with high catastrophizing [31]. DMN/insula connectivity has also been linked with pain widespreadness, assessed by the number of painful body regions, in chronic pain patients undergoing an auditory test of cognitive function [4], suggesting that both pain catastrophizing and widespreadness may be mediated by such cross-network connectivity. Interestingly, we recently found that DMN nodes, in particular the PCC, were implicated in encoding externally cued pain catastrophizing thoughts in fibromyalgia [35]. While these studies support the notion that pain catastrophizing underpins the association between chronic pain and connectivity between SLN with DMN nodes (e.g. insula and PCC, respectively), the role of widespread bodily distribution of pain in this association is unknown.

Whole-body pain drawings have a relatively long history as pen-and-paper assessments, with the advantage of providing a richer account of patients’ pain experience compared to basic numeric/visual pain ratings, often aiding clinicians’ qualitative understanding of patients’ pain. Recently, dedicated apps for touch-screen tablets have been developed for digital pain drawing acquisition, enabling quantitative analyses of fine-grained spatial patterns across patients, beyond qualitative inquiry of individual patient drawings and potentially oversimplified metrics such as the number of a coarse set of body regions reported as painful [48].

Here, we applied digital pain drawings and fMRI in fibromyalgia patients to investigate the associations between catastrophizing, resting brain network connectivity, and pain widespreadness. We hypothesized that increased SLN connectivity to DMN nodes (e.g. PCC) would be associated with increased pain catastrophizing and widespreadness across specific body sites in fibromyalgia.

Methods

Participants

We enrolled 113 female patients (mean age±SD: 41.70±12.39 years old, Race: 91 Caucasian, 9 African-American, 3 Asian, 10 other/multiracial), meeting the American College of Rheumatology (ACR) diagnostic criteria for fibromyalgia [57], through online Clinical Trials listings (clinicaltrials.partners.org), a Partners Healthcare medical records database, and physician referral. Of these, 9 patients were not able to complete study procedures (e.g. due to MRI contraindications, scanner discomfort, and scheduling issues), 20 patients had no usable pain drawing data within allowable time window (i.e. within 14 days of the MRI visit), and 2 patients did not complete the necessary questionnaires. Thus, a sample of 82 fibromyalgia participants were included for initial data analysis. Of these, we excluded 3 participants due to excessive head motion during fMRI, based on the following exclusion criteria: 1) >2° head rotation in any direction, and 2) >2 mm frame-by-frame displacement [31], using motion parameters calculated by MCFLIRT (see “fMRI preprocessing” below). Consequently, the final analyzed data sample consisted of 79 participants (Table 1). All patients completed the Brief Pain Inventory [16] prior to testing (mean score±SD: Pain Severity: 5.12±1.87; Pain Interference: 5.58±2.47). The study was carried out at the Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital (MGH), in Boston, MA. All study protocols were approved by MGH and Partners Human Research Committee and all participants provided written informed consent.

Table 1:

Patient demographics (final sample)

	Mean±SD (Range)
Age	41.5±12.6 (18–65)
Time since diagnosis (Months)	78.5±74.1 (2–309)
Pain severity (BPI, 0–10)	5.1±1.9 (1.3–10)
Pain interference (BPI, 0–10)	5.6±2.5 (1.0–10)
Race (Count)
Caucasian	69
African-American	3
Asian-American	2
Other/multiracial	5

BPI=Brief Pain Inventory

Inclusion criteria were as follows: age 18–65 and female (since fibromyalgia is highly female-predominant, and since it is unknown whether there might be sex differences in pain biomarkers for fibromyalgia, we aimed to reduce the potential variability added by not including a likely small group of male participants in this study); meeting the ACR diagnostic criteria for fibromyalgia [57] for at least 1 year; receive stable doses of medication prior to entering the study; average fibromyalgia pain intensity of at least 4 (on a scale from 0 to 10), and reported fibromyalgia pain for at least 50% of days during the preceding month; ability to provide written informed consent; and fluency in English. Exclusion criteria were as follows: comorbid acute or chronic pain conditions rated as more painful than fibromyalgia pain; use of stimulant medications for fatigue associated with sleep apnea or shift work; psychiatric disorders with a history of psychosis; psychiatric hospitalization within 6 months prior to enrollment; current or recent use of recreational drugs; active suicidal ideation; current participation in other therapeutic trials; lower limb vascular surgery or current lower limb vascular dysfunction; pregnant or nursing; history of significant head injury (e.g. with substantial loss of consciousness); history of anxiety disorders interfering with MRI procedures (e.g. panic); and contraindications to MRI.

Pain catastrophizing

Participants completed the validated pain catastrophizing scale (PCS) questionnaire [49], which assesses the tendency to apply an ‘exaggerated negative “mental set” brought to bear during actual or anticipated pain experience’[50]. Items included: ‘I worry all the time about whether the pain will end’; ‘I feel I can’t go on’; ‘It’s terrible and I think it’s never going to get any better’; ‘It’s awful and I feel that it overwhelms me’; ‘I feel I can’t stand it anymore’; ‘I become afraid that the pain will get worse’; ‘I keep thinking of other painful events’; ‘I anxiously want the pain to go away’; ‘I can’t seem to keep it out of my mind’; ‘I keep thinking about how much it hurts’; ‘I keep thinking about how badly I want the pain to stop’; ‘There’s nothing I can do to reduce the intensity of the pain’; ‘I wonder whether something serious may happen’. While the PCS comprises 3 subscales – rumination, helplessness, and magnification – we calculated the sum of all items as a total score, since we did not have separate hypotheses regarding specific subscales. Importantly, while the pain catastrophizing scale is formalized as a trait-based scale, as opposed to measures of situational pain catastrophizing (see e.g. [25]), dispositional catastrophizing can also depend on specific context and be malleable over time (see [18] for a recent critical discussion of the concept of “pain catastrophizing”).

Pain drawings (acquisition)

During a behavioral study visit prior to MRI, patients reported current clinical pain using the recently developed software ‘Symptom Mapper’^a for a digital tablet [42]. Pain was reported in terms of body location (free-hand drawing over a body outline, 1000×1084 pixel resolution), quality of pain (aching, shooting, burning, or tenderness), tissue type (skin, muscle, bone), and pain intensity. Specifically, for each ‘pain sensation’ they were first asked to select the quality of pain by choosing a descriptor from a list (i.e. aching, shooting, burning, or tenderness), pain intensity using a Visual Analog Scale (VAS 0–10), and tissue type for where they thought the depth of the pain was focused (skin, muscle, or bone). When this information had been entered, patients drew the location of experienced pain over a front and back view of a neutral body outline. Patients were asked to shade every point of the body outline where the pain sensation was present, and to avoid hatching, ticking, or use of symbols (e.g. arrows). They were able to ‘undo’ markings if erroneously drawn. Patients were able to enter as many pain sensations as they needed to accurately indicate their fibromyalgia pain.

Processing and analyses of pain drawings

Each individual’s 2-dimensional pain map for each pain quality was converted to NIFTI image format, and analyzed using image analysis tools amenable to voxel or pixel-wise intensity mapping (FMRIB Software Library, FSL, v6.0).

Descriptive analyses of pain drawings:

To investigate the overall spatial bodily distribution of pain across patients, we calculated the frequency of pain sensations (percent of patients reporting pain, irrespective of intensity) pixel-by-pixel for the entire body. This pain frequency body map was calculated individually for each specific sensation category (aching, tenderness, burning, and shooting), and overall for any pain sensation, collapsed across categories (Fig. 1).

Figure 1:

Topographic distribution of fibromyalgia pain qualities. Body maps represent the percentage of patients (N=79) indicating respective body locations as painful (pixel-by-pixel, 1000×1084 pixel resolution), for any pain sensation irrespective of type (a), and for each pain sensation separately (b). Pain drawings were made digitally using a pain sensation mapping app on a touch-screen tablet.

We also calculated the overall percent of participants reporting any pain sensation, and the pain intensity (mean±SD), for each pain quality.

Overall pain widespreadness was operationalized as the percent of the total body area marked as painful (irrespective of pain sensation category, see Supplementary Methods and Fig. S1 for more details on this metric of widespreadness) and used as a regressor for functional brain network connectivity analyses (see below). In order to evaluate the association between widespreadness and pain catastrophizing, we performed a Spearman-Rank coefficient (a Shapiro-Wilk test indicated that pain widespreadness deviated from a normal distribution: W=0.96, P=0.025).

Pain extent and quality within different body regions:

A key facet of fibromyalgia is pain widespreadness. To investigate how the spread of pain might differ across body regions and pain quality, we employed pre-defined lateralized body Regions Of Interest (ROI) based on the Widespread Pain Index (WPI [55]). In addition to the 19 WPI regions, we also included ROIs for the face (above the jaw) and back of head, which are not defined labels in the WPI scale, as there were pain reports also for these regions. This resulted in a total of 21 body regions, which were used for this descriptive analysis of regional pain extent only, and not for the sake of calculating a “WPI score” (number of regions checked as painful). Masking these regions of each patient’s body map, we calculated the extent of pain (percent of body region marked as painful) for each sensation type and body region, and calculated the group mean. Importantly, a given region was only included in this calculation if it contained pixels marked as painful. Thus, the resulting values reflected the extent of painful sensations for respective body regions and pain qualities, which was not affected by the relative difference of total pain sensations between different pain qualities.

Laterality of spatial extent of pain:

We explored whether spatial extent of pain (percent of body region marked as painful) was lateralized for any body region. Specifically, we performed a repeated measures ANOVA (corrected for sphericity using the Greenhouse-Geisser correction) with factors ‘Laterality’ (right, left) and ‘Body part’ (lower leg, upper leg, hip, lower arm, upper arm, shoulder, jaw, face above jaw). For this analysis, we only included body ROIs that were not solely midline (i.e. the neck, chest, abdomen, upper back, and lower back were not included in this analysis).

Catastrophizing’s body region-specific associations with pain qualities:

Although previous studies have linked pain catastrophizing with pain intensity or severity, these investigations have typically considered catastrophizing in relation to one-dimensional (overall) pain ratings. To explore how pain catastrophizing might be differentially associated with pain intensity depending on pain quality and body location, we performed a whole-body pixel-wise regression analysis, both for each specific pain sensation and for pain overall (irrespective of sensation type), with pain catastrophizing (PCS, total score) as a regressor of interest. Due to the large number of zero values for any given pixel (i.e. not marked as painful), the distribution of intensity scores across patients were negatively skewed. We therefore carried out a pixelwise non-parametric regression analysis (FSL’s randomise tool), employing Threshold-Free Cluster Enhancement with 2D optimization, and using 5000 permutations.

MRI acquisition

Blood oxygen level-dependent (BOLD) fMRI data were collected during rest (3.0T Siemens Skyra, Siemens Medical, Erlangen, Germany, 32-channel head coil), using a whole-brain, simultaneous multi-slice, T2*-weighted gradient echo BOLD echo-planar imaging pulse sequence (multi-band acceleration factor = 5, repetition time = 1250 ms, echo time = 33 ms, flip angle = 65°, voxel size = 2 mm isotropic, number of axial slices = 75 (FOV = 200 × 200 mm).

A high-resolution T1-weighted volume (multi-echo MPRAGE) was also collected to facilitate anatomical localization and spatial registration of individual fMRI BOLD volumes to MNI152 standard space (repetition time = 2530 ms, echo time = 1.69 ms, inversion time (TI) = 1100 ms, flip angle = 7°, field of view = 256 × 256 mm, spatial resolution = 1 mm isotropic).

A gradient echo sequence for mapping of magnetic field (B₀) inhomogeneities was collected (repetition time = 800 ms, echo time 1 = 5.29 ms, echo time 2 = 8.75 ms, flip angle = 60°, voxel size = 2 mm isotropic, number of axial slices = 75 (FOV = 200 × 200 mm).

MR-compatible cardiac pulse and respiration signals were collected continuously (500 Hz, MP150; Biopac Systems) throughout the fMRI run, and used for physiological noise correction. Respiratory data were collected using an MR-compatible belt system constructed in-house, based on the system devised by Binks et al. [10].

fMRI preprocessing

Preprocessing of individual fMRI data was performed using tools from the FMRIB’s Software Library (FSL, v6.0.0; www.fmrib.ox.ac.uk/fsl), Freesurfer (http://surfer.nmr.mgh.harvard.edu), and MATLAB (The MathWorks, Natick, MA). Since cardiac and respiratory activity is known to influence the BOLD signal [11,14], individual fMRI data were first corrected for cardiorespiratory artifacts using the retrospective image correction algorithm (RETROICOR) [24]. Due to inadequate cardiorespiratory data quality, FMRI data for 5 participants were preprocessed without RETROICOR. Annotation of heart beats was performed using an in-house semi-automated algorithm (MATLAB), while respiratory volume (per timepoint) was calculated using automated algorithms and in-house software (MATLAB). Physiological data were resampled at 40 Hz before RETROICOR application. To further correct for cardiorespiratory artifacts in the BOLD fMRI signal, we applied respiratory and cardiac response functions to the respiration and cardiac data, respectively, using previously determined convolution kernels [11,14], resulting in nuisance regressors used in the functional connectivity analyses (see below). This approach has been shown to be efficient in removing additional cardiorespiratory noise beyond the RETROICOR step. Preprocessing included removal of the first 3 fMRI volumes, slice-timing correction, motion correction using MCFLIRT, field map-based echo-planar imaging unwarping (FSL’s PRELUDE and FUGUE), and nonbrain voxel removal (BET). Physiological and scanner-related noise was then filtered out using signals from non-parenchyma tissue (aCompCor [8]), which complements the RETROICOR algorithm by also targeting additional sources of artifacts beyond those introduced by cardiac and respiratory signals. Specifically, principle component analyses were performed on the preprocessed data, separately within White Matter (WM) and Cerebrospinal Fluid (CSF) regions of interest. These regions were determined using probabilistic segmentation of WM and CSF at 90% probability (SPM 12, http://www.fil.ion.ucl.ac.uk/spm/), and then eroded by 1 voxel in order to minimize partial volume effects. The top 5 components were regressed out of the preprocessed data. The global signal timeseries were not included as a regressor [13]. For each individual, the preprocessed data were registered to standard space (Montreal Neurological Institute, MNI152). First, each individual’s structural volume was registered to the MNI152 standard space using Boundary Based Registration for the functional to high-resolution transformation (bbregister, Freesurfer, v6.0.0) and FSL’s Linear registration tool (FLIRT, 12 degrees of freedom) [28], followed by FSL’s non-linear registration tool (FNIRT) [1] for the high-resolution to standard-space transformation. Finally, the preprocessed fMRI time series were spatially smoothed (full width at half maximum = 4 mm) and temporally high-pass filtered (f_high = 0.008 Hz as computed by FSL’s cutoffcalc).

fMRI analyses

A group-level Independent Component Analysis (ICA) was carried out in order to identify functional resting state networks, using the FSL tool Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC [7]). First, all individual preprocessed fMRI data were concatenated into a single 4D dataset. We then performed MELODIC with a set limit of 25 independent components, consistent with multiple previous studies that have yielded reliable estimation of resting state network connectivity [21,40]. This was followed by a dual regression analysis. As in our previous studies, we calculated spatial correlation with previously defined templates [5] for identification of functionally relevant components corresponding to canonical networks. In line with the notion that pain catastrophizing is linked with exaggerated or misattributed saliency for incoming signals, we specifically investigated functional connectivity with the SLN (Fig. S2, [61]), which has been previously linked with chronic pain dysfunction, and hypothesized to be linked with pain widespreadness. These component maps were then used as spatial regressors for General Linear Model (GLM) analyses to estimate temporal dynamics associated with each independent component for each individual participant. These timeseries were then used as regressors in a second GLM, together with several nuisance regressors (6 motion parameters calculated by MCFLIRT and cardiorespiratory functions defined by convolving each participant’s cardiac and respiratory variation time series with the cardiac and respiratory transfer functions [11,14]).

To investigate the association between pain widespreadness and SLN connectivity with specific brain regions, reconstructed independent spatial maps for SLN were passed up to a group-level regression analysis with pain widespreadness as a regressor of interest, using FMRIB’s Local Analysis of Mixed Effects (FSL-FLAME1+2). The whole-brain group map was then corrected for multiple comparisons using FSL’s ‘cluster’ tool, which applies Gaussian Random Field GRF theory to determine initial minimal extent threshold, using a voxel-wise cluster forming statistical threshold of z = 2.3, and cluster significance threshold of P = 0.05 [6,58].

Mediation analysis

Previous studies have suggested that SLN connectivity with DMN regions plays a role in centralized (chronic) pain [2,26,41], especially for patients displaying high pain catastrophizing [31]. We therefore investigated whether SLN connectivity with the PCC, a key DMN node we have previously suggested encodes pain catastrophizing [35] and which found to be linked with pain widespreadness in the current study, statistically mediated the association between pain catastrophizing and widespreadness. Individual PCS scores were used as the independent variable. SLN connectivity to PCC was used as the mediator variable. Specifically, we extracted each individual’s mean Zstat value from the PCC region, using a binarized mask containing statistically significant voxels from the whole-brain SLN connectivity vs. pain widespreadness linear regression analysis, intersected by an anatomical mask of the posterior cingulate (Cingulate gyrus, posterior division, probability>10%, from the Harvard-Oxford Cortical Structural Atlas). This approach, which combines anatomical and functional definitions for PCC may provide improved sensitivity compared to a purely anatomically atlas-defined mask for large cortical regions. Pain widespreadness values (ranked, Shapiro-Wilk normality test: W=0.96, P=0.02) were used as the dependent variable. The R package ‘Mediation’ was used for mediation analyses. We tested for statistical significance using a non-parametric boot strapping approach (1000 iterations, α=0.05), and considered the mediation significant if the total indirect effect (a*b) was statistically significant, while the previously significant direct effect (path c) became non-significant after controlling for the mediator (c’) [27].

Results

Descriptive analyses of pain drawings

Figure 1 shows a pixelwise representation of the frequency of pain markings across patients. The most common locations for fibromyalgia pain (any pain sensation, Fig. 1a) were localized in the neck/upper back (74.68%) and the lower back (72.15%). The same pattern was seen for aching (neck: 53.09%, lower back: 56.79%). The most common areas for tenderness were the neck (28.40%), the left upper arm (24.69%), and the left upper shoulder (23.46%). Burning pain sensations were most commonly reported in the shoulders (16.05%), neck (14.81%), feet (13.58%), upper back (13.58%), lower back (13.58%), lower arms (11.11%), and left thigh (11.11%). Shooting pain was most commonly present in the neck (17.28%), followed by the lower back (16.04%), knee (13.58%), thigh (12.34%), back of head (11.11%) and hands (11.11%).

Not every patient reported pain sensations of all categories. The most commonly reported sensation was aching (94.94% reporting any aching sensation, pain intensity (mean±SD): 58.76±18.44, scale of 0–100) followed by tenderness (74.68% reporting sensation, pain intensity: 53.70±19.54), shooting (67.09% reporting sensation, pain intensity: 64.23±16.90) and burning (54.43% reporting sensation, pain intensity: 60.60±16.23) (Fig. 1b).

For pain drawings overall, irrespective of specific sensation (i.e. collapsing across aching, burning, shooting, or tenderness sensations), fibromyalgia patients reported relatively high pain widespreadness, calculated for each individual as the percent of the entire body marked as painful (mean±SD: 26.95±21.91 percent of total body area). A Spearman-Rank coefficient (a Shapiro-Wilk test indicated that pain widespreadness deviated from a normal distribution: W=0.96, P=0.025) indicated a positive correlation between pain widespreadness (i.e. % of entire body area for any sensation) and pain catastrophizing (r=0.26, P=0.02).

Pain extent and quality within different body regions

The calculation of pain extent within individual body regions indicated that “aching” sensations had the highest area coverage of all pain sensations (mean percent of the body±SD: 16.62±16.35), followed by “tenderness” (13.29±17.68), “burning” (8.37±11.07), and “shooting” (5.02±4.78) sensations (Table 2 and Fig. 2).

Table 2:

Widespreadness (Percent of total area marked as painful) by body location and pain sensation (mean (SD))

Body location	Any sensation	Aching	Burning	Shooting	Tenderness
Any location	26.95 (21.91)	16.62 (16.35)	8.37 (11.07)	5.02 (4.78)	13.29 (17.68)
Back of head	16.13 (29.09)	11.50 (24.01)	2.94 (9.00)	6.42 (18.74)	5.96 (19.25)
Face, left	18.14 (30.09)	10.68 (23.38)	3.01 (15.12)	9.43 (22.21)	5.15 (16.85)
Face, right	15.79 (27.52)	9.50 (20.47)	2.39 (14.03)	7.40 (19.82)	5.53 (17.43)
Jaw, left	14.48 (26.77)	7.87 (19.58)	5.20 (15.97)	4.75 (16.16)	5.14 (17.79)
Jaw, right	12.63 (26.40)	6.83 (19.57)	2.53 (11.89)	4.98 (18.49)	5.42 (17.19)
Neck	36.74 (29.89)	25.69 (26.79)	10.31 (19.63)	9.26 (16.35)	18.39 (26.83)
Shoulder, left	44.31 (29.79)	27.99 (25.90)	16.08 (21.77)	6.23 (12.56)	22.88 (25.44)
Shoulder, right	36.59 (29.75)	23.96 (25.30)	11.09 (16.61)	4.06 (10.05)	17.18 (24.18)
Upper arm, left	29.91 (30.38)	13.56 (20.29)	11.5 (22.85)	3.76 (10.45)	19.13 (28.53)
Upper arm, right	26.65 (29.97)	13.11 (20.79)	10.33 (21.24)	1.04 (2.89)	17.14 (28.94)
Lower arm, left	25.92 (27.76)	14.78 (21.25)	8.67 (17.54)	5.73 (11.23)	14.05 (22.83)
Lower arm, right	23.61 (28.08)	13.94 (22.45)	7.89 (17.52)	4.74 (10.59)	12.00 (20.89)
Upper back	46.38 (33.93)	32.43 (30.73)	16.26 (28.18)	5.99 (11.31)	21.31 (28.13)
Lower back	43.10 (31.46)	33.29 (28.67)	10.56 (21.81)	10.39 (18.1)	13.78 (25.31)
Chest	19.41 (27.71)	9.13 (19.47)	5.01 (14.46)	3.24 (10.31)	13.16 (22.54)
Abdomen	13.33 (24.47)	8.12 (19.26)	2.29 (8.93)	1.77 (6.37)	7.66 (21.95)
Hip, left	28.78 (30.18)	21.64 (27.01)	7.03 (13.80)	7.64 (12.90)	13.64 (23.48)
Hip, right	26.24 (29.37)	19.10 (26.30)	5.05 (14.23)	5.67 (11.88)	11.93 (20.96)
Upper leg, left	23.64 (25.93)	14.01 (20.39)	7.50 (15.53)	5.08 (8.75)	11.21 (22.44)
Upper leg, right	21.83 (26.39)	14.18 (21.14)	5.51 (13.37)	3.07 (6.27)	10.28 (21.68)
Lower leg, left	27.05 (26.77)	14.53 (19.34)	10.77 (18.14)	5.65 (9.31)	13.12 (21.81)
Lower leg, right	26.18 (25.96)	15.12 (20.41)	9.03 (17.18)	4.71 (8.74)	12.75 (21.08)

Figure 2.

Pain extent within body regions. Body maps show the mean extent of pain (% of pixels within each region marked as painful), for different body regions and sensation type.

Since body regions differ in size, it is possible that differences in within-region pain extent between different body regions could be confounded by region size (e.g. smaller regions having larger relative coverage). Importantly, we did not find any correlation between pain extent and region size across body regions (r=0.08, P=0.48).

Laterality of spatial extent of pain

A repeated measures ANOVA with factors ‘Laterality’ and ‘Body part’ indicated a significant main effect of Laterality, in which the spatial extent of pain (percent of region marked as painful) was higher overall on the left side of the body (F(1,78)=14.56, P<0.001) and a less strong, but statistically significant, interaction term Laterality*Body part (F(4.69, 365.50)=2.46, p=0.036, which may have been driven by a larger laterality difference for the shoulder (Fig. S3).

Catastrophizing’s body region-specific associations with pain qualities

The pixelwise non-parametric regression analysis indicated that pain catastrophizing (PCS total score) was positively correlated with pain intensity – irrespective of sensation type – in the lower back, neck/upper back, shoulders/upper arms, thighs, and legs (Fig. 3a). For individual pain sensations, pain catastrophizing was positively correlated with aching pain in the lower back, neck, and shoulders, and with leg-related burning and shooting pain (Fig. 3b). Further, pain catastrophizing was negatively associated with burning and shooting pain in the back and the face, and with shooting pain in the back of the head. While pain catastrophizing was not strongly associated with tenderness, there was a negative correlation between pain catastrophizing and tenderness in the upper abdomen.

Figure 3:

Association between pain catastrophizing and chronic pain intensity across the body. A) A whole-body pixel-wise regression analysis of pain intensity irrespective of sensation type (any pain sensation) indicated that pain catastrophizing was significantly associated with pain intensity in the neck/upper back, lower back, shoulders/upper arms, thighs, and legs. B) Separate regression analyses for each sensation type indicated different patterns for different pain sensations. While pain catastrophizing was positively associated with low-back aching pain, it was negatively associated with low-back burning pain. Furthermore, pain catastrophizing was positively associated with burning and shooting pain in the legs, while there was no strong relationship with leg aching pain. Tenderness intensity did not show strong associations with pain catastrophizing.

As shown in Fig. 3b (left panel), PCS showed an opposite correlation with aching (r=0.43) and burning pain (r=−0.34) in the low back, although not all participants reported aching or burning pain in the low-back. In an alternative analysis with non-zero values for low-back pain excluded, these correlations remained consistent with the original statistics, and, in fact, showed stronger associations (Aching: r=0.50, Burning: r=−0.74).

Association between pain widespreadness and SLN connectivity

A whole-brain regression analysis indicated that pain widespreadness was positively associated with SLN connectivity to PCC (peak voxel [MNI152, x y z] = 12 −52 40 mm), a key node of the DMN, in addition to dorsolateral prefrontal cortex (dlPFC, peak voxel, right hemisphere [MNI152, x y z] = 42 30 40 mm; left hemisphere = −32 25 42 mm), left primary somatosensory cortex (S1, peak voxel [MNI152, x y z] = −44 −38 40 mm), and the anterior Insula (aINS, peak voxel [MNI152, x y z] = −32 6 −10) (Fig. 4, Table 3). There were no voxels showing a significant negative correlation with pain widespreadness.

Figure 4:

Association between Salience Network connectivity and pain widespreadness. A whole-brain regression analysis showed that chronic pain widespreadness (percent of the digital body map indicated as painful) was significantly correlated with SLN connectivity with the right PCC, bilateral dlPFC, and right S1. The lower left panel illustrates the association between whole-body widespreadness and SLN connectivity with the PCC (mean Zstat scores extracted from a binary mask corresponding to the significant PCC cluster. Moreover, SLN – PCC connectivity was positively correlated with pain catastrophizing (lower right panel). XYZ coordinates refer to MNI152 standard space. SLN=Salience Network; PCC=Posterior Cingulate Cortex; dlPFC=dorsolateral Prefrontal Cortex, S1=primary somatosensory area; *P<0.05.

Table 3:

Whole-brain regression analysis with whole-body widespreadness

	Side	Size (voxels)	Coordinates (MNI 152)			Peak z-stat
			X	Y	Z
PCC	R	297	12	−52	40	4.54
S1	L	276	−44	−38	40	4.36
aINS	R	231	32	6	−10	3.84
dlPFC	R	218	42	30	40	4.02
dlPFC	L	167	−32	26	42	4.54

PCC: Posterior Cingulate Cortex; S1: Primary Somatosensory Cortex; aINS: anterior Insula; dlPFC: dorsolateral Prefrontal Cortex.

Mediation of the association between pain catastrophizing and widespreadness by SLN-PCC connectivity

There was a significant direct linear association between PCS and pain widespreadness (path c coefficient(SEM): 0.46(0.20), t=2.26, P=0.026, Fig. 5a and Fig. S4), which became non-significant when controlling for SLN-PCC connectivity (path c’: 0.25(0.19), t=1.36, P=0.179, Fig. 5b). There were significant linear direct associations between PCS and SLN-PCC connectivity (path a: 0.02(0.01), t=2.15, P=0.034) and between SLN-PCC connectivity and Pain widespreadness (path b: 10.47(2.02), t=5.20, P<0.001). A bootstrapped mediation analysis indicated a significant effect of the indirect path (a*b coefficient=0.21, P=0.01, CI=0.05, 0.39), suggesting that SLN-PCC connectivity statistically mediated the association between pain catastrophizing (PCS score) and pain widespreadness (% of total body area reported) (Fig. 5).

Figure 5:

SLN-PCC connectivity mediated the association between pain catastrophizing and widespreadness. A) Pain catastrophizing showed a direct linear association with pain widespreadness (percent of the digital body map indicated as painful). B) A mediation analysis showed a statistically significant effect of the indirect path (the coefficient estimate for path a*b is shown with 95% confidence intervals in parentheses), indicating that SLN-PCC resting connectivity statistically mediated the association between pain catastrophizing and pain widespreadness. Coefficients for direct paths a, b, c, and c’ are reported with the Standard error of the mean (SEM) in parentheses. PCS=Pain Catastrophizing Scale; SLN=Salience Network; PCC=Posterior Cingulate Cortex; *P<0.05; **P<0.01; ***P<0.001

To investigate the possibility that overall pain intensity affected this mediation, we performed an additional alternative mediation model, with variables PCS (independent variable), SLN-PCC connectivity (mediator), and Pain widespreadness (Dependent variable), as above, but with ‘mean pain intensity’ (across the entire body) as a covariate. The results indicated a significant effect of the indirect path (a*b=0.16, P=0.014, CI=0.03, 0.33), similar to the original mediation, thus not indicating that the mediation was driven by overall pain intensity.

Discussion

In this study, we collected digital pain drawings in combination with assessment of resting state functional MRI to investigate the association between pain widespreadness, catastrophizing, and functional brain network connectivity in patients suffering from fibromyalgia. We found that pain catastrophizing was associated with increased pain widespreadness (i.e. number of pixels endorsed on pain drawings). Specifically, pain catastrophizing was positively correlated with aching pain in the lower back, neck, and shoulders, but negatively associated with burning and shooting pain in the back and face. Furthermore, pain widespreadness was also associated with increased brain SLN connectivity to the PCC, a key node of the Default Mode Network. In fact, we found that the strength of SLN-PCC connectivity statistically mediated the association between pain widespreadness and catastrophizing.

The usage of digital pain drawings has increased in recent years. Patients’ drawings provide unique information about clinically meaningful parameters, such as the precise location and extent of pain, that are not accurately captured by one-dimensional pain ratings [48]. They also enable patients to describe spatial patterns of their pain in more fine-grained detail compared to surrogate indices of pain widespreadness, such as the number of major body regions endorsed for pain [55]. While there has been recent progress in utilizing digital pain drawings for more sophisticated outcomes such as pain area/widespreadness, pain clustering, segmental involvement, and compound metrics combining spatial and intensity information[48], the application of these techniques in combination with neuroimaging to assess potential brain mechanisms supporting these aspects of the pain experience has been lacking. Such information may be key to understanding chronic pain conditions such as fibromyalgia, for which these pain parameters can be highly heterogeneous between patients [15]. Here, we applied pixel-based analysis using conventional software for analyzing multi-voxel fMRI data (FSL) in order to assess individual topography and spatial extent of pain loci. Comparing different categories of pain sensations, we found that aching sensations were the most commonly reported, followed by tenderness, shooting, and burning sensations. For aching sensations, pain frequency maps revealed prominent clusters in the lower-back, upper-back, neck, shoulders, and knees, similar to other reports of fibromyalgia pain properties [34]. This pattern was also seen for shooting pain, while tenderness and burning were less uniform across patients. Notably, back and neck pain were prominent across all pain categories.

Pain catastrophizing is common in patients with widespread chronic pain, and is a key risk factor for increased pain severity, co-morbid depression, more widespread pain, and impaired mental health [20]. Indeed, a recent meta-analysis identified preoperative pain catastrophizing as the strongest predictor of persistent pain (>3 months) after knee replacement surgery [36]. Our digital pain drawings data showed differences between body regions in how catastrophizing was associated with pain intensity. In particular, catastrophizing was inversely associated with aching and burning pain in the lower back. Those patients with the highest pain catastrophizing scores reported higher aching intensity, but lower burning intensity, in the lower-back. Interestingly, burning sensation is associated with neuropathic pain, while aching is a more non-specific functional pain sensation [17], raising the possibility that overlapped neuropathic pain syndromes in fibromyalgia that affect the back (e.g. radicular low back pain) may lead to reduced sensitivity to pain catastrophizing for some of the affected patients. On the other hand, burning was one of the least commonly reported pain qualities, thus this inverse association between burning and pain catastrophizing in the low back may be driven by a small subset of patients, as many patients reported no burning sensation. Moreover, some sensations may be reflexively associated with specific body regions, such as “aching” for the low back, “shooting” in the legs, and “throbbing” in the head. Ultimately, future studies should replicate these findings in new patient cohorts before the inverse association reported in our study is considered for further speculation.

Amplified resting brain cross-network connectivity – or reduced anti-correlation – between key nodes of the SLN and the DMN is a reproducible characteristic of chronic pain [2,26,31,41]. Recent studies have suggested that altered DMN connectivity [32] and activity [3,35] in chronic pain might reflect self-referential cognitive and affective processes such as pain rumination and catastrophizing, and that increased SLN connectivity with the PCC, a key DMN node, underpins subjective chronic pain severity rather than physical impairment [26]. Using experimental induction of depressed mood, PCC was activated in a group of healthy individuals in response to evoked pain when reading statements related to catastrophizing [9]. Notably, a recent study applied Support Vector Machine learning to classify fibromyalgia patients from healthy controls based on fMRI response to evoked pain and multisensory non-pain stimuli, and found that PCC activation during multisensory stimuli contributed strongly to classification while PCC responses to evoked pain stimuli were not a significant predictor [38]. This is consistent with the notion that, rather than PCC playing a direct role in nociceptive processing, the PCC may play a higher-level role in chronic pain, underpinning cognitive processes such as self-referential cognition and catastrophizing. Our group recently found that cLBP patients, relative to healthy controls, showed increased SLN connectivity with PCC after back pain exacerbation [31]. Notably, in a subgroup of patients with the highest pain catastrophizing scores, increased DMN-aINS connectivity induced by pain exacerbation, was correlated with individual worsening of low back pain. Furthermore, in another recent study, fibromyalgia patients, relative to healthy controls, activated PCC while reflecting upon catastrophizing-related statements, with greater PCC activation associated with greater self-rated applicability of pain catastrophizing statements [35]. In the present study, we found that resting SLN-PCC connectivity was positively associated with both pain catastrophizing and whole-body widespreadness of fibromyalgia pain. Interestingly, elevated resting connectivity between key nodes of the SLN and DMN has also been previously linked with pain widespreadness in rheumatoid arthritis [4], further suggesting shared neuropathophysiology. In our study, the association between pain catastrophizing and whole-body widespreadness was statistically mediated by SLN-PCC connectivity, suggesting a potential brain mechanism for how pain catastrophizing can engender greater clinical pain widespreadness in fibromyalgia. These findings may contribute to an increasingly mechanistic understanding of how sensory and psychological processes interact during the development of chronic pain, which will ultimately aid the development of new therapeutic approaches.

We also found that pain widespreadness was correlated with resting SLN connectivity to other brain regions – e.g. S1 and dlPFC. S1 connectivity to anterior insula, a key hub of the SLN, has been reported with somatotopic specificity for evoked pain in fibromyalgia [30] and was associated with clinical pain in cLBP [31]. These results suggest that SLN/S1 connectivity supports saliency and focused attention on both regionally-specific pain intensity as well as more generalized pain widespreadness. Furthermore, a recent study in patients with chronic pelvic pain reported that those with widespread pain, relative to more localized pain, showed increased SLN connectivity with S1, in addition to increased gray matter volume in this primary sensory region [33]. In turn, the dlPFC is a functionally diverse region, with different subregions constituting central nodes of the SLN, DMN, and Frontoparietal Control (or Executive Control/Attention) Networks [60]. In our study, the peak coordinates for correlation with pain widespreadness were within the boundaries of the Frontoparietal Control Network. The dlPFC is commonly activated in response to evoked pain [47], and has been shown to play an important role in endogenous down-regulation of pain [54]. Interestingly, regions of the SLN and Frontoparietal control network frequently coactivate as a task-positive ensemble for many experimental tasks, such as working memory [46]. Our study suggests that greater pain widespreadness is linked with more broadly distributed SLN connectivity, encompassing nodes of the closely related Frontoparietal control network.

One limitation of our study is the lack of a non-fibromyalgia control group, which limits inferences about whether these findings generalize to other chronic pain conditions. Furthermore, future studies should include men with fibromyalgia to disentangle possible sex/gender differences. Another important limitation, although we found that SLN-PCC connectivity statistically mediated the association between pain catastrophizing and pain widespreadness, the cross-sectional nature of this investigation limits inferences about causal directionality. One possibility is that increased SLN-PCC connectivity mediates an effect in which increased pain widespreadness exacerbates pain catastrophizing, or that they interact in a mutually-reinforcing cycle, consistent with the reciprocal relationship between fear/anxiety, avoidance, and the spread of pain to previously non-painful body locations, analogous to the rendering of previously nonpainful sensory signals as painful, as described by the fear-avoidance model of chronic pain [53]. Future longitudinal clinical studies, or experimental models allowing for causal inferences, may shed further light on the directional dynamics between intrinsic brain connectivity, pain catastrophizing, and pain widespreadness in fibromyalgia. Specifically, such studies may discover whether aberrant coupling of the SLN, which supports stimulus salience processing, with DMN nodes such as the PCC, is primarily a driver or a consequence in this development. Moreover, the contribution of networks other than the SLN should also be investigated further. Finally, the body map template used in this study could be interpreted as masculine, which stands in contrast to the all-female sample. Future studies should strive to apply body depictions that resemble the study sample as closely as possible with regard to age, sex, and BMI.

In conclusion, we identify a putative brain process underlying the linkage between greater pain catastrophizing and greater pain widespreadness in fibromyalgia. Our study further defines a specific role for intrinsic brain SLN-DMN cross-connectivity in mediating cognitive-affective and sensory dimensions of the clinical pain experience.