Journal Logo

Research Paper

Early changes in brain network topology and activation of affective pathways predict persistent pain in the rat

Sperry, Megan M.a; Granquist, Eric J.b; Winkelstein, Beth A.a,c,*

Author Information
doi: 10.1097/j.pain.0000000000002010

1. Introduction

Multiple chronic disorders of the central nervous system (CNS), including pain, are associated with altered brain activity and structure, which are hypothesized to contribute to symptom persistence.11,12,18 However, the transition from acute to chronic symptoms remains poorly understood.42 Although persistent pain alters the expression of synaptic receptors and strengthens synaptic activity to greater extents than is observed with transient pain,10,29 it is unknown if persistent pain differentially engages subcircuits, or even entire networks, of the brain compared to conditions when pain resolves. Identifying brain adaptations that are unique to persistent pain may inform clinical interventions by identifying anatomical targets for neuromodulation17,36 or drug delivery,59 while leaving acute pain pathways intact.

Network analysis, in which the brain is treated as a network of interacting components,45,46 has identified altered brain connectivity in pain patients with chronic lower back pain, complex regional pain syndrome, osteoarthritis, and temporomandibular joint (TMJ) disorder.6,21,31 For example, associations between brain regions are strengthened in chronic back pain, particularly between the insular and cingulate cortices, and between those regions and the prefrontal cortex.18 Although changes in functional networks are evident in patients with pain lasting for 5 months to 11 years,21,31 it is likely that brain network adaptations may occur even earlier than those identified in patients. For example, mild functional brain network alterations appear 5 days after the onset of persistent neuropathic pain in the rat, network structure is more fragmented 12 to 14 days after pain onset, and the corticolimbic system is substantially altered by day 28.3,24 However, brain networks have not been compared as conditions differentially develop into either persistent or transient pain.

Because clinical studies are inherently limited by heterogeneous patient histories, lack of baseline data, and challenges in acquiring data early after pain onset, this study used a tunable rat model to induce either transient or persistent TMJ pain22,54 to assess if brain networks adapt after pain onset and/or can be modulated by peripheral treatment. Using 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) imaging, resting-state brain scans were acquired at times: (1) before they developed either transient or persistent pain, (2) when pain is detectable in both groups (day 7), and (3) later when pain remains only in the persistent pain group (day 14).22,54 Correlation distribution, node strength, clustering coefficient, and community structure were measured to evaluate if, and how, local and global brain network properties adapt in different pain states. Structural equation modeling (SEM) independently tested the activation of specific pain circuits in each state. To confirm that any adaptations are associated with pain development, a third group was pretreated with intra-articular etanercept (anti-tumor necrosis factor [TNF]) before undergoing procedures to impose persistent pain. Because etanercept inhibits TNF-α, which sensitizes nociceptive neurons in the TMJ and is acutely increased in the TMJ only with persistent pain,22,49 this approach enabled identifying brain network and subcircuit features that are unique to a persistent pain state and may represent early prognostic biomarkers of chronic pain.

2. Methods

2.1. Animal procedures

All procedures were approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania. Experiments were performed using female Holtzman rats (Envigo, Indianapolis, IN) weighing 271 ± 12 g at the beginning of the study (approximately 10 weeks of age). Rats were housed with conditions recommended by the Association for Assessment and Accreditation of Laboratory Animal Care with a 12/12-hour light/dark cycle, environmental enrichment, and free access to food and water. Rats were randomly assigned to TMJ loading and treatment groups. Repeated mouth-open jaw loading (1 hours/day for 7 days) under isoflurane anesthesia (4%-5% induction; 2%-3% maintenance) was used to induce TMJ sensitivity that is either transient (2 N loading; n = 16) or persistent (3.5 N loading; n = 22), as previously detailed.22,37,54 A separate group of rats (n = 11) received the TNF-α inhibitor etanercept (Amgen, Thousand Oaks, CA) intra-articularly 1 day before the start of the 3.5 N TMJ loading paradigm. Rats were anesthetized using isoflurane (4% induction; 2%-3% maintenance) and a 27 G needle was placed into the bilateral TMJs under computed tomography (CT) guidance. A single dose of etanercept (100 µg) dissolved in saline (20 µL) was delivered to each bilateral TMJ.

Temporomandibular joint pain was assessed by mechanical reflex testing in the TMJ region22,37 and the rat grimace scale (RGS) to quantify facial expression.41,51,54 Temporomandibular joint sensitivity was assessed by mechanical reflex testing by a single operator at baseline and every other day during and after loading (days 1, 3, 5, 7, 9, 11, 13, and 14) for rats exposed to 2 or 3.5 N loading (n = 10/group).22,37 Spontaneous pain was separately characterized for rats with persistent pain at baseline, day 7, and day 13 and compared to rats with transient pain at baseline and day 7 (n = 6/group) using the RGS to quantify facial expression through video recording and blinded scoring by a single operator of 4 action units: orbital tightening, nose and cheek flattening, ear curling, and whisker bunching.41,51,54 Rats pretreated with etanercept before the 3.5 N loading were assessed for sensitivity at baseline and days 1, 3, 5, and 7 (n = 10). The threshold for eliciting a head withdrawal was measured using von Frey filaments of increasing strengths from 0.6 to 60 g to stimulate the TMJ region (Stoelting, Wood Dale, IL).22,37 Spontaneous pain was assessed at baseline and on day 7 in a subset of pretreated rats (n = 7) and a separate group of rats with persistent pain (n = 8) using the RGS.41,51,54

2.2. Positron emission tomography imaging and image processing

18F-FDG PET images were acquired for rats (n = 30) in all groups at baseline before any other procedures and on day 7 (1 day after cessation of jaw loading). Positron emission tomography images were also acquired at day 14 (8 days after cessation of jaw loading) in rats from each of the transient and persistent pain groups (n = 10/group).53 Using previously described procedures,53 rats were injected with 18F-FDG (0.73 ± 0.22 mCi; University of Pennsylvania Cyclotron Facility) through tail vein catheter under brief exposure to isoflurane anesthesia (4% induction; 2% maintenance). Rats were transitioned to dexmedetomidine sedation (0.075 mg/kg in 0.9% saline; DEXDOMITOR; Zoetis, Parsippany, NJ). Three-dimensional PET images were acquired 1 hr after injection of 18F-FDG using either the Philips MOSAIC HP Small Animal PET scanner or β-CUBE PET scanner (Molecubes, Gent, Belgium) (15-minute single-frame acquisition). Complementary CT scans were acquired immediately after PET scanning using either the ImTek microCT scanner (Freiburg, Germany) or X-CUBE CT scanner (Molecubes). After scanning, dexmedetomidine sedation was reversed using atipamezole (ANTISEDAN; Zoetis) and rats were transferred to isolated housing until radioactivity was below detectable limits.

18F-FDG PET images were segmented into 50 brain regions and normalized by the mean uptake of the whole brain.5318F-FDG PET and CT data were reconstructed, coregistered, and cropped around the head.26,47 The Small Animal Molecular Imaging Toolbox anatomical magnetic resonance imaging template was spatially normalized to CT scans by affine registration (Advanced Normalization Tools)2 and the same affine transform was applied to the corresponding Small Animal Molecular Imaging Toolbox stereotaxic rat brain atlas.4818F-FDG PET images were segmented into 50 brain regions as defined by the rat brain atlas; the 18F-FDG uptake was averaged across voxels in the left and right brain sides for each brain region of each rat, normalized by the mean uptake of the whole brain (Supplementary Table S1, available at http://links.lww.com/PAIN/B142).53

2.3. Network construction and analysis

Weighted, undirected intersubject brain networks were constructed separately for each group at each imaging time point; nodes denote brain regions and edges represent the functional relationship between brain regions.53 Because networks constructed using an intersubject approach have been shown previously to exhibit stable local network properties for group sizes of 10 or more,53 the groups in this study are expected to be large enough to minimize data variability. The interregional Spearman correlation was calculated for the 18F-FDG uptake in each pair of brain regions, resulting in a (50 × 50) correlation matrix with correlation coefficients ranging from −1 to 1. Node strength, clustering coefficient, and community structure were calculated using the Brain Connectivity Toolbox.45 Minimum spanning trees (MSTs) were calculated to assess each network in its most simplified form and are insensitive to connection strength. Leaf fraction, which is the fraction of nodes within the MST with a degree of one and is inversely related to clustering coefficient, was calculated for each MST. Due to the nondeterministic nature of community detection calculations, the consensus community structure was identified from 100 partitions of the network using iterative thresholding.7 To verify clustering outcomes, MSTs were identified, which are insensitive to alterations in connection strength, and leaf fraction was calculated.55 Networks were visualized using the Brain Connectivity Toolbox and Octave Networks Toolbox (MIT; http://dx.doi.org/10.5281/zenodo.22398). Network constructions, analyses, and visualizations were performed in Matlab R2018b (Mathworks, Natick, MA).

2.4. Structural equation modeling

Fifteen known and anatomically correct circuits were fit to 18F-FDG uptake data using SEM; hypothesized circuits were composed of brain regions associated with sensory function only (2 circuits), affective function only (5 circuits), or a combination of sensory and affective brain regions (8 circuits) (Supplementary Fig. S1, available at http://links.lww.com/PAIN/B142).10,14,52 Models were fit to the 18F-FDG uptake data using the lavaan package in R version 3.5.1 (R Foundation for Statistical Computing)44; the fit was considered strong if the comparative fit index (CFI) was greater than 0.9, the root mean square error of approximation (RMSEA) was less than 0.08, and the standardized root mean square of the residual (SRMR) was less than 0.08.25

2.5. Data analyses and statistics

All data are expressed as mean ± SD. The mean and SD of head withdrawal thresholds and RGS scores were calculated across all rats in each group for each time point. The SD, σ, was calculated for correlation distributions by fitting a normal curve to the distribution in Matlab. The mean and SD of node strength and clustering coefficient were calculated across all 50 regions in the brain network for each group at each time point. Clustering coefficient, node strength, and RGS scores were separately compared between days and groups using two-way repeated-measures analyses of variance with group and day as factors and a post hoc Tukey test in R version 3.5.1 (R Foundation for Statistical Computing). For mechanical reflex testing, head withdrawal thresholds were averaged across each group, log-transformed before statistical testing to normalize the distribution,43 and compared using a repeated-measures analysis of variance with a post hoc Tukey test. A t test was used to compare if the injected 18F-FDG tracer activity was different between groups. The threshold for significance was set at P < 0.05.

3. Results

3.1. Brain network connectivity distribution is altered in persistent, but not transient, pain

Correlation networks were constructed from anatomically partitioned, resting-state 18F-FDG PET images to evaluate the brain as a network of functionally associated regions53 in rats with persistent sensitivity and spontaneous pain detected by the RGS and in a separate group with only transient sensitivity (n = 10 rats/group) (Figs. 1 and 2A). On days 1, 3, 5, 7, and 9, head withdrawal thresholds do not differ between the transient and persistent pain groups, and are higher in the transient pain group on days 11, 13, and 14 (P < 0.0001) than for the persistent pain group (Fig. 1A). RGS scores increase (indicating more pain) only in the persistent pain group at day 7 (0.85 ± 0.17; P = 0.009) but are not different from baseline (0.30 ± 0.08) in the transient pain group (0.44 ± 0.12) at that time point (Fig. 1B). RGS scores do not differ from baseline (0.37 ± 0.27; P = 0.91) in the persistent pain group at day 13 (0.50 ± 0.21) (Fig. 1B). No difference (P = 0.58) in the injected radiotracer activity was detected between the groups with persistent (0.81 ± 0.22 mCi) and transient (0.77 ± 0.21 mCi) sensitivity.

Figure 1.
Figure 1.:
Differential pain responses are observed in groups exhibiting transient pain and persistent pain. (A) Head withdrawal threshold does not differ between the transient and persistent pain groups on days 1, 3, 5, 7, and 9, and is higher (indicating lower TMJ sensitivity) in the transient pain group than in the persistent pain group on days 11, 13, and 14 (F1,8 = 31.86, ^P < 0.0001, two-way repeated-measures ANOVA). (B) RGS scores increase at day 7 (F1,2 = 3.65, *P = 0.009, two-way repeated-measures ANOVA) from baseline levels only in the persistent pain group, but are not different from baseline in the transient pain group at day 7 (n = 6 rats/group). At day 13, RGS scores are not different from baseline levels in the persistent pain group and drop below the previously reported analgesic intervention score for the RGS (dashed line).38 ANOVA, analysis of variance.
Figure 2.
Figure 2.:
Brain network correlation strength distributions shift from baseline control responses in the persistent pain state. (A) Brain network heat maps for transient and persistent pain states exhibit correlations in metabolic activity between each brain region. Heat maps are constructed from 18F-FDG PET brain images acquired at baseline (before pain), day 7 (sensitivity present in both groups), and day 14 (sensitivity remains only for the persistent group). (B) Based on estimation of a normal distribution (gray lines), the SD of the distribution is consistent across time points for transient pain but increases on days 7 and 14 for persistent pain. (C) At day 14, the strong positive connections (R > 0.5) increase in the network for persistent pain compared to that for transient pain, with strongest correlations within the limbic system. PET, positron emission tomography.

To assess the stability of network connection strengths in relationship to pain state, the distribution of correlation strengths was evaluated at baseline, 7 days after pain onset in both groups, and at a delayed time (day 14) when the pain is differentiated and only detected in the persistent pain group. Normal distributions fit to each network's correlations reveal a consistent SD across time points in the transient pain group (σ = 0.45-0.46), whereas it increases on days 7 (σ = 0.49) and 14 (σ = 0.52) for persistent pain (Fig. 2B). The number of strongly positive correlations (R > 0.5) increases in persistent pain (680 correlations) at day 14 over that observed in transient pain at that same time point (434 correlations) and over baseline levels (518 correlations). Compared to transient pain, persistent pain demonstrates differentially strong correlations within the limbic system, with strong interconnections between the nucleus accumbens, amygdala, hypothalamus, and entorhinal cortex (Fig. 2C). However, strong associations are not identified between the hippocampus and other components of the limbic system, a structure traditionally associated with persistent pain.1,58

3.2. Network metrics reveal differential adaptations of network properties between transient and persistent pain

Although shifts in correlation distributions as early as day 7 (Fig. 2) suggest that functional relationships between brain regions may adapt in persistent pain, they do not evaluate regional properties in the context of the entire brain network. So, we also measured the clustering coefficient, a topological measure of segregated information transfer, and node strength, a measure of a brain region's connectivity to other regions of the network, to assess if there are adaptations in mesoscale and local network structure.46 Amongst positive correlations, the clustering coefficient increases from baseline (0.41 ± 0.06) only with persistent pain at days 7 (0.47 ± 0.09; P = 0.0004) and 14 (0.47 ± 0.11; P = 0.004) and is higher (P < 0.00001) in persistent pain than transient pain (0.37 ± 0.04) at day 14 (Fig. 3A). For transient pain, clustering decreases (P = 0.004) at day 14 (0.37 ± 0.04) from baseline levels (0.42 ± 0.08). Clustering is markedly weaker amongst negative correlations; the clustering coefficient decreases (P = 0.03) amongst negative correlations only in the persistent pain group on day 14 (0.04 ± 0.04) compared to baseline (0.07 ± 0.03) and does not differ between pain groups at any time (Fig. 3B).

Figure 3.
Figure 3.:
Network diagnostics differ amongst positive correlations in brain networks of transient and persistent pain at day 14. (A) Clustering coefficient increases from baseline only in the persistent pain group at days 7 (F1,2 = 15.62, *P = 0.0004) and 14 (**P = 0.004). Clustering is also higher at day 14 in persistent (+P < 0.00001) compared to transient pain, for which clustering decreases from baseline levels (***P = 0.004). (B) Clustering decreases amongst negative correlations only in the persistent pain group on day 14 compared to baseline (F1,2 = 2.29, *P = 0.03) but does not differ between transient and persistent pain groups. (C) Node strength increases only in the persistent pain group at days 7 (F1,2 = 16.29, *P = 0.01) and 14 (**P = 0.03), whereas it decreases in the transient pain group at day 14 (***P < 0.0001) compared to baseline. Node strength is higher in the transient pain group over the persistent pain group at baseline (+P = 0.02) and is stronger in the persistent pain group at day 14 (++P < 0.0001). (D) Node strength amongst negative correlations increases on day 14 relative to baseline in both transient (F1,2 = 0.302, *P = 0.008) and persistent (**P = 0.0001) pain groups, but does not differ between groups at any time point. (E) Brain regions with the strongest (≥90th percentile) and weakest (≤10th percentile) clustering are highlighted by anatomical representations at days 7 and 14 for the transient (blue) and persistent (red) pain groups (insets). Group sizes (N = 50 brain regions) are the same at each time point and differences between groups are tested using two-way repeated-measures ANOVAs. ANOVA, analysis of variance.

Node strength among positive correlations follows a similar pattern to clustering and increases from baseline (11 ± 4) only in the persistent pain group at days 7 (13 ± 5; P = 0.01) and 14 (13 ± 6; P = 0.03), whereas decreasing (P < 0.0001) from baseline (13 ± 6) in the transient pain group at day 14 (10 ± 3) (Fig. 3C). Although node strength is higher in the transient pain group over the persistent pain group at baseline (P = 0.02), by day 14, node strength is stronger (P < 0.0001) in persistent pain compared to transient pain. Node strength amongst negative correlations does not differ between groups at any time, but increases on day 14 in both transient (8 ± 3; P = 0.008) and persistent (9 ± 5; P = 0.0001) groups relative to their corresponding baseline configurations (Fig. 3D).

Evaluating the local clustering coefficient for individual brain regions at day 7 shows strong clustering (≥90th percentile) amongst cortical regions in both transient and persistent pain (Fig. 3E). The somatosensory, parietal, retrosplenial, and visual cortices are among the most highly clustered brain regions for both groups. The transient and persistent pain groups share minimal overlap in weakly clustered regions at day 7; in transient pain, multiple thalamic subregions and the caudate putamen exhibit a low clustering coefficient, whereas the cerebellum, caudate/putamen, and entorhinal cortex are weakly clustered with persistent pain at day 7. At day 14, clustering is strongest in the limbic system (amygdala, nucleus accumbens, hypothalamus, and ventral pallidum) and weakest in the hippocampus, association cortex, and medial geniculate for the persistent pain group. In comparison, the transient pain group has weaker clustering amongst limbic structures, with a high clustering coefficient only at the ventral pallidum and shell of the nucleus accumbens (Fig. 3E). Clustering outcomes are verified using MSTs, which are simplified network representations that are insensitive to changes in edge weight distribution; MSTs were evaluated using the leaf fraction metric, which is inversely proportional to clustering coefficient.55 Changes in leaf fraction between groups and days are consistent with clustering coefficient findings (Supplementary Fig. S2, available at http://links.lww.com/PAIN/B142).

3.3. Brain networks exhibit greater functional segregation only in persistent pain

Because evaluation of the clustering coefficient identified strong local clustering in select cortical regions at day 7 and in limbic regions at day 14 only for persistent pain (Fig. 3E), network community structure was assessed to evaluate if the highly clustering regions are segregated into functionally associated modules. Brain networks were iteratively partitioned into communities using the Louvain community detection algorithm (resolution parameter of γ = 1.0) and a consensus structure was identified for each group at each time point. At both day 7 and day 14, the transient pain group partitions into only 3 communities compared to 4 in the persistent pain group, which exhibits greater functional segregation (Fig. 4). Despite differences in the number of communities identified in transient and persistent pain, brain regions associated with pain processing overlap in community affiliation between groups. At both days 7 and 14, the cortical regions with high local clustering, including the somatosensory, parietal, retrosplenial, and visual cortices, are partitioned into the same community for both groups (Fig. 4). At day 14, the amygdala, hypothalamus, entorhinal cortex, and ventral tegmental area regions of the limbic system are partitioned into the same community in both transient and persistent pain (Fig. 4). However, the shell and core of the nucleus accumbens are partitioned into the limbic system module only in persistent pain (Fig. 4). The hippocampus is not associated with the limbic system module for either pain state at day 14; instead, the hippocampus is associated with posterior structures (periaqueductal gray [PAG] and mesencephalic region) and the prefrontal cortex in transient pain and the cortex and thalamus in persistent pain. The medulla and pons are completely isolated from the larger brain network in the persistent pain state (Fig. 4). Although brain circuits seem more segregated in persistent pain, community associations within key subsystems (limbic and cortical) are not substantially reorganized.

Figure 4.
Figure 4.:
Brain networks are functionally segregated in persistent pain. (Left) Brain networks are organized into their optimized community structure and color coded by functional community. On both days 7 and 14, a three-community structure is observed in the transient pain group and a more segregated four-community structure is identified in the persistent pain group. (Right) Corresponding anatomical plots show the communities with color designating community affiliation.

3.4. Pathway modeling identifies a descending affective circuit that is activated only when pain is present

Although increased brain network clustering and segregation suggest that functional adaptations occur early in the development of pain and may predict its persistence, network analysis does not independently test contributions of specific brain circuits. Testing multiple anatomically correct pathways composed of sensory, affective, and combined sensory/affective brain regions by SEM highlights 2 related pathways that are activated in the persistent pain state (Fig. 5). The descending pathway from the cingulate cortex to the prefrontal cortex and ending at the PAG exhibits a strong fit (CFI > 0.9; RMSEA < 0.08; SRMR < 0.08) to 18F-FDG uptake data only in cases when pain is present. Strong fits were identified for the persistent pain group at days 7 and 14 and also the transient pain group at day 7 (Fig. 5). However, this pathway is not activated when pain is absent, such as at baseline (control) or at day 14 in the transient pain group (Fig. 5). The same descending pathway with an additional bidirectional connection between the PAG and the amygdala exhibits a strong fit to 18F-FDG uptake data only in the persistent pain group on day 7 (Fig. 5). Notably, that strong bidirectional connection occurs only in the instance when spontaneous pain is detected (Fig. 1B) and is not detected in the groups without facial grimace. Although the model fit remains strong at day 14 for the persistent pain group (CFI = 0.94), there is substantial fitting error (RMSEA = 0.18). As such, the descending circuit is not classified as strongly active at this time point.

Figure 5.
Figure 5.:
Structural equation modeling detects the activation of 2 related descending pathways when pain is present. (A) The descending pathway between the cingulate cortex, prefrontal cortex, and PAG exhibits a strong fit (CFI > 0.9; RMSEA < 0.08; SRMR < 0.08) to 18F-FDG uptake data in these regions when pain is present (persistent pain group at days 7 and 14; transient pain group at day 7), but not when pain is absent (baseline, transient pain group at day 14). (B) The same pathway with an additional bidirectional connection between the PAG and the amygdala exhibits a strong fit to 18F-FDG uptake data only in the persistent pain group on day 7. Model fitting that does not converge is denoted by d.n.c.

3.5. Pretreatment that attenuates pain modulates network adaptations and prevents activation of pain-associated circuitry

To test if the changes identified by network and SEM analyses are associated with the development of pain, rats were pretreated with intra-articular etanercept (anti-TNF) before the procedures that induce persistent pain to prevent TMJ sensitivity (Fig. 6A). In the group undergoing etanercept pretreatment before loading to induce persistent pain, head withdrawal thresholds are higher during the loading period on days 1, 3, and 5 (P < 0.0001), and after loading on day 7 (P < 0.0001) compared to the untreated persistent pain group (Fig. 6A). The spontaneous pain, evaluated by the measure of pain-associated facial expressions (RGS), is lower (P = 0.004) in the pretreated group at day 7 compared to both untreated rats and is not different from baseline levels (P = 0.052). Notably, the RGS for the persistent pain group, but not the pretreated group, surpasses the previously reported analgesic intervention score and is significantly higher than baseline (P < 0.0001) (Fig. 6B).38

Figure 6.
Figure 6.:
Pretreatment with intra-articular etanercept attenuates the development of both pain and early brain network alterations but does not affect community structure. (A) Head withdrawal thresholds are higher in the group undergoing pretreatment compared to the group with persistent pain (F1,4 = 13.72, #P < 0.0001, two-way repeated-measures ANOVA) on days 1, 3, 5, and 7 (n = 10 rats/group). (B) RGS scores are higher (+P < 0.0001, two-way repeated-measures ANOVA) on day 7 than baseline only in the persistent pain group (N = 8) and are lower (++P = 0.004) in the pretreated (N = 7) group at day 7 compared to the persistent pain group. The mean score for the pretreated group is below the previously reported analgesic intervention score (dashed line) for the RGS.38 (C) At day 7, more strong positive (red) and negative (blue) connections are present in the persistent pain brain network compared to the treatment-attenuated pain network. (D) The clustering coefficient amongst positive correlations only increases at day 7 from baseline for the untreated group with persistent pain (F1,1 = 14.62, *P < 0.0001, by two-way repeated-measures ANOVA). At day 7, clustering is lower (**P < 0.0001) in the etanercept-treated pain-free group compared to the persistent pain group (N = 50 brain regions per group per time point.) (E) Node strength amongst positive correlations similarly increases (F1,1 = 4.97, ^P = 0.008, by two-way repeated-measures ANOVA) in the untreated, persistently painful group at day 7 compared to baseline. Node strength is lower (^^P < 0.0001) in the etanercept-treated group compared to persistent pain group (N = 50 brain regions per group per time point.) (F) Overlapping 4-community structure is identified in both the persistent and attenuated pain states at day 7. (G) Descending circuitry between the cingulate cortex, prefrontal cortex, PAG, and amygdala does not fit to 18F-FDG PET data for the pretreated group. ANOVA, analysis of variance; PAG, periaqueductal gray; PET, positron emission tomography.

Construction of brain networks for the pretreated group identifies similar trends as the transient pain group, particularly stability in the SD of correlation distributions between baseline (σ = 0.44) and day 7 (σ = 0.42). A greater number of very strong positive and negative connections (R > 0.75 or R < −0.75) is present in the persistent pain network (374 correlations) compared to the pain-free pretreated network (194 correlations) (Fig. 6C), with a concentration of strong positive correlations within the limbic system in persistent pain and far less dense connectivity within that system in the pretreated group (Fig. 6C). Evaluating brain network metrics reveals that the clustering coefficient amongst positive correlations increases (P < 0.0001) from baseline (0.41 ± 0.06) only in persistent pain (0.47 ± 0.09). Similar to the transient pain group (Fig. 3), the clustering coefficient is lower (P < 0.0001) in pain-free pretreated rats at day 7 (0.36 ± 0.07) compared to those with persistent pain (Fig. 6D). Clustering coefficient in pretreated rats is also lower (P < 0.0001) than that detected in the transient pain group. Node strength amongst positive correlations follows a similar trend to the clustering coefficient; node strength in the pretreated group (10 ± 4) is lower than the persistent pain group at day 7 (13 ± 5; P < 0.0001) (Fig. 6E). Node strength only increases from baseline (11 ± 4; P = 0.008) in the persistent pain group (Fig. 6E). Brain networks in pretreated rats also have lower node strength (P = 0.03) compared to the transient pain group. Four-community structure is identified in both the persistent and attenuated pain states at day 7 and similar community associations are identified, especially between portions of the hippocampus and amygdala as well as within the cortex (Fig. 6F). Pathway modeling reveals that neither descending affective circuit, which fit strongly to 18F-FDG uptake in persistent pain at day 7 (Fig. 5), fits data for the pretreated group at day 7 (CFI < 0.9 for both models) (Fig. 6G).

4. Discussion

Although reorganization of functional brain networks is critical for healthy brain development, learning, and memory formation,8,9,13 it can also be pathologic and implicated in disease.11,12,18 Maladaptive alterations in brain networks and subcircuits are increasingly recognized as contributing to chronic pain disorders.42 In this study, we provide evidence that differential brain network features can be identified early after the onset of pain that persists, but not transient pain that resolves. In particular, increased network clustering and engagement of descending affective pathways inclusive of the amygdala are found early only in rats with pain that persists and are not present in rats with acute transient pain. Network clustering and node strength are lowest in those that are pain-free after etanercept treatment. At the time when pain has resolved or remains, differences in affective circuits are amplified, with enhanced clustering within the limbic system and maintained activation of a key descending pain pathway only evident for persistent pain. With evidence from several outcomes across groups, results support using network organization as a prognostic biomarker of the eventual development of persistent pain. Because acute pain medications, such as opioids, are less effective for chronic pain patients due to the development of tolerance and dependence,50 network outcomes could be used to guide therapeutic development and predict which patient groups will benefit from therapy.16 Network data from PET imaging, magnetic resonance imaging, and electroencephalography could be particularly useful as selection tools for enriched clinical trials or for screening before invasive procedures.

4.1. Early adaptations in brain networks and pathways discern persistent and transient pain

Altered network and pathway features indicate that cortical segregation and activated affective circuits are identifiable as early as 7 days after persistent pain onset, which is earlier than the functional and structural changes identified by clinical imaging within months of pain onset6,21,31 and similar to changes at 1 to 4 weeks after the onset of neuropathic pain in the rat.3,24 The increased clustering and divided community structure that are observed on day 7 only in the case when pain persists but at a time when both models exhibit pain suggests that information transfer across the brain network is more segregated in persistent pain; this feature has also been identified with spinal nerve injury and osteoarthritis in rodents.20,24 Although there is consensus that brain network properties are altered in chronic pain,3,18,24 there is controversy around which specific features adapt. Other studies of neuropathic and TMJ pain have not found changes in clustering, but report changes in connectivity strength within brain subcircuits or hemispheres.3,28 Network differences may depend on the length of time since pain onset or the type of chronic pain, both of which may affect the extent of structural and functional brain adaptations.3,18 Variability in findings could also be a function of the methods applied for brain network construction.3,57 Standardized network analyses could possibly address these discrepancies, so too could interrogating brain responses at multiple scales: regional, circuit paths, and global.

The pathway connecting the prefrontal cortex, cingulate cortex, and PAG is acutely activated in both transient and persistent pain and similarly stimulated in acute and chronic pain in humans.10,14,40 Its conservation across species and pain states14,52 reinforces its role in supraspinal pain processing, but suggests that it is not an early indicator of persistent pain. Yet, the amygdala is activated along this pathway at day 7 only in rats that develop persistent pain, suggesting that early engagement of the prefrontal-limbic pathway may be specifically related to pain maintenance. Corticolimbic circuits are hypothesized as critical for the transition to chronic pain56; altered white matter and gray matter are observed along corticolimbic pathways in TMJ disorder35,58 and increased functional connectivity is detected between frontal and limbic regions with increasing orofacial pain.23 The prefrontal-limbic pathways are also activated specifically in patients who report spontaneous pain from chronic knee OA, postherpetic neuralgia, and low back pain.4,19,40 Because spontaneous pain was detected at day 7, persistent TMJ sensitivity may be associated with early activation of this pathway.

4.2. Limbic system clustering is amplified in pain that persists

The altered network distributions with persistent pain align with brain networks being altered in patients with chronic lower back pain, chronic fatigue syndrome, and osteoarthritis,6 and in mice with osteoarthritis.20 Highly correlated activity is strongest in portions of the limbic system; activation of this system has been proposed as a predictor and determinant of chronic pain conditions.56 However, the hippocampus, a key component of the limbic system, does not have strong positive correlations with other brain regions within the network. Although it has been suggested to play a central role in emotional learning in pain,56 studies posit that it becomes disconnected from the larger brain network in neuropathy and back pain, thereby reducing the information flow into hippocampal circuits responsible for memory formation and consolidation.15,29 The lack of strong connectivity between the hippocampus and the rest of the limbic system supports hippocampal disconnection in persistent pain.

As suggested by highly correlated activity in the limbic regions, clustering is strongest in the limbic system with persistent pain, particularly in key anatomical brain regions associated with persistent pain: the amygdala, nucleus accumbens, ventral pallidum, and hypothalamus.5,20,24 All subregions of the hippocampus have low clustering, suggesting that they are poorly integrated with other brain regions and further supporting that this structure is disconnected from the larger brain network.15 Not only are limbic regions the most highly clustered nodes in the brain network with pain, but they also cluster together into the same community, indicating the limbic system is strongly interconnected but weakly integrated with other brain subsystems.34 Local encapsulation of neuronal activity is normal and advantageous for specialized functions, such as detecting visual motion, but can prevent efficient data transfer required for consciously effortful tasks, such as working memory.34 With persistent pain, the brain may direct more incoming nociceptive signals to the limbic system in an attempt to normalize communication within CNS structures, such as the cortex, where pain is interpreted.20

4.3. Altered brain communication predicts pain persistence

Brain network approaches have been proposed to predict longitudinal symptom changes in patients with chronic pelvic pain30 and behavioral responses to sensory and cognitive events.27 However, it was unknown that brain network topology and pathway activation are capable of distinguishing sustained from transient pain early after pain onset. Although brain networks can be influenced by anesthesia39 and stress,32 the inclusion of a pain-free pretreatment group confirms that the observed brain network adaptations are due to the development of pain. Prognostic imaging biomarkers of pain persistence are potentially valuable both in clinical settings and as translational research tools. Brain communication could be assessed to predict the likelihood of recovery from intractable pain symptoms and/or forecast responsiveness to therapies, similar to approaches for psychiatric conditions.16,33 Those approaches propose using a combination of brain dynamics measurements and digital phenotyping data from smartphones to develop novel signatures of psychopathology for use in diagnosis, prognosis, and treatment selection.16,33

Although this study provides insights into brain network adaptations, this study does not investigate if brain network changes are reversible. Etanercept treatment at a later time point would help determine if brain adaptations can be reversed after the onset of pain as well as establish if network changes are maintained by input from peripheral nerves. In addition, a static, resting-state brain scan was used to construct networks. Measuring brain activity over time would allow assessment of dynamic features of the brain network, such as changes in community affiliation of regions over time, known as brain network “flexibility.”12 In the static resting state, the persistent pain group seems not to substantially reorganize compared to the transient pain group. However, brain regions may exhibit changes in connectivity strength and community affiliation over time based on the subject's task,13 which are not captured in the static brain network. For example, patients with schizophrenia exhibit very high network flexibility during working memory, with unstable network modules and widely distributed associations that are not regionally circumscribed.12 Although dynamic brain organization has not been assessed in chronic pain states, fluctuations in the experienced intensity and quality of chronic pain27 may be explained by temporal variations in regional connectivity and community structure.

Despite these considerations for future work, our findings provide the first evidence that brain networks differ between transient and persistent pain states early after pain onset. The presence of activated prefrontal-limbic circuits early after pain develops and strong clustering within the limbic system at a delayed time point differentiate persistent and transient pain. Although this study focused on the development of TMJ pain, it is expected that this experimental approach could be extended to other pain disorders and diseases of the CNS to investigate critical shifts in disease states that promote chronic symptoms.11,12,18

Conflict of interest statement

The authors have no conflicts of interest to declare.

Appendix A. Supplemental digital content

Supplemental digital content associated with this article can be found online at available at http://links.lww.com/PAIN/B142.

Supplemental video content

A video abstract associated with this article can be found at http://links.lww.com/PAIN/B130.

Acknowledgements

This project was funded by the Catherine D. Sharpe Foundation, Oral and Maxillofacial Surgery Foundation, and Oral and Maxillofacial Surgery Schoenleber Research Fund. The authors thank Eric Blankemeyer from the Penn Small Animal Imaging Facility for his technical assistance with PET and CT imaging, Sonia Kartha and Dr Ya-Hsin Yu for help with imaging, Rachel Welch for scoring of the rat grimace scale, and Dr Gordon Barr for helpful discussions. E.J. Granquist is a consultant for Zimmer Biomet but has no conflict.

References

[1]. Apkarian AV, Mutso AA, Centeno MV, Kan L, Wu M, Levinstein M, Banisadr G, Gobeske KT, Miller RJ, Radulovic J, Hen R, Kessler JA. Role of adult hippocampal neurogenesis in persistent pain. PAIN 2016;157:418–28.
[2]. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. A reproducible evaluation of ANTs similarity metric performance in brain image registration. Neuroimage 2011;54:2033–44.
[3]. Baliki M, Chang PC, Baria AT, Centeno MV, Apkarian AV. Resting-state functional reorganization of the rat limbic system following neuropathic injury. Sci Rep 2014;4:6186.
[4]. Baliki M, Chialvo DR, Geha PY, Levy RM, Harden RN, Parrish TB, Apkarian AV. Chronic pain and the emotional brain: specific brain activity associated with spontaneous fluctuations of intensity of chronic back pain. J Neurosci 2006;26:12165–73.
[5]. Baliki MN, Petre B, Torbey S, Herrmann KM, Huang L, Schnitzer TJ, Fields HL, Apkarian AV. Corticostriatal functional connectivity predicts transition to chronic back pain. Nat Neurosci 2012;15:1117–19.
[6]. Baliki MN, Schnitzer TJ, Bauer WR, Apkarian AV. Brain morphological signatures for chronic pain. PLoS One 2011;6:e26010.
[7]. Bassett DS, Porter MA, Wymbs NF, Grafton ST, Carlson JM, Mucha PJ. Robust detection of dynamic community structure in networks. Chaos 2013;23:013142.
[8]. Bassett DS, Wymbs NF, Porter MA, Mucha PJ, Carlson JM, Grafton ST. Dynamic reconfiguration of human brain networks during learning. Proc Natl Acad Sci 2010;108:7641–6.
[9]. Betzel RF, Byrge L, He Y, Goñi J, Zuo XN, Sporns O. Changes in structural and functional connectivity among resting-state networks across the human lifespan. Neuroimage 2014;102:345–57.
[10]. Bliss TVP, Collingridge GL, Kaang BK, Zhuo M. Synaptic plasticity in the anterior cingulate cortex in acute and chronic pain. Nat Rev Neurosci 2016;17:485–96.
[11]. Bonnelle V, Ham TE, Leech R, Kinnunen KM, Mehta MA, Greenwood RJ, Sharp DJ. Salience network integrity predicts default mode network function after traumatic brain injury. Proc Natl Acad Sci 2012;109:4690–5.
[12]. Braun U, Schäfer A, Bassett DS, Rausch F, Schweiger JI, Bilek E, Erk S, Romanczuk-Seiferth N, Grimm O, Geiger LS, Haddad L, Otto K, Mohnke S, Heinz A, Zink M, Walter H, Schwarz E, Meyer-Lindenberg A, Tost H. Dynamic brain network reconfiguration as a potential schizophrenia genetic risk mechanism modulated by NMDA receptor function. Proc Natl Acad Sci 2016;113:12568–73.
[13]. Braun U, Schäfer A, Walter H, Erk S, Romanczuk-Seiferth N, Haddad L, Schweiger JI, Grimm O, Heinz A, Tost H, Meyer-Lindenberg A, Bassett DS. Dynamic reconfiguration of frontal brain networks during executive cognition in humans. Proc Natl Acad Sci U S A 2015;112:11678–83.
[14]. Bushnell MC, Čeko M, Low LA. Cognitive and emotional control of pain and its disruption in chronic pain. Nat Rev Neurosci 2013;14:502–11.
[15]. Cardoso-Cruz H, Lima D, Galhardo V. Impaired spatial memory performance in a rat model of neuropathic pain is associated with reduced hippocampus-prefrontal cortex connectivity. J Neurosci 2013;33:2465–80.
[16]. Cornblath EJ, Lydon-Staley DM, Bassett DS. Harnessing networks and machine learning in neuropsychiatric care. Curr Opin Neurobiol 2019;55:32–9.
[17]. Crosby ND, Weisshaar CL, Smith JR, Zeeman ME, Goodman-Keiser MD, Winkelstein BA. Burst and tonic spinal cord stimulation differentially activate gabaergic mechanisms to attenuate pain in a rat model of cervical radiculopathy. IEEE Trans Biomed Eng 2015;62:1604–13.
[18]. Farmer MA, Baliki MN, Apkarian AV. A dynamic network perspective of chronic pain. Neurosci Lett 2012;520:197–203.
[19]. Geha PY, Baliki MN, Chialvo DR, Harden RN, Paice JA, Apkarian AV. Brain activity for spontaneous pain of postherpetic neuralgia and its modulation by lidocaine patch therapy. PAIN 2007;128:88–100.
[20]. Hess A, Axmann R, Rech J, Finzel S, Heindl C, Kreitz S, Sergeeva M, Saake M, Garcia M, Kollias G, Straub RH, Sporns O, Doerfler A, Brune K, Schett G. Blockade of TNF-α rapidly inhibits pain responses in the central nervous system. PNAS 2011;108:3731–6.
[21]. Ichesco E, Quintero A, Clauw DJ, Peltier S, Sundgren PM, Gerstner GE, Schmidt-Wilcke T. Altered functional connectivity between the insula and the cingulate cortex in patients with temporomandibular disorder: a pilot study. Headache 2012;52:441–54.
[22]. Kartha S, Zhou T, Granquist EJ, Winkelstein BA. Development of a rat model of mechanically induced tunable pain and associated temporomandibular joint responses. J Oral Maxillofac Surg 2016;74:54.e1–54.e10.
[23]. Khan SA, Keaser ML, Meiller TF, Seminowicz DA. Altered structure and function in the hippocampus and medial prefrontal cortex in patients with burning mouth syndrome. PAIN 2014;155:1472–80.
[24]. Kim C-E, Kim YK, Chung G, Jeong JM, Lee DS, Kim J, Kim SJ. Large-scale plastic changes of the brain network in an animal model of neuropathic pain. Neuroimage 2014;98:203–15.
[25]. Kline R. Principles and practice of structural equation modeling. New York, NY: Guilford Publications, 2015.
[26]. Krishnamoorthy S, Blankemeyer E, Mollet P, Surti S, Van Holen R, Karp JS. Performance evaluation of the MOLECUBES β-CUBE—a high spatial resolution and high sensitivity small animal PET scanner utilizing monolithic LYSO scintillation detectors. Phys Med Biol 2018;63:155013.
[27]. Kucyi A, Davis KD. The neural code for pain: from single-cell electrophysiology to the dynamic pain connectome. Neuroscientist 2017;23:397–414.
[28]. Kucyi A, Moayedi M, Weissman-Fogel I, Goldberg MB, Freeman BV, Tenenbaum HC, Davis KD. Enhanced medial prefrontal-default mode network functional connectivity in chronic pain and its association with pain rumination. J Neurosci 2014;34:3969–75.
[29]. Kuner R, Flor H. Structural plasticity and reorganisation in chronic pain. Nat Rev Neurosci 2016;18:20–30.
[30]. Kutch JJ, Labus JS, Harris RE, Martucci KT, Farmer MA, Fenske S, Fling C, Ichesco E, Peltier S, Petre B, Guo W, Hou X, Stephens AJ, Mullins C, Clauw DJ, Mackey SC, Apkarian AV, Landis JR, Mayer EA. Resting-state functional connectivity predicts longitudinal pain symptom change in urologic chronic pelvic pain syndrome: a MAPP network study. PAIN 2017;158:1069–82.
[31]. Loggia ML, Kim J, Gollub RL, Vangel MG, Kirsch I, Kong J, Wasan AD, Napadow V. Default mode network connectivity encodes clinical pain: an arterial spin labeling study. PAIN 2013;154:24–33.
[32]. Low LA, Bauer LC, Pitcher MH, Bushnell MC. Restraint training for awake functional brain scanning of rodents can cause long-lasting changes in pain and stress responses. PAIN 2016;157:1.
[33]. Lydon-Staley DM, Barnett I, Satterthwaite TD, Bassett DS. Digital phenotyping for psychiatry: accommodating data and theory with network science methodologies. Curr Opin Biomed Eng 2018;9:8–13.
[34]. Meunier D, Lambiotte R, Bullmore ET. Modular and hierarchically modular organization of brain networks. Front Neurosci 2010;4:1–11.
[35]. Moayedi M, Weissman-Fogel I, Salomons TV, Crawley AP, Goldberg MB, Freeman BV, Tenenbaum HC, Davis KD. White matter brain and trigeminal nerve abnormalities in temporomandibular disorder. PAIN 2012;153:1467–77.
[36]. Muldoon SF, Pasqualetti F, Gu S, Cieslak M, Grafton ST, Vettel JM, Bassett DS. Stimulation-based control of dynamic brain networks. PLoS Comput Biol 2016;12:e1005076.
[37]. Nicoll SB, Hee CK, Davis MB, Winkelstein BA. A rat model of temporomandibular joint pain with histopathologic modifications. J Orofac Pain 2010;24:298–304.
[38]. Oliver V, De Rantere D, Ritchie R, Chisholm J, Hecker KG, Pang DSJ. Psychometric assessment of the rat grimace scale and development of an analgesic intervention score. PLoS One 2014;9:e97882.
[39]. Pan WJ, Billings JCW, Grooms JK, Shakil S, Keilholz SD. Considerations for resting state functional MRI and functional connectivity studies in rodents. Front Neurosci 2015;9:1–17.
[40]. Parks EL, Geha PY, Baliki MN, Katz J, Schnitzer TJ, Apkarian AV. Brain activity for chronic knee osteoarthritis: dissociating evoked pain from spontaneous pain. Eur J Pain 2011;15:843.e1–843.e14.
[41]. Philips B, Weisshaar C, Winkelstein B. The rat grimace scale can evaluate spontaneous neuropathic pain in a model of cervical radiculopathy. Comp Med 2016;67:34–42.
[42]. Price TJ, Basbaum AI, Bresnahan J, Chambers JF, de Koninck Y, Edwards RR, Ji RR, Katz J, Kavelaars A, Levine JD, Porter L, Schechter N, Sluka KA, Terman GW, Wager TD, Yaksh TL, Dworkin RH. Transition to chronic pain: opportunities for novel therapeutics. Nat Rev Neurosci 2018;19:1–2.
[43]. Rolke R, Magerl W, Campbell KA, Schalber C, Caspari S, Birklein F, Treede RD. Quantitative sensory testing: a comprehensive protocol for clinical trials. Eur J Pain 2006;10:77–88.
[44]. Rosseel Y. Lavaan: an R package for structural equation modeling. J Stat Softw 2012;48:1–36.
[45]. Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010;52:1059–69.
[46]. Rubinov M, Sporns O. Weight-conserving characterization of complex functional brain networks. Neuroimage 2011;56:2068–79.
[47]. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, Tinevez JY, White DJ, Hartenstein V, Eliceiri K, Tomancak P, Cardona A. Fiji: an open-source platform for biological-image analysis. Nat Methods 2012;9:676–82.
[48]. Schwarz AJ, Danckaert A, Reese T, Gozzi A, Paxinos G, Watson C, Merlo-Pich EV, Bifone A. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: application to pharmacological MRI. Neuroimage 2006;32:538–50.
[49]. Sessle BJ. Peripheral and central mechanisms of orofacial inflammatory pain. Int Rev Neurobiol 2011;97:179–206.
[50]. Skolnick P, Volkow ND, Arnold LM, Russell IJ, Diri EW, Duan WR, Young JP, Sharma U, Martin SA, Barrett JA, Haig G, Cao L, McDonnell A, Nitzsche A, Alexandrou A, Saintot PP, Loucif AJ, Brown AR, Young G, Mis M, Randall A, Al E, Compton WM, Jones CM, Baldwin GT, Derry S, Gill D, Phillips T, Moore RA, Edlund MJ, Martin BC, Russo JE, DeVries A, Braden JB, Sullivan MD, Finnerup NB, Attal N, Haroutounian S, McNicol E, Baron R, Dworkin RH, Gilron I, Haanpää M, Hansson P, Jensen TS, Al E, Miller G, Paul SM, Mytelka DS, Dunwiddie CT, Persinger CC, Munos BH, Lindborg SR, Schacht AL, Scannell JW, Blanckley A, Boldon H, Warrington B, Schnitzer TJ, Ekman EF, Spierings EL, Greenberg HS, Smith MD, Brown MT, West CR, Verburg KM, Volkow ND, McLellan AT. Re-energizing the development of pain therapeutics in light of the opioid epidemic. Neuron 2016;92:294–7.
[51]. Sotocinal SG, Sorge RE, Zaloum A, Tuttle AH, Martin LJ, Wieskopf JS, Mapplebeck JC, Wei P, Zhan S, Zhang S, McDougall JJ, King OD, Mogil JS. The rat grimace scale: a partially automated method for quantifying pain in the laboratory rat via facial expressions. Mol Pain 2011;7:55.
[52]. Sperry MM, Kandel BM, Wehrli S, Bass KN, Das SR, Dhillon PS, Gee JC, Barr GA. Mapping of pain circuitry in early post-natal development using manganese-enhanced MRI in rats. Neuroscience 2017;352:180–9.
[53]. Sperry MM, Kartha S, Granquist EJ, Winkelstein BA. Inter-subject FDG PET brain networks exhibit multi-scale community structure with different normalization techniques. Ann Biomed Eng 2018;46:1001–12.
[54]. Sperry MM, Yu Y-H, Welch RL, Granquist EJ, Winkelstein BA. Grading facial expression is a sensitive means to detect grimace differences in orofacial pain in a rat model. Sci Rep 2018;8:13894.
[55]. Tewarie P, van Dellen E, Hillebrand A, Stam CJ. The minimum spanning tree: an unbiased method for brain network analysis. Neuroimage 2015;104:177–88.
[56]. Vachon-Presseau E, Centeno MV, Ren W, Berger SE, Tetreault P, Ghantous M, Baria A, Farmer M, Baliki MN, Schnitzer TJ, Apkarian AV. The emotional brain as a predictor and amplifier of chronic pain. J Dent Res 2016;95:605–12.
[57]. Vachon-Presseau E, Tétreault P, Petre B, Huang L, Berger SE, Torbey S, Baria AT, Mansour AR, Hashmi JA, Griffith JW, Comasco E, Schnitzer TJ, Baliki MN, Apkarian AV. Corticolimbic anatomical characteristics predetermine risk for chronic pain. Brain 2016;139:1958–70.
[58]. Younger JW, Shen YF, Goddard G, Mackey SC. Chronic myofascial temporomandibular pain is associated with neural abnormalities in the trigeminal and limbic systems. PAIN 2010;149:222–8.
[59]. Zhang S, Kartha S, Lee J, Winkelstein BA. Techniques for multiscale neuronal regulation via therapeutic materials and drug design. ACS Biomater Sci Eng 2017;3:2744–60.
Keywords:

Brain networks; Pain; Temporomandibular joint; Modeling; Brain imaging

Supplemental Digital Content

© 2020 International Association for the Study of Pain