Corticolimbic anatomical characteristics predetermine risk for chronic pain

Abstract

SEE TRACEY DOI101093/BRAIN/AWW147 FOR A SCIENTIFIC COMMENTARY ON THIS ARTICLE: Mechanisms of chronic pain remain poorly understood. We tracked brain properties in subacute back pain patients longitudinally for 3 years as they either recovered from or transitioned to chronic pain. Whole-brain comparisons indicated corticolimbic, but not pain-related circuitry, white matter connections predisposed patients to chronic pain. Intra-corticolimbic white matter connectivity analysis identified three segregated communities: dorsal medial prefrontal cortex-amygdala-accumbens, ventral medial prefrontal cortex-amygdala, and orbitofrontal cortex-amygdala-hippocampus. Higher incidence of white matter and functional connections within the dorsal medial prefrontal cortex-amygdala-accumbens circuit, as well as smaller amygdala volume, represented independent risk factors, together accounting for 60% of the variance for pain persistence. Opioid gene polymorphisms and negative mood contributed indirectly through corticolimbic anatomical factors, to risk for chronic pain. Our results imply that persistence of chronic pain is predetermined by corticolimbic neuroanatomical factors.

Keywords: brain network; chronic pain; diffusion tensor imaging (DTI); limbic system; magnetic resonance imaging (MRI).

© The Author (2016). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved. For Permissions, please email: journals.permissions@oup.com.

Figures

See Tracey (doi:10.1093/brain/aww147) for a scientific commentary on this article. Why some individuals progress from acute to chronic pain is unclear. Vachon-Presseau et al. monitor patients with subacute pain for three years, and show that pre-existing structural and functional connectivity within dorsal mPFC-amygdala-accumbens corticolimbic circuitry, as well as amygdala and hippocampal volumes, represent risk factors for transition to chronic pain.

Figure 1

Whole-brain structural connectivity identifies corticolimbic regions conferring risk for chronic pain. (A) Timeline of brain scans and type and number of subjects studied. (B) Recovering (SBPr) and persisting (SBPp) patients were dichotomized based on 20% reduction in back pain intensity at Week 56. The two groups showed time-dependent divergence (Weeks 0–56) in pain intensity [F(2.52, 140.92); P < 0.001 repeated measure ANOVA] and pain disability [F(2.58, 110.74) = 3.44; P = 0.025 repeated measure ANOVA], sustained in the 3-year subsamples [Week 156 pain intensity t(1,33) = 3.67; P = 0.001; pain disability t(1,36) = 3.28; P = 0.002 two-sided unpaired t-tests]. *P < 0.05 between group comparison; +P < 0.05 within group comparison relative to Week 0. (C) Structural white matter-based networks were constructed from probabilistic tractography between nodes covering the whole brain. (D and F) Group comparison at link density of 0.1 at Week 0 was studied from network constructed from 480 and 264 parcellation schemes. In both networks SBPp show more white matter connections between corticolimbic nodes (P < 0.05, FDR corrected). (E) In 480-parcellation, location and number of nodes within pain and corticolimbic systems are illustrated. The SBPp > SBPr contrast indicates denser structural connections in SBPp across corticolimbic nodes, but not for pain-related nodes, for all link densities. Observed incidence of connections compared to expected incidence in a random network was significantly higher for corticolimbic nodes (chi-square test, P < 0.0001 for all link densities) but not for pain-related nodes (P > 0.28 for all link densities). Histogram is the cumulative probability (total connectivity = 1.0) for corticolimbic nodes and pain-related nodes compared to expected connectivity in a random network, averaged for all densities. (G) Concordant results were observed in 264-parcellation. Observed incidence of connections in SBPp was much higher than expected across corticolimbic nodes (chi-square test, P < 0.0001 for all link densities) but not for pain-related nodes (P > 0.28 for all link densities). Number of subjects is shown in parentheses. Error bars indicate SEM.

Figure 2

Intra-corticolimbic white matter connectivity identifies three modules, only the dorsal medial PFC–amygdala–nucleus accumbens module imparts risk for chronic pain. (A) Display of the intra-corticolimbic white matter-based structural networks generated from 1068 nodes (white matter networks) at 0.1 link density. (B) Modularity analysis performed on this network separated the system into three communities: (i) dorsal medial PFC–amygdala–nucleus accumbens (mPFCdorsal-amy-NAc); (ii) ventral medial PFC–amygdala (mPFCventral-Amy); and (iii) orbitofrontal cortex–amygdala–hippocampus (OFC-amy-hipp). (C) Normalized mutual information was used to quantify similarity between individual subject community structure and the mean group community structure presented in (B) SBPp, SBPr and healthy controls displayed similar modular organization across all time points, due to absence of group differences [F(2,69) = 2.34; P = 0.10 repeated measure ANOVA] or a time × group interaction [F(5.73, 200.65) = 0.72; P = 0.63 repeated measure ANOVA]. (D) Structural connection probability for SBPp, SBPr and healthy within the dorsal medial PFC–amygdala–nucleus accumbens module calculated across link densities ranging between 0.02 and 0.2. Incidence of white matter connections was higher in SBPp at all densities [F(2,78) = 6.10; P = 0.003 repeated measure ANOVA]. The histogram displays group differences at 0.1 link density. (E) Per cent structural connections for SBPp and SBPr and healthy controls over 3 years at 0.1 link density. Incidence of connections in the dorsal medial PFC–amygdala–nucleus accumbens module was higher in SBPp [F(2,69) = 4.74; P = 0.012 repeated measure ANOVA] for visits 0–56 weeks. Concordant results were seen at Week 156 [t(1,34) = 1.79; P = 0.08 two-sided unpaired t-tests]. Scatterplot shows 3-year stability of dorsal medial PFC–amygdala–nucleus accumbens module connectivity for all SBP. (F) Incidence of functional connections in the dorsal medial PFC–amygdala–nucleus accumbens module was higher in SBPp across 0.02 to 0.2 link densities [F(1,60) = 4.00; P = 0.05]. The histogram displays group differences at 0.1 link density. (G) SBPp showed higher incidence of functional connections in the dorsal medial PFC–amygdala–nucleus accumbens module [F(1,60) = 8.00; P = 0.006 repeated measure ANOVA], for visits 0–56 weeks at 0.1 link density. This functional connectivity, however, was not different between SBPp and SBPr subsamples at Week 156, and the scatter plot showed no within subject relationship between Weeks 0 and 156 functional connections. *P < 0.05 post hoc comparisons between SBPp and SBPp. &P = 0.08. Number of subjects is shown in parentheses. Error bars indicate standard error of the mean (SEM).

Figure 3

Amygdala and hippocampus volumes are risk factors for chronic pain. (A and B) Subcortical structures were segmented, corresponding volumes summed across hemispheres, and compared between groups as a function of time (Supplementary Tables 4 and 5). Heat maps display overlap of automated segmentation across SBPp, SBPr, and healthy controls at week 0. (C and D) Repeated measure ANCOVAs (group and time effects; covariate for age, gender, and grey matter volume) revealed that SBPp showed smaller amygdala [F(2,63) = 7.94; P < 0.001] and hippocampus [F(2,70) = 6.35; P = 0.003], but no main effect of time (P-values > 0.69) or group by time interaction (P-values > 0.23). Bar graphs illustrate group differences at 3 years. Scatterplots show volume reproducibility over 3 years across all SBP. *P < 0.05 SBPp versus SBPr, and SBPp versus healthy controls. (E and F) Age-matched chronic back pain patients showed significant smaller amygdala [t(1,42) = 2.71; P = 0.009 two-sided unpaired t-tests] and hippocampus [t(1,43) = 1.99; P = 0.05 two-sided unpaired t-tests] than SBPr, matching the volumes observed in SBPp. Similarly, osteoarthritis patients displayed at least marginally smaller amygdala [t(1,52) = 2.21; P = 0.03 two-sided unpaired t-tests] and hippocampus [t(1,52) = 1.95; P = 0.057 two-sided unpaired t-tests] when compared to age matched healthy controls. *P < 0.05, &P < 0.06. (G) Vertex-wise shape analyses of the amygdala and the hippocampus indicated that SBPp had thinner right amygdala across all visits. Number of subjects is shown in parentheses. Error bars indicate SEM.

Figure 4

Corticolimbic characteristics determine risk for chronic pain but symptom severity depends only on initial pain intensity. (A) Path analysis was used to identify the contribution of rs678849 SNP, pain at Week 0, and brain parameters on risk for developing chronic pain (SBPp/r at Week 56). The white matter network (White matter) and functional connectivity network (Functional) connections of dorsal medial PFC–amygdala–nucleus accumbens (mPFCdorsal-amy-NAc) module and bilateral total amygdala volume, at Week 0, were independent predictors of pain persistence or recovery (SBPp/r) at Week 56 (all P-values ≤ 0.01). The relationship between OPRD1 rs678849 and pain persistence/recovery was fully mediated through the bilateral amygdala volume (indirect pathway 95% confidence interval: −0.021 and −0.173; R2 = 0.04 of unique variance). Covariance coverage within and between variables exceeded 88%. Numbers of subjects are shown in parentheses. (B) Scatterplots show that pain intensity at Week 0 significantly correlated with pain intensity and pain disability at Week 156 (P-values ≤ 0.001).