Fig 1.
Psychological and personality factors identify four chronic pain traits.
A. The study design consisted of six visits to the lab, with four of these including brain imaging sessions. Questionnaires were administered at V1 and pain was manipulated with placebo treatment (T1) followed by a washout period (W1), repeated twice (T2, W2). Sixty-two patients completed all six visits (Group 1) and 46 patients completed the questionnaires at V1 but did not finish the rest of the study (Group 2). B. Each covariance matrix (Group 1 and Group 2; ordered based on PCA results for Group 1) showed associations between the questionnaire outcomes used to probe CBP patients’ profiles (r values shown in bar from blue to red). C. A network constructed from the Group 1 covariance matrix displays the strongest correlations (top 10%) and topological properties indicated the presence of four communities. Size of the nodes are scaled based on degree count. D. PCA performed on all 36 questionnaire outcomes (Group 1) identified four orthogonal components. Given the observed loadings and network clustering, we label these components as: Pain-trait; Character-trait; Aware-trait; and Emote-trait. Only components loading >0.4 or <−0.4 are depicted. Abbreviations of the psychological factors: PASS and its four subscales: Avoidance (pass/a), Cognitive (pass/c), Fear (pass/f), and Physiological (pass/p); PCS and its three subscales: Rumination (pcs/r), Magnification (pcs/m), and Helplessness (pcs/h); CPAQ and its two subscores: Activity Engagement (cpaq/a) and Pain Willingness (cpaq/pw); MAIA: Not Distracting (maia/nd), Attention Regulation (maia/a), Emotional Awareness (maia/e), and Self-Regulation (maia/sr), Noticing (maia/n), Not Worrying (maia/nw), Body Listening (maia/bl), and Trusting (maia/t); PSQ and its two subscales: Painful (psq/p) and Non-Painful (psq/np); PANAS with Positive (panas/p) subscale; FFM with its five subscales: Observe (ffm/o), Describe (ffm/d), Act with Awareness (ffm/a), Non-Judge (ffm/nj), and Non-React (ffm/nr); ERQ with its Reappraisal (erq/r) and Suppression (erq/s) subscales; ACS; EACS; LOT-R; and LAQ. For each figure, the numerical data are available S1 Data, and a description of the analyses and P values are reported in S4 Table. ACS, Attentional Control Scale; CBP, chronic back pain; CPAQ, Chronic Pain Acceptance Questionnaire; EACS, Emotional Attentional Control Scale; ERQ, Emotional Regulation Questionnaire; FFM, Five Facets of Mindfulness; LAQ, Loss Aversion Questionnaire; LOT-R, Life Orientation Test, Revised version; MAIA, Multidimensional Assessment of Interoceptive Awareness; PANAS, Positive and Negative Affect Scale; PASS, Pain Anxiety Symptoms Scale; PCA; PCS, Pain Catastrophizing Scale; PSQ, Pain Sensitivity Questionnaire.
Fig 2.
Relating trait dimensions to pain characteristics.
Correlation matrices indicate the extent of the association of each trait with back pain intensity (EMAs, Memory, NRS), quality (MPQs/a, PainDetect), and negative affect (PANAS/n, BDI). A. In group 1, Pain-trait scores were related to almost all pain characteristics queried at V1 and V2, while Emote-trait scores were negatively related to negative affect at V1 and V2. B. Consistent findings were observed in group 2 (data only available at V1). C. Consensus was obtained when the effects were replicated within participant (across V1 and V2 in Group 1) and between cohort of patients (Group 1 and Group 2). The consensus map shows that Pain-trait related to almost all pain assessments and Emote-trait related specifically to negative affect. The consensus across visits and groups was established using uncorrected P values *P < 0.05; **P < 0.01; ***P < 0.001 (significant even after Bonferroni correction for all comparisons [56 comparisons in Group 1 and 28 comparisons in Group 2]). For each figure, the numerical data are available S1 Data, and a description of the analyses and P values are reported in S4 Table. BDI, Beck Depression Index (only collected at V1); EMA, ecological momentary assessment (phone app daily pain ratings); NRS, numerical rating scale; MPQa, McGill pain questionnaire/affective; MPQs, McGill pain questionnaire/sensory; PANASn, Positive and Negative Affect Scale/negative; SF-12p, 12-item short-form survey/physical health.
Fig 3.
Resting state functional connectivity determined Pain-trait and Emote-trait.
A. Resting-state fMRI was used to construct functional connectivity covariance matrices, and a 3-fold cross-validation procedure was implemented to identify links related to trait scores. Scatterplots show the normalized score of each individual and their chronic pain traits when they were in the left-out fold. Pain-trait was determined from both positive and negative links and Emote-trait could be determined only by positive but not negative links. B, D, F. The most stable links were the common links consistently selected by the models across all three folds. C, E, G. The matrices display within- and between-community connections of all links, irrespective of their weights, when the brain networks are represented by 14 communities. The full connectivity profile for each trait is also shown across all 272 nodes. *P < 0.05; **P < 0.01. For each figure, the numerical data are available S1 Data and a description of the analyses, and P values are reported in S4 Table.
Fig 4.
Stability assessment of neurotraits from repeat brain scans over 5 weeks.
A. Strength of relationships between weighted patterns of connectivity that were trained at V2 with corresponding psychology/personality trait scores. B. Relationship between neurotraits (derived from connectivity based on subsequent brain scans) and trait scores determined by questionnaires at V1. The trait–neurotrait relationship is preserved. *P < 0.05; **P < 0.01; ***P < 0.001. For each figure, the numerical data are available S1 Data, and a description of the analyses and P values are reported in S4 Table.
Fig 5.
Association between neurotraits and pain measurements.
A. Data reduction was performed to extract three principal components reflecting pain intensity, pain quality and affect/physical health measured at visit 1 and visit 2. The component matrix displays the correlations between the pain measurements and each component. B. The mean neurotrait represents the averaged sum of the positive and the negative z(r) values across all four visits (n = 56). The covariance matrix shows an association of the neurotraits with pain intensity and affect/health but not with pain quality. C. The covariance matrices show the associations between pain components (visit 1 and visit 2) and the neurotraits extracted at each visit separately (visits 2–5). D. Partial correlations show that Neurotrait 1 and 2 were strongly correlated with affect/negative health when controlling for Neurotrait 3. E. The effect was reliable when extracting the neurotraits from separate visits. F. Neurotrait 3 was also strongly correlated with affect/negative health when controlling for Neurotrait 1 and 2. G. The effect was reliable across each visit. +P = 0.053; *P < 0.05; ** P < 0.01 (significant after Bonferroni corrections for nine comparisons); ***P < 0.001. For each figure, the numerical data are available S1 Data, and a description of the analyses and P values are reported in S4 Table.
Fig 6.
Association between socioeconomic factors and traits and neurotraits.
A. Income differentiated patients on Pain-trait after controlling for ethnicity and years of education (while the inverse was not true), suggesting that socioeconomic status (rather than ethnicity or years of education) contributed to Pain-trait 1. Annual income of >25,000 seemed to represent a socioeconomic threshold distinguishing vulnerability from protection. B. Applying the 25,000 threshold yielded similar results in Group 2. C–D. Self-reported income further differentiated Neurotrait 1 and Neurotrait 2 (marginally significant; P = 0.06). E. Spearman correlations were used to display subscales associated with income (only significant relations are displayed (uncorrected P < 0.05). All abbreviations are defined in Fig 1. Post-hoc comparisons were Bonferroni corrected for six comparisons &P ≤ 0.1; *P < 0.05. For each figure, the numerical data are available S1 Data and a description of the analyses and p-values are reported in S4 Table.