Quantifying cerebral contributions to pain beyond nociception

Abstract

Cerebral processes contribute to pain beyond the level of nociceptive input and mediate psychological and behavioural influences. However, cerebral contributions beyond nociception are not yet well characterized, leading to a predominant focus on nociception when studying pain and developing interventions. Here we use functional magnetic resonance imaging combined with machine learning to develop a multivariate pattern signature-termed the stimulus intensity independent pain signature-1 (SIIPS1)-that predicts pain above and beyond nociceptive input in four training data sets (Studies 1-4, N=137). The SIIPS1 includes patterns of activity in nucleus accumbens, lateral prefrontal and parahippocampal cortices, and other regions. In cross-validated analyses of Studies 1-4 and in two independent test data sets (Studies 5-6, N=46), SIIPS1 responses explain variation in trial-by-trial pain ratings not captured by a previous fMRI-based marker for nociceptive pain. In addition, SIIPS1 responses mediate the pain-modulating effects of three psychological manipulations of expectations and perceived control. The SIIPS1 provides an extensible characterization of cerebral contributions to pain and specific brain targets for interventions.

Figure 1. Identifying cerebral contributions to pain beyond nociception.

(a) The left panel provides an example of pain ratings for different levels of noxious stimuli and the right panel shows that there still remain large variation in pain ratings even after controlling for noxious stimulus intensity and the neurologic pain signature (NPS) response. (b) The main goal of the current study is to develop a multivariate model of endogenous cerebral contributions to pain beyond nociception. Some of the cerebral contributions may interact with nociceptive brain systems (red nodes), whereas others contribute to pain independent of nociceptive processing (green nodes). (c) The multivariate pattern of fMRI activity predictive of residual pain ratings after removing the effects of the stimulus intensity and NPS response, termed the stimulus intensity independent pain signature-1 (SIIPS1). The map shows thresholded voxel weights (at q<0.05 false discovery rate (FDR), equivalent to uncorrected voxel-wise P<0.0025) based on weighted t-tests across maps for 137 subjects in the training data sets (Studies 1–4). Thresholding was performed for display only; all weights were used in the subsequent analyses. Some examples of unthresholded patterns are presented in the insets; small squares indicate individual voxel weight. aINS, anterior insula; CB, cerebellum; dmPFC, dorsomedial PFC; dpINS, dorsal posterior insula; HC, hippocampus; MCC, mid-cingulate cortex; midINS, middle insula; NAc, nucleus accumbens; SMA, supplementary motor area; TP, temporal pole; vmPFC, ventromedial PFC; vlPFC, ventrolateral PFC. (d) Z-scored quartile residual pain ratings versus cross-validated (leave-one-participant-out) prediction (also z-scored and quartile binned) with the SIIPS1. Each coloured line represents a fitted line for each individual. The violin plot in the right panel shows the distribution of the slopes from regression analyses for the prediction–outcome relationship. All participants except for two individuals (98.5%) showed positive prediction–outcome relationships. Each coloured dot represents an individual's slope.

Figure 2. Deconstructing the SIIPS1.

(a) Each dot of the scatter plots represents a contiguous brain region from the SIIPS1 thresholded at q<0.05, FDR corrected (see Fig. 1c). Red dots represent regions with positive predictive weights, and blue dots represent regions with negative predictive weights. The y axis of the scatter plots shows the mean correlations between the local pattern expression (with absolute pattern weights) and trial-by-trial noxious stimulus intensity across 183 participants from Studies 1–6. The x axis of the scatter plots shows the mean pattern weights of contiguous regions. Dashed gray lines indicate one-sample t-test results that are corrected for multiple comparisons using family-wise error rate<0.05 (Bonferroni methods; equivalent to uncorrected P<0.0011). Therefore, dots above the dashed lines indicate regions significantly correlated with noxious input intensity (temperature) and dots below the dashed line indicate regions independent of noxious input intensity. Brain region maps for (b) regions that showed significant non-zero correlations with noxious input intensity and for (c) regions that showed no correlations with noxious input intensity, but still contributed to the prediction of single-trial level pain ratings. Region labels, mean weight values and mean correlation values with noxious stimulus intensity (and their t- and p-values) can be found in Supplementary Table 4. aINS, anterior insula; Caud, caudate; CB, cerebellum; dmPFC, dorsomedial PFC; dlPFC, dorso-lateral PFC; dpINS, dorsal posterior insula; HC, hippocampal area; LG, lingual gyrus; MCC, middle cingulate cortex; midINS, middle insula; MTG, middle temporal gyrus; NAc, nucleus accumbens; PHC, parahippocampal area; Precen, precentral cortex; Precun, precuneus; S2, secondary somatosensory cortex; SMA, supplementary motor area; SMC, sensory motor cortex; STG, superior temporal gyrus; Thal, thalamus; TP, temporal pole; vlPFC, ventrolateral PFC; vmPFC, ventro-medial PFC.

Figure 3. Unthresholded patterns of SIIPS1 predictive weights for some regions-of-interest (ROIs).

The ROIs include the nucleus accumbens (NAc), hippocampus, amygdala, PHG and caudate. The unthresholded pattern map used to make predictions included both positive and negative weights in each region, suggesting more complex, fine-grained mapping between these regions and pain. (a) Serial coronal views of the predictive weights within the NAc, showing positive predictive weights in a shell-like region and negative weights in a core-like region, as identified in a previous fMRI-based parcellation study. Differential roles of the NAc shell (pro-pain) versus core (anti-pain) subdivisions have been shown in animal literature. (b) Serial coronal views of the hippocampus and amygdala ROIs. Positive weights are apparent in the superficial and central subdivisions in the amygdala (as defined by ref. 47), and negative weights is the laterobasal group. A recent meta-analysis found that the superficial sub-region is often reported in experimental pain studies. In the hippocampus, positive weights were found in some areas covering cornu ammonis and dentate gyrus, and also near caudate tail. (c) Serial axial views of the PHG ROI show positive weights in the entorhinal cortex (as defined by ref. 47) and a peri-amydaloid areas, and negative weights in other parahippocampal areas. (d) three-dimensional surface map of the un-thresholded SIIPS1 pattern for the ROIs. The pattern showed differential roles of caudate tail (positive) versus head (largely negative, but mixed), as suggested in animal and metaanalysis studies, which associate caudate tail with stable, learned stimulus value and sensorimotor functions, and caudate head with more flexible, context-dependent stimulus value.

Figure 4. Joint contributions of two brain signatures to pain (predicting single-trial pain ratings).

(a) We estimated the unique and shared contributions to pain of the NPS and the SIIPS1 using multilevel general linear model. The trial-by-trial responses of the NPS and the SIIPS1 were independent variables, and trial-by-trial pain report was the outcome variable. (b) The pie chart and the bar plots show mean explained variance across six pain studies and dots on the bar plots show mean explained variance for each study. ‘NPS+SIIPS1' indicates total variance explained by the NPS and the SIIPS1, and the ‘NPS' and ‘SIIPS1' indicates variance explained by the NPS and SIIPS1 separately. ‘Univariate' indicates variance explained by the univariate map alone (for the univariate map; see Supplementary Fig. 5). Grey lines between ‘SIIPS1' and ‘Univariate' connect the same study, demonstrating that the univariate models consistently explain less variance in pain ratings than the multivariate models in cross-validated analyses. ***P<0.001, one-sample t-test, which treats study as a random effect. (c) Regression coefficients (and s.e.m. in parenthesis) and explained variance by the NPS and the SIIPS1 in training data sets (Studies 1–4). Cross-validated signature response (leave-one-participant-out cross-validation) was used in the analysis. (d) Results from the testing data sets (Studies 5–6). Violin plots show the distribution of standardized β-coefficients for the NPS and SIIPS1 response using kernel density estimation and the red horizontal lines indicate Empirical Bayes weighted mean coefficients. (e) Results from the negative control data set (vicarious pain task with the same subjects of Study 2; see Methods). ***P<0.001; Bootstrap (10,000 iterations) and permutation (5,000 iterations) tests were used for significance testing of regression coefficients and explained variance, respectively. For more details, see Methods.

Figure 5. Mediation of psychological pain modulation (Studies 5–6).

(a) Study 5's experimental design and the behavioral findings: In Study 5, the expectancy level was manipulated by two different cues associated with two levels of heat intensity (high and low). After participants learned the cue-heat intensity association, the low-pain cue was followed by a low (LL trial type) or medium pain (LM) with 50% chance and the high-pain cue was followed by a medium (HM) or high pain (HH) with 50% chance. Violin plots show participants' averaged pain ratings for two medium pain conditions (LM and HM) and grey lines connect the same individuals' pain ratings. (b) The significant expectancy effects on pain were mediated by the SIIPS1, not by the NPS. The path coefficients and s.e.m. (in parenthesis) for the mediation effects (Path a × b) are reported here. (c) Study 6's experimental design and the behavioural findings: in Study 6, we manipulated the levels of perceived control and expectancy with a 2-by-2 design (for more details, see Methods). The bar plots show participants' averaged pain ratings for four different experimental conditions and the right ones show the multilevel general linear model results (beta coefficients). Error bars represent s.e.m. (d) The perceived control effects were mediated only by the SIIPS1 (that is, not by the NPS) and the expectancy effects were mediated by both the SIIPS1 and the NPS. The path coefficients and s.e.m. (in parenthesis) for the mediation effects (Path a × b) are reported here. For more detailed methods and statistics of path coefficients, see Methods and Supplementary Table 5. ⁺P<0.05, one-tailed; *P<0.05, **P<0.01 and ***P<0.001, two-tailed. Significance tests in this figure include paired t-test, multi-level mediation analyses (bootstrap test) and multi-level general linear model (bootstrap test).