Fig. 6. (A) Histogram of [11-C] carfentanil receptor occupancy (DVR). Color-coded categories were derived from k-means clustering with 10 categories. The highest 6 categories (DVR ~> 1) are shown in color, and only these high-binding voxels were analyzed for placebo effects. The red-white color map shows binding categories corresponding to the top row in B. The sepia shading shows the same categories in a transparent color scheme, and corresponds to the bottom rows of B. (B) (Top) Regional binding levels. (Bottom) Regions of interest (green) overlaid on binding levels (transparent sepia colormap). Amy, amygdala; aINS, anterior insula; aOFC, anterior orbitofrontal cortex; DLPFC, dorsolateral prefrontal cortex; lOFC, lateral OFC/inferior frontal border; mOFC, medial OFC; NAC, nucleus accumbens; PAG, periaqueductal gray; pgACC, pregenual anterior cingulate; rACC, rostral anterior cingulate; thal, thalamus. C) Pain ratings across trials in each condition. Results show stable pain ratings in all conditions, indicating that pain did not habituate or sensitize across the session (individual ratings were similarly stable) and that placebo effects remained roughly constant in magnitude throughout the session.

Fig. 7. Robust regression scatter plots for activated voxels from group analysis showing significant correlations between placebo-induced opioid activation in heat vs. warm [(CH - PH) - (CW - PW)] with reported placebo during heat (CH - PH reports). Behavioral data were correlated with opioid receptor availability data from voxels showing a significant group effect for opioid activation during [(PH - CH) - (PW - CW)], described in Table 1. Darker circles indicate higher weights for that participant in the iteratively reweighted regression, with black indicating the maximum weight of 1. Lighter circles indicate reduced weights, with white (unfilled) indicating zero weight. Additional analyses exploring the interpretation of these negative correlations are shown in SI Fig. 8.

Fig. 8. Localization and scatter plots for voxels within ROIs showing significant correlations with reported placebo (RP). Regions showed group placebo effects either during simple heat [(CH - PH), Top], or simple warm stimulation [(CW - PW), Bottom]. Correlations between placebo opioid activation during heat and RP are shown. Partial correlations reported below control for order of C and P administration. R amygdala: XYZ = [25,-2,-27], max. partial r = -.61, P < 0.05; R mOFC: XYZ = [14,54,-20], partial r = -.78, P < 0.006; R lOFC: XYZ = [50,27,-14], partial r = -0.74 P < 0.007. Two additional regions were significant with small volume correction (SVC): R DLPFC: XYZ = [34, 11, 50], partial r = -0.65, P < 0.02; L ventral aIns: XYZ = [-29,9,-18], partial r = 0.67, P < 0.006. The large dorsal cingulate region apparent in the Top Middle was similarly negatively correlated, but the scatterplot is not shown because it did not meet the criterion of group activation in (CH - PH) and so was not part of the search volume for correlation analyses.

Fig. 9. Detailed analysis of correlations between reported placebo and opioid binding in right lateral orbitofrontal cortex (lOFC). (A) Significant lOFC region of interest shown in blue, with the a priori region of interest shown in green underlay. (B) Correlations between reported placebo (CH - PH) and placebo-induced opioid activation (CH - PH) (Left) and [(CH - PH) - (CW - PW)] (Right). (C) Correlations between opioid binding (high binding signifies low endogeous activation) and reported placebo in each condition. CW: control warm; CH: control heat; PW placebo warm; PH: placebo heat. Strong placebo responders showed lower binding on average across task conditions (r = .56, P = 0.03) and lower binding on the control conditions specfically (for CH, r = .72, P = 0.003). Note that the overall magnitude of binding in heat and warm is not interpretable, as warm trials preceded heat and it is possible that the scale may vary somewhat. The difference in slopes between CH and PH drives the correlations in difference scores shown in B. (D) Voxel-wise maps of high opioid-binding regions showing t-values for correlations between reported placebo and binding in control conditions. Show are correlations between reported placebo and CH binding (Left) and correlations with CW binding (Right). The threshold was P < 0.05, indicated by the black bar on the color scale, although the significance in many regions exceeded this threshold. Correlations in medial frontal, insular, and lateral orbitofrontal cortex were reliable and negative, indicating that stronger placebo responders showed lower control-condition binding across many opioid-rich regions. These correlations are important because they indicate that strong placebo responders are likely to either release opioids across experimental conditions, perhaps in response to the experimental context, or have higher opioid receptor binding affinity, contributing to the negative correlations with [CH - PH] binding differences shown in B.

Fig. 10. Detailed analysis of midbrain nuclei showing significant small volume corrected opioid placebo effects. To examine activation extent, the threshold shown here was P < 0.05, uncorrected. Top of figure shows areas activated more by placebo during heat than warm stimulation [(CH - PH) - (CW - PW)] in yellow and simple placebo effects during heat (CH - PH) in red. Distinction between midbrain nuclei are made according to the atlas of Duvernoy (1) [*adapted by Zambreneau et al., Pain 2005; 114:397-407, used with permission; see also ref. 2], as shown on the Top Right. The Temperature ´ Placebo interaction was significant in both dorsal and mid-ventral PAG, extending into DRN as well. (Bottom) Dissociation between heat-specific placebo effects (CH - PH) in red and anticipatory placebo opioid decreases (PW - CW) in blue. Anticipatory effects were found in the dorsal PAG and extended into the superior colliculus and surrounding brain. Placebo effects during heat were found in mid-ventral regions of the PAG and most strongly in the NCF and DRN and surrounding reticular formation, extending into VTA but largely inferior to the red nucleus. Prior localizations of the DRN (3) and nCF [reproduced with permission from ref. 4 (copyright 2006, by the Society for Neuroscience)] are provided for comparison.

1. Duvernoy, H. M. (1995) The human brain stem and cerebellum (Springer-Verlag, New York).

2. Zambreanu, L., Wise, R. G., Brooks, J. C. W., Iannetti, G. D., & Tracey, I. (2005) Pain 114, 397-407.

3. Seymour, B., O'Doherty, J. P., Dayan, P., Koltzenburg, M., Jones, A. K., Dolan, R. J., Friston, K. J., & Frackowiak, R. S. (2004) Nature 429, 664-667.

4. Keltner, J. R., Furst, A., Fan, C., Redfern, R., Inglis, B., & Fields, H. L. (2006) J Neurosci 26, 4437-4443.

Fig. 11. Details of multivariate connectivity analysis performed on opioid binding data across conditions. (A) Optimal solution for opioid activity correlations across regions included 7 networks (permutation P value for clustering < .0001, Z = 4.0). (B and C) Chosen solution of 7 networks provides best improvement in fit over permuted data. (D) Likelihood of observed pattern between networks occurring given 10,000 permutations (P < 0.001).

Fig. 12. The same multidimensional scaling graph shown in Fig. 4, but with lines connecting pairs of regions for which the inter-region correlation (Spearman's rho across subjects) is significantly greater for placebo than control during heat (black, rho(PH) > rho(CH) P < 0.01; light gray, P < 0.05; light blue, rho(PH) < rho(CH), P < 0.05). (Inset 1) As predicted a priori, the PAG-rACC relationship is stronger with placebo. This is also true of a number of other regions, including insula-NAC (Inset 2), amygdala-DLPFC (Inset 3), and others. The correlation scatter plots (on ranked data) for inter-region correlations during placebo (Left) and control (Right) for these three pairs are shown for pairs numbered 1-3 on the central plot. 26 regions showed stronger correlations with placebo (black lines), whereas only five showed stronger correlations with control (blue lines). (Inset 4) Shown is the null hypothesis distribution (gray bars) for the number of positive - the number of negative significant correlations at P < 0.05. The black line shows the observed number (P < 0.0001). Overall, placebo increased functional integration among regions.