Local brain oscillations and interregional connectivity differentially serve sensory and expectation effects on pain

Pain emerges from the integration of sensory information about threats and contextual information such as an individual’s expectations. However, how sensory and contextual effects on pain are served by the brain is not fully understood so far. To address this question, we applied brief painful stimuli to 40 healthy human participants and independently varied stimulus intensity and expectations. Concurrently, we recorded electroencephalography. We assessed local oscillatory brain activity and interregional functional connectivity in a network of six brain regions playing key roles in the processing of pain. We found that sensory information predominantly influenced local brain oscillations. In contrast, expectations exclusively influenced interregional connectivity. Specifically, expectations altered connectivity at alpha (8 to 12 hertz) frequencies from prefrontal to somatosensory cortex. Moreover, discrepancies between sensory information and expectations, i.e., prediction errors, influenced connectivity at gamma (60 to 100 hertz) frequencies. These findings reveal how fundamentally different brain mechanisms serve sensory and contextual effects on pain.


Supplementary Materials
. Frequentist analysis of the effects of stimulus intensity, expectations, and prediction errors on local brain activity (top three panel rows) and inter-regional functional connectivity (bottom three panel rows). Effects were assessed by a frequentist rmANOVA with factors intensity and expectation. The top number in each tile shows the uncorrected p-value, the bottom number the FDR-adjusted p-value. The adjustment was performed across all 15 connections and 6 ROIs in the case of inter-regional connectivity and local activity, respectively. The color of the tiles scales with the uncorrected p-value. Corresponding F-values and η 2 -value are shown in Figure S2.  Figure S2. Frequentist analysis of the effects of stimulus intensity, expectations, and prediction errors on local brain activity (top three panel rows) and inter-regional functional connectivity (bottom three panel rows). Effects were assessed by a frequentist rmANOVA with factors intensity and expectation. The top number in each tile corresponds to the Fvalue, the bottom number is the η 2 -value. Corresponding uncorrected and FDR-adjusted p-values are shown in Figure S1. The color of the tiles scales with the uncorrected p-value.  Figure S3. Direction and strength of intensity, expectation, and prediction error effects on local brain activity (top three panel rows, [V 2 s]) and inter-regional functional connectivity (bottom three panel rows [ -]). The figure shows average mean differences between the two levels of intensity, expectation, and prediction error. The top number in each tile indicates the mean difference between two types of conditions, the bottom number indicates the standard deviation of these mean differences across participants. The tile color scales with the quotient of the mean difference and standard deviation, thus indicating both the direction of a condition difference (blue and red indicating smaller and larger values in the first relative to the second condition, respectively) as well the as the magnitude of the difference relative to its variability across subjects (more intense colors indicating a larger difference). For the intensity contrast (hi -li), the mean difference is defined as the mean across all participants of the difference between the averaged hi-conditions [ -] std. dev.
(mean [HEhi, LEhi]) and averaged li-conditions (mean [HEli, LEli]). Accordingly, for the expectation contrasts (HE -LE), the mean difference is the mean across all participants of mean[HEhi, HEli] -mean [LEhi, LEli]. For the prediction error contrast (hPE -lPE), the mean difference is the mean across all participants of mean[HEli, LEhi] -mean [HEhi, LEli]. Figure S4. Control analysis for the effects of stimulus intensity, expectations, and prediction errors on local brain activity. Power at alpha, beta, and gamma frequencies was quantified using the time windows 500-900 ms, 300-600 ms, and 150-350 ms, respectively. Heat maps indicate Bayes factors of a Bayesian rmANOVA with factors intensity and expectation. The color of the heat map tiles scales with the log of the Bayes factor. It ranges from blue (BF < 1/3, at least moderate evidence against an effect) to yellow (BF > 3, at least moderate evidence for an effect). Brain schematics display ROIs in yellow which exhibit at least moderate evidence for an effect (BF > 3).  . with = P C power model (pow) @ connectivity model (conn).
For the computation of the Bayesian model evidence a prior distribution over the model parameters and must be specified. Here, we select a bivariate standard normal distribution: where and are the zero vector and identity matrix, respectively. Figure S3 shows graphs of logistic functions for several parameter values drawn from the prior distribution. Figure S8. Graphs of logistic functions for several parameter values drawn from the prior distribution specified above. To show the prior graphs in relation to the data, by way of example, power values of ROI S1 are depicted as black circles. Specifically, the x-and y-values of the data points correspond to the values in vectors # and % , respectively.
According to Bayes' rule, the probability of the data given the model (a.k.a. model evidence) is model = ∬ 8 @ , | , ; S5 6 ; , V , where for the power and connectivity models, is substituted by C and @ , respectively. In our implementation, we compute this integral using standard Monte Carlo integration with 10samples.
The described procedure results in pow = 6 power and conn = 15 connectivity values per frequency band. The model evidence is computed for all individual power and