Expectation violation and attention to pain jointly modulate neural gain in somatosensory cortex

Abstract

The neural processing and experience of pain are influenced by both expectations and attention. For example, the amplitude of event-related pain responses is enhanced by both novel and unexpected pain, and by moving the focus of attention towards a painful stimulus. Under predictive coding, this congruence can be explained by appeal to a precision-weighting mechanism, which mediates bottom-up and top-down attentional processes by modulating the influence of feedforward and feedback signals throughout the cortical hierarchy. The influence of expectation and attention on pain processing can be mapped onto changes in effective connectivity between or within specific neuronal populations, using a canonical microcircuit (CMC) model of hierarchical processing. We thus implemented a CMC within dynamic causal modelling for magnetoencephalography in human subjects, to investigate how expectation violation and attention to pain modulate intrinsic (within-source) and extrinsic (between-source) connectivity in the somatosensory hierarchy. This enabled us to establish whether both expectancy and attentional processes are mediated by a similar precision-encoding mechanism within a network of somatosensory, frontal and parietal sources. We found that both unexpected and attended pain modulated the gain of superficial pyramidal cells in primary and secondary somatosensory cortex. This modulation occurred in the context of increased lateralized recurrent connectivity between somatosensory and fronto-parietal sources, driven by unexpected painful occurrences. Finally, the strength of effective connectivity parameters in S1, S2 and IFG predicted individual differences in subjective pain modulation ratings. Our findings suggest that neuromodulatory gain control in the somatosensory hierarchy underlies the influence of both expectation violation and attention on cortical processing and pain perception.

Fig. 1

A) Experimental task. Top-down attentional set was manipulated at the block level by presenting a verbal instruction to attend away from pain and towards the visual stimulation on the screen (i.e., cue=“CROSS”; unattended pain) or attend to the painful stimuli perceived on the dorsum of the hands (i.e., cue=“HAND”; attended pain). In each block, a total of 25 trains of painful stimuli were delivered to the dorsum of one hand at the time, using an oddball roving sequence. Each train included 3–7 stimulus repetitions at a constant inter-stimulus interval of 1 s. Each painful stimulus consisted of two rapidly square-wave pulses of 50 μs duration, with an inter-pulse interval of 5 ms. The same stimulus intensity was used for both left and right stimuli. We considered a deviant (d) the first stimulus in each train (i.e., change in stimulus location). To match the number of trials of deviant and standard stimuli, we only modeled the last repetition before a change as a standard (s). While painful stimuli were delivered, a fixation cross on the screen changed in color from black to white or vice versa every 2–5 s. The visual change never occurred at the same time as a change in the painful stimulus location. When instructed to pay attention to the visual stimulation, participants had to silently count the number of time the cross changed in color from white to black or vice versa (Block A). Instead, when instructed to pay attention to the painful stimuli, participants had to silently count the number of times the stimulation switched from the left to the right hand or vice versa (Block B). Block order was counterbalanced across participants. At the end of each block, participants were required to report the number of switches, as well as to rate the average pain intensity experienced for each hand. B) The probability of repetitions between 3–7 times was 5%, 15%, 60%, 15%, 5%, respectively. C) Mean and standard error of pain ratings for left and right hand and attended (white) and unattended (grey) pain, separately. Participants reported less intense pain when the painful somatosensory stimuli were unattended.

Fig. 2

ERF and source reconstruction results. Summary of laterality (separately for left and right), attention, expectation violation main effects, and attention by expectation violation interactions. In each panel, the first row depicts the timing and topography of event-related field effects at the scalp level. The left figure represents the posterior-anterior displacement of the effect as a function of time (y axis, from 20 to 400 ms). The red arrow indicates the ERF maximal peak; e.g., in central-anterior locations, at around 80 ms (left stimulation) and 50 ms (right stimulation) and in a central location at around 100 ms (attention). The central figure represents the left-right displacement as a function of time (y axis, from 20 to 400 ms). Again, the red arrow indicates the ERF maximal peak; e.g., in the right hemisphere (left stimulation), left hemisphere (right stimulation) or close to the midline (attention). Finally, the right figure depicts the topography of the ERF effect at the peak time point; e.g., over anterior right sensors (left stimulation), anterior left sensors (right stimulation), widespread across posterior and anterior sensors (attention). The second row illustrates the topography of reconstructed sources. The analysis identified four bilateral sources (S1, S2, IFG, and IPC). The specific locations of left and right S1 sources (MNI coordinates: left [−26, −36, 58]; right [32, −40, 64]) were derived by comparing right vs. left and left vs. right stimulation (i.e., laterality main effect). We established bilateral S2 coordinates (MNI coordinates: left [−62, 14, 20]; right [62, 24, 26]) by comparing attended vs. unattended pain, regardless the stimulation side. Finally, we identified left inferior frontal and right inferior parietal regions (MNI coordinates: left IFG [−54, 8, 16]; right IPC [36, −66, 40]) in the expectation violation main effect, as well as right inferior frontal and left inferior parietal regions (MNI coordinates: right IFG [54, 0, 10]; left IPC [−32, −64, 46]) in the attention by expectation violation interaction. The identified sources were then entered into a dynamic causal modelling specifying alternative connectivity architectures. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

A) Architectures of 24 alternative models of somatosensory and fronto-parietal regions, fitted to grand-averaged ERF data. The models differed with respect to the inclusion of a frontal (IFG) and/or parietal (IPC) node, as well as the hierarchical architecture and connections between somatosensory and fronto-parietal areas. B) Contextual modulations of intrinsic connectivity by attention and expectation violation were optimized with respect to 1 null model, 12 bilateral alternative models (attention and expectation violation) and 16 contralateral somatosensory models (expectation violation). In the figure, the contralateral models are shown for left expectation violation. Right expectation violation models were identical but with left lateralization of somatosensory regions. C) Both contralateral S1 and S2 were specified as cortical targets of thalamic input. D) All DCMs were tested using a canonical microcircuit model. Each source was thus modeled as compromising 4 neuronal populations (superficial and deep pyramidal cells, spiny stellate and inhibitory interneurons). E) The winning model structure (M10), identified using fixed-effect Bayesian model selection, included all bilateral regions, with IPC at the highest hierarchical level, as well as connections between S1 to both fronto-parietal nodes and connections between S2 and the frontal node. E) The winning model of connectivity modulation, identified using fixed-effect Bayesian model selection, included changes in gain in bilateral primary, secondary somatosensory cortex, as well as and IFG by attention (M10). Further, the winning model revealed changes in contralateral primary and bilateral secondary somatosensory cortex by expectation violation (M13).

Fig. 4

Bayesian Model Averaging (BMA) of the contextual modulation of intrinsic connectivity by attention, as well as intrinsic and extrinsic connectivity by expectation violation. Increased connectivity is showed in red, while decreased connectivity in blue. Attention increased (disinhibition) somatosensory gain, while decreasing (increased inhibition) frontal gain. Expectation violation increased the gain of contralateral primary and bilateral secondary somatosensory cortex. Further, expectation violation increased forward connectivity, in a right-lateralized fashion irrespective of violation location, while decreasing backward connectivity mostly between contralateral regions. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5

A) Parametric empirical Bayes analysis revealed that inter-individual variability in pain modulation ratings correlated with intrinsic and extrinsic connectivity changes driven by attention and expectation violation. The neural effect of attention on self-inhibitory connections in left S1 and right S2 correlated with the degree to which participants experienced pain enhancement by attention. Further, following left expectation violation, pain modulation ratings were predicted by the strength of backward connectivity to somatosensory regions primarily in the right hemisphere. Instead, following right expectation violation, pain modulation ratings were predicted by the strength of forwards connectivity from S1 and S2 to higher order regions in the left hemisphere. This pattern of results complemented hemispheric functional asymmetries observed at the group level. B) Prediction accuracy at the leave-one-out cross validation, which allowed us to quantify the ‘out of sample’ effect size in terms of the correlation between pain modulation ratings and DCM parameters. The analysis revealed a subset of intrinsic and extrinsic parameters that successfully predicted pain modulation ratings (r(20)=.55, p<.01). The x-axis represents single participants (N=22), while the y-axis depicts the actual vs. predicted z-scored pain modulation ratings.