Dynamic causal modelling of anticipatory skin conductance responses

Abstract

Anticipatory skin conductance responses [SCRs] are a widely used measure of aversive conditioning in humans. Here, we describe a dynamic causal model [DCM] of how anticipatory, evoked, and spontaneous skin conductance changes are generated by sudomotor nerve activity. Inversion of this model, using variational Bayes, provides a means of inferring the most likely sympathetic nerve activity, given observed skin conductance responses. In two fear conditioning experiments, we demonstrate the predictive validity of the DCM by showing it has greater sensitivity to the effects of conditioning, relative to alternative (conventional) response estimates. Furthermore, we establish face validity by showing that trial-by-trial estimates of anticipatory sudomotor activity are better predicted by formal learning models, relative to response estimates from peak-scoring approaches. The model furnishes a potentially powerful approach to characterising SCR that exploits knowledge about how these signals are generated.

Fig. 1

Example for how the skin conductance signal during two trials is generated (simulated data). In this example, a CS is presented at 0 and 15 s, eliciting two anticipatory sudomotor activity bursts (aSCR neural input). The US is received at 4 and 19 s and each time evokes a sharper firing burst (eSCR neural input). Both these neural input functions are summed up (SCR neural input) and convolved with the response function specific to the skin/sweat gland system (RF for SCR) to cause the observed SCR. Similarly, a small number of spontaneous sudomotor bursts occurs in the inter-trial interval (SF neural input), which is convolved with its own response function (RF for SF) to cause the observed SF. A small baseline change in the inter-trial interval is modelled here with another neural input function (SCL neural input) that is simply cumulatively integrated to yield the observed SCL change. All response components are then added to yield the observed compound skin conductance (SC) signal.

Fig. 2

Schematic summary of the model inversion scheme. First, all eSCRs from one dataset are summarised by their first principal component, and a response function (i.e. the RF for SCR) is approximated to this data. This, together with the RF for SF, knowledge about the timing of experimental events, and assumptions about the form of the neural input, is then used to estimate the most likely underlying neural inputs, given the data. These neural input functions are estimated to optimise the data fit.

Fig. 3

Variance ratio R² of the aSCR estimates that can be explained by formal learning models (i.e. Rescorla–Wagner) under optimal parameters, suggesting that DCM estimates bear a closer relation to central processes than estimates from GLM with one regressor per trial, or peak-scoring estimates.