HighlightsWe validate a Psychophysiological model (PsPM) to infer anticipatory sympathetic arousal from changes in skin conductance.We optimise the inversion of this PsPM in terms of a constrained non-linear dynamic causal model.This method allows a quantification of fear memory in humans.

Across species, cued fear conditioning is a common experimental paradigm to investigate aversive Pavlovian learning. While fear‐conditioned stimuli (CS+) elicit overt behavior in many mammals, this is not the case in humans. Typically, autonomic nervous system activity is used to quantify fear memory in humans, measured by skin conductance responses (SCR). Here, we investigate whether heart period responses (HPR) evoked by the CS, often observed in humans and small mammals, are suitable to complement SCR as an index of fear memory in humans. We analyze four datasets involving delay and trace conditioning, in which heart beats are identified via electrocardiogram or pulse oximetry, to show that fear‐conditioned heart rate deceleration (bradycardia) is elicited and robustly distinguishes CS+ from CS−. We then develop a psychophysiological model (PsPM) of fear‐conditioned HPR. This PsPM is inverted to yield estimates of autonomic input into the heart. We show that the sensitivity to distinguish CS+ and CS− (predictive validity) is higher for model‐based estimates than peak‐scoring analysis, and compare this with SCR. Our work provides a novel tool to investigate fear memory in humans that allows direct comparison between species.

During fear conditioning, pupil size responses dissociate between conditioned stimuli that are contingently paired (CS+) with an aversive unconditioned stimulus, and those that are unpaired (CS‐). Current approaches to assess fear learning from pupil responses rely on ad hoc specifications. Here, we sought to develop a psychophysiological model (PsPM) in which pupil responses are characterized by response functions within the framework of a linear time‐invariant system. This PsPM can be written as a general linear model, which is inverted to yield amplitude estimates of the eliciting process in the central nervous system. We first characterized fear‐conditioned pupil size responses based on an experiment with auditory CS. PsPM‐based parameter estimates distinguished CS+/CS‐ better than, or on par with, two commonly used methods (peak scoring, area under the curve). We validated this PsPM in four independent experiments with auditory, visual, and somatosensory CS, as well as short (3.5 s) and medium (6 s) CS/US intervals. Overall, the new PsPM provided equal or decisively better differentiation of CS+/CS‐ than the two alternative methods and was never decisively worse. We further compared pupil responses with concurrently measured skin conductance and heart period responses. Finally, we used our previously developed luminance‐related pupil responses to infer the timing of the likely neural input into the pupillary system. Overall, we establish a new PsPM to assess fear conditioning based on pupil responses. The model has a potential to provide higher statistical sensitivity, can be applied to other conditioning paradigms in humans, and may be easily extended to nonhuman mammals.

Learning to predict threat depends on amygdala plasticity and does not require auditory cortex (ACX) when threat predictors (conditioned stimuli, CS) are simple sine tones. However, ACX is required in rodents to learn from some naturally occurring CS. Yet, the precise function of ACX, and whether it differs for different CS types, is unknown. Here, we address how ACX encodes threat predictions during human fear conditioning using functional magnetic resonance imaging (fMRI) with multivariate pattern analysis. As in previous rodent work, CS+ and CS- were defined either by direction of frequency modulation (complex) or by frequency of pure tones (simple). In an instructed non-reinforcement context, different sets of simple and complex sounds were always presented without reinforcement (neutral sounds, NS). Threat encoding was measured by separation of fMRI response patterns induced by CS+/CS-, or similar NS1/NS2 pairs. We found that fMRI patterns in Heschl's gyrus encoded threat prediction over and above encoding the physical stimulus features also present in NS, i.e. CS+/CS- could be separated better than NS1/NS2. This was the case both for simple and complex CS. Furthermore, cross-prediction demonstrated that threat representations were similar for simple and complex CS, and thus unlikely to emerge from stimulus-specific top-down, or learning-induced, receptive field plasticity. Searchlight analysis across the entire ACX demonstrated further threat representations in a region including BA22 and BA42. However, in this region, patterns were distinct for simple and complex sounds, and could thus potentially arise from receptive field plasticity. Strikingly, across participants, individual size of Heschl's gyrus predicted strength of fear learning for complex sounds. Overall, our findings suggest that ACX represents threat predictions, and that Heschl's gyrus contains a threat representation that is invariant across physical stimulus categories.