Modeling fear‐conditioned bradycardia in humans

Abstract

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.

A pupil size response model to assess fear learning

“…SCR and ECG data from this experiment were published previously (Castegnetti et al, ; Staib et al, ). Two sets of CS+/CS‐ were used (simple stimuli: sine tones with a constant frequency of 400 or 800 Hz over the entire 4‐s interval; complex stimuli: a train of four frequency modulated sounds of 1 s each, which were rising from 400 to 800 Hz or falling from 800 to 400 Hz).…”

“…Overall, estimating neural processes via model‐based approaches tends to improve the signal‐to‐noise ratio (Bach & Friston, ). For example, model‐based analysis of SCR (Staib, Castegnetti, & Bach, ), heart period responses (Castegnetti et al, ), respiration responses (Castegnetti, Tzovara, Staib, Gerster, & Bach, in press), and possibly also startle eyeblink responses (Khemka, Tzovara, Gerster, Quednow, & Bach, ) provides better discrimination of CS+/CS‐ responses than peak scoring measures in fear conditioning paradigms.…”

“…For some data types (e.g., skin conductance), responses to CS+ and CS‐ look very similar and only differ by their amplitude (Boucsein, ). For others, such as heart period, responses appear to be composed of several components: one reflecting a response to the CS, which is the same for CS+ and CS‐, and an added component with different time course for the CS+ (Castegnetti et al, ). Crucially, in this study, it appeared that the common heart period response to the CS differs across experiments and may depend on specific perceptual characteristics that do not necessarily generalize across studies.…”

“…Crucially, in this study, it appeared that the common heart period response to the CS differs across experiments and may depend on specific perceptual characteristics that do not necessarily generalize across studies. However, we found that by only modeling the CS+‐related component, CS+ and CS‐ could be reliably distinguished, such that there is no need to model common CS responses (Castegnetti et al, ). Indeed, additionally modeling such responses would influence parameter estimates for CS‐ and CS+ equally, such that their contrast would be unchanged.…”

A linear model for event-related respiration responses

“…In order to estimate the psychological input into this system, we used a general linear convolution model for each participant and experiment (GLM), in line with previous approaches to analysis of skin conductance responses (Bach et al, 2009), heart period responses (Castegnetti et al, 2016, Paulus et al, 2016), or functional magnetic resonance imaging (Friston et al, 1994). A GLM can be written as$Y=X\beta +\mathrm{ϵ}\text{,}$where X is design matrix in which each column is obtained by convolving impulse functions at event onset with each component of the RF.…”