Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2018 Aug 31;14(8):e1006243.
doi: 10.1371/journal.pcbi.1006243. eCollection 2018 Aug.

Human Pavlovian Fear Conditioning Conforms to Probabilistic Learning

Affiliations
Free PMC article

Human Pavlovian Fear Conditioning Conforms to Probabilistic Learning

Athina Tzovara et al. PLoS Comput Biol. .
Free PMC article

Abstract

Learning to predict threat from environmental cues is a fundamental skill in changing environments. This aversive learning process is exemplified by Pavlovian threat conditioning. Despite a plethora of studies on the neural mechanisms supporting the formation of associations between neutral and aversive events, our computational understanding of this process is fragmented. Importantly, different computational models give rise to different and partly opposing predictions for the trial-by-trial dynamics of learning, for example expressed in the activity of the autonomic nervous system (ANS). Here, we investigate human ANS responses to conditioned stimuli during Pavlovian fear conditioning. To obtain precise, trial-by-trial, single-subject estimates of ANS responses, we build on a statistical framework for psychophysiological modelling. We then consider previously proposed non-probabilistic models, a simple probabilistic model, and non-learning models, as well as different observation functions to link learning models with ANS activity. Across three experiments, and both for skin conductance (SCR) and pupil size responses (PSR), a probabilistic learning model best explains ANS responses. Notably, SCR and PSR reflect different quantities of the same model: SCR track a mixture of expected outcome and uncertainty, while PSR track expected outcome alone. In summary, by combining psychophysiological modelling with computational learning theory, we provide systematic evidence that the formation and maintenance of Pavlovian threat predictions in humans may rely on probabilistic inference and includes estimation of uncertainty. This could inform theories of neural implementation of aversive learning.

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1
Fig 1. Single trial estimates for SCR and PSR for three experiments, for CS+ / CS- trials.
The estimates are derived from psychophysiological models (PsPMs) for anticipatory SCR and PSR. As in our analyses, only US- trials are shown. Because there are twice as many CS- (dark gray) than CS+US- (light gray) trials, we only display CS- responses for every second CS- trial to simplify presentation, but include all CS- responses in our analyses. Shaded bars reflect the standard error across participants.
Fig 2
Fig 2. Predictions of probabilistic and non probabilistic RL models, grouped in families according to their predictions.
For reasons of consistency with the ANS data displayed on Fig 1, we only show predictions of non-reinforced trials for CS+ (solid lines) and an equal number of CS- trials (equal to 40, dashed lines). (a) Model predictions for probabilistic models. Model predictions are illustrated according to whether they reflect a measure of the estimated CS value (green lines, outcome), the degree of surprise/model update (orange lines, surprise), or a combination of two quantities (purple lines, combination). (b) Model predictions for non probabilistic models, similar to -a-. This figure shows predictions for an exemplar trial sequence, to illustrate the trial-by-trial dynamics of each model. See S1 Fig for averages over 100 trial sequences, illustrating the general trends of the predictions.
Fig 3
Fig 3
Model selection results (a) across three families of observation models (b) for the entire model space and (c) for previously proposed non probabilistic models for threat learning. The three model families of panel (a) collapse across all models and reflect observation functions related to value (models RW, HM2, BM), surprise (model HM1) and combinations (noted Bayesian Combination model, BC). The bars in all panels indicate model evidence, based on protected exceedance probabilities (p.x.p., RFX).
Fig 4
Fig 4. Proportion of variance in the physiological data explained by each of the models across the four experiments.
Bar illustrate the explained variance for each model and experiment.
Fig 5
Fig 5. Confusion matrices for family and model recovery rates, using BIC as the evidence metric.
(a) Recovery rates across families. Rows represent true models/families (used to simulate data), and columns represent fitted models/families. (b) Recovery rates for single models, similar to -a-. Results are based on 256 simulated experiments, each with 20 participants.

Similar articles

See all similar articles

Cited by 7 articles

See all "Cited by" articles

References

    1. LeDoux JE. Coming to terms with fear. Proc Natl Acad Sci U S A. 2014;111: 2871–8. 10.1073/pnas.1400335111 - DOI - PMC - PubMed
    1. LeDoux JE. Emotion circuits in the brain. Annu Rev Neurosci. 2000;23: 155–184. 10.1146/annurev.neuro.23.1.155 - DOI - PubMed
    1. Ciocchi S, Herry C, Grenier F, Wolff SBE, Letzkus JJ, Vlachos I, et al. Encoding of conditioned fear in central amygdala inhibitory circuits. Nature. 2010;468: 277–282. 10.1038/nature09559 - DOI - PubMed
    1. McNally GP, Johansen JP, Blair HT. Placing prediction into the fear circuit. Trends Neurosci. Elsevier Ltd; 2011;34: 283–292. 10.1016/j.tins.2011.03.005 - DOI - PMC - PubMed
    1. Bach DR, Weiskopf N, Dolan RJ. A Stable Sparse Fear Memory Trace in Human Amygdala. J Neurosci. 2011;31: 9383–9389. 10.1523/JNEUROSCI.1524-11.2011 - DOI - PMC - PubMed

Publication types

Feedback