A common neural network differentially mediates direct and social fear learning

Highlights

• We compared Direct (Pavlovian) and social (vicarious) fear learning within participants.

• Direct and social fear learning activates a common neural network (Amygdala, AI, ACC).

• The US enters this common network via the Amygdala in direct fear learning, and through the AI in social fear learning.

• AI more connected to TPJ during social fear learning.

• Associability gates network connectivity.

Abstract

Across species, fears often spread between individuals through social learning. Yet, little is known about the neural and computational mechanisms underlying social learning. Addressing this question, we compared social and direct (Pavlovian) fear learning showing that they showed indistinguishable behavioral effects, and involved the same cross-modal (self/other) aversive learning network, centered on the amygdala, the anterior insula (AI), and the anterior cingulate cortex (ACC). Crucially, the information flow within this network differed between social and direct fear learning. Dynamic causal modeling combined with reinforcement learning modeling revealed that the amygdala and AI provided input to this network during direct and social learning, respectively. Furthermore, the AI gated learning signals based on surprise (associability), which were conveyed to the ACC, in both learning modalities. Our findings provide insights into the mechanisms underlying social fear learning, with implications for understanding common psychological dysfunctions, such as phobias and other anxiety disorders.

Introduction

Humans, and many other animals, can acquire fears through observing conspecifics being subjected to aversive events (Bandura and McClelland, 1977, Debiec et al., 2017, Olsson and Phelps, 2007, Rachman, 1977). This capacity for social learning without direct exposure to aversive consequences is an adaptation that allows the organism to avoid the costs of individual learning, such as predation and poisoning (Laland, 2004, Lindström et al., 2016). Social learning might, however, not always be adaptive: a large proportion of human fears and phobias is acquired through social means, speaking to the generality of this learning mechanism (Askew and Field, 2008, Rachman, 1977).

Despite the cross-species importance of social fear learning, its neural and computational underpinnings are poorly understood, which stands in contrast to our quickly advancing knowledge about fear learning through direct, Pavlovian, conditioning (henceforth direct fear learning). Animal studies of direct fear learning have delineated the amygdala as critical for the acquisition, storage and expression of conditioned fear (LeDoux, 2012). The amygdala is thought to be the site where information about the stimulus predicting danger (the conditioned stimulus, CS) becomes associated with information about the aversive event (the unconditioned stimulus, US). Similarly, research has demonstrated that the amygdala is critical for direct fear learning also in our species (Delgado et al., 2006, LaBar et al., 1998, Phelps and LeDoux, 2005).

Recent studies in rodents (Debiec and Sullivan, 2014, Jeon et al., 2010, Knapska et al., 2006), and fMRI studies in humans (Meffert et al., 2015, Olsson et al., 2007), suggest that the amygdala also plays a central role in acquiring fears through social fear learning. Together with a range of overt behavioral similarities (Olsson and Phelps, 2007), this shared neural circuitry suggests that partially overlapping mechanisms are involved in direct and social fear learning. However, because the available studies have not directly contrasted direct and social forms of fear learning within the same participant, this conclusion has been tentative. This situation mirrors the ongoing debate within the wider field of learning and decision making, where many conclusions about the neural and computational overlap between individual and social experiences are constrained by comparison between studies or reverse inference (Ruff and Fehr, 2014).

In the present study, we addressed the dearth of knowledge about the neural underpinnings of social fear learning by investigating shared and unique features of hemodynamic responses (fMRI) during direct and social fear learning within the same participant. Such a direct comparison within participants is crucial for analyzing the similarities and differences between direct and social fear learning. Moreover, we combined formal learning theory and dynamic causal modeling (DCM) to better understand how the computational mechanisms involved in direct and social fear learning converge and diverge.

The computational mechanisms underlying direct fear conditioning can be well characterized by classical formal learning theory, such as the Rescorla-Wagner (R-W) (Rescorla and Wagner, 1972a) and Pearce-Hall (P-H) models (Pearce and Bouton, 2001) that describe how the CS and US signals are combined algebraically to associate cues with aversive events. The R-W model encompasses the idea of error-driven learning, where the mismatch (i.e., prediction errors) between delivered reinforcements and the organism's predictions of reinforcement leads to updated associations. One basic assumption in the R-W model is that the rate of learning is constant (Rescorla and Wagner, 1972a). In contrast, the P-H model explicitly describe how more surprising outcomes change the rate of learning (termed associability) (Pelley, 2004). Associability increases in proportion to the absolute prediction error on the last interaction with a stimulus, allowing the agent to adapt to changing environments, which by definition, leads to larger prediction errors, and thereby higher associability. These two accounts have recently been combined in a hybrid model, which is closely related to optimal (Bayesian) statistical inference (Roesch et al., 2012). The computations posited by the hybrid model have been linked to neural signals within the amygdala and insular cortex in humans (Boll et al., 2013, Li et al., 2011), as well as the amygdala in rodents (Roesch et al., 2012). These structures implement a surprise-based associability signal, which gates learning from prediction errors when the environment is changing. Whether the amygdala and insular cortex subserve similar roles in social fear learning is presently unknown.

Based on a previously suggested neural model of social fear learning (Olsson and Phelps, 2007), we hypothesized that social fear learning would involve many of the same brain regions as direct fear learning, representing a cross-modal (self/other) core “aversive learning network”. We predicted that the amygdala would be at the center of this aversive learning network, and that one key computational role of the amygdala would be to gate learning based on associability, or attention, signals, in both direct and social fear learning. In addition to the amygdala, this aversive learning network should involve regions linked to the aversive value of pain, both self-experienced and empathic, such as the anterior insula (AI) and the anterior cingulate cortex (ACC), because the US response driving fear learning rests on such representations. Furthermore, because social fear learning by definition must involve distinctly social processes (simply because another individual is the recipient of the US), we predicted that regions commonly involved in social cognition and theory of mind, such as the dorsomedial prefrontal cortex (DMPFC), the superior temporal sulcus (STS), and the temporo-parietal (TPJ), would be uniquely involved in social fear learning (Olsson and Phelps, 2007). This involvement might either be additive, so that the “aversive learning” network underlying social fear learning would involve additional nodes relative to direct fear learning, or modulatory, so that the same network nodes would receive differential input depending on the modality (self/other).