Placing prediction into the fear circuit

Abstract

Pavlovian fear conditioning depends on synaptic plasticity at amygdala neurons. Here, we review recent electrophysiological, molecular and behavioral evidence suggesting the existence of a distributed neural circuitry regulating amygdala synaptic plasticity during fear learning. This circuitry, which involves projections from the midbrain periaqueductal gray region, can be linked to prediction error and expectation modulation of fear learning, as described by associative and computational learning models. It controls whether, and how much, fear learning occurs by signaling aversive events when they are unexpected. Functional neuroimaging and clinical studies indicate that this prediction circuit is recruited in humans during fear learning and contributes to exposure-based treatments for clinical anxiety. This aversive prediction error circuit might represent a conserved mechanism for regulating fear learning in mammals.

Figure 1

Fear learning and US evoked responding in rat LA and PAG neurons is attenuated when the shock US is expected and depends on MOR in the vlPAG. (a) Rats were trained in a two-stage fear conditioning paradigm. In the first stage ( ‘Stage I’), animals were trained with CS(A) US pairings (A+) over three days and the percentage of freezing behavior during the 30-second CS presentationswas recorded. During ‘Stage II’, animals received either vehicle (Sal) or CTAP (MOR antagonist) administered into the vlPAG prior to .combined pairings of either CS(AB)-US [in which the US was already predicted by CS(A)] or CS(CD) -US (in which the US was not predicted). During the ‘Test’ stage, behavioral freezing responses to 30 s presentations of CS(B) (blue) and CS(D) (black) were assessed drug free and without the shock US. Blocking of fear learning (i.e. reduced fear learning) to CSB was observed in animals which had previously received intra-vlPAG saline, as exemplified by a lower freezing during the 30 s presentations of CS(B) compared with CS(D) (‘Saline’). However, the blocking effect was abolished in animals which had previously received intra-vlPAG CTAP (‘CTAP’). (b) & (c) Rats were trained in a Pavlovian fear conditioning involving electrophysiological recordings of LA and PAG neurons during CS and US presentations The US-evoked neural response was significantly inhibited in both the LA and PAG when it is predicted by a well-trained CS. Population averaged (Y-axis) peri-event time histograms showing inhibition of US-evoked responding in (b) LA and (c) PAG neurons when the US is predicted (Pred, blue) by a previously trained CS compared with when it is presented unpredictably (Unpred, black). Time during the US presentation (individual 2 ms shock pulses over 2 seconds) is shown on the X-axes with individual shock pulses indicated by red hash marks. Note that statistical analyses compared averaged firing rates during the US period in the Pred and Unpred conditions. Panel A is reproduced, with permission, from [28]. Panel B is reproduced, with permission, from [21].

Figure 2

Working model of proposed neural circuitry that is involved in teaching signal processing during fear conditioning. For simplicity, ascending projections are shown in the left hemisphere only and descending projections in the right hemisphere only. The putative US pathway conveying information about the actual shock US (λ) is shown in black. The putative CS pathway conveying information about the expected outcome (-∑V) is shown in red. The putative prediction error modulated teaching signal (λ - ∑V) is shown in green. Shock USs (λ) activate spinal and trigeminal dorsal horn neurons which project to the PAG and from there through midline and intralaminar thalamus to the dmPFC and LA to produce depolarization of LA pyramidal cells. During fear conditioning, weaker auditory thalamic and cortical CS afferent inputs to LA pyramidal neurons (broken black lines) are strengthened (ΔV) when they are co-active with US-evoked depolarization of the same cells. Following conditioning, CS inputs to LA activate projection neurons to the CeA, which send inhibitory projections to the vlPAG (-∑V) that: 1) produce freezing (possibly by relieving PAG output neurons from tonic inhibition); and, 2) inhibit shock-US responsive neurons in the PAG to attenuate US processing. Disinhibited output from the PAG may also be relayed to the rostroventromedial medulla to inhibit shock US processing at the level of the dorsal horn. Thus, the ascending projection from the dorsal horn to the PAG may also be a prediction error modulated teaching signal. Pathways for CS-mediated inhibition would serve to inhibit US-evoked depolarization of LA neurons when the US is predicted (such as during blocking) thereby limiting associative plasticity of CS inputs onto the same cells.

Figure 3

Expectation modulates thalamic and cortical responses to aversive USs. (a) Human participants received CS presentations followed by a loud noise unconditioned stimulus (UCS) on 100% (‘CS100’) or 50% (‘CS50’)of trials. The unexpected UCS elicited a greater BOLD response in the anterior cingulate and dorsolateral PFC as compared to the expected UCS, as revealed by fMRI recordings of subjects during the fear conditioning experiment (AUC: area under curve for BOLD signal in anterior cingulate on CS50 and CS100 trials [left bottom]). The magnitude of the US-elicited fMRI signal (AUC) was negatively correlated with the self-report of UCS expectancy in both dorsolateral PFC and anterior cingulate (anterior cingulate shown in right bottom ). (b) Time course data for fMRI signal in thalamus. fMRI signal was similar during the 10 s CS period (CS onset and duration indicated by grey background) for both CS50 (solid line) and CS100 (dashed line) but responses to the loud noise UCS (UCS onset and duration indicated by black bar) were greater on the CS50 (i.e. UCS unexpected) than CS100 (i.e. UCS expected) trials ((c & d) Rats received auditory CS and shock US presentations. These were either expected (‘Expected’) due to prior CS – US pairings or unexpected due to the absence of such prior pairings (‘Unexpected’). Unexpected CS – US presentations produced significantly greater neuronal activation (as measured by c-Fos expression) in midline thalamic neurons identified as projecting to the dmPFC, compared to expected presentations or CS alone presentations (‘Control’). Black dots in (c) indicate c-Fos positive nuclei; brown dots indicate neurons retrogradely labeled from the dmPFC with the neural tracer cholera toxin B (CTb); blue dots indicate dual labeled CTb/c-Fos positive neurons. Panels A and B are reproduced, with permission, from [82]. Panels C and D are modified with permission from [83].

Box 1 Figure I

Schematic diagram depicting the main areas of the amygdala and its involvement in fear learning and memory.

Box 3 Figure I

Schematic diagram depicting the main anatomical subregions of the PAG.