Amygdala Inhibitory Circuits Regulate Associative Fear Conditioning

Only 20% of BLA neurons are GABAergic interneurons (Figure 1A), yet they tightly control the activity of the remaining large population of principal neurons (, , , ). In contrast to excitatory principal neurons, GABAergic interneurons have thin, aspiny dendrites, and their axonal arbor is usually restricted to the BLA (, ). Intriguingly, a small subset of GABAergic BLA neurons with yet unknown function have long-range projections to remote regions including the basal forebrain or entorhinal cortex (, , ). Different classes of BLA interneurons are typically defined by their expression pattern of calcium binding proteins or neuropeptides, which correlates with neuronal morphology, postsynaptic targets, and cellular physiology (). These groups are remarkably similar to interneuron populations observed in other brain areas such as the cortex or hippocampus ().

Based on earlier immunohistochemical studies, two major, nonoverlapping groups of BLA interneurons can be defined by the expression of the calcium binding proteins calbindin (CB) and calretinin. CB interneurons account for about 60% of BLA interneurons and can be further subdivided into parvalbumin (PV)-, somatostatin (SOM)-, neuropeptide Y-, and cholecystokinin (CCK)-expressing cells (, , , ) (Figure 1). Substantial overlap of SOM and neuropeptide Y expression has been reported, while the other groups are largely separate (, , ). Approximately 20% of BLA interneurons are calretinin positive, many of which coexpress CCK or vasoactive intestinal peptide (VIP), both individually and with some overlap (, ).

During auditory fear conditioning, CS and US information converges in the LA, which is a major site of neuronal plasticity (, , , , ). Behavioral changes upon fear conditioning are thought to rely on synaptic plasticity at sensory afferents onto principal neurons, causing enhanced neuronal responses to the auditory CS (, , , , , ). According to the classical Hebbian learning model, these synaptic changes are induced by converging and temporally coincident CS and US inputs to the same postsynaptic cells (, ). While the auditory CS alone would elicit only a weak postsynaptic response, the US induces a strong postsynaptic depolarization, thereby triggering associative, activity-dependent synaptic plasticity mechanisms leading to the induction of long-term potentiation (LTP) at coactive CS inputs. This suggests that the US could act as a teaching signal, instructing associative plasticity at CS-activated synapses (). Accordingly, pairing an auditory CS with artificial depolarization of LA neurons in vivo induces, albeit only weak, fear learning (). Consistent with this finding, additional inputs including neuromodulatory transmitters have been implicated (, , , , ). A recent study using a deep brain imaging approach of large neuronal ensembles in the BLA found that only a fraction of principal neurons with potentiated CS responses displayed somatic US responses, and that fear conditioning–induced changes in the encoding of the CS involved both up- and downregulation of CS responses (). Together, these studies support more complex scenarios involving additional learning rules and/or plasticity of inhibitory circuit elements. Importantly, understanding the function of cell type– and compartment-specific inhibitory circuits, and their dynamic regulation through synaptic and neuromodulatory mechanisms, will be key for obtaining a mechanistic model of neuronal circuit plasticity underlying associative fear conditioning.

Repeated presentations of a conditioned but nonreinforced auditory cue will gradually lead to the extinction of a conditioned fear response. Rather than forgetting, fear extinction creates a new inhibitory memory competing with the original CS-US association (, , ). The CS responsiveness of BLA principal neurons changes during extinction training, and similar to the so-called fear or CS-up neurons, which specifically increase their CS responses during fear conditioning, extinction neurons have been reported, which display enhanced activity to the auditory cue after extinction training (, , , , ). These two functional classes of BLA principal neurons preferentially project to the prelimbic or infralimbic (IL) subdivision of the medial prefrontal cortex, respectively, two brain regions with opposite roles during fear extinction (, , ). The acquisition of fear extinction thus can be regulated by the balance of activity between distinct output pathways targeting the different medial prefrontal cortex subdivisions ().

Inhibitory cells, particularly cannabinoid receptor type 1–expressing CCK_L interneurons, have been proposed to mediate the switch between high and low fear states and the concomitant changes in neuronal activity in the BLA during extinction learning (, , , , ). Although CCK_L interneurons connect to an equal extent with prelimbic- or IL-projecting BA principal cells, their synapses display target-specific short-term plasticity (). CCK_L-synapses onto IL-projecting cells display stronger depolarization-induced suppression of inhibition, an endocannabinoid-dependent form of short-term plasticity (, ). This effect was attributed to differential expression levels of the postsynaptic cannabinoid-synthesizing enzyme diacylglycerol lipase α in IL- and prelimbic-projecting cell populations () (Figure 2D). Therefore, it is conceivable that during extinction, when IL-projecting cells are strongly activated, CCK_L inhibitory input is rapidly suppressed, which may enhance the contrast between BLA fear and extinction pathways and promote rapid behavioral adaptations (Figure 2E, F). This endocannabinoid-mediated disinhibition could further provide a time window for integration of excitatory stimuli and synaptic plasticity in extinction cells.

Furthermore, plasticity of inhibitory interneurons has been suggested as a key mechanism in fear extinction (). In contrast to fear learning (), excitatory drive onto distinct subtypes of LA interneurons may be increased by LTP following fear extinction (, , ). In accordance with a shift in excitability of BLA fear and extinction pathways, it has been demonstrated that perisomatic PV⁺ terminals increase around then silent BA fear cells after extinction training (). Correspondingly, ablation of axoaxonic inhibitory synapses in the BLA impairs fear extinction (). This implies that PV-mediated silencing of fear neurons might be needed for the expression of extinction.

In addition to BLA interneurons, the medial cluster of ITCs (mITCs) provides an inhibitory gate for the information flow between the BLA and CEA, thereby regulating fear extinction. Extinction training is associated with increased expression of immediate early genes in mITCs (, ), and ablation of these cells interferes with extinction retrieval (). Furthermore, extinction leads to a potentiation of excitatory BA inputs to mITCs, which in turn increases inhibition onto CEA output neurons to reduce fear responses (). This effect depends on activity in the IL subdivision of the medial prefrontal cortex, which can directly activate cells in the mITC cluster in vivo (, , ).

Taken together, these data demonstrate the importance of BLA inhibitory circuit elements to create a competitive extinction memory suppressing fear responses. Nevertheless, in vivo studies addressing the role of distinct BLA interneurons or ITCs during and after extinction are still lacking. Therefore, further efforts investigating the precise activity patterns and recruitment of different BLA interneuron groups and ITCs with both recordings and perturbations in freely moving animals will be necessary to delineate their individual contributions to the acquisition and expression of extinction memories.

The increasing numbers of cell type-specific genetically modified mouse lines, together with technical advances in optogenetics as well as tracing and recording techniques, have led to major progress in understanding the role of defined inhibitory microcircuits in the amygdala (, , ). However, unequivocal identification and recording of neuronal subtypes in deep tissues during complex behavioral tasks is typically low yield. Few studies managed to reliable record and analyze BLA interneuron activity in vivo, which could reveal first general principles of the importance of inhibitory mechanisms (, ). Nevertheless, even within a genetically defined group, individual cells exhibit an intriguing response diversity to different environmental stimuli. This presses for future studies that further investigate the functional role of interneuron classes upon learning, ideally from large populations in single animals. Furthermore, it is important to emphasize that conventional optogenetic and pharmacogenetic manipulations of genetically defined cell types override this response diversity and can thus only provide limited conclusions about causal links between neuronal activity and behavioral adaptations.

Notably, the diversity observed in interneuron populations resembles principal neuron activity. Fear acquisition induces both upregulation and downregulation of CS responses in LA principal cells, and both mechanisms are equally important to increase the similarity of CS representation to the US on the ensemble level in support of learning (). Disinhibition of excitatory principal neurons as described above is a popular explanation for Hebbian plasticity models of learning (), while circuit mechanisms that lead to the downregulation of CS responses, potentially involving synaptic depression and/or plasticity of inhibitory circuits, are less well defined. Furthermore, simple disinhibitory models have been challenged recently, as strong reciprocal connectivity of PV, SOM, and VIP interneurons has been described in neocortical areas (, ). Mutually inhibitory interneuron groups can change their response patterns depending on stimulus or context, leading to differential activity in principal neuron ensembles (, , ). This reciprocal connectivity most likely also exists in BLA inhibitory circuits (, , ) and could be important to regulate network activity in different behavioral states or contexts.

Finally, the observed diversity of in vivo activity patterns within a genetically defined interneuron population calls for a more functionally driven definition of an interneuron subtype. Indeed, widely used marker proteins for transgenic mouse models such as PV, SOM, or VIP show little or no overlap in the BLA (, ), and at least partially correlate with postsynaptic targets and activity patterns. However, within the same group, further subtypes with specific subcellular targeting of postsynaptic cells or with different short-term dynamics of inhibitory synapses exist (, , , ), suggesting a more divergent spatial and temporal control of postsynaptic activity. Furthermore, it is unclear if the same cells target principal neurons and other groups of interneurons, or even different projection neurons as previously shown in other brain areas (, ). It would be crucial to test whether specific in vivo activity patterns could be assigned to distinct subtypes based on morphological features or afferent and efferent connectivity. This can now be studied with the help of additional marker proteins in combination with novel intersectional genetic tools such as Cre and Flp recombinase–driven expression systems (, , , ).