An amygdala-cortical circuit for encoding generalized fear memories

Abstract

Generalized learning is a fundamental process observed across species, contexts, and sensory modalities that enables animals to use past experiences to adapt to changing conditions. Evidence suggests that the prefrontal cortex (PFC) extracts general features of an experience that can be used across multiple situations. The anterior cingulate cortex (ACC), a region of the PFC, is implicated in generalized fear responses to novel contexts. However, the ACC's role in encoding contextual information is poorly understood, especially under increased threat intensity that promotes generalization. Here, we show that synaptic plasticity within the ACC and signaling from basolateral amygdala (BLA) inputs during fear learning are necessary for generalized fear responses to novel encountered contexts. The ACC did not encode specific fear to the training context, suggesting this region extracts general features of a threatening experience rather than specific contextual information. Together with our previous work, our results demonstrate that generalized learning about threatening contexts is encoded, in part, within an ascending amygdala-cortical circuit, whereas descending ACC projections to the amygdala drive generalized fear responses during exposure to novel contexts. Our results further demonstrate that schematic learning can occur in the PFC after single-trial learning, a process typically attributed to learning over many repeated learning episodes.

Fig. 1. Arc expression in the ACC is increased during strong context fear conditioning.

A: Timeline for behavior and Arc immunohistochemistry experiment. Mice underwent contextual fear conditioning as described. Sixty minutes after training, mice were perfused, and brain tissue was processed for immunohistochemistry to assess the expression of the plasticity-associated protein Arc. B: Post-shock freezing during acquisition increased after successive presentations of the 5 un-signaled foot shocks (p < 0.0001). C: Quantification of Arc expression in the ACC after strong training, home cage, or immediate shock with the same parameters as strong training (5 shocks, 1 mA). Arc expression was significantly greater in the ACC in mice trained with 5 unsignaled foot shocks (5S) compared to home cage controls (HC) and immediate shock controls (IS) (p < 0.0001). D: Representative images of Arc immunohistochemistry in the ACC. From left to right, a representative image of the 5-shock training, home-cage controls, and immediate shock controls. *p ≤ 0.05; ***p ≤ 0.001; ****p ≤ 0.0001.

Fig. 2. Generalization requires associating a general context representation and threat.

A: Timeline for the alternative context training procedure and the immediate shock procedure. The boxes below the timeline represent the context in which the mice are placed during each part of the experiment. Mice were fear-conditioned in an alternative training context (triangle, C), then exposed to the standard training context (square, A) six hours later. Mice were then tested for fear in a counterbalanced design in the standard training (square, A) and novel contexts (circle, B). They were then tested in the alternative training context (triangle, C). For the immediate shock procedure, mice were placed in the training context (square, A) and immediately received 5 footshocks and were immediately removed from the context. Twenty-four hours later, mice were exposed to the training context for nine minutes. One day later, mice were tested for fear generalization in a novel context (circle, B). B: Acquisition for the alternative context training experiment. Mice display higher freezing in the alt. (alternative, C) training context during training compared to the std (standard, A) training context exposure [t (10) = 10.56, p < 0.0001]. C: Mice did not generalize fear to the novel context (circle, B) when trained using the alternative training context procedure. Freezing was higher during the test in the standard (std, square, A) training context than in the novel (circle, C) context (p = 0.0275), likely due to some transfer of learning during the unpaired context exposure on the acquisition day. There was a significant difference between the standard training context and the alternative training context (p = 0.0132), as well as the novel context and the alternative training context (p < 0.0001). Freezing in the standard training context did not differ between training and the expression test (p = 0.1004). Mice that received immediate shock in the standard training context and tested in a novel context also displayed low freezing, suggesting that generalization is not strictly due to shock exposure without a representation of context. *p ≤ 0.05; ****p ≤ 0.0001.

Fig. 3. Inactivation of NMDAR in the ACC blocks the acquisition of generalized context fear.

A: Timeline of behavioral experiments. B: Acquisition of context fear conditioning during the experiment in which mice were infused with DL-AP5 before training. There were no differences between DL-AP5 and vehicle-treated mice during acquisition (p = 0.7049). C: NMDAR blockade in the ACC during learning significantly attenuates contextual fear generalization. Mice receiving DL-AP5 during training displayed significantly less freezing to the novel context than vehicle-treated mice (p = 0.0008). Freezing to the training context was not different between the groups (p = 0.5155). D: Generalization index for pre-training DL-AP5. AP5-treated mice had a significantly greater difference score compared to vehicle-treated mice, indicating a greater difference between their training context freezing compared to their novel context freezing [t (24) = 3.14, p = 0.0044]. E: Post-training NMDAR blockade in the ACC does not affect context fear generalization or context-specific fear. There were no significant differences in freezing between AP5- and vehicle-treated mice in the novel (p = 0.9629) or training context (p = 0.9719). F: Schematic of cannula placements in the ACC. Placements for pre-training and post-training DL-AP5 experiments are combined into one schematic. G: Post-training inactivation of the ACC with lidocaine blocked context fear generalization but had no effect on context-specific fear. Mice infused with lidocaine immediately after training showed a significant reduction in freezing to the novel context compared to vehicle-treated mice (p = 0.0025). There were no differences in freezing in the training context (p = 0.9531). **p ≤ 0.01; ***p ≤ 0.001.

Fig. 4. Chemogenetic inactivation of the prelimbic cortex during training does not reduce context fear generalization.

A: Timeline of behavioral experiment. Mice were infused with an hM4Di-mCherry or EGFP-expressing control AAV and received injections of CNO before context fear conditioning. Mice were tested for fear responses in the training (square, A) and novel (circle, B) contexts in a counterbalanced design. B: Schematic representation of hM4Di-mCherry expression in the prelimbic cortex (PL). Dark pink represents minimum virus spread, medium pink represents average, and light pink represents maximum virus spread. C: Representative image of PL hM4Di-mCherry expression. D: During acquisition, mice expressing hM4Di or EGFP showed no differences in their post-shock freezing (main effect of treatment; p = 0.7150). E: Mice were tested for fear expression in the training and novel context with 72-h between tests. There were no differences in freezing in the training (p = 0.9970) or novel context (p = 0.9854) between hM4Di- and EGFP-expressing mice. F: Generalization index for PL inactivation. There were no differences in the generalization index between hM4Di-expressing and EGFP -expressing mice (p = 0.8709).

Fig. 5. BLA Inputs to the ACC are necessary to encode context fear generalization.

A: Schematic of viral inactivation strategy. A retrograde AAV expressing cre recombinase was bilaterally injected into the ACC and a cre-dependent AAV expressing hM4Di or mCherry (DIO hM4Di-mCherry or DIO mCherry) was bilaterally injected into the BLA. This enabled specific inactivation of BLA neurons projecting to the ACC. B: Timeline of the behavioral experiment. After viral infusions, mice underwent context fear conditioning and were tested for fear responses in the training (square, A) and novel (circle, B) contexts in a counterbalanced design. C: Representative image of fiber expression in the ACC. A 10X confocal image was acquired. Image shows hM4Di-mCherry fiber expression in the ACC demonstrating fibers from BLA projection neurons. D: Representative image of hM4Di-mCherry expression in the BLA. Images were obtained on a stellaris confocal microscope with a 10x objective (main image). Inset is a 20x image of neurons within the BLA. E: Schematic of hM4Di viral spread in the BLA. Dark pink represents minimum virus spread, medium pink represents average, and light pink represents maximum virus spread. F: Inactivation of BLA-to-ACC circuit during learning. All mice showed increased post-shock freezing with each shock delivery and there were differences between the groups (main effect of treatment; p = 0.1379). G: Inactivation of the BLA-to-ACC circuit during context fear learning is necessary for context fear generalization. BLA-to-ACC circuit inactivation eliminated generalization in the novel context, but specific fear to the training context was unaffected (main effect of context [F (1, 22) = 9.014, p = 0.0066] Sidak’s post hoc, p = 0.0031). H: Generalization index for hM4Di-expressing and mCherry-expressing control mice. hM4Di-expressing mice had greater freezing difference scores between the training and novel context compared to mCherry-expressing controls [t (22) = 2.693, p = 0.0133]. I: Inactivation of BLA-to-ACC circuit did not alter locomotion in the open field. Mice received i.p. injections of CNO (5 mg/kg) 30 min prior to being placed in the open field. We saw no significant difference in distance traveled between hm4Di- and mCherry-expressing mice [t (17) = 0.2473, p = 0.8076]. *p ≤ 0.05; **p ≤ 0.01.

Fig. 6. BLA inputs to the ACC drive context fear generalization under training conditions that normally do not support generalization.

A: Schematic of circuit activation strategy. A retrograde AAV expressing hM3Dq-mCherry or EGFP was bilaterally injected into the ACC. Five weeks later, mice were bilaterally cannulated over the BLA to enable local infusions of CNO in the BLA. One week after cannulations, mice underwent fear conditioning. Infusions of CNO via guide cannula enabled direct activation of BLA neurons that project to the ACC. B: Timeline of behavioral experiments. Mice received a bilateral infusion of CNO into the BLA immediately before undergoing context fear training. They were then tested in the training and novel context in a counterbalanced design. C: Mice were trained with either a three-shock, 0.6 mA protocol (weak training) or a 5 shock, 1.0 mA (strong training). There was a significant difference in freezing to the novel context (p = 0.0233). Mice trained with weak fear conditioning displayed less freezing to the novel context than mice with strong training. Mice that underwent the strong training procedure displayed the expected generalization to the novel context. D: When the BLA-ACC circuit was activated using chemogenetics during training, both groups of mice acquired context fear as expected, with no differences between treatment groups (main effect of shock [F (2.289, 22.89) = 18.00, p < 0.0001]). E: During a fear expression test, hM3Dq-expressing mice froze significantly more in the novel context compared to EGFP-expressing mice (main effect of context [F (1, 10) = 43.02, p < 0.0001], Sidak’s post-hoc (p = 0.0410)). F: The generalization index indicated a greater difference in freezing between the training context and novel context in EGFP-expressing mice compared to hM3Dq-expressing mice [t (10) = 3.089, p = 0.0115]. G: Schematic of cannula placements in the BLA and hM3Dq-mCherry in the BLA. *p ≤ 0.05.