Encoding of contextual fear memory in hippocampal-amygdala circuit

The study addresses how contextual fear memories are encoded within the hippocampal-amygdala circuit. Prior work established that the hippocampus (particularly ventral CA1, vCA1) conveys contextual representations to the amygdala and that the pathway is critical for retrieval of contextual fear memory. However, whether and how associative learning induces synapse-specific changes in this circuit remained unknown. The authors hypothesized that fear memory associated with a particular context is encoded by selective strengthening (input-specific synaptic potentiation) of hippocampal vCA1 inputs that convey that context to a subset of basal amygdala (BA) neurons, enabling selective fear responses to the threat-predictive context. This question is important for understanding the circuit and synaptic basis of adaptive fear learning and the mechanisms that confer stimulus/context specificity.

• Contextual fear learning requires coordinated hippocampus-amygdala activity, and vCA1 neurons project monosynaptically to the amygdala and can drive defensive behaviors.
• The vCA1–amygdala pathway has been implicated in retrieval of contextual fear memory, but encoding mechanisms at these synapses had not been tested.
• Memory engram cells have been identified in hippocampus and amygdala, but how hippocampal engrams connect to amygdala engrams and how synaptic strengths are modified during contextual fear learning remained unclear.
• Prior studies showed input-specific LTP in amygdala for discriminative cue-fear learning and learning-induced plasticity in hippocampal circuits, suggesting sparse, synapse-specific encoding might generalize to contextual fear circuitry.

• Animals and genetics: Heterozygous Fos-CreERT2, Arc-CreERT2, Fos-tTA/Fos-shGFP, and double transgenic Fos-CreERT2 × ROSA-LSL-tdTomato or Fos-CreERT2 × Fos-tTA mice were used to enable activity-dependent neuronal labeling. Both sexes, 6–8 weeks old.
• Viral tools and tracers: AAVs for Cre-dependent reporters/effectors (DIO-eYFP, DIO-mCherry), hM4Di-mCherry for chemogenetic silencing, ChR2/Chronos for optogenetics, TRE-driven constructs for tTA-dependent labeling, CAV2-Cre for retrograde Cre delivery from BA, HSV-mCherry for retrograde labeling, and glycoprotein-deleted EnvA-AG-RV-mCherry for monosynaptic retrograde trans-synaptic tracing from BA starter cells expressing TVA-G.
• Anatomical mapping: Anterograde labeling of vCA1 axons in amygdala; retrograde labeling from BA to identify vCA1:BA projectors; c-Fos immunohistochemistry to identify neurons active during context exposure, fear conditioning, or recall.
• Activity-dependent labeling: Context exposure after tamoxifen/4-OHT or off-Dox windows to permanently tag vCA1 neurons active in a given context (Context A or B) and BA neurons active during fear conditioning (BA fear neurons). Multiple labeling sessions were used to boost ChR2 expression.
• Behavioral paradigms: Single-trial contextual fear conditioning; discriminative contextual fear conditioning across days with Context A (shocked) and B (non-shocked); optogenetic pairing (20 Hz vCA1:BA projector stimulation) with footshock; habituation and photostimulation tests; chemogenetic silencing during training with CNO.
• Chemogenetics: hM4Di expression in vCA1:BA projectors via CAV2-Cre or in context-specific vCA1 ensembles via Arc-CreERT2; CNO administration before conditioning to silence.
• Optogenetics: Chronos/ChR2 expression in vCA1:BA projectors or in context-specific vCA1 ensembles; in vivo 20 Hz blue light stimulation paired or unpaired with shocks; in slices, photostimulation of ChR2+ vCA1 axons in BA.
• Electrophysiology: Whole-cell patch-clamp recordings from BA principal neurons in acute slices. Measured AMPAR EPSCs at −80 mV and NMDAR EPSCs at +40 mV to compute AMPA/NMDA ratios. Additional measures: paired-pulse ratio (PPR), progressive block of NMDAR EPSCs by MK-801 to assess presynaptic release, TTX + 4-AP to isolate monosynaptic inputs.
• Pharmacology: Anisomycin post-training to induce retrograde amnesia and block protein synthesis-dependent consolidation; MK-801 pre-training to block NMDARs during learning.
• Trans-synaptic tracing: In Arc-CreERT2 mice, BA neurons active during FC or home cage were labeled with TVA-G-GFP; after RV-mCherry injection, quantified vCA1 neurons monosynaptically projecting to these BA populations and their c-Fos activation during context exposure.
• Statistics: ANOVA (including repeated measures), t-tests, Kruskal–Wallis for non-normal data, Pearson correlations; significance at p<0.05.

• vCA1→BA activity is required for acquisition of contextual fear: Chemogenetic silencing of vCA1:BA projectors (hM4Di + CNO) during training reduced freezing 24 h later compared with vehicle; no effect in mCherry controls (two-way ANOVA interaction p<0.05).
• Artificial CS from vCA1:BA projectors: Pairing 20 Hz optogenetic activation of vCA1:BA projectors with footshocks produced photostimulation-elicited freezing in a neutral context during tests in Chronos mice but not eYFP controls; unpaired stimulation did not produce freezing (two-way ANOVA, group × session p<0.01; paired vs unpaired, test p=0.39 in unpaired).
• Context-specific vCA1 ensembles identified: Activity-dependent labeling tagged a small subset of vCA1:BA projectors (e.g., ~4.9% of mCherry+ vCA1:BA projectors labeled by context exposure). Context re-exposure preferentially reactivated the same ensemble (higher c-Fos in A–A vs A–B).
• Sparse, heterogeneous inputs to BA: Photostimulation of context-specific vCA1 axons (Context A labeling) evoked glutamatergic EPSCs in BA with heterogeneous amplitudes; robust inputs present only in a subset of BA neurons. Monosynaptic nature confirmed with TTX block and 4-AP rescue.
• Learning induces input- and context-specific synaptic potentiation: After discriminative FC where Context A predicted shocks, AMPA/NMDA ratio in Context A vCA1→BA synapses increased vs no-shock controls (p=0.024). Only ~11.5% of BA neurons in FC exceeded NS mean + 2 SDs, indicating potentiation in a subpopulation. No change in Context B pathway in discriminators (p=0.68). Across FC mice including generalizers, Context B AMPA/NMDA correlated positively with freezing in B (r=0.71) and negatively with discrimination index (r=0.69).
• No global vCA1→BA potentiation: With global vCA1 ChR2/Chronos expression, AMPA/NMDA did not differ between FC and NS groups; PPR and MK-801 block rates also unchanged.
• Potentiation targets BA fear neurons: In mice with Context A vCA1 labeling and BA fear neuron labeling, BA fear neurons (tdTomato+) showed higher AMPA/NMDA ratios and larger AMPAR EPSCs than neighboring unlabeled BA neurons; no differences in intrinsic excitability metrics 5 days post-conditioning. Random vCA1 inputs to BA fear neurons (global ChR2) did not show potentiation.
• Consolidation blockade prevents synaptic potentiation: Post-training anisomycin reduced freezing (retrograde amnesia) and abolished the difference in AMPA/NMDA between BA fear neurons and neighbors; across mice, the tdT+–tdT− AMPA/NMDA difference correlated with freezing during recall (r=0.75, p=0.007). Pre-training NMDAR blockade (MK-801) also blocked acquisition and vCA1→BA strengthening (Supplementary).
• BA fear neurons preferentially sample context-specific vCA1 inputs: Rabies tracing from BA fear neurons (anisomycin used to minimize learning-induced plasticity confounds) showed that vCA1 neurons monosynaptically connected to BA fear neurons were more likely to be c-Fos+ upon Context A exposure than those connected to BA neurons active in home cage (c-Fos+ proportion significantly higher, p=0.001). Electrophysiology under anisomycin indicated tdT+ BA fear neurons had larger monosynaptic EPSCs from Context A vCA1 inputs than tdT− neighbors (repeated measures two-way ANOVA, p<0.001), implying more convergent context-specific vCA1 inputs.
• Necessity of context-specific vCA1 activity: Silencing vCA1 neurons labeled in Context A (hM4Di + CNO) during conditioning reduced freezing vs vehicle; no effect when silencing neurons labeled in irrelevant Context B. Silencing Context A vCA1 ensemble during conditioning also eliminated potentiation differences between BA fear and neighboring BA neurons (AMPA/NMDA, p=1.00), linking learning impairment to lack of synaptic strengthening.

The findings support a Hebbian, synapse-specific encoding mechanism for contextual fear memory within the vCA1→BA circuit. When a particular context predicts danger, vCA1 neurons that encode that context provide inputs to a subset of BA neurons that are engaged during fear conditioning. Learning selectively strengthens synapses at the intersection of these presynaptic context-encoding vCA1 inputs and postsynaptic BA fear neurons, enabling efficient and selective activation of the amygdala during recall in the threat-predictive context. This input specificity explains how fear responses remain selective to relevant contexts. Additionally, BA neurons recruited into the memory trace appear to receive more context-specific vCA1 inputs than other BA neurons, potentially biasing their allocation to the engram. Pharmacological disruptions of consolidation (anisomycin) or acquisition (MK-801) prevented both behavioral memory and synaptic potentiation in the identified pathway, reinforcing the functional link between synaptic LTP and memory formation. The approach also demonstrates that examining functionally defined synapse populations can reveal sparse, learning-induced plasticity that would be missed by random sampling of synapses. The results align with broader models in which specific input ensembles and postsynaptic cell selection (via input convergence and transient excitability states) determine engram allocation and retrieval efficacy.

This study demonstrates that contextual fear memory is encoded by selective synaptic strengthening of vCA1 hippocampal inputs conveying context-specific information onto a subset of BA fear neurons. The authors provide converging behavioral, optogenetic, chemogenetic, electrophysiological, and tracing evidence that: (1) vCA1→BA activity is necessary for learning; (2) pairing vCA1:BA projector activation with shock is sufficient to form a fear association; (3) discriminative learning potentiates synapses specifically in the context-relevant vCA1→BA pathway and specifically onto BA fear neurons; and (4) preventing consolidation or silencing context-specific vCA1 ensembles blocks both memory and potentiation. These results support a Hebbian, input- and cell-specific LTP mechanism for encoding selective fear memories. Future work could delineate molecular mechanisms underlying this LTP, define the time course of synaptic changes, determine contributions of interneuron circuits and other projection-defined BA subpopulations, and assess generalization to other forms of associative learning and to disease models of maladaptive fear.

• Activity-dependent labeling has an extended temporal window and potential background labeling; only a subset of labeled neurons reflect those active exclusively during the intended events.
• Trans-synaptic tracing and input abundance findings are correlational; while anisomycin minimized plasticity during tracing, causal manipulations of input number were not performed.
• Intrinsic excitability of BA fear neurons was assessed 5 days post-conditioning; transient excitability changes during/soon after learning may have been missed.
• Experiments were conducted in mice with ex vivo slice electrophysiology; generalization across species and in vivo synaptic dynamics were not directly measured.
• Potentiation was assessed primarily via AMPA/NMDA ratio and EPSC amplitudes; complementary structural measures (e.g., spine changes) were not reported here.