The Representational Geometry of Emotional States in Basolateral Amygdala

Why It Matters

The findings reveal that complex emotional information can be decoded from mixed‑selectivity ensembles, reshaping theories of amygdala function and offering new targets for treating affective disorders.

Key Takeaways

• Mixed-selectivity neurons dominate BLA encoding of emotion
• Population-level geometry enables specialized readouts without single-variable neurons
• BLA inhibition reduces aversive tremble frequency and ingress magnitude
• Decoding of valence, identity, tremble, and ingress exceeds 90% accuracy
• Tremble epochs align with 4–8 Hz respiratory cycles

Pulse Analysis

The basolateral amygdala has long been viewed as a valence hub, where distinct neuronal ensembles drive approach or avoidance behaviors. Classic lesion and optogenetic studies established this binary framework, yet they left unanswered how the amygdala represents the richer tapestry of emotional states that go beyond simple positive‑negative labels. Recent advances in high‑speed calcium imaging and behavioral assays now allow scientists to probe the fine‑grained dynamics of BLA circuits in real time, opening a window onto the neural geometry that underlies emotional complexity.

In the new study, mice were trained on a Pavlovian odor‑shock paradigm and then tested in a virtual burrow that captured tremble (freezing‑like) and ingress (flight‑like) responses with millimeter precision. Simultaneous two‑photon imaging revealed that most BLA neurons responded to multiple variables—valence, odor identity, and the animal's behavioral state—exhibiting mixed selectivity. Despite this apparent ambiguity at the single‑cell level, population analyses showed that the ensemble’s firing‑rate space organized into distinct geometrical patterns, enabling specialized readouts for each variable. Linear classifiers decoded valence, stimulus identity, tremble, and ingress with accuracies ranging from 90% to 100%, and decoding performance saturated after sampling just 25‑50% of the recorded cells, indicating a sparse yet efficient code.

These results have broad implications for computational neuroscience and psychiatric research. The ability of mixed‑selectivity populations to generate unambiguous readouts suggests a flexible neural substrate for emotional abstraction, potentially explaining how the brain generalizes fear to novel cues. Moreover, the demonstrated impact of BLA inhibition on specific defensive behaviors points to circuit‑level interventions that could modulate maladaptive emotional responses in anxiety and trauma‑related disorders. Future work will likely explore how downstream structures integrate these geometric codes and whether similar principles apply to human amygdala function.