Learned and inferred valence arise from interactions between stable and dynamic subnetworks

Adaptive behavior requires assigning emotional value to sensory cues and inferring valence for novel stimuli to guide appropriate generalization. The prelimbic cortex (PL) is critical for threat expression and discrimination, yet its neuronal ensembles undergo pronounced turnover over time. How stable memory representations emerge from such network dynamism--and how they support inference to previously unexperienced stimuli--remains unresolved and central to debates on systems consolidation and the neural basis of generalization. Using longitudinal calcium imaging in freely moving mice, we tracked PL population activity for 30 days during two opposing versions of tone-discriminative fear learning and probed responses to conditioned and novel tones at recent and remote time points. Despite substantial ensemble reorganization, PL population dynamics reliably encoded graded emotional valence. Stimulus-evoked population similarity scaled precisely with behavioral generalization, and consistent population states emerged only for tones associated with shock or those eliciting strong generalized freezing, indicating that population-level similarity predicts inferred threat. Network analyses identified two functionally distinct subnetworks. Dynamic tone-selective ensembles encoded sensory features independent of learning and exhibited substantial turnover. In contrast, a valence-coding subnetwork whose neurons responded to all frequencies, integrated learned and inferred emotional value along a graded axis. Strikingly, only these graded valence neurons preserved cellular identity and response structure across time. These findings reveal that persistent valence-encoding subnetworks form a stable scaffold embedded within dynamic cortical ensembles. This architecture reconciles cortical turnover with long-term memory stability and provides a circuit-level mechanism for maintaining the emotional "gist" of experience while enabling flexible generalization.