THE ECTORHINAL CORTEX TRANSMITS ATTENTION-GATED TEACHING SIGNALS TO THE LATERAL PERIAQUEDUCTAL GRAY TO DRIVE FEAR LEARNING

Adaptive behavior relies on teaching signals, such as prediction error (PE), to guide learning, yet how cortical circuits compute and selectively route these signals to downstream effectors remains unclear. Here, we investigate the ectorhinal cortex (ECT) as a source of such signals for the lateral periaqueductal gray (lPAG), a key center for fear expression. Combining circuit mapping, in vivo electrophysiology, two-photon imaging, and optogenetic manipulations in mice, we first establish that the ECT→lPAG projection is necessary for fear conditioning. We demonstrate that ECT computes a canonical PE signal, with responses to conditioned stimuli (CS) potentiating while those to unconditioned stimuli (US) depress, and transmits this information to lPAG. Crucially, single-cell imaging combined with an attention-gated reinforcement learning model (the PH-RL model) reveals a more sophisticated mechanism: ECT harbors a multiplexed representation, dissociating PE, attentional weight (α), and expected value (V). Remarkably, the lPAG-projecting subpopulation is highly enriched for value-encoding neurons, indicating a selective output channel for action-relevant information. With the manipulation of the PE signal using a blocking paradigm, neural dynamics of ECT precisely match the PH-RL model predictions. Our findings uncover a fundamental principle of cortical computation: integrating multiple learning variables and outputting to selective, target-dependent routes. This provides a unified circuit and computational mechanism governing how instructive signals drive defensive behavior.