STABLE NETWORK STATISTICS EMERGE FROM DYNAMIC CELLULAR COMPOSITION IN MEC AND CA3 POPULATION CODES

To navigate the world, mammals rely on an internal representation of space, a cognitive map that is able to generalize spatial metrics across environments while distinguishing distinct experiences. This balance is thought to emerge from complementary population codes in the medial entorhinal cortex (MEC) and CA3. MEC provides a structured, low-dimensional population code where neuronal responses covary across contexts, supporting generalization of geometric relationships. In contrast, CA3 exploits a high-dimensional population code with decorrelated representations across environments, thereby enabling episodic specificity. A key question is whether differences in population code dimensionality reflect intrinsic circuit constraints, or whether they are shaped by task engagement, sensory structure, or experience. To address these questions, we used longitudinal calcium imaging of MEC and CA3 neuronal populations while mice explored two distinct virtual reality environments and a “no VR” condition that minimizes sensory cues and task structure upon self-paced locomotion. We found that in MEC, population activity showed conserved correlation patterns across all conditions, stemming from a highly interconnected subnetwork characterized by neurons with high functional connectivity (high-degree cells). In contrast, CA3 exhibited only sparse, isolated correlated pairs embedded within a largely disconnected network. Despite stability in correlation structure, the specific identities of correlated neuronal pairs turned over rapidly and to a similar extent in both areas, with few pairs persisting beyond two consecutive sessions. Our results therefore suggest that in MEC and CA3, population code dimensionality emerges from intrinsic circuit constraints. These constraints persist despite experience-driven turnover in cellular composition, maintaining stable network statistics.