Circuit regulation along an activity continuum shapes flexible and stable representations during learning

Learning requires neural representations to flexibly incorporate new information while progressively stabilizing to support enduring memories. How neural circuits balance these demands remains unclear. Here, we identify complementary mechanisms supporting representational flexibility and stability in the medial entorhinal cortex (MEC), a key component of the cognitive map for memory-guided navigation. Longitudinal calcium imaging in mice learning novel virtual environments revealed a persistent axis of excitatory neuronal activity associated with distinct representational dynamics. Low-activity neurons exhibited intermittent activation that varied with learning stages and greater learning-dependent representational refinement, flexibly shifting their encoding from landmarks toward reward locations. By contrast, high-activity neurons exhibited more persistent activation and maintained more stable spatial representations. Dopaminergic pre-reward ramping was associated with the emergence of reward encoding in low-activity neurons, whereas interactions with inhibitory interneurons contributed to the stability of high-activity neurons. Notably, interneurons exhibited activity changes opposing those of excitatory neurons during learning, and their transient activation upon novelty exposure delayed excitatory representational remodeling, stabilization, and behavioral learning, suggesting that inhibitory circuits gate learning-dependent representational dynamics. Together, these findings reveal how activity-dependent excitatory neuronal organization and circuit-level regulation enable complementary forms of representational flexibility and stability to coexist, facilitating the incorporation of novel experiences into stable, context-specific cognitive maps that support learning.