Speed-dependent place- and time-field shifts do not require explicit temporal coding

Place and time cells are widely thought to provide complementary representations of spatial location and elapsed time in the hippocampus. Recent experiments reported CA1 neurons whose place and time fields shift systematically with running speed, suggesting that representations of space and time are integrated and compete within a common neural code. Here we show that speed-dependent place- and time-field shifts can arise without explicit temporal coding. Using a hierarchy of computational models, we demonstrate that these phenomena emerge when the velocity of the internal estimate of position becomes progressively less sensitive to increases in the animal's speed. Recurrent neural networks trained exclusively for path integration spontaneously developed this behavior. Their analysis revealed a circuit mechanism in which weakly direction-selective neurons stabilize the activity bump at low speeds and progressively release this brake as speed increases. Finally, recurrent networks trained to jointly encode position and elapsed time distinguished genuine spatiotemporal representations from apparent temporal tuning generated by the task-induced correlation between space and time. Together, our results provide an alternative interpretation of speed-dependent place- and time-field shifts, identify a computational mechanism that extends continuous bump attractor models, and generate experimentally testable predictions.