A shared manifold for scalable temporal representations in self-paced timing

Abstract

Adaptive behaviour often relies on tracking the passage of time, yet how distinct brain regions generate coherent temporal representations remains unclear. Using a self-paced interval timing task in mice, we show that heterogeneous single-neuron temporal firing profiles distributed across regions are organized within a shared ring manifold. Within this low-dimensional space, population activity evolves along a common trajectory across different intervals and encodes elapsed time in a relative reference frame. Different durations are not represented by separate neural states, but by modulation of traversal speed. These scalable dynamics arise from coordinated population-wide co-scaling of single-neuron activity and support trial-by-trial adjustments in timing behaviour. A cross-regional assembly of start neurons predicts, at interval onset, upcoming waiting duration and behavioural adjustments, linking initial population states to trajectory evolution. Together, these findings identify population traversal of a shared activity manifold as a mechanism for scalable temporal representation across brain regions.