Neural activity in awake organisms shows widespread, spatiotemporally diverse correlations with behavioural and physiological measurements1–4. We propose that this covariation reflects in part the structured, nonlinear dynamics of an underlying arousal-related process that organizes brain-wide and body-wide physiology on the timescale of seconds. By framing this interpretation within dynamical systems theory, we arrive at a surprising prediction: a single, scalar measurement of arousal (for example, pupil diameter) should suffice to reconstruct the continuous evolution of multidimensional, spatiotemporal measurements of large-scale brain physiology. Here, to test this hypothesis, we perform multimodal cortex-wide optical imaging5 and behavioural monitoring in awake mice. We demonstrate that the seconds-scale spatiotemporal dynamics of neuronal calcium, metabolism and brain blood oxygen can be accurately and parsimoniously modelled from a low-dimensional, nonlinear manifold reconstructed from a time delay embedding6,7 of pupil diameter. Extending this framework to behavioural and electrophysiological measurements from the Allen Brain Observatory8, we demonstrate the ability to integrate diverse experimental data into a unified generative model via mappings from a shared arousal manifold. Our results support the hypothesis9 that spontaneous, spatially structured fluctuations in brain-wide physiology on timescales of seconds—widely interpreted to reflect regionally specific neural communication10,11—are in large part expressions of a low-dimensional, organism-wide dynamical system. In turn, reframing arousal itself as a latent dynamical system offers a new perspective on fluctuations in brain, body and behaviour observed across modalities, contexts and scales.
