Reconstruction and control of endocrine–metabolic dynamics

Mammalian physiology control networks integrate fast neuromodulators with slow endocrine signals, yet we lack a quantitative, control-oriented view of these dynamics. We developed a method to monitor chemical and behavioral states in freely moving mice that combines chronic jugular microdialysis with video-tracked locomotion. We measured hundreds of amines and small peptides every 7.5 minutes for ~8 hrs in three mice and asked three questions:

(1) How are interactions across molecules and timescales organized?

(2) What minimal set of signals makes the system observable?

(3) Which nodes best modulate behavior-relevant states?

We fit biologically constrained, sparse vector autoregressive models with exogenous inputs (VARX) across animals to infer directed, lagged molecular networks and identify where predictive information about locomotion resides. Using the Sparse Identification of Nonlinear Dynamics (SINDy) framework, we extracted candidate ODE motifs and computed empirical Gramians to assess observability and controllability. Across animals, models recovered known physiological relationships, and a compact panel of analytes predicted locomotion minutes in advance. These results frame endocrine-metabolic dynamics as an observable, controllable system and suggest testable design principles for sensing and perturbation.