A FRAMEWORK FOR CLOSED-LOOP ALL-OPTICAL CONTROL OF CORTEX-WIDE NEURAL DYNAMICS

While characterising neural population trajectories has provided insights into population codes and their correlation with behaviour, establishing causal roles requires closed-loop control to steer neural activity along desired manifolds. However, implementing closed-loop control of the brain requires a perturbational framework for precise manipulation of neural circuits incorporating cross-region interactions.
We combined experimental and theoretical approaches to enable closed-loop control of neural dynamics across the mouse dorsal cortex. First, we developed an all-optical system combining widefield calcium imaging and patterned optogenetic inhibition in freely behaving mice. Mapping cortex-wide responses to local perturbations of each region generated a cortical ‘perturbome’, a causal matrix substantially different from correlation-based functional connectivity, revealing circuit organisation invisible to observational measures. Second, we developed a linear dynamical system model which achieved high accuracy in predicting cortex-wide perturbational responses. This demonstrates that complex activity patterns following local perturbations can be captured by principled dynamical models. Third, using optimal control theory, we identified the controllability manifold: the subspace of maximally controllable trajectories. This enables computing multidimensional stimulation patterns to steer brain dynamics along any desired trajectory within this manifold, but not in manifolds with low predicted controllability. Preliminary experimental results demonstrate successful closed-loop control of single brain areas, paving the way for real-time control of multi-dimensional trajectories. Our work provides a causal, perturbational approach to cortex-wide neural dynamics beyond observational models. Together, this work has important implications for closed-loop brain stimulation and control.