BALANCING STABILITY AND FLEXIBILITY: DYNAMICAL SIGNATURES OF LEARNING IN IN-VITRO NEURONAL NETWORKS

CL1 is a novel system that bridges biological intelligence and adaptive neuronal traits by integrating in-vitro neuronal networks with in-silico computational elements via micro-electrode arrays (MEAs). These neuronal ensembles exhibit self-organized adaptive intelligence in dynamic gaming environments through closed-loop stimulation and recording, yet the network dynamics underlying this real-time learning remain underexplored.
We inferred pairwise causal interactions between channels using Granger causality, reconstructing connectivity networks from statistically significant links and identifying influential nodes via outgoing and incoming connections. To probe dynamics, we reconstructed the phase space of multichannel spiking activity using state-space methods, selecting embedding dimensions via false nearest neighbors and time delays via mutual information, and analyzed temporal structure using recurrence plots.
We analyzed 45-minute spiking recordings from 23 neuronal cultures (111 rest and 133 gameplay) and observed distinct dynamics between influential and influenced nodes. Gameplay showed higher recurrence (RR) and determinism (DET) than rest, while influenced nodes exhibited lower RR and more negative Lyapunov exponents, indicating more ordered dynamics farther from the edge of chaos. In contrast, influential nodes showed higher RR and small negative Lyapunov exponents, consistent with recurrent dynamics near the edge of chaos.
Our results reveal a functional dichotomy in in-vitro neuronal networks, where influential channels operate near the edge of chaos with cyclic dynamics that drive adaptability, while influenced channels remain more ordered and stable. This interplay between near-chaotic drivers and stable receivers enables neuronal cultures to balance flexibility and robustness, shedding light on how coordinated dynamics support adaptive behavior.