Model-based analysis of stability in networks of neurons (PhD)

Neural circuits in all areas of the brain are subject to ongoing remodelling due to the plasticity of individual synapses and neurons. Experience-driven learning, spontaneous fluctuations and homeostatic regulation all contribute to this continuous reconfiguration of neuronal networks. However, despite the malleability of individual cells, both behaviour and function of neural circuits are generally found to be stable and reliable. This suggests that neuronal plasticity is somehow orchestrated across the network, in order not to destabilize the global functioning. To date, however, a mechanism for maintaining such group stability has not been identified. The present work was aimed to address this issue by using long-term in vitro recordings and subsequent model-based analysis of stability.

Cultured rat hippocampal neurons were recorded with high-density multi-electrode arrays over several days, revealing that in vitro preparations exhibit the desired global stability accompanied by local fluctuations. Subsequently, pairwise maximum entropy models were used to characterize activity patterns of groups of neurons at each time point. The particular analytic form of the employed model allowed for a quantification of parameter sensitivity in each group by using the Fisher Information Matrix.
The analysis of obtained matrices has shown that the models exhibit two important properties: sloppiness and sparsity. Together, these properties result in models that are insensitive to changes in many of their parameters, and only sensitive to a few or a combination of a few parameters. A comparison of group activities and model parameters over days further revealed that changes in those are related to sensitivity. In particular, neurons with parameters classified as insensitive changed relatively more than neurons with sensitive parameters. Interestingly, parameter sensitivity was found to be highly correlated with firing rates.

These results suggest that the highly active neurons, through being the most influential on group behaviour and at the same time the most stable over time, are responsible for global stability. At the same time, the remaining majority of less active neurons are free to explore the parameter space and undergo learning, without de-stabilizing the group behaviour.

Related Themes

Related Publications and Presentations

  • Dagmara Panas, Alessandro Maccione, Luca Berdondini, and Matthias Hennig, “What can MaxEnt reveal about high-density recordings and what can high-density recordings reveal about MaxEnt?”, BMC Neuroscience, 2011, 12(Suppl 1).
  • Dagmara Panas, Alessandro Maccione, Luca Berdondini, and Matthias Hennig, “Towards tracking homeostasis on high-density multielectrode arrays”, Bernstein Conference 2012, 2012.
  • Dagmara Panas, Alessandro Maccione, Luca Berdondini, and Matthias Hennig, “Homeostasis in large networks of neurons through the Ising model – do higher order interactions matter?”, Computational Neuroscience (CNS), Paris 2013, 2013.

Related People