Final Report Summary - CIFINE (Controlling information flow in multi-layered neuronal networks.) The first highlight of my time as a Marie Curie Reintegration Fellow at the EPFL is the discovery of an intriguingly simple, Hebbian plasticity rule on inhibitory synapses that leads to robust and self-organized balance of excitation and inhibition. It requires virtually no fine-tuning and captures a surprising number of recent experimental findings. It lead to a first author publication in Science (Vogels et al., 2011). The precision of the learned balance depends on the degree of correlation between the excitatory and the inhibitory inputs to the cell, ranging from a global balance in the absence of correlated inputs to a detailed balance for strong correlations. The phenomenon is robust to the shape of the learning rule, as long as it obeys two fundamental requirements: postsynaptic activity must potentiate activated inhibitory synapses, while in the absence of postsynaptic firing inhibitory synapses must decay. We discussed its relevance with regard to biological plausible rules in a review in Frontiers of Neural Circuits (Vogels et al. 2013)Inhibitory synaptic plasticity, through its self-tuning qualities, opened many avenues of research. With inhibitory synaptic connectivity that maintains its own strength, it is now possible to build more complex signal processing chains that have been touched upon in the literature, but could not be build due to complex tuning requirements. It will be possible to build balanced feed forward chains with intertwining pathways, compare their performance to experimental results and scrutinize them analytically, both of which I have begun to work on in my new position at Oxford University.More excitingly, the discovery of inhibitory synaptic plasticity also led to the second, unexpected highlight of my time at the EPFL, the generation of network architectures that can autonomously perform orchestrated complex motor dynamics. This work is described in two publications, "Nonnormal amplification in random balanced neuronal networks” by Hennequin, Vogels and Gerstner, published in Physics Review E and “Rich transient dynamics in optimally stabilized cortical network models”, by Hennequin and Vogels/Gerstner (co-senior authors), published in NEURON. In this work, we explored the dynamics and properties of recurrent, optimally inhibition stabilized neuronal networks. Quiescent in the idle state, these networks can engage in complex transient dynamics of large amplitude that mirror experimental phenomena and suggest inhibitory control of complex excitatory recurrence as a generic organizational principle in cortex. An additional branch of work worth mentioning is work that resulted in the paper“Connection-type specific biases make random network models consistent with cortical recordings”, by Tomm, Avermann, Petersen, Gerstner and Vogels, in the Journal of Neurophysiology. It examines the validity of the ubiquitously used random sparse network models. In high-dimensional searches the connectivity parameters of thousands of model networks were evaluated for their fidelity regarding two experimental data sets. The authors could falsify the random sparse connectivity hypothesis for 7 of 36 tested connectivity parameters, but also confirmed the hypothesis in 8 cases. More importantly, they show how conceptually simple changes can create networks that produce biologically realistic dynamics. This work is further developed and elaborated on as well, in my new lab at the University of Oxford. It should be noted that the overlap between the proposed aims and the delivered papers is not 100%, because it became clear that some of the ground work to build elaborate processing networks was yet to be done before hand. For example, my first paper on inhibitory synaptic plasticity addressed Aim II - to control a multi-stream gating network through neuronal mechanisms within the same network - of the original goals. Additionally it provided crucial techniques and background for the later work. Clearly we don’t understand all principles that guide cortical structural development, but inhibitory plasticity promises to play a prominent rule. Similarly, Aim I - to derive general principles for connectivity schemes - was addressed in my work with Christian Tomm, in which we scanned thousands of models for their biological fidelity and finally identified a network that reproduced experimental results with high accuracy.Aim 3, building networks that can compute biologically relevant stimuli, was addressed in my work with Guillaume Hennequin, in which we showed that precisely inhibition stabilised networks can produce neuronal dynamics that reproduce the neuronal dynamics of motor cortex in surprising detail.In summary, it can be said that while my work did not overlap completely with the anticipated deliverables, but it matched to a high degree the general research plan and addressed most of the interesting points of my original proposal. We published a total of five papers in 3 years on the subject and I have developed several new projects during my time at the EPFL. It should also be noted that I successfully re-integrated in Europe and now hold an independent position at the University of Oxford.
