Spike propagation in driven chain networks with dominant global inhibition Wonil Chang 1,2 and Dezhe Z. Jin 1, * 1 Department of Physics, The Pennsylvania State University, University Park, Pennsylvania 16802, USA 2 Department of Bio and Brain Engineering, KAIST, Daejeon 305–701, Korea Received 13 August 2008; published 20 May 2009 Spike propagation in chain networks is usually studied in the synfire regime, in which successive groups of neurons are synaptically activated sequentially through the unidirectional excitatory connections. Here we study the dynamics of chain networks with dominant global feedback inhibition that prevents the synfire activity. Neural activity is driven by suprathreshold external inputs. We analytically and numerically demon- strate that spike propagation along the chain is a unique dynamical attractor in a wide parameter regime. The strong inhibition permits a robust winner-take-all propagation in the case of multiple chains competing via the inhibition. DOI: 10.1103/PhysRevE.79.051917 PACS numbers: 87.18.Sn, 05.45.Xt, 84.35.+i Synfire chain activity, in which synchronous spikes propa- gate along a chain of successive groups of neurons connected unidirectionally via excitatory synaptic connections 1, has been extensively studied 2and suggested as the underlying mechanism for precisely timed sequential firings of neurons observed in a number of neural systems, including songbirds 35, cortical activity 6, and primate motor cortex 7. Synfire activity requires the excitation be in a restricted re- gime: the excitation must be strong enough for the synaptic activity to evoke spikes in subsequent groups of neurons, but weak enough to avoid runaway instability 4,8. In this paper, we demonstrate that precisely timed spike propagation in chain networks can be robustly established beyond the synfire regime. Instead of the synaptic activation, neural activity is sustained by suprathreshold external inputs. The activity is controlled by a strong global feedback inhi- bition and shaped by the unidirectional excitatory connec- tions between the groups. The inhibition dominates the exci- tation and the synfire activity is suppressed. We show that spike propagation is a unique attractor to which the dynamics flows from all initial conditions when the external inputs are on. This mechanism is robust, with a large working param- eter regime for the excitation and inhibition strengths. The strong inhibition also permits a robust winner-take-all selec- tion of a single chain for spikes to propagate when there are multiple chains competing for the activity, which could be a mechanism for action selection if each chain encodes an ac- tion element such as a song syllable in songbirds 4,5. Our results in the “driven-chain” regime are obtained through analytical analysis and numerical simulations of chain networks of leaky integrate-and-fire neurons. The ana- lytical analysis is aided with three simplifications: the groups of neurons are replaced by single neurons, the global inhibi- tion is modeled with all-to-all inhibitory connections be- tween the neurons, and the synaptic interactions are approxi- mated as pulse coupling. We prove that sequential spiking along the chain with precise timings is the unique global attractor in a wide parameter regime of the excitation and inhibition; furthermore, in the same parameter regime, the spike propagation selects a single chain if multiple chains compete. The analytical results are confirmed numerically with the simplifications removed and noise added. Many models of biological and physical systems includ- ing heart cells, fireflies, earthquakes, and neural networks belong to a broad class of models consisting of systems of pulse-coupled oscillators 911; our simplified neural net- work model fits into this class as well. Our analytical analy- sis should add insights into the relationship between the structure of coupling and the dynamics, a key for under- standing these diverse systems. Sequential spiking in chains of pulse-coupled oscillating single neurons has been investi- gated before 1214and it has been shown that spike se- quences are stable in generic inhibition-dominant networks 15. However, these works do not show that the dynamics of a given network is attracted to a unique spike sequence at- tractor regardless of the initial conditions. A unique attractor is robust against perturbations and noise since the basin of attraction is large. This is an important characteristic if the pattern drives a single motor action such as a song syllable 3,4. Our analysis establishes that the spike propagation in chain networks in the driven-chain regime is a unique stable attractor to which the dynamics converges from all initial conditions. The dynamics of the neurons in the simplified model is as follows: dV j t dt = E R + I - V j t+ I s , 1 where I s is the synaptic current and is given by I s = n=1 - G E j,s n V j t+ G I E I - V j tt - t n . 2 Here is the membrane time constant; V j tis the membrane potential of neuron j ; E R and E I are the resting membrane potential and the reversal potential of the inhibitory synapse, respectively the reversal potential of the excitatory synapse is 0; I is the constant external input; G E j,i is the excitatory conductance from neuron i to neuron j , which is G E 0 if j = i +1 is the neuron next to neuron i down the chain and 0 otherwise; G I 0 is the global inhibitory conductance be- * djin@phys.psu.edu PHYSICAL REVIEW E 79, 051917 2009 1539-3755/2009/795/0519175©2009 The American Physical Society 051917-1