Sequence Generation by Model Neural Networks

Cyclic patterns of motor neuron activity are involved in the production of many rhythmic movements, such as locomotion, mastication and scratching. These movements can occur in the absence of both internal pacemakers and external feedback. Typically, they are controlled by neural circuits referred to as central pattern generators (CPGs). We describe a class of model neural networks whose behavior mimics that of CPGs. Pattern generation by these model networks depends upon the interplay between synaptic connections that act on two different time scales. Short term excitation between pairs of neurons is followed by long term inhibition, and short term inhibition by long term excitation. The long term effect depends on the time averaged activity of a neuron. The change in sign of the next synaptic input to neuron as a function of time causes the network to make transitions between patterns. This formal description of sequence generation predicts the strength of all possible connections between pairs of neurons based solely on the output of the CPG. It also predicts the resting activity of the neurons in terms of the measured synaptic strengths between all pairs of neurons. We illustrate our theory by applying it to the CPG controlling escape swimming in the mollusk Tritonia diomedea.