Connectionist Models of CPG Networks

Abstract

Central Pattern Generators (CPGs) are neural networks that can produce coordinated patterns of rhythmic activity to orchestrate repetitive behaviors such as feeding, locomotion, and respiration. CPGs are known to be composed of phylogenetically conserved connectional units that are typically organized into chains to enable coordinated activation of the effector organs (muscles). Through integration of experimental and computational approaches at the cellular and network levels, essential connectional elements of CPGs have been revealed (Grillner 2006). Network Topology: At the connectional level, a CPG network consists of “unit CPGs” representing the nodes of the network, and the reciprocal connections between unit CPGs correspond to the branches of the network. Physiologically, a unit CPG represents a group of neurons that can generate a recurrent burst of activity (Grillner 2006). Simple inputs such as tonic excitation can produce rhythmic outputs in these networks. Increasing the stimulus strength typically increases the frequency of the network rhythm. Two unit CPGs are reciprocally coupled by mutual inhibition to generate alternating activity patterns (Fig. 1a) or by mutual excitation to generate synchronized bursts (Fig. 1b). Many unit CPGs connected in a chain with mutual excitation can generate a progressive phase lag in the bursts of activity along the CPG chain (Fig. 1c). It is generally accepted that rhythmogenesis is intrinsic to the neurons forming the unit CPG. However, connectionist models of CPG networks have unequivocally demonstrated that rhythmic patterns can be a resultant of a topology of connections among unit CPGs (Buchanan 1992). Hence, based on the nature of connections between unit CPGs, a CPG network can generate and shape patterns of activity. Reconfiguration of CPGs: Central pattern generators can reconfigure to switch from one activity pattern to another to produce multiple forms of the same behavior (e.g., walking, running, and galloping as different forms of locomotion) or functionally distinct behaviors (e.g., normal breathing versus gasping). Mathematical and simulation models suggest that the mechanisms underlying the reconfiguration of CPGs may include (1) altered strength and duration of the excitatory drive to the CPG network (e.g., Jung et al. 1996), (2) altered strength and duration of the mutual connections between unit CPGs (e.g., Venugopal et al. 2007), and (3) recruitment and de-recruitment of unit CPGs (e.g., Nasse et al. 2008). Such reconfiguration of a given CPG network architecture can confer multifunctionality. Unit CPG Models: Biophysical models of unit CPGs are typically based on conductance-based Hodgkin-Huxley formalism to model the constituent neurons with one or more compartments (e.g., see review Grillner et al. 2005). In models that exclusively focused on network properties, nonlinear oscillators have been widely used to represent unit CPGs (e.g., see review Ijspeert 2008). It appears that the network activity pattern of interest can be produced irrespective of the type of the oscillator model (e.g., Collins and Richmond 1994). This provides various options for choosing a suitable oscillator to represent the unit CPGs/nodes, particularly when mathematical analyses are adopted for