Neural dynamics of motor sequencing in lateral prefrontal cortex

L.R. Pearson¹ and S. Grossberg¹

1Department of Cognitive and Neural Systems

Boston University, Boston, MA, USA

A detailed laminar cortical model is proposed to explain the cortical interactions responsible for various aspects of working memory function, motor plan learning and sequential item selection. Simulations of this system replicated the results of neuro-physiological recording during the performance of a sequential motor task (Averbeck et al., PNAS 2002; Exp Brain Res 2003) in which macaque monkeys copied geometrical figures on an LCD screen using an X-Y joystick showing (a) simultaneous coding of all items in a planned sequence prior to sequence initiation which was highly predictive of eventual sequential order and (b) stronger coding for concurrently executed segments. Both of these properties are central assumptions of competitive-queuing models of serial order (Grossberg, Journal of Mathematical Psychology 1978; Houghton et al., PSYCHE 1995).

Among the areas widely associated with many functions of working memory storage, sequence planning and sequential motor execution are a set of interconnected structures in granular prefrontal cortex. The model proposes functional roles for the layers of lateral prefrontal cortex, showing how their structure, proposed circuitry and known connectivity with both sensory and motor systems enables them to play a role in the retention, processing and organized storage of multi-modal temporally ordered items for voluntary planned sequential behavior. The model demonstrates computational means by which dynamic cycles of working memory item storage, sequential plan learning, updating and resetting (recall) can be implemented by the network of feedforward, feedback and horizontal connections which link cortical layers together, within and across cortical areas. The model also tests the hypothesis that similar interconnected laminar circuitry in lower-level temporal and parietal areas forms a unified system of hierarchical processing of information by which items can be grouped (chunked), updated and learned at multiple processing levels.

This work was supported in part by the AFOSR, DARPA, NSF and the ONR.