ABSTRACT
391The neuronal cytoskeleton, scaffolding for neuronal cell shape and framework for intracellular communication, adds a unique and powerful dimension to the capabilities of biological neurons to compute, adapt, and learn. Potential capabilities of this network of protein polymer strands bridge the gap between detailed and complex biological neurons and their simplified counterpart, the biologically inspired artificial neural network. Intracellular signaling, biologically plausible by several cytoskeleton-mediated mechanisms, could play a key role in neuronal learning paradigms and synaptic adaptation, and have the potential of providing signals from axon to dendrite, from synapse to nucleus, or between membrane sites along dendrites. We propose a model for neuronal learning via cytoskeletal signaling, and identify specific biophysical mechanisms that could plausibly implement this model. Cytoskeletal signaling mechanisms are proposed for transmission along protein polymer strands such as microtubules, actin filaments and neurofilaments, and cross-bridge proteins transfer signals between strands. The proposed cytoskeletal signaling provides a key element in a biologically plausible model for back-propagation of learning signals as synaptic weights are trained. Biophysical sites for each step in an error-correcting learning paradigm are suggested, and cytoskeletal signaling can be modeled by cellular automata or moving ionic waves. Thus the cytoskeleton, anchored to synaptic proteins, furnishes a protein molecular level of processing in the neural network and could imbue the network with key learning and adaptational capabilities.
