Soc. Neurosci. Abstr. 161.2, 2000.

Axonal propagation of spike-based Hebbian learning is a key mechanism to the formation of temporal feature-maps on a sub-millisecond time scale

R. Kempter¹ ², C. Leibold², H. Wagner³ and J. L. van Hemmen²

¹Keck Center, UCSF, San Francisco, CA 94143-0732,
²Physik Dept, TU Munich, 85747 Garching, Germany,
³Inst Biologie II, RWTH Aachen, 52074 Aachen, Germany.

Computational maps are of central importance to the brain's representation of the outside world. Nevertheless it is hardly understood how temporal-feature maps, e.g., in the auditory system, arise. We address this problem by analyzing the barn owl's nucleus laminaris (NL) and show how its experimentally observed map of interaural time differences (ITDs) can emerge from a combined action of homosynaptic spike-based Hebbian learning and the propagation of synaptic modifications along the presynaptic axon. Both mechanisms are necessary to map formation but the latter may be orders of magnitude weaker than the former. Other interaction mechanisms between NL neurons are unsuitable in the present context, e.g., electrical coupling by gap junctions, electrical crosstalk, extracellular diffusion of a messenger, recurrent synaptic coupling. Lateral inhibition is not needed for a diversity of ITD tuning. The Hebbian learning rule has to be sensitive to the order of pre- and postsynaptic spikes on a sub-millisecond time scale. We demonstrate why the growth of the owl's head does not impair map formation and that a reduction of the degree of phase locking in the input, e.g., by means of an ear occluder, delays the development of the ITD map. The learning algorithm selects axonal arbors from nucleus magnocellularis to NL with proper delays. In this way, the emergence of the neurophonic can be explained. In a NL lamina (best frequency f>2kHz) of adult owls, we predict that the distribution of latencies from the ear to the border of NL has peaks separated by multiples of 1/f. The resulting structure is consistent with Jeffress' proposal of a place code and in agreement with physiological and anatomical data on NL. In the chick or in mammals, similar physiology based on different anatomical arrangements than in the owl has been described. The development of such maps may also be explained by our algorithm.

R.K. supported by the Deutsche Forschungsgemeinschaft, Kl 608/10-1 and Ke 788/1-1.

keywords: nucleus laminaris, sound localization, barn owl, computational map