| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Leigh Marine Laboratory, University of Auckland, Auckland, New Zealand
1 Department of Biology, Wesleyan University, Middletown, CT
The cerebellum is considered a motor control structure, yet it shares strong anatomical similarities with the cerebellar-like nuclei of the fish hindbrain, which are clearly sensory and process electrosensory and lateral line information. Can these contrasting roles be brought together in a single framework, providing insight into cerebellar function?
The striking anatomical similarity between the cerebellum and the cerebellar-like sensory nuclei is that they share a molecular layer organization in which the apical dendrites of the projection neurons receive input from many thousands of parallel fibers (1,2). Our hypothesis is that, in both systems, plasticity of the synapses between the parallel fibers and the projection neurons allows the formation of a parallel fiber composite that can subtract unwanted activity in the projection neurons and their downstream targets. In cerebellar-like structures, this subtraction can be directly observed as a negative sensory image (1). Here we report a motor learning paradigm in fish that exhibits an analogous negative motor image.
The cerebellar-like sensory nuclei receive primary afferent input from octavolateralis sensory systems, and relay that activity to midbrain levels. In these nuclei, spike activity of a projection neuron reduces the synaptic strength of concurrently active parallel fibers (3,4,5), a learning rule termed anti-Hebbian plasticity. The composite signal thus formed subtracts re-afferent noise generated on the sensory system by the animal’s own movements. Experiments show that the projection neurons learn to cancel an external stimulus that is paired with the animal’s own activity; and a defining feature of these adaptive filters is the negative sensory image, or cancellation signal, that is revealed when the stimulus is withheld (6). One way of describing cerebellar-like function is that the ongoing activity of the projection neurons themselves is, in effect, the error that selects the parallel fiber composite that forms the required negative sensory image. External signals are inherently unpredictable and are not cancelled by this mechanism because they are not represented in the parallel fiber system.
In the cerebellum, the projection neurons are called Purkinje cells. Like the principal cells in cerebellar-like nuclei, Purkinje cells receive many thousands of parallel fiber inputs. In addition, however, Purkinje cells receive a unique, strong synaptic input from climbing fibers. In many current models of cerebellar function, climbing fibers signal error in motor control (7,8), and their activity drives learning at the parallel fiber to Purkinje cell synapses in such a way as to reduce that error in a manner not unlike the learning that takes place in cerebellar-like sensory structures. However, learning in this case has become uncoupled from the ongoing activity of the Purkinje cell and only occurs in response to "instruction" from the climbing fiber. Can error-driven motor learning and the adaptive sensory filters be drawn into a single functional framework? We have sought to do this by developing a motor-learning paradigm that we predicted would exhibit an explicit negative motor image comparable to the negative sensory image of the adaptive filter.
From first principles, it should be possible to perturb almost any motor control system, provide a cue to allow the animal to anticipate and hence to reduce the impact of the perturbation, and thence reveal the negative motor image by providing the cue but withholding the perturbation. We have chosen to work with the postural control system of fish. Fish swimming in an upright position maintain their posture with a classical feedback control system. The vestibular system detects roll-error, which is then corrected by the fins. To provide a repeatable perturbation we attached magnets to the dorsal and ventral edge of the fish with their poles facing in opposite directions. These fish then swam in a small flume, outside of which were situated magnetic coils. Switching the coils on provided a rotational perturbation to the fish. Light-emitting diodes situated within the fish’s visual field provided the learning cue. The roll of the fish was quantified by processing the images from a digital video camera positioned above the flume. In these experiments, the cue and the field were on simultaneously for a period of 1 s, and were repeated at 10- to 20-s intervals. A minimum of 5 records were taken at the beginning and the end of the learning period, which lasted for a minimum of 1 h. The magnetic stimulus was then turned off, and a comparable number of records were taken of the learning cue presented alone. The peak roll at the end of the 1-s stimulus was quantified in each case. Experiments were performed on four rainbow fish (Glossolepis incisus) and three scup (Stenotomus chrysops). With rainbow fish, the learning period reduced the normalized roll from 62% ± 3.3 (mean ± standard error of the mean) to 46.2% ± 3.8, and the size of the negative roll on the presentation of the learning cue alone was -8.3% ± 2.1, which is significantly less than zero (z = 2.02, P < 0.05). In preliminary experiments with scup, a similar pattern was seen in two out of three fish.
These experiments provide an example of error-driven motor learning that demonstrates an explicit negative motor image comparable to the negative sensory image of cerebellar-like adaptive filters. Although we have no direct evidence that the site of learning in this paradigm is cerebellar, these experiments do suggest how cerebellum (at least in its simplest form) and cerebellar-like sensory structures can be considered within a single conceptual framework. In both systems, synaptic plasticity between the parallel fibers and the projection neurons may generate a parallel fiber composite that subtracts unwanted activity in the respective projection neurons. On this view, the essence of the molecular layer organization is that it provides a vast array of potential information from which to generate a cancellation signal, or error correction signal, via the anti-Hebbian learning mechanism.
Funded in part by an NSF grant to DB.
Literature Cited
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
