Richard Thompson, M.D
Department of Brain and Cognitive Sciences
University of Southern California
Los Angeles, California
March 11, 2002
A Memory Trace Found
Some years ago, we adopted classical conditioning of the eyeblink response as a prototypic model of Pavlovian conditioning and the rabbit as our initial preparation, largely because so much work had been done with this animal and paradigm. All mammals studied, including human, exhibit the same basic properties of associative learning in eyeblink conditioning.
Several brain systems become massively engaged in the paradigm, particularly the hippocampus and cerebellum; if the US (unconditioned stimulus) is sufficiently aversive, the amygdala is also engaged. However, using the basic delay procedure (conditioned stimulus and unconditioned stimulus overlap and coterminate) only lesions of the cerebellum abolish the CR (conditioned response). The critical lesion is to the anterior cerebellar interpositus nucleus ipsilateral to the trained eye. The lesion has no effect on the reflex eyeblink. Neuronal unit activity increases massively in this critical region of the nucleus with training and electrical stimulation of the region elicits eyeblink in untrained animals; the circuit is hard-wired from interpositus to behavior.
In a long series of studies using eletrophysiological recordings, lesions, stimulation, and anatomical pathway tracing, we identified the essential (necessary and sufficient) circuit for this form of learning and memory. Reversible inactivation (drugs or cooling) of the following structures does not prevent learning at all: motor nuclei, red nucleus, 9 superior cerebellar peduncle (output from cerebellum). Reversible inactivation of the anterior interpositus nucleus completely prevents learning and inactivation of the cerebellar cortex impairs but does not prevent learning. So the anterior interpositus appears to be the critical structure and the probable locus of the basic or primary memory trace. Blocking protein synthesis in the interpositus nucleus completely prevents learning.
Use of mutant and KO (knock out) mice has demonstrated that eyeblink conditioning can develop, albeit with some impairment, in the complete absence of functional cerebellar cortex (ped mouse) so long as the interpositus nucleus is not lesioned (if it is, there is no learning at all). In the cerebellar cortex, several mutant and KO mice studies support the following observation: if cerebellar cortical LTD is impaired, so is eyeblink learning.
Recordings from identified cerebellar Purkinje neurons in the behaving animals are convenient because the CS evoked mossy-parallel fiber activation is recorded as simple spikes and the US evoked climbing fiber activation is recorded as complex spikes and the two types are easily separable. In trained animals, although several patterns of learning- induced simple spike responses are observed (and many Purkinje neurons are of course not interested in the form of learning), the most common pattern is a decrease in simple spike frequency in the CS period. This result is completely consistent with ]to's phenomenon of LTD (Long Term Depression) in cerebellar cortex (decrease in parallel fiber synaptic efficacy on Purkinje neuron dendrites as a result of repeated pairing of the CS (parallel fiber activation] and US [climbing fiber activation]).
For complex spikes, those Purkinje neurons that are influenced by the US consistently show evoked complex spikes to the onset of the US on US alone trials and to the US impaired trials early in training. As learning develops, these US evoked complex spikes are suppressed, just as is US-evoked activity in the inferior olive. Since we argue that this circuit (trigeminal to inferior olive to cerebellum as climbing fibers) is the essential reinforcing or teaching pathway-these results are completely consistent with the elegant formulation by Rescoria and Wagner for acquisition of classically conditioned responses.
Cerebellar cortical lesion studies and studies on mutant and KO mice with cerebellar cortical abnormalities all produce the same two effects: (1) the CR peak latency is no longer adaptive, i.e., no longer occurs at the onset of the US; instead it has a significantly shorter latency, as we showed many years ago (McCormick and Thompson, 1984, Science, 223, 296-299); and (2) acquisition of the behavioral CR is slower and develops to a lesser extent. We have argued as noted above, that the basic or primary memory trace is established in the interpositus nucleus; we suggest that secondary traces develop in cerebellar cortex that serve to modulate the interpositus (via Purkinje neuron inhibition of interpositus neurons) to achieve adaptive timing and normal learning.
In a series of experiments we were able to identify the brain circuit necessary for the behavioral phenomenon of blocking, discovered by Kamin. If animals are first trained, e.g., to a tone CS (corneal airpuff US) until they are well trained and then given additional training to a compound tone-light CS (corneal airpuff US), and then tested to the light CS, they show no learning to the light. In contrast, if animals are only given training to compound tone-light CS, they learn to respond to the tons and the light. Prior training to the tone blocks subsequent learning to the light in the compound stimulus training. In cognitive terms, the light adds no new information and so is ignored.
We are not satisfied with such "mentalistic" explanations and set out to identify the essential circuit for blocking in eyeblink conditioning. As it happens, the interpositus has a strong direct GABAergice inhibitory projection to the inferior olive, as well as a strong excitatory projection to the red nucleus (and from there to motor nuclei). Since neuronal activity in the interpositus grows markedly over training in the CS period preceding the onset of the behavioral CR, we reasoned that the growing inhibition of the inferior olive would shut down its climbing fiber projection to the cerebellum normally evoked by the US onset; so after the animal is well trained to tone, additional training to tone-light will not result in additional learning because the reinforcing or teaching input to the cerebellum, the climbing fiber system, is shut down at the inferior olive. This argument is consistent with the more general formulation of the Rescoria-Wagner algorithm. Recordings from the inferior olive supported this possibility, as did recordings of complex spikes from Purkinje neurons in cerebellar cortex (evoked by climbing fibers).
In the critical test, we completed the blocking paradigm with an additional group given tone training first, then compound tone-light training for 5 days with infusions of picrotoxin in the inferior olive to block GABA inhibition from the interpositus (a control group received the same training but with only vehicle infused in the inferior olive). The control for blocking received no prior tone training. The result was as expected. The group not given prior tone training showed clear learning to the light CS; the blocking group with vehicle infusion in the inferior olive showed blocking, i.e., no learning to light. The critical group, receiving picrotoxin in the inferior olive during compound tone-light training after tone training showed responding to light just like the control group that had no prior tone training. Blocking GABA inhibition in the inferior olive completely blocked the behavioral phenomenon of blocking. This is a most satisfying result in that we were able to show that the cerebellar circuit itself instantiated the phenomenon of blocking-blocking is an emergent property of the network itself and not a result of some specialized molecular processes.