linking back to brembs.net






My lab:
lab.png

ResearchBlogging.org
When we discovered a novel learning system in the fruitfly Drosophila (Brembs & Plendl, 2008) and then found out how it interacted with the one learning system which is described in all relevant textbooks (Brembs 2009), we weren't quite sure how general these findings would be for other animals and humans. In the subsequent years, genetically similar processes were discovered in the marine snail Aplysia, songbirds and mice, so we started to be quite confident that we had discovered something quite profound. When it turned out that the fly orthologue of the well-known human gene implicated in language, FOXP-2, was also involved in this learning process, we became quite sure that we had something rather evolutionary ancient in our hands. The key to discover this learning process in the diverse species was to exclude learning of environmental stimuli and instead force the subject to learn about thir own behavior - which is why we called this learning process self-learning, as opposed to world-learning, the process that associates environmental stimuli.

For instance, we use the Drosophila flight simulator to train the animals to either attempt right turns or left turns. Whenever we indicate to them, which direction is punished by using different colors, the flies learn about the colors and not about the learning directions, i.e., the world-learning process is activated and no self-learning is involved. If we take the colors away, they learn about their own behavior, i.e., the self-learning process is engaged. Similarly, if mice are allowed to learn the location of a submerged platform in a Morris Water Maze task using the visual cues around the arena, their self-learning process is irrelevant. However, if the lights are switched off then the animals have to learn the location of the platform using self-motion cues, then the same genetic manipulations we use in flies to manipulate self-learning, also affect the learning performance of the mice (Rochefort et al., 2011, see also this researchblogging post for more details).

Now, a new paper from the lab of John Krakauer has been published, which indicates that analogous interactions between self- and world-learning may potentially be taking place in humans as well. As in the other studies, the authors here removed predictive environmental stimuli during training to uncover an additional learning process, which is usually hidden in the background. Because this interaction appears to be strikingly similar to an interaction between learning systems in the fruit fly Drosophila and other animals, the fascinating possibility of an evolutionary ancient arrangement of multiple learning systems is raised.

The authors are able to distinguish two types of learning processes that appear to be engaged simultaneously during certain reaching tasks. In these experiments, the subjects move their hand towards a target, but they are barred from seeing their arms due to a mirror, which hides their arms and displays the contents of a computer screen on which the researchers control the amount of visual feedback the participants get. The basic principle of the training is to compensate for a quick displacement of the target when the participants have already started moving their hand towards it. Participants who are allowed to use visual cues during the learning phase, show a rapid decay back to baseline, if they are tested without the cues after training. If, however, training occurs without visual cues, the decay is attenuated. Thus, similar to the other animals, removing visual cues engages a learning system which is otherwise not triggered, even if the behavior generated during the task is otherwise completely identical.
In flies, we interpreted this finding as the learning of visual cues (world-learning) inhibiting or suppressing the learning of self-motion cues (self-learning), which is also one of the interpretations the authors offer in this new article. In flies, snails, birds and mice, it was shown that self-learning requires different biochemical pathways than world-learning. The authors here refer to world-learning as a model-based learning process and to self-learning as a model-free, reinforcement learning process, which is in agreement with the animal data so far.

However, one finding is difficult to reconcile between the human and the animal experiments: in mice, the self-learning process (also isolated by removing visual cues) is dependent on protein kinase C in the cerebellum (Rochefort et al., 2011), whereas in humans, the model-free (i.e., self-learning) process appears to be cerebellum-independent (cited by the authors of this article).

Taken together, these experiments suggest that learning of behavior and learning of external stimuli are not only handled by evolutionary ancient pathways, spanning the entire bilaterian branch, but that also the rules of their interactions have been conserved over the last 500 million years of evolution.


Shmuelof, L., Huang, V., Haith, A., Delnicki, R., Mazzoni, P., & Krakauer, J. (2012). Overcoming Motor "Forgetting" Through Reinforcement Of Learned Actions Journal of Neuroscience, 32 (42), 14617-14621 DOI: 10.1523/JNEUROSCI.2184-12.2012
Posted on Friday 28 December 2012 - 13:40:01 comment: 2
{TAGS}

Comments
DrGaryG posted on28 Dec 12: 20:00:
Guest


There's a classic book by Vincent Dethier that addresses this, To Know a Fly, (1962, Holden-Day)

Julien Colomb posted on06 Jan 13: 10:05:
Comments: 12

Registered: 06 Aug 08: 11:04

Localisation of the PKC requirement is probably task dependent such that the discrepancy between the different studies is to be expected. The motor programs affected are different in the navigation task in the mouses, and in the "arm targetting" in humans. For instance, another self-learning task is the change in the amplitude of a reflex, when this reflex amplitude is rewarded or punished. We may expect a PKC dependent change in motorneurons in that case.


You must be logged in to make comments on this site - please log in, or if you are not registered click here to signup
Render time: 0.0989 sec, 0.0059 of that for queries.