The flies 'get' the task after only about 8 minutes of training and keep clear of the punished color/turning direction even if the heat was switched off afterwards. When I analyzed what the flies have learned, the turning, the colors or both, it turned out they learned only the colors. I tested the color memory by asking them to choose colors with a different behavior. The experiment for the flies can be imagined like first learning to drive a car in the US or the European continent, only to then have to drive a car in the UK or Australia, with everything reversed. After 8 minutes of training, the flies could switch from one car to another. When I did the same analysis after 16 minutes of training, I found that both the behavior and the colors had been learned. In the transfer test, the flies now failed to choose the previously unpunished color with the novel behavior. In the car example, if you practice car driving long enough, driving in a different car in a different traffic system becomes much harder then if you've just started learning how to drive. Basically, it becomes difficult to kick the habit, even if you're a fly. These results don't only match well with vertebrate and human studies, it also dovetails nicely with our recent discovery that the behavior-learning and the world-learning (for the flies: turning and color-learning, respectively) systems can be distinguished at the molecular level.
Once we saw the interaction between behavior and world-learning, we started to use the genetic toolkit that comes with the model system Drosophila and found that a very well-studied insect brain region, the mushroom-bodies, is involved in some of the interactions underlying the regulation of habit formation in flies.
These results now allow us for the first time to tease apart the mechanisms of interacting memory systems at the molecular and the circuit level. If you think that's interesting, you can get the new paper in the download section.