| This research was done by the Matthew Wilson research team at the MIT department of Brain and Cognitive Sciences.
From studies in rats it has been known for a while that certain cells in the hippocampus can encode for the location of the animal during a maze task. These are called "place cells" because, in the most robust examples, the cells are active when and only when the animal is at a single location in the maze or task field - while other cells are exclusively active at other locations in the maze.
Matt Wilson and friends repeated this type of experiment with a few variations.
First, they recorded activity from multiple hippocampal cells at once (called ensemble recording) and ran the rats through portions of a very simple circular track several times. Thus, instead of a location in the maze being linked to the activity of one cell in the hippocampus, locations were linked to temporal activity patterns distributed over several cells - unique temporal patterns that were confirmed through many repetitions of the same trial in the maze.
Second (the neat part), when they observed the activity in the same cells while the animals were sleeping, they found that the unique patterns of hippocampus activity seen in the maze were *replayed during REM sleep*.
Were the rats really dreaming about their day in the lab? That much is not known (and, frankly, were I a rat I *know* I'd be dreaming about sex,) but what this experiment does is lend support to a relatively old theory that hippocampal activity during REM sleep is responsible for memory encoding - a theory that has not had much experimental support so far, but is none-the-less very intuitive and attractive to some neuroscientists.
The original article is "Temporally structured Replay of Awake Hippocampal Ensemble Activity during Rapid Eye Movement Sleep", in the journal Neuron, volume 29, pp 145-56. | 
