Almost all neuroscientists believe that memories are stored in the connections between neurons: synapses. Learning, then, consists of the strengthening of some synapses, the weakening of others, and maybe even the formation of entirely new ones. But a paper from Catherine Miller and colleagues suggests that changes to DNA are also involved: Cortical DNA methylation maintains remote memory.
DNA is a series of bases, and fundamentally there are just four: C, A, T and G. However, the Cs and the As can be
i.e. modified by the addition of a very simple methyl chemical group. They then stay that way until they get demethylated in the reverse process. Methylating a gene generally reduces its expression.
It's a bit like writing notes in pencil on top of a printed document: it doesn't change the underlying genetic sequence, but it's a semi-permanent change and it can be inherited by dividing cells. Methylation is a classic example of an epigenetic change, and epigenetics is very hot right now.
Miller et al found that learning induces the methylation of a gene called calcineurin (CaN) in the cells of the frontal cortex of rats. These changes appeared within 1 day of the learning event, and they persisted for at least 30 days (the longest time studied - they could well last much longer). Methylation of another gene, reelin, was also increased, but only for a few hours.
When they blocked these changes by injecting a DNA methylation inhibitor into the frontal cortex, it caused amnesia - even if the drug was given 30 days after the learning had taken place. In other words, the methylation inhibitors somehow erased the memory traces. These authors have previously reported that the same kind of learning causes a short-lived increase in methylation in the hippocampus. Taken together with these data, this fits with the well-known theory that memory traces start off being stored in the hippocampus and are then somehow transferred to the cortex later.
This kind of research has a bit of a history. The idea that memories are stored in DNA has led some to theorize that memories can be inherited. It also reminds me of the work of psychologist and Unabomber-victim James McConnell, who claimed that planarian worms can learn information by eating the ground-up remains of other worms who knew something...
These data are very interesting, but they don't imply anything quite so exciting. The pattern of methylation seemed entirely random (except in the sense that it was targeted at certain genes) - so rather than encoding information per se, the DNA changes were acting as a way of reducing CaN gene expression. Most likely, the reduction in CaN was limited to certain cells, and these were the cells that formed the connections that encoded the information.
Miller, C., Gavin, C., White, J., Parrish, R., Honasoge, A., Yancey, C., Rivera, I., Rubio, M., Rumbaugh, G., & Sweatt, J. (2010). Cortical DNA methylation maintains remote memory Nature Neuroscience, 13 (6), 664-666 DOI: 10.1038/nn.2560