The paper, formally published in Nature (abstract visible, article behind paywall), describes the mechanism by which this happens.
I have written to the lead researcher to get confirmation that this is actually a change in the DNA sequence itself and not a change in the epigenome that alters what the DNA transcribes to. He has kindly written back to confirm the findings. It IS a change in the DNA. Every brain cell in you has a genome unique to itself.
In short, the brain is a cluster where each node is running self-modifying code — a practice that no computer scientist or software engineer would dream of trying to do, considering it way too fragile, too unpredictable and too difficult. The university I went to, you'd be murdered in the hall if you'd proposed even single-threaded self-modifying algorithms, never mind a few trillion tightly-coupled threads.
The hope in genetics is that this will lead to a better understanding of genetic diseases, such as the various forms of dementia. My fear is that it will have the opposite effect — you can't exactly sequence every cell in the brain of a live patient to see what is going on, which may lead to geneticists ruling the problem too hard.
The other consequence of this find is that we are all chimera. Human DNA can no longer be regarded as a single thing in a single person, with only a few exceptional cases. The terms "Chimera" and "Gestalt" apply to everyone on a fantastic scale. Which makes them meaningless, unless they get redefined to work around the problem.
Arguably, that's an academic point at the moment. Nomenclature is nothing too serious and there's no actual hard evidence that it would cause problems in DNA forensics, though the mere fact that there's a possibility might cause problems in the courtroom whatever the science itself says about the impact."