Brain scans read memories

FORMATION of a memory is widely believed to leave a 'trace' in the brain - a fleeting pattern of electrical activity which strengthens the connections within a widely distributed network of neurons, and which re-emerges when the memory is recalled. The concept of the memory trace was first proposed nearly a century ago, but the nature of the trace, its precise location in the brain and the underlying neural mechanisms all remain elusive. Researchers from University College London now report that functional magnetic resonance (fMRI) can be used to decode individual memory traces and to predict which of three recently encoded memories is being recalled.

The new study, led by Eleanor Maguire of the Wellcome Trust Centre for Neuroimaging, builds on earlier work which demonstrates that fMRI can be used to predict simple mental states from brain activity. Last year, Maguire and her colleagues showed that it is possible to predict an individual's position in a virtual reality environment from patterns of activity in the hippocampus, and researchers from Vanderbilt University showed that activity in the visual cortex can be decoded to predict which image is being retained in working memory. Even more remarkably, Japanese researchers have reconstructed visual images from brain activity, including novel ones that their participants had never seen before.

This time, ten participants were shown three short film clips, each featuring a different woman performing a series of actions. After viewing each film 15 times, the participants were placed in a brain scanner and asked to recall them as vividly as possible. In the first condition, a recall cue was presented on a small screen inside the scanner, indicating which of the three films they should recall. The researchers recorded the brain activity associated with recollection of each. Afterwards, while still in the scanner, the participants performed 30 'free recall' trials. Each of these started with a short period in which they decided which of the three films they would recall. After recalling their chosen memory, the participants then indicated their choice using a keypad.

It was found that the traces of individual memories could be detected, decoded and distinguished from one another. The researchers were therefore able to accurately predict which of the three films was being recalled from the activity of the hippocampus. They also found that the activity associated with memory recall was remarkably consistent across all ten participants, and added the data together to produce a 'frequency heat map' (below). This clearly shows three distinct regions of peak activity: the anterior (or front) portion of the left and right hippocampi and the posterior (back) portion of the right hippocampus.

The researchers were surprised to find that the activity associated with recollection of each memory was stable over repeated recall trials performed soon after the memories were encoded. It is, however, now known that memories encoded in the hippocampus are eventually transferred to the frontal cortex for long-term storage, so that their recollection become less dependent on the former structure, and more so on the latter, with time. With this in mind, it would have been interesting to see how the activity patterns changed some time later, by asking the participants to return to have their brains scanned at a later date while recalling the same memories.  

Experiments like this are now routinely referred to as 'mind reading' in the mass media, but are nothing of the sort. This particular study, and others like it, involve using specially designed computer algorithms to distinguish between a very limited number of known activity patterns. Our memories are of course infinitely more diverse than those of the short film clips used here. They enable us to perform mental time travel, to recollect not only what we had for breakfast yesterday but also childhood events that took place decades ago. True 'free recall' would involve asking participants to recollect any real life event, and it is unlikely that brain scanning techniques could ever be used to determine what memory was being recollected in such a situation, even with the inevitable technological advances.

Nevertheless, the study does reveal some details about the functional anatomy of the hippocampus. The frequency heat map shows that information related to recollection of episodic memories is not distributed randomly across the human hippocampus, but rather is concentrated within specific regions. Recollection of recently encoded episodic memories appears to be largely dependent upon anterior regions of the left and right hippocampi, and the posterior region of the rigth hippocampus. By contrast, spatial memory appears to be dependent upon only the posterior regions. The researchers therefore speculate that activation of the posterior right hippocampus might be related to the spatial locations in the memories.

Exactly how the hippocampus encodes memories is still unknown. Single cell recordings performed on epileptic patients about to undergo surgery show that the same neurons involved in encoding a memory are also activated during subsequent recollection of that memory. But memories are not encoded and retrieved by single neurons. Instead, the cells are thought to be part of a wide network distributed throughout the brain. The network supports extremely rich representations, containing numerous disparate elements - such as sights and smells - that are sewn together during recollection. Successful memory recall therefore requires that the representations of all the memory elements are re-activated at the same time.

In the early 1970s, the pioneering computational neuroscientist David Marr proposed that the hippocampus performs the function of activating all the neural representations needed for recollection. It does so by storing a 'memory index', a simple formation that is linked to all the other representations, and which can activate them together to aid recollection. Future research using computer algorithms similar to the one used here could bridge the gap between single cells and the patterns of activity in distributed neuronal networks, enabling researchers to gain a better understanding of how memories are encoded.

Related:

• Tracing memories
• Memory lessons from Homer Simpson 
  
• Reading the contents of working memory
• Visual images reconstructed from brain activity
  
• Where do you think you are? A brain scan can tell

Chadwick, M. J., et al. (2010). Decoding Individual Episodic Memory Traces in the Human Hippocampus. Curr. Biol. DOI: 10.1016/j.cub.2010.01.053.

Marr, D. (1971). Simple memory: A theory for archicortex. Philos. Trans.
R. Soc. Lond. B Biol. Sci. 262: 23-81. [PDF]