Glimpsing memory traces in real time

MEMORY is one of the biggest enduring mysteries of modern neuroscience, and has perhaps been researched more intensively than any other aspect of brain function. The past few decades have yielded a great deal of knowledge about the cellular and molecular mechanisms of memory, and it is now widely believed that memories are formed as a result of biochemical changes which ultimately lead to the strengthening of connections between nerve cells.

It is, however, also clear that memories are not encoded at the level of single neurons. Instead, the memory trace is thought of as a flurry of electrical activity within a scattered population of cells. Yet, very little is known about how memories are encoded and retrieved by populations of cells. Using a new large-scale recording technique, researchers from the Medical College of Georgia have now directly observed, for the first time, the population-level activity associated with encoding and retrieval of memory traces.

Joe Tsien and his colleagues trained 10 adult mice using the classical fear conditioning paradigm. The animals were first put through a habituation session, in which they were placed in a triangular chamber, and then a square one, for 5 minutes each on three consecutive days. Whilst in the square chamber (or conditioning chamber), they were played a brief tone. On the fourth day, after a short rest period in their home cages, the mice were placed back into the conditioning chamber, and repeatedly presented with the same tone. This time though, each tone was followed, 20 seconds later, with a mild electric shock administered by means of a metal floor grid.

This procedure is known to produce several different kinds of memories. The mice quickly learned to associate the tone with shock, so that they exhibited fear behaviour (freezing) as soon as they heard it. And, as expected, they also froze when the tone was played back to them in the triangular chamber an hour after the training session, showing that the memory trace was persisted. The procedure also produced a memory of the context in which the electric shock was administered. When placed back into the conditioning chamber after completing the training trials, they froze immediately, even though they were not played the tone. However, in the absence of electric shocks, they no longer showed any fear - the contextual memory quickly faded, in a process referred to as extinction.

Throughout the training and recall trials, implantable electrode arrays were used to simultaneously record the activity of ensembles of more than 200 neurons in the CA1 region of the hippocampus, which is known to be critical for associative memory. The researchers could therefore observe and decode, in time windows of one hundredth of a second, the dynamic patterns of neuronal activity associated with the formation and subsequent retrieval of the memories. The recording technique, which was developed recently in Tsien's lab, also enabled the researchers to investigate how individual hippocampal neurons respond to various features of the events and stimuli.

Very few cells (less than 1%) had a generalized response to all types of stimuli, such as the tone before, during and after pairing with the electric shock. Nearly 7% of cells responded selectively to the tone, as well as to the electric shock during and after pairing with the tone, although a significant proportion of these (~19%) responded to each stimulus with a different firing rate. One set of cells (~2% of the total) responded to the tone and to the electric shock after pairing; another small set (~2%) responded only to electric shock before it was paired with the tone, and a third (~4%) only to the shock after pairing. One larger set of cells (>18%) was observed to fire selectively in response to the foot shock but not to the tone.

Initially, the tone presented to the mice did not elicit a significant response from the population of hippocampal neurons. (By contrast, the first foot shock, which was delivered 20 seconds after the tone, did elicit an immediate response.) However, distinct patterns of activity were recorded over successive trials, normally emerging during the second or third pairing of the stimuli. The same patterns were found to be reactivated between 6-14 times per minute following each presentation of the tone and shock, but to change quickly with each successive pairing, from the pattern elicited by the former to the pattern elicited by the latter, perhaps reflecting learning of the association between the two stimuli and also consolidation of the memory trace.

Once the mice had learned to associate the tone with the foot shock, presentation of the tone was found to elicit a string a reactivated memory traces during the ensuing 60 seconds. About half of these corresponded to the pattern of activity elicited by the tone, and a further ~30% to the activity pattern elicited by the shock. Importantly, the number of reactivated patterns was found to increase in proportion to the number of conditioning trials, so that increasing numbers of traces were reactivated with each successive trial. Furthermore, the number of patterns was strongly correlated with extent of the subsequent freezing behaviour, and the reactivation of the foot shock trace was found to peak almost exactly 20 seconds after the tone was played, suggesting that the neuronal ensemble had encoded the time interval between the tone and the shock.

In the contextual recall tests, re-exposing the mice to the environment in which they had received the electric shocks led to a rapid re-emergence of the various previously observed activity patterns. The reactivated memory traces were found to consistently precede freezing behaviour by an average of 1.4 seconds. Again, the number of reactivated patterns was found to be tightly correlated with the freezing, so that the more patterns that were recorded, the greater amount of time the animal spent frozen during the 5 minutes of the recall test.

This study shows that laying down and recalling a memory trace involves rich and diverse patterns of neuronal activity within ensembles of neurons in the CA1 region of the hippocampus ,and that there are distinct patterns of activity associated with different stages of memory processing (such as learning and retrieval). These patterns often appeared in tandem, and with no apparent order in time, although a higher-order organization may be revealed as the recording techniques become more sophisticated. The study also shows that there is a meaningful relationship between these activity patterns and the behaviour of the animals. Although non-invasive techniques for recording such neuronal activity in humans do not yet exist, the findings could some day help researchers to gain a better understanding of, and develop treatments for, efficient disorders involving memory impairments.


Chen, G., et al. (2009). Neural Population-Level Memory Traces in the Mouse Hippocampus PLoS ONE, 4 (12) DOI: 10.1371/journal.pone.0008256

More like this

Hi! Fascinating post.

I have a kind of tangential question...

You say: "However, in the absence of electric shocks, they no longer showed any fear - the contextual memory quickly faded, in a process referred to as extinction."

I'm wondering whether you really mean that the memories are thought to "fade" or if that was just a loose way of speaking.

It seems to me that any experiment that support the hypothesis that the memories faded also supports the hypothesis that the memory doesn't fade, but that further memories are produced, leading the mouse to lose it's fear of the context.

Is that right?


The fear doesn't disappear immediately, but when the tone is presented several times without the electric shock, the memory of the association between the two stimuli fades quickly.

Is that to say that during the extinction process the pattern of activity decreases to a point where it is subthreshold and thus does not elicit the freezing behavior?

It seems that there may be a residual (subthreshold) trace that allows for reinstatement to occur.
Perhaps via changes in synaptic strength among the neuronal assemblies involved.