Whose Engram Is it Anyway?



Neocortical ensembles become engram-bearing during memory acquisition and may just lack the machinery to drive memory expression at recent timepoints

Systems consolidation is the process by which memories are stabilized into our brain’s long-term memory stores. The standard model of systems consolidation argues that recent memories are retained in the hippocampus, whereas remote memories are retained in the neocortex. In the interim period, which is thought to take anywhere from weeks to months, the hippocampus is said to “entrain” the neocortical ensembles that will ultimately house the memory trace (also called the engram), thereby rendering that memory hippocampus-independent.

This idea came about from clinical observations from patient H.M. and others after him: damage to the hippocampus resulted in temporally-graded amnesia – the inability to recall recent memories but still retaining old ones. Many experiments have since lent support to the model by demonstrating the double dissociation between hippocampus-dependent recent memories and neocortex-dependent remote memories (1).

If the neocortex is dispensable for recent memory retrieval and it needs to be entrained by the hippocampus then one may assume that neocortical neurons do not house the memory trace until consolidation has occurred. According to a poster from Dr. Michel van den Oever’s lab, this is not exactly the case. I learned that medial prefrontal cortex (mPFC) neurons may in fact house the memory trace very early on but lack the machinery to drive its expression.

To demonstrate this, they used a strategy that allows them to tag neurons that are active at encoding. Cells that are active during this time-window are presumed to be memory trace-housing cells if turning them back on results in retrieval of the memory. Therefore, tagged neurons were also made to express specialized receptors that can be targeted with a specialized drug.

They accomplished this by injecting a cocktail of two viruses into the mPFC: One virus drove cre recombinase expression to cells expressing the immediate early gene, cFos, in a tamoxifen-dependent manner, while the other virus expressed a designer receptor exclusively activated by designer drug (DREADD) in a cre-dependent manner. Using this dual virus strategy meant the researchers could use tamoxifen to express DREADDs exclusively in cells that are active during a specific time-window (i.e., during memory encoding). Once the specialized receptors were in place, clozapine-n-oxide (CNO) was injected systemically to selectively activate or inhibit the relevant cell population.

Mariana Matos gave me a crystal-clear run-through of the poster. The goal behind the project was to see whether CNO-mediated excitation of mPFC neurons active during encoding would elicit retrieval of the memory, and whether this would vary based on the age of the memory. But first, Mariana and her labmates had to demonstrate that their DREADDs were working.

They were! CNO resulted in decreased excitability when the inhibitory DREADD, hM4Di, was used. This illustrates that the construct was doing what it needed to do! But what about the labeling of engram-harboring cells? Was that working? The poster had some images showing that an mCherry reporter was expressed whenever a cell was tagged with hM4Di. What’s interesting is that the number of cells tagged varied based on what the animal was doing during the tagging period. Contextual fear conditioning (CFC) elicited more hM4Di expression in mPFC neurons than being in the home cage. This is probably because there’s not much to learn in the home cage and thus, fewer cells will be activated, leading to fewer hM4Di-tagged neurons.

So the methods work. Now, Mariana and co tested whether inhibiting the task-relevant mPFC cells (i.e., the ones that were active during encoding and therefore labeled with hM4Di) would block the freezing response in a CFC paradigm. Normally, mice will freeze when they’re placed in an environment that was previously paired with a footshock. This freezing response (i.e., memory expression) was observed whether the test was done 4 days or 30 days following CFC. The prediction here is that inhibiting the mPFC neurons should only impact remote memory expression (i.e., at 30 days post-training), but not recent memory expression. Such a finding would recapitulate the classical double dissociation of the standard model. This is precisely what they found.

But what if the expression of hM4Di is lower on day 4 than day 30? This would mean that the CNO is less efficacious at inhibiting the mPFC neurons in the former case, which would explain the lack of behavioural effect. Mariana was quick to ease these concerns by showing that hM4Di expression was stable across the 30-day period.

The above experiments indicate that the mPFC neurons activated during CFC are not necessary for recent memory expression, only remote. It was rather interesting to me that these cells would be active so early on but then seemingly lay dormant for a month, after which they became functionally important. My understanding had been that mPFC neurons needed to be entrained over this month-long period and that they’d only become engram-bearing cells after that time had elapsed. Mariana’s poster suggested that mPFC neurons may be recruited at the same time as hippocampal neurons.

If that’s actually the case, then what happens if we stimulate the mPFC neurons? If they are recruited early on – which seems to be the case – then activating them should elicit memory retrieval irrespective of timepoint. After all, the fact that they’re not necessary for the retrieval of recent memory doesn’t tell us much about whether activating them is sufficient to drive memory expression! Mariana and co decided to express hM3Dq, an excitatory DREADD, and asked whether exciting the mPFC neurons would lead to freezing behaviour at recent and remote timepoints.

It did! The mice froze in neutral contexts upon activation of the CFC-tagged mPFC neurons. In fact, CNO elicited freezing whether it was given 4 or 31 days post-acquisition. Note that the neutral contexts were never paired with a footshock so control mice didn’t freeze at all. Freezing only happened when the CFC-engram-harboring neurons were activated.

The above findings present a picture of mPFC neurons being recruited to the memory trace around the time of acquisition. In other words, CFC elicits recruitment of mPFC neurons and this seems to happen almost immediately after conditioning because the tagging, thus far, had been done right after CFC. Yet, it seems these cells are not needed for recent memory retrieval since inhibiting them has no effect on freezing at the 4-day timepoint. This led Mariana to pose the following question: what if the tagging is done at recent or remote retrieval? Are the mPFC neurons active during recent retrieval the same ones that were active at acquisition? If so, tagging the neurons during the day 4 retrieval test and subsequently activating those neurons should lead to freezing.

It turns out that reactivating these mPFC neurons has no effect. The mice did not exhibit any freezing behaviours regardless of whether CNO was given at the recent or remote timepoint. However, CNO did elicit freezing if the mPFC neurons were tagged at remote retrieval. This suggests that the cortical ensembles recruited during training are not called upon during recent retrieval, only during remote retrieval.

Mariana argued that a different cortical ensemble may be activated during recent retrieval; an intriguing idea, to say the least! Her findings are consistent with recent work from Dr. Susumu Tonegawa’s lab showing that neocortical ensembles could harbor the engram but lack the endogenous machinery to drive retrieval. They could show that with optogenetics, one could drive the expression of these memories, consistent with Mariana’s findings (2).

Poster title: “Early memory allocation to cortical ensembles is crticial for remote memory expression”
Presenter: Mariana Matos

1: For a review of systems consolidation, check out this recent paper written by Howard Eichenbaum: https://www.nature.com/articles/nrn.2017.74

2: Kitamura et al., 2017: http://science.sciencemag.org/content/356/6333/73.long

Geith Maal-Bared
PhD Candidate
van der Kooy Lab
University of Toronto
Neuronline: @GeithMB
Twitter: @GeithMB
Website: geithisms.com