The memory trace, the engram, and the big confusion

I firmly believe that many untold assumptions about how the brain works bias the questions we address in memory research.

To me, understanding how the brain forms persistent memories is one of the most fascinating goals of neuroscience research.

Two (or one?) key concepts in memory research

I’ve recently come across many papers using two important concepts for memory research: Memory trace and Engram.

While these two interrelated concepts are meant to help us in the study of memory, academics do not agree on their definitions this leads to awkward, paradoxical and counterintuitive uses of engram and memory trace. So, what is a memory trace? and what is the engram?

What is a memory trace?

According to many online dictionaries (I’ve sampled one here)

The“memory trace”is:
“the hypothetical structural alteration in brain cells following learning. See also engram.”

While“the engram”is:
“a presumed encoding in neural tissue that provides a physical basis for the persistence of memory; a memory trace.”

These definitions are pretty straightforward. However, they make an important assumption (actually there are many assumptions – some are wrong – but I’ll focus on the most important for the sake of this discussion): Memory trace and engram are equivalent (or synonyms, or, at least, very very similar).

It would be very silly to limit our search for a definitive answer to the question“what is a memory trace” to just a random online dictionary search …

So I turned to PubMed hoping to find a more formal definition (or at least some sort of academic approval to the definitions above).

Two interesting papers about the engram

I found two magnificent reviews on the topic “what is the engram and how to track it down”: Finding the engram – by the Josselyn (@sjo09) and Frankland (@franklandlab) Labs
Memory Engram Cells Have Come of Age – by Tonegawa Lab (Steve Ramirez one of the authors – @Okaysteve – is on twitter)

These two papers are an excellent read but I came to a brutal halt on the first two pages:

Tonegawa et al, 2015 – page 1: ” ‘Engram’ is roughly equivalent to “memory trace”, the term used by some contemporary neuroscientists.”

Josselyn et al, 2015 – page 2: “… the engram can be considered similar to a memory trace (and engraphy is the process used to form the engram)…”

WOW these two sentences really threw me off! In particular, the use of the adverb ROUGHLY associated with the adjective EQUIVALENT – it almost sounded like anoxymoron – in my head.

The same concept was present in @sjo09 and @franklandlab‘s sentence: the similarity between memory trace and engram. So they are equivalent but not really equivalent.

This kind of language reveals the hidden turmoil of these scientists, and the whole academic community, facing complex topics in memory research

At this point I was completely hooked up! I decided to dig deeper to let emerge (if existed) the uncertainty on the definition of these two key concepts in memory research.

Pubmed text analysis – Methods

– The abstracts were downloaded from PubMed using the following query “memory trace”[All Fields] AND “animals”[MeSH Terms:noexp] (n = 604 abstracts)

– Each abstract was split into sentences and only the sentences containing the words “memory trace(s)” were retained for the analysis (n = 734 sentences).

– Each sentence was manually classified into 3 mutually exclusive categories:Cognitive, Physical or Ambiguous.

Simply put:

The memory trace can be viewed as a physical trace, a change, occurring on the neural tissue functional to the production of a persistent memory.

Alternatively, a memory trace can be viewed as a chunk of information (“A memory”). This piece of information (“A memory”) can then be “processed” or “stored”. This definition implies the existence of stores – see the Atkinson-Shiffrin model – and it is firmly grounded in the cognitive area.

Hence, sentences can be classified as
COGNITIVE (e.g. “the storage of the memory trace”, “the retrieval of the memory trace”, “the neural correlates of the memory trace”)
PHYSICAL (“the formation of the memory trace in < brain region>”, “neurons are part of the memory trace”)
AMBIGUOUS (e.g. “reactivation of the memory trace” – this could either be a reactivation of the circuit or the reinstatement/retrieval of the memory trace through exposure of the original stimuli that induced the memory in the first place)

Pubmed text analysis – Results

Long story short, after classifying 624 (out of 734) sentences this is what I found:


I see a clear trend in the data. There is a significant increase in the usage of “physical” sentences (i.e. the “memory trace” is a physical trace of a memory process) (red dots + red regression line).


I believe this trend reflects our (almost obsessive) efforts to find “the” neural basis of memory processes

We started off by using “memory trace” in its cognitive conception. However, the evolution of techniques we have witnessed in the past 10-15 years has fundamentally changed the way we interrogate the brain tissue and this urged scientists to “unconsciously revise” the cognitive concepts that we use to guide our research.

The present/future

I’m not saying that this revision is bad or wrong. On the contrary, this is REALLY exciting! I believe we are reaching a turning point! I think we should let this hidden force emerge and not let it unconsciously haunt our experiments!

The evolution of new techniques to study the brain are changing our point of view: are the traditional cognitive concepts still adequate in face of this unprecedented level of spatial/temporal/multiple-dimensional resolution of brain tissue analysis?

Should we invest more time and energy rethinking the models we use to approach the brain?

Well yes! That is why I felt really connected to Tonegawa’s paper and the “tentative” formality of the engram technology approach (this ‘formality’ alone is worth a blog post on its own …)

Im not ready to start a cognitive re-revolution (see the cognitive revolution)… but we should start with something simple: How about a simple distinction between memory trace and engram? The memory trace is the chunk of information (“A” memory) that can be encoded in an engram (neural substrate supporting the persistence of the memory trace).

This is far from being formal – I already see many problems with this, but it’s a first step 🙂

I think Tonegawa’s lab almost got it right … until I stumbled upon their recent paper on the topic:

Memory engram storage and retrieval
Tonegawa, Pignatelli, Roy and Ryan. Curr. Opinion Neurobiol 2015

If the engram is a physical trace (e.g. “the enduring physical and/or chemical changes that were elicited by learning” – as stated in their 2015 paper) how is it possible to store it (i.e. can you “store” chemical changes)?


The cognitive conception of “memory trace” is way more prominent in russian papers than in English/American papers

In some old papers (1952-1960) memory trace and engram occur together with different conceptions


[0] This entire blog is haunted by the Cartesian Dualism. You deal with it …

[1] The analysis is limited to the animal literature. This may bias the results towards a “physical interpretation” of the memory trace

[2] It’s assumed that the abstract is the condensed knowledge of the paper

[3] I haven’t classified all the sentences… there’s ~100 more to go. I’ll do it eventually …

[4] The classification is arbitrary, I should sit down with more people and agree on a classification scheme – but I think I got it “almost right” 🙂

[5] There are more sentences than papers: this means that the some abstracts may contain several “memory trace” sentences. This is actually very nice because sometimes I would find in the same abstract contrasting conceptions (physical vs cognitive) of memory trace – material for future posts (?)

[6] Papers published before the 60’s have been excluded from this quick analysis- there weren’t many paper and they may bias the analysis. I’ll figure out a way to include them in a more formal way in the analysis

[7] This is far from being an in depth analysis – there is plenty of interesting stuff to look for in this dataset – it’s just material for future posts 🙂

[8] linear regression may not be the best tool to detect trends in publications

[9] I haven’t done any analysis on the word “engram”. Understanding the co-occurence of the two may reveal interesting stuff – note to self: do the analysis and write another blog post!

[10] No sentences have been reported in this blog post: I’ll post some examples once I find a clever way to sum up all this information!


Leonardo Restivo

Behavioral Neuroscientist, M.Sc., Ph.D. - Passionate about Behavior, Data Visualization & Psychology. Read my CV+résumé. Follow me on twitter @scipleneuro