Engrams: The Physical Representation of Memory in the Brain

I am working as a baccalaureate researcher at the Genzel Lab in Donders Institute for Brain Cognition and Behavior for the past three months. The lab investigates the role of sleep on memory formation and processing. One of the activities I look forward to every week is the Schema Meeting. The Schema Meeting as the name suggests is like a journal club where we discuss papers on memory schemas, their formation, their interactions with new memory, and their effect on memory retrieval. Briefly, memory schema is an organized group of past experiences and associations which become active depending on the context to help inform decisions and make predictions (Ghosh and Gilboa, 2014). The paper for one of the meetings was ‘Making memories last: the synaptic tagging and capture hypothesis’ by Roger L. Redondo and Richard G.M Morris, Nature, 2011. So while reading, I stumbled upon the concept of engrams, and as with every new terminology, and trust me there are more than a dozen every time, I googled. So what are engrams? The physical representation of a memory. I think my heart skipped a beat, that blew my mind.

Now to give some context, I am an engineer trying to leverage my computational skills to explore my passion to understand the brain. Being new to the field, quite often things blow my mind. For the past few months, I have been studying hippocampal and cortical activities that if observed in a sequence are biomarkers for memory consolidation and how different learning tasks affect these activities. I have been reading how immediate early genes are biomarkers of learning but so far I have been asking myself ‘ but how is memory really stored? What is the physical representation of a memory? Is there a one-to-one mapping between this physical entity and a memory? and if so, then is there a quantification of the maximum capacity of memories that I can store at a time? and how are they related to these activities that I am studying?’ So yeah many questions are yet to be answered but Engrams are a beginning and that too is a huge one.

So what are Engrams?
Engrams are cell assemblies that form preferential synaptic connections to each other during a learning experience ( as a response to a stimulus ) and are reactivated during retrieval of that memory ( in response to the stimulus present during the learning experience ). These cell assemblies undergo enduring offline physical and/or chemical changes that were elicited by learning and underlie the newly formed memory associations. An engram is the physical representation of a memory. Boom. But is it not that simple. Let’s read further.

What does this mean?
This is the first evidence towards tying learning-based behaviour and thoughts to a physical basis. There have been studies that have reported the manipulation of these engrams that has led to loss of a memory, retrieval of a memory in the absence of a cue or stimulus present during learning, combining of two memories, and the formation of artificial memory. For example, in one experiment, the tagged neurons of the Amygdala were surgically removed after the mouse was trained and tested to check if it learnt the memory. When the mouse was presented with the cue present during learning, it failed to retrieve the memory and respond accordingly. In another experiment, hippocampal neurons, responsible for storing information associated with context, tagged during exploration of context A were optogenetically stimulated while the animal was undergoing contextual fear conditioning in a disparate context B. Upon testing the memory, the mice responded to the cue in context A whereas it only trained in context B. It also responded to context B but not a new context C (different from A and B) thereby showing that an artificial memory was created.

What took us so long to find such an important yet seemingly simple phenomenon?
This is not a new concept. In 1904, an evolutionary zoologist turned memory researcher, Richard Semon first coined the term ‘engram’ and defined it as ‘… the enduring though primarily latent modifications in the irritable substance produced by a stimulus… ‘. Further, he proposed that an engram is formed during a learning experience, the involved cells undergo offline and enduring changes physically and chemically to become a part of an engram, and the engram as a whole is reactivated during memory retrieval. However, his hypothesis was discarded due to a lack of technology and knowledge to study the brain at the cell level. A few decades later, Donald Hebb, a psychologist, known for his cell assembly theory, and work on explaining learning-induced behavior in terms of connections between neuron assemblies, hypothesized that a group of cells form a cell assembly if they are active simultaneously during a learning experience and are characterized by growth processes and metabolic changes that strengthen the connections between the cells, and affect each other’s firing. This was similar to the concept proposed by Semon. However, over the last century, there have been tremendous advances in technology required to image and manipulate the brain at a cellular level and engrams or cellular assemblies are being studied to validate their existence and define its properties.

How are they even detected?
Experiments in memory research are designed to have 2 main sections, training and testing. In the first, the subject is trained to learn a task/ experience through multiple repetitions, and then the memory of that task is tested after a day(s) by presenting a cue/ stimulus present during learning. Researchers mark cells active during a memory test using Immediate Early Gene (IEG) immunohistochemistry whereas cells active during a training experience are ‘tagged’ through the use of temporally inducible IEG promoters that drive the expression of more enduring fluorescent (or other) reporter proteins. The above chance overlap between these two cell populations is suggestive of an engram.

Is only one region responsible for storing information associated with a memory?
While cell assemblies are localized to one brain region, experiments have proven that engrams in different brain regions interact with each other during memory retrieval and support different aspects of an experience. For example, in contextual fear memory, the hippocampus, amygdala, and the cortex all have activated cell assemblies that are said to support context, affectual information, and sensory information respectively. These connected engrams spanning different brain regions form an engram complex which is the entire brain-wide engram supporting a memory that is stored in sets of engram cell ensembles in different brain regions connected via an engram cell pathway.

How are multiple memories linked?
We all rarely remember isolated memories but rather remember experiences in connection to one other and form memory schemas and generalizations. What does this mean from a cellular point of view? While there are several cells in a region, the probability of a cell becoming a part of an engram is directly proportional to its intrinsic excitability. When a memory is retrieved, its corresponding cell assembly gets excited. During this period if the subject has another learning experience, these excited cells have a higher probability of getting co-allocated to the engram associated with the second experience, and the two experiences get linked. The retrieval of one memory leads to the retrieval of the other memory due to the activation of the overlapping memory engrams.

How are the cortical and hippocampal activities like ripple, deltas, and spindles that we study as biomarkers of memory consolidation related?
Hippocampus is the brain region associated with encoding new memories for a short term before they are consolidated to become long-term memories upon which it shifts to the cortex. This conversion of memory from its labile to stable form is supported by sharp-wave ripples in the hippocampus and spindles in the cortex and a dialogue between them during nonREM sleep is considered to be a biomarker of memory consolidation. Further, the hippocampal and cortical neurons active during training are reactivated during the sharp-wave ripples. The exact role of sharp-wave ripples with respect to engram cell assemblies is unknown. However, sharp-wave ripples promote synaptic depression of hippocampal neurons. A study suggests that hippocampal engram cells tagged during context exploration are more likely than non-engram neurons to participate in sharp-wave ripple events, perhaps allowing these engram cells to escape this SWS-induced synaptic depression. Thus, sharp-wave ripples are said to contribute to the formation of memory engrams and erasure of irrelevant information during slow-wave sleep by increasing the net synaptic depression necessary to increase neuronal responsiveness.

Do engrams have one-to-one mapping to memories?
Well no. The cell assemblies in an engram change over time and a study shows that the retrieval of a remote memory (13 days) activated a different proportion of cells than the retrieval of that memory 2 days after training the mice. Moreover, the tagged Prefrontal Cortex (PFC) neurons were active 13 days after the training but not 2 days after training, On the other hand, tagged hippocampal neurons were active 2 days after training but not on the 13th day. This is also in agreement with studies that discuss how system consolidation shifts the memory from the HPC to the PFC as it stabilizes from its labile state. But in spite of this time-dependent shift, are these cortical engrams stable, and can they be mapped precisely to a memory? We just discussed how related experiences might be co-allocated cells that belong to engrams associated with the two experiences. Studies have yet to compare engrams from 2 similar but separate memories for the degree of dissimilarity and if stimulations of the assembly or a portion of it can only lead to retrieval of only one memory. There have also been no studies regarding engrams associated with the memory of the same experience resembling across mice.

What’s the current state of the research?
Most of the studies related to engrams are restricted to rodents. Engrams have not been detected in humans. The development of non-invasive methods to conduct such studies in humans may lead to progress in understanding engrams and their properties in humans but till then the research on engrams will focus on rodents which can subsequently be used to understand information storage, processing, and retrieval in humans. A lot of ongoing work focuses on understanding why there is not a 100 % overlap between the cell tagged during training and those activated during testing. This involves questioning the methods used to identify the cells during these two processes. Researchers are also trying to manipulate these engrams for different types of memories, focussing on multiple brain regions to study engram complexes and their interactions during the retrieval of a memory. Studies are also trying to identify the enduring changes that the cells undergo to become a part of an engram and how it changes over time and affects the memory associated with it. From an AI perspective, the structure of the engram, cell assemblies, and the neural mechanisms underlying these engrams that are responsible for linking memories and forming schemas can be used to build algorithms that are better able to generalize learnings.

So there is clearly a lot of work going on. This article addresses just the tip of the iceberg of the research, both past and ongoing, on engrams and the physical representation of memories. There are many questions both in the scientific community and in my not so updated and informed mind. Like, do dreams lead to reactivation of engrams or fusion of two or more engrams to give to artificial memories which are latent? Can they be stimulated artificially? What is the neural basis for recurring dreams? I have only attempted to introduce the concept and answer the questions I had when I first read about engrams in a very layman language for all enthusiasts and curious minds out there to understand and explore further. A great paper that helped me get started, and also one that explains the idea of engrams and the related research comprehensively, without a lot of jargon and a lot more in-depth is Memory Engrams: Recalling the past and imagining the future by Dr. Sheena Josselyn and Dr. Susumu Tonegawa.

Hope you guys enjoyed reading it as much as I have enjoyed assimilating the vast literature on engrams and writing about it. Do suggest additional questions you have and/or corrections to the text.