Notes on Lashley's search for the engram

26 May, 2024

What is a memory? What does it mean for us to remember something? How do we remember our grandparents? Is there some specific neuron somewhere in my brain that has "grandmother" encoded in its chemical structure?

The engram is a theoretical idea that memories are encoded in some physical unit. In the engram, my grandmother is indeed a singular or group of neurons located in my brain, and if I were to remove or damage that neuron, I would lose the memory of my grandmother. We've observed amnesia in patients with physical brain damage, and so some notion of a physical relationship to memory must exist... right?

No doubt understanding the engram would provide a gigantic boost to the quest for artificial general intelligence. If we could diagram out the physical encoding of a memory, then it's an easy logical step to replicate those same exchanges in an alternative physical structure. Yet it remains elusive. We don't yet understand how memory works.

In the latter years of his career, psychologist Karl Lashley went "In Search of the Engram," seeking to narrow down the physical region of the brain where memories get stored. My notes today are on this paper. As the writing style and approach of a 1950s psychology paper are vastly different than a modern computer science paper, I'll give some spoilers: he doesn't find the engram.

Introduction and Discussion of Methodology

Descartes attempted to describe the notion of memory as the pineal gland vibrating, which jostled "animal spirits" toward the brain, which stored the traces of whatever is being remembered. More recent memories are easier to access as the "pores" are more open, leaving more accessible channels for these animal spirits. Descartes was seemingly close: replace animal spirits for nerve impulses and pores for synapses, and the general idea is there. Descartes didn't know about electricity yet.

Over time, neuroscience (though it wasn't called that yet) started leaning toward the idea of specific areas of the brain holding different functions. Sensory areas were distinct from motor control areas, and loss of speech or movement was suspected to be due to damage to those areas of the brain. As a result, the idea that the engram might reside in the area of the brain closest to the nature of the memory gained traction. A ballet dancer might store the memory of a pirouette in the motor control areas of the brain, for example. We still couldn't find the engram, and some extreme positions took the idea that the engram might a single brain cell. If we couldn't find specific memories, we just got the cell wrong.

Physiologists, meanwhile, were concerning themselves with conditioned reflexes. Was a reflex stored in the brain, in a muscle, in a collection of nerves, or something else? When we learn a reflex, where do we put it? This conditioned reflex idea was the focus of Lashley's research as an idea of targeting a specific memory. He took two approaches. First, he generated specific sensory responses in animals and analyzed the reactions that were taken, similar to the famous Pavlovian responses in dogs. Second, (and this is the type of experiment which likely wouldn't fly today) he trained animals to perform certain tasks, then destroyed different parts of their brains to see how the memory was affected.

Motor Cortex

Lashley trained mice to jump to a white triangle and avoid a white X when the two figures are set against a black background, but to do the reverse when the figures are set against a striped background. (This was the most difficult task mice could be taught.) After firmly encoding this response into the mice, Lashley then removed the motor cortex or severed its connections with the sensory areas of the brain. Whether this surgical removal was performed before or after the learning, it had no effect: mice were still perfectly capable of learning the response. The same idea was tried with monkeys, which were trained to open various lockboxes. The motor areas were subsequently removed. The monkeys required a recovery period after the operation, but after 12 weeks, the monkeys were still able to open the lockboxes in the same manner as before.

Lashley noted that the traditional view of the motor cortex was that it was responsible for integrating all voluntary movements. The experiments proved that view to be incorrect. Lashley hypothesized that the conditioned reflex was part of the cerebellum and basal ganglia, but could not prove his hypothesis with existing experiments. Some additional areas were probed, but results were inconsistent. There was no single consistent location where memories seemed to be reliably stored.

Transcortical Conduction

Lashley trained mice to run through a particular maze, then sliced through the cortex at different angles, positing that the conditioned reflex might be stored across the brain in different areas. This yielded a bit more insight: if the incision separated sensory areas from other parts of the cortex, the habit was lost, but any other slices did not have an effect on memory.

There were similar results observed in the monkeys. The experimenters repeated this experiment while being careful to maintain barebones connections between sensory inputs and the rest of the brain (so at least the input from the eyes and hands could reach the rest of the brain). When this was done, the monkeys could still remember the habits, even if the rest of the brain connections were severed.

Lashley is troubled by these results, as they point to almost no additional understanding of the brain. His results are somewhat confirmed by human observations: when patients received similar severance to help with epilepsy conditions, it also did not affect overall conditioned responses. Humans stayed coordinated and could easily remember what their bodies were supposed to do. Lashley comes to the following conclusion:

It is difficult to interpret such findings, but I think that they point to the conclusion that the associative connexions or memory traces of the conditioned reflex do not extend across the cortex as well-defined arcs or paths. Such arcs are either diffused through all parts of the cortex, pass by relay through lower centres, or do not exist.

Short answer: I have no idea.

The Problem of the Association Areas

Prefrontal and frontal areas were considered to be locations of higher functions, as they are more prominent regions of the brain in more evolved organisms and simpler in lesser organisms. Thus, our consciousness, personality, complex memories, etc. must all be stored here. Lashley notes this does not appear to be true, either. In the latchbox experiment, some monkeys had their prefrontal areas separated from the rest of the brain. Despite this, they showed no problems opening the latchboxes. Tasks that required sequential operations might be lost, as the monkeys showed difficulty jumping from one task to the next. Different tasks were tried around recognizing shapes, with no loss in the memory.

Lashley concludes that, after removing any different parts of the frontal and prefrontal cortex in monkeys, that memory does not appear to be stored in some particular complex region of the brain. Even destruction of the prefrontal, parietal, occipital, and temporal areas doesn't have an effect. Lashley supposes that memory could somehow be stored differently in humans, but considers it very unlikely: the similarities are just too strong between monkeys and humans in these areas.

The Role of Subcortical Structures

Perhaps we don't form memories in our cortex at all, but rather in some other structure within the brain. As most of these structures are buried deep within the brain, it is extremely challenging to perform a direct experiment on them. Lashley notes that some animals with no cortexes have been shown to learn basic associations, though with extreme slowness relative to a cortical animal.

The basal ganglia are associated with a number of degenerative diseases in human beings, allowing Lashley to observe the effect of reflex formation in humans with impaired basal ganglia. (Epilepsy, Huntington's, and Parkinson's all affect the basal ganglia) When the basal ganglia are compromised, motor skills decline, but it does not affect our ability to learn motor coordination: meaning it's unlikely that physical memories are entirely stored here.

It is a long-held belief that while memories may be initially formed and stored in the cortex, over time, they drop into subcortical areas. The evidence for this is the resistance of old habits to erasure, while new habits are more fragile. The logical idea is that consciousness is a function of the cortex, old habits eventually become subconscious and automatic, therefore they must become part of the subcortex. This is a leap of faith that is not quite proven with the evidence.

Lashley cites an earlier experiment training mice to discriminate based on light: a dark alley should trigger a different reaction than a well-lit one. When the visual cortex is removed in this case, it completely removes the habit. However, the animals are able to re-learn the habit. If parts of the thalamus and the optic tectum are removed, the mouse is no longer able to learn the habit at all. This points to an idea that if the cortex is undamaged, learning occurs there, but if the cortex is removed, learning moves elsewhere in the brain.

The Engram Within Sensory Areas (Equipotential Regions) and the Law of Mass Action

We now get into the theory for which Lashley is best known: equipotentiality. It is the capacity of the brain to distribute functionality across different parts of the brain, such that if one part is damaged, functions can be taken over by other parts of the brain. This is subject to the law of mass action: that if the brain is damaged, the capability of the entire brain is damaged proportionately, not any one part more than others. Habits might be temporarily impaired, but over time, the brain will redistribute the load.

Lashley proved that this capability could be stretched to the extreme. In one experiment, 98% of the visual cortext was damaged, yet the brain still maintained the ability to distinguish between visual features. In another, monkeys had most of their occipital lobes removed, including most of the ability to process visual images. As long as some capacity to do so was retained, the monkeys maintained their visual memories.

If mice trained in the maze have some small portion of their cortex removed, then their ability to navigate the maze remains mostly intact. If 50% or more is removed, the mouse may have to completely relearn the maze, but it will still maintain the capacity to learn. This does not matter which parts of the cortex are removed: the result is the same.

However... this is not entirely consistent. Lashley notes that if specific regions of the cortex associated with processing sensory information are removed, the ability to learn the habit is severely diminished. A specific set of blind mice were taught to navigate the maze using no visual clues. When their visual cortex was removed, though, they lost the ability to learn the maze! Even though their minds never processed visual information, it appears that some manner of spatial information is processed in the visual cortex regardless. No part of the brain is wasted.

The Complexity of Memory Traces

Lashley notes that the experiments conducted deal only with a very small portion of memory: that of conditioned habits, which can be learned by animals and observed in experimental data. More complex memories, such as your memory of reading this blog post, are different. Given the complexity of observations in even these small memories, no doubt more complex memories have even more complicated brain patterns.

Additionally, the notion of stimulus and response is entirely too simple for this type of memory formation. The experimenter may consider the structure of a maze to be the variable, but there's far more: the smell of the wood, light in the lab, feel of the ground may all play a role in forming memory. Lashley specifically notes an example of a rat trained to respond to a triangle, but forgetting that training once the triangle was rotated. Additionally, some memory of the training methods themselves must be retained, whether they be a negative (electric shock) or positive (cheese reward) method. We constantly make associations between one set of memories and our entire corpus of available knowledge.

Summary

Lashley notes six learnings from the experimental data that have now become commonplace in cognitive science studies and our understanding of the brain:

1. Reflex paths going straight from sense organ to motor cortex do not appear to be required. The motor cortex can be removed and reflexes still formed.
2. Isolated memories cannot be demonstrated. The entire brain appears to weigh in for any given memory.
3. Associative areas do not appear to be storehouses for individual memories.
4. Reflex traces are not isolated connections between sensory and motor elements: there is an entire substratum of activity forming a reflex.
5. Equivalent reflex traces appear to be established in multiple parts of the brain at once, or the brain is capable of filling in the gaps somehow.
6. There does not appear to be any region of cells held in reserve for memories. All parts of the brain appear to be firing at once.