SCARED MICE AND THE WORKINGS OF THE MIND
(Image from godrunning.com)
A while ago, I posted an essay explaining how a technique called optogenetics is helping us gain more understanding of how the brain works. However, it will only allow a partial understanding. If you think of the brain as being sort of like a computer, optogenetics helps us figure out the ‘hardware’ but we also need the ‘software’. This requires ‘cracking the neural code’. That is, the rules the brain follows to convert collections of electrical impulses into perception, memory, knowledge and behaviour. Scared mice are being used to help crack this code.
Among the teams working on cracking the neural code is a group lead by Joe T Tsien. As Tsien himself explained, “we study the questions that many people are curious about. How the brain works, how memory works. We then take it down to different levels. What is the molecular basis for the memory level process? That means what genes are involved in laying down memory at a very fundamental level?”.
Investigating memory at this level, Tsien had determined what molecules are critical to the process, and had used this knowledge to genetically-engineer a genius mouse that the team nicknamed ‘Doogie’. But while this achievement leant more weight to the materialist philosophy of mind, ‘memory’ was still rather mysterious. As Tsien said, “nobody knew how, exactly, the activation of nerve cells in the brain represents memory”. So, he and his team set out to “find a way to describe, mathematically and physiologically, what memory is”.
(Joe Tsien and Doogie. Image from the Augusta Chronicle.)
Along with Longian Ling, Tsien developed a recording device that would enable them to monitor the activities of hundreds of nerve cells. This probe was set up to record activity in a region of the hippocampus called CA1, which was already known to be key in memory formation. With their brains rigged to record any activity going on, mice were put through a series of experiments designed to be mildly alarming while not causing actual harm. The reason why startling events were chosen is because they tend to produce strong and lasting memories, and that requires a large number of cells in the hippocampus. This made it more likely that the team “would be able to find cells activated by the experience and gather enough data to unravel any patterns involved in the process”.
The mice were subjected to an ‘earthquake’ (put inside a container, which was shaken), an ‘owl attack’ (simulated by a puff of air to the mouse’s back) and an ‘elevator drop’ (put in a box that was allowed to free-fall a short distance). Each mouse was put through each event seven times, with each episode separated by several hours of rest. At all times, the brain was monitored for any activity.
The data gathered during the experiment was then analyzed using pattern recognition methods, especially one called ‘Multiple Discrimimant Analysis’. MDA can be thought of as a kind of translation tool that converts the native language of neurons (which we do not understand) into a visual format we can make sense of. Tsien himself described MDA as “a mathematical subspace capable of discriminating distinct patterns generated by different effects”. The team projected the data they had gathered from an individual mouse onto MDA’s 3D graph, and it showed four distinct “bubbles”, three for each startling episode and one for when the mouse was resting. What these bubbles represented was distinct patterns of activity in the CA1 neural ensembles.
(The hippocampus, which is involved in memory formation. Image from
The MDA analysis was repeated again and again for different times of that animal’s experience, thus enabling the team to see how the patterns evolved dynamically, allowing them to see more clearly how the animal was laying down memories of each event. They then used another method called ‘Hierarchical Clustering Analysis’ with the sequential MDA method in order to figure out how the network of neurons were encoding different events.
Using these methods, the team discovered that a particular group of neurons was firing for every event. It seemed reasonable to assume this group were responding to something all events had in common: The fact that they were startling. Tsien calls these distinct subsets of neural populations ‘neural cliques’ explaining, “a clique is a group of neurons that respond similarly to a select event and thus operate collectively as a robust coding unit”. It was also determined that each event was represented by a set of neural cliques encoding different features of an event. Because some cliques were activated by all three episodes, others by fewer, and others by only one specific kind of event, the team theorised that “information about those episodic events is represented by neural clique assemblies that are invariantly organized hierarchically (from general to specific)”.
Say you put a mouse through the elevator drop. You get a neural clique that also appears in the ‘earthquake’ and ‘owl attack/air puff’. Call this a ‘Startle Clique’. You also get a clique that is present during the ‘earthquake’ but not the ‘owl attack/ air puff’. What does the earthquake have in common with the elevator drop that is not shared by the owl attack? Well, the former two involve some kind of motion whereas the latter does not. So, call this a “general motion clique”. You get a clique that is activated only by the elevator drop. It must therefore be encoding specific details of motion not shared by the “earthquake” event: A ‘drop clique’.
Drawing on this discovery, Joe Tsien explained, “the brain relies on memory-encoding cliques to record and extract different features of some event, and it essentially arranges the information relating to a given event into a pyramid whose levels are arranged hierarchically from the most general, abstract features to the most specific. We also believe that each such pyramid can be thought of as a component of a polyhedron that represents all events falling into a shared category”.
(Image from researchgate)
One useful aspect of this way of organizing memory is that information extracted from a novel experience can be integrated with past experiences that have something in common with it (whether it be specific details or more, general abstract ones). What the brain essentially does, is it substitutes the specific cliques that sit on the apex of the memory pyramid. So, what this combinatorial, hierarchical approach to memory formation provides, is a way for the brain to encode key features of specific episodes while simultaneously extracting general information that it can apply to future events- ones that may share some essential details but differ in other ways.
Having uncovered the basic mechanism of memory formation, the team then set about devising a method that would allow them to compare patterns from one brain to another; pass information from a brain to a computer and even decipher what someone remembers and thinks. To do this, they used a mathematical treatment called ‘Matrix Inversion’. This enabled them to translate neural clique assemblies into a string of binary code, where a 1 shows a clique is active and a 0 shows inactivity. Because each memory pyramid generates a unique string of binary code, simply by scanning the code the team could infer what experience the mouse had been through and where it had happened, with up to 99% accuracy.
(Image from wikimedia commons)
Considering the future applications of this work, Tsien said, “realtime processing of memory might, one day, lead to memories downloaded directly to computers for permanent digital storage…Someday, intelligent computers and machines…with a logical architecture similar to the hierarchical organization of memory-encoding units in the hippocampus might…even exceed our human ability to handle complex cognitive tasks”.
REFERENCES: “World Wide Mind” by Michael Chorost
“The Memory Code” by Joe Tsien
This essay was originally posted on my blog.