What is the Hippocampus? Part I: Episodic Memory and Index Theory
It seems appropriate to begin this blog with some background information on the hippocampus, since it will be receiving the lion’s share of my focus. This article will begin with basic hippocampal physiology and function (i.e. its shape, place, and purpose in the brain). I will also introduce episodic memory, which can best be described as the kind of memory that can play back in your mind. This will lead to a longer discussion of how the hippocampus stores these memories; some parts of its functionality are widely agreed upon, while others require further research. Some examples will get us started:
As depicted in the illustration at the left, the hippocampus is a small horn-like structure within the brain’s temporal lobe. As a bilateral neural structure, the hippocampus extends into both hemispheres of the brain. As is hopefully made clear by this image (and the following), it inhabits a significant portion of the lateral ventricle, one of several open spaces within the brain filled mostly with fluid.
The image below comes from an MRI scan of a real human brain. MRI provides a much more faithful depiction of neural structures than CT scans or other representations. You can clearly recognize the long stem ending in a bulb from the picture to the left. The cornu ammonis (or CA divisions) contains the majority of cells that “store” episodic memory in the hooked bulb at the base of the stem. Even the small ridges portrayed at left are visible along the stem, which extends away from the CA into the lateral ventricle. Also evident is its close physical connection to surrounding structures: this image only shows one slice of the brain, but some important structures include the parahippocampal gyrus (imagine a little closer laterally to you at the point of label 1) and the entorhinal cortex (just below the CA). Along with the hippocampus proper, these structures make up the medial-temporal hippocampal system (MTH), largely responsible for storing and retrieving our memories.
Memory is a slippery term in neuroscience: multiple kinds of memory have been identified, including episodic, semantic, spatial and so on. Research suggests that the hippocampus is the seat of our episodic memory, or those memories that can be played in back in the mind; this takes place either by our own internal direction or upon re-activation in reaction to an outside stimulus. The naturally high plasticity of hippocampal cells – especially in CA 3 and an area called the dentate gyrus at the tip of the hippocampal hook – means that new connections can strengthen quickly. This creates a suitable environment for a place in the brain that must constantly update itself. The hippocampus is the center of our memories of childhood, last New Year’s day, last night’s dinner, or even what you were doing five minutes ago. We usually aren’t trying explicitly to encode some or all of those memories though, so how come we remember them at all?
The brain acquires most episodic memories passively – most times we aren’t explicitly trying to remember something as it happens – which implies that the design of hippocampal cells lends itself to the rapid and regular acquisition of new patterns of associations. These patterns form when the brain binds the various elements of our present environment as we become increasingly aware of them. We can, of course, actively seek to acquire memories by focusing our attention on the present moment, or simply a part of it. However, this extra focus does not guarantee retention of the memory, as many a disappointing exam score or forgotten anniversary demonstrate. Despite the extra attention and perhaps meaning we can ascribe via effortful memorization, the kind of binding taking place during the formation of episodic memories appears to be essentially passive.
Another key point about episodic memory is that it is unique. More precisely, individual episodic memories are unique. This point may seem obvious, but it carries important implications. Imagine (you may not have to) that you work the checkout counter at a grocery store, or any job where the days seem to repeat themselves without the slightest variation. At the end of the day you would still be able to separate that day’s events from the one prior, and so on. There are a variety of factors that affect this distinction, but the idea is that these memories are at least somewhat separable.
Perhaps the clearest way to begin understanding how the hippocampus achieves separable encoding is to understand what it is not doing. The hippocampus does not store entire memories; this is impractical. Content information is stored by neurons in the neocortex, the outer neural area starting just at the back of your ears that extends forward and around above the eyes. The neocortex is much larger and less plastic, which means that more information can be stored in an environment subjected to less constant changes than in the hippocampus. Furthermore, pathways from the neocortex containing the many elements of an episodic memory eventually converge and form a single, unified pathway as they approach and enter the hippocampus. A neuron can only send one piece of “information” at a time, so the information being sent would have to compress as the many paths unified. Storing memory content in the hippocampus would result in a great deal of information loss, making our most vivid memories all but impossible (we have them though, so this is clearly not the case).
How then does the hippocampus work? According to prominent neuroscientists David Marr and Tim Teyler, it works a great deal like one of these:
Think of the brain as an organic library made of gray (and white!) matter. The neocortex would be the stacks, shelf after shelf and row upon row of information accessible by simply opening each book (i.e. activating a neuron). The problem? Finding your target book among the billions of neural shelves is nearly impossible without a schema to sort through and localize each book. The hippocampus serves this function. As mentioned above, a group of neurons that fires simultaneously connects as a unified pattern through the hippocampus. This pattern forms what is known as the memory trace: a single and unique set of synapses within the hippocampus that binds the various elements making up a memory. (Just to be sure, a synapse is the junction between the end of one neuron – its axon – and the beginning of another neuron’s cell body or dendrites extending from its cell body.) This system is described by the Indexing Theory of Hippocampal Memory (Index Theory for short). Like a card catalog, or even a book’s index [gasp], this theory suggests that the hippocampus contains no real information beyond the synapses that trace back to the associated elements of a memory.
There are a number of advantages to such a system. First, it affords the brain the ability to keep track of associations powerful enough to bind for a relatively low cost; a single connecting pathway rather than binding the associated elements within the cortex, which would require a much greater number of connective neurons. This leads into a second advantage known as Pattern Separation, to which I referred earlier. As an index, a well-functioning hippocampus creates discrete memory traces, each one objectively distinct from all others. The image below depicts what would happen without the hippocampus: without the orderly connections between points A-D and C-F established by the index in (A), activating point B would unpredictably activate the other members of the two sets as displayed in (B). Closely related memories like family gatherings and trips to a favorite vacation spot (or even multiple similar spots) would be nearly indistinguishable.
It could be suggested that, even with the index, points E and F could be activated by activating point B. Activation at node B lights up the set [A,D], and because C and D make up part of set [C,F], it could trigger as well given significant activation of nodes C and D. In all likelihood, this depends on how closely related to each other those memories are. Accidental activation across memories with a large number of overlapping elements could help explain mistakes in specific memory content. The integrity of the original memory pattern would be damaged by incorporating elements bound together at a different time. Interestingly, simultaneous memory activation can and does occur, but a great deal more control is involved. We frequently activate and maintain two or more memories to compare and contrast their content, either intentionally or subconsciously (potentially while dreaming). This process brings a fairly clear benefit though.
The difference between these possibilities relates to a third advantage of the hippocampal index: By creating biological stimulus-binding points, it naturally abstracts information into concepts. The idea here is that because the memory trace represents a set of points in the cortex, it becomes another data point altogether. To use a visual metaphor, an episodic memory plays like a movie, linking the elements that you remember in a replay-able fashion. Rather than strengthening a number of strong connections between each cortical point, those points each maintain one strong connection through their hippocampal index (the memory trace). And rather than spending the energy to activate sufficient elements of the cortical pattern to trigger a specific memory, the brain can simply activate a single point in the hippocampus that sets off a chain of activation within those elements. With sufficient interaction between cortex and hippocampus, a biological analog of the trace itself extends into the neocortex that can form connections as an independent element with other cortical events.
Keep this in mind: Hippocampal representations are relational: properties of a given stimulus or set of stimuli relate to each other, which includes the context in which they were bound. For this reason, the index is crucial for attaching meaning to a given set of stimuli in its context. Old or new, we come to understand the stimuli that make up our environment by integrating our previous experiences with that set.
This description hints at a number of supporting connections that form over time as the memory trace reactivates, or if some part of the whole cortical pattern activates the trace. It also might suggest that only one index or trace is created for a given cortical pattern. The former problem will have to wait until a more thorough discussion of cells in the MTH system. The latter will be addressed in the next article, which will describe the neural hierarchy of activation (i.e. the direction[s] of neural activation in the brain) and introduce Multiple Trace Theory. Steve McQueen will assist.
[This article draws from Chapter 13 of The Neurobiology of Learning and Memory, titled “The Hippocampus Index and Episodic Memory” by Dr. Jerry W. Rudy.]