In the previous article we talked about the fundamental issues in the computational modeling of brain functions. With the hippocampus being the structure that we would like to understand - and ultimately model - we also took a look at what the hippocampus is involved in when it comes to humans. This gave us the first major clue about the generic hippocampal function across mammals. In this part we'll take a look at the second major clue: spatial processing in rodents.
Spatial processing in the hippocampus.
In 1971 Professor John O'Keefe and his group published their initial report on the so called hippocampal place cells found in rodents (rats and mice, primarily) - arguably one of the crucial milestones in hippocampal research and a discovery that resulted in the 2014 Nobel Price in Physiology and Medicine. Place cells are a distinct type of neurons found in the hippocampal CA3 and CA1 areas (O'Keefe and Dostrovsky, 1971). As the name already suggests they are primarily active when an animal is located within a specific region of a given environment called the cell's place field.
When placed in an unknown setting it usually takes rats about 10-12 minutes of exploration to map an open area of 1-2 square meters (Yan et al., 2003; Wilson and McNaughton, 1993). "Map" hereby refers to the process of the place cells acquiring their individual place fields - in order words, the process of the hippocampus of establishing a neural representation of the surrounding environment. Once such an internal map is established the place fields remain stable and place cells fire reliably if the animal finds itself within them. When placed in a different setting place fields remap to new locations seemingly at random and cells active in one environment may or may not be reused while the animal maps the newly discovered space (Leutgeb et al., 2005). Different mappings - or "spatial representations" as they are usually referred to - can be remembered for weeks at a time (Ziv et al., 2013) which directly links the hippocampal place cell research done in rodents with the memory research done in humans. However, since place cell activity is per definition recorded from individual cells, place field data from humans is very difficult to acquire. In fact, studies that do present such data are usually performed with epilepsy patients who agree to invasive monitoring in order to confirm place cell-like activity in the human hippocampus (Ekstrom et al., 2003).
Direction and distance
The hippocampal place code does not solely consist of place cell activity. In 1990 Dr. Jeffrey Taube published the findings of his group about head direction cells found in the postsubiculum (Taube et al., 1990). As you might imagine, these cells are sensitive to the direction the animal is facing at any given time, irrespective of the actual location of the animal. While this signal is measured in hippocampal neurons, however, it is not computed there and originates in other parts of the brain (the vestibular system is relevant here).
Yet another type of cell involved in spatial processing was discovered in 2005 and shared the already mentioned 2014 Nobel Price in Physiology and Medicine: the grid cells. These neurons are found in the entorhinal cortex, i.e., upstream of the place cells in the hippocampus proper. Each individual grid cell displays multiple firing fields that are aligned with the vertexes of an imagined regular hexagonal grid spanning the environment. These regular grids expressed by each grid cell vary in scale along the entorhinal cortex and thus are able to encode the surrounding space at different levels of resolution (Brun et al., 2008). However, while they differ in scale, all grids are locked in orientation throughout the whole grid cell population; this only changes if something causes the internal spatial representation to "remap" - as is the case if the animal is put into an unknown environment.
One particularly interesting property of grid cells activity is their compartmentalized nature: When entering distinct sections of an environment - for example by entering a room through a small door - the grid population seemingly remaps as well. In other words, while individual grid cells cover an empty space by a continuous grid, if there is a door connecting two such spaces then the grid representation does not simply continue through that opening. Instead, the grid cell population remaps to a different encoding of the neighbouring space, despite there being no physical obstacle in between the two sections (Derdikman et al., 2009). Note how this might also hold true for the place field encoding of a space, but due to the already non-continuous nature of place cells this would be impossible to observe.
In contrast to place cells, both grid cells and head direction cells do not require a explicit period of time in order to stabilize themselves when entering a new environment (Hafting et al., 2005). This suggests that the processes responsible for driving these different neuron populations are not the same and we don't necessarily have to look for, or model, a single function able to produce the activity of grid cells, place cells, and head direction cells at the same time.
At this point, let's take a moment to review what we have just read: There are multiple different sub-populations of neurons within the overall hippocampal formation, each providing a different aspect of encoding spatial information. We have place cells that encode individual locations; we have head direction cells that encode the current facing; and we have grid cells that encode a sort of metric across an environment. In other words, we can identify where we are; we have a continuously maintained compass; and we can track our trajectory. As long as these systems work reliably, this set of tools completely solves the problem of exploration and navigation.
The role of path integration.
Rats have evolved to be able to navigate under low-light conditions and controlling for their sensory capabilities is a difficult task in itself. It is usually assumed that navigation in darkness - i.e., in the absence of access to visual information - largely relies on the ability of an animal to integrate self-motion information, a process known as path integration or dead-reckoning (Etienne and Jeffery, 2004; McNaughton et al., 2006). A fundamental consequence of such a system that relies primarily on tracking its own actuators is the accumulation of a certain drift over time - and any spatial representation that relies on such a system is destined to inherit this drift (McNaughton et al., 1996; Müller and Wehner, 1988).
In order to maintain an accurate and stable representation the system is thus assumed to make use of visual cues in order to continuously anchor the representation and counteract any inherent drift (McNaughton et al., 1996). Consequently, when examining path integration experiments usually take place in darkness in order to observe and measure any accumulating drift. In addition, experimentators have to make sure that animals truly lose their access to any salient sensory cues that would help them to stabilize their path integration system. This includes cues outside of the human senory range, such as the small urine marks that rats commonly leave behind when traversing a space. While the smell of such droplets quickly dissipates, urine is visible under ultraviolet light and rat vision is know to extend into the ultraviolet end of the spectrum, thus quite possibly turning urine markers into visible navigation aids (Desjardins et al., 1973). In addition, most laboratories include a variety of electronic equipment that emits electromagnetic fields that rodents have been shown to be sensitive to (Mather and Baker, 1981). The hearing capabilities of rats also include ultrasound (Kelly and Masterton, 1977; Burn, 2008) which is often emitted from speakers and undetectable by human ears. Ironically, speakers are often used to produce white noise in order to prevent animals from making use of any auditory cues in order to orient themselves. However, this is not to discredit any experimental findings - experimental protocols are usually well aware of these issues - but rather to demonstrate (a) how hippocampal spatial encoding remains stable under different sensory conditions, and (b) that high level features such as orientation, awareness of place, and grid cell functionality do not rely on visual cues alone but rather seem to derive from a whole hierarchy of different sensory modalities.
To close this introduction in spatial encoding within the hippocampus it is important to note that individual neurons are not the exclusive carriers of spatial information. Theta rhythm activity can be measured to travel through the hippocampus as a wave and has been intrinsically linked to place field activity (Buzsaki, 2005; Mizuseki et al., 2012). In addition, the selective strenghening and weaking the links between neurons - known as long term potentiation and long term depression, respectively - are fundamental mechanisms that underly not only to maintaining spatial representations but hippocampal learning in general (Bliss and Lomo, 1973; Bliss and Collingridge, 1993; Lynch et al., 1977; Dayan and Willshaw, 1991; Andersen et al., 2006, p.343-420).
- intermediate summary
- modelling efforts (including our own)
- summary (this is how theory and practice works together in order to solve neuro-questions)
- Scannability (pictures, more visible structure, animations, etc.)
- Author info w/ picture(!)
Continue reading in part IV where we discuss the role of the hippocampus in navigation and gather more evidence for the hippocampus to perform the same universally useful function across different species.
Disclaimer: This series of articles is heavily based on the introduction chapter of my PhD thesis.