Mar 12 2010
Memory and the Hippocampus
Neuroscientists are making steady progress in mapping the brain using fMRI and other new techniques. Researchers at the Wellcome Trust Centre for Neuroimaging at UCL (University College London) have been publishing a steady stream of interesting results.
It has been known for some time that the hippocampus, a small structure in the medial temporal lobe, is important for learning and memory. The structure gets its name from the seahorse, because it looks curled up like the tail of a seahorse. Many details of the anatomy and function of the hippocampus remain to be explored, and the new technology is providing a useful window.
Recently it was discovered that the hippocampus contains what are called place cells – neurons that are activated according to our location in three-dimensional space. These neurons, in essence, process information relating to our location. However, it was not known whether or not these cells are laid out in the hippocampus in a predictable pattern, or if they are essentially random from person to person.
Demis Hassabis and Professor Eleanor Maguire at the Wellcome Trust Centre published a study about a year ago looking at human volunteers while walking around a virtual space. By analyzing the activity of their hippocampus with a computer algorithm they developed, they were able to predict greater than chance where the subjects had been. This strongly implies that there is a regular pattern to the layout of place cells in the hippocampus. Like most anatomical structures, there is almost certainly a great deal of individual variation around a basic layout. This means that greater success rates might be achieved if the algorithm were first calibrated to an individual – perhaps a subject of future research.
Now the same team has published follow up research involving subjects looking at three films, all involving women on a city street performing some task, like mailing a letter. Afterwards they were placed in an fMRI and asked to recall each of the films in turn. The computer algorithm was then able to analyze the pattern of activity in their hippocampuses (there is one on each side) and predict greater than chance which of the three films they were watching. This supports the hypothesis that there is some regularity to the layout of neurons in the hippocampus.
Of course these preliminary results led to extrapolation of this technology to the potential of “reading minds”. This may be theoretically possible, but don’t hold your breath. We are a long way away from the Matrix, where an almost flawless reality can be jacked directly into, and read from, the brain.
What is certainly true is that we are in the middle of an exciting time in neuroscience where new tools have accelerated the research, allowing us to map and understand the brain better than ever before. It seems likely that we will continue to make incremental improvements in our ability to model the brain, and even interpret brain activity. There are researchers mapping out the visual cortex, and able to use fMRI to see what a person is reading. And these researchers are able to see where a person has been or what location they are thinking of (from a limited set of choices).
However, these applications represent the low-hanging-fruit of this research – areas of the brain that are visuo-spatial where the brain has some degree of topical mapping. In other words, the neurons in the brain are laid out physically in a way that maps to the visual or spatial data they represent. The same is true of the motor and sensory cortex – there is, in fact, a homunculus – a representation of the body laid out along the motor cortex and the sensory cortex.
As we get to other areas of the brain, however, it remains to be seen how much this somatotopic mapping will hold up. Abstract areas of the brain involved with math, language, emotion, and similar functions may not be so conveniently organized. I also imagine that the level of complexity will increase not linearly but geometrically or even exponentially. Further, the amount of variation will likely also increase significantly. Systems used to read brain activity may require exhausting calibration.
But there is nothing inherently impossible about such mind-reading technology, it is just likely to become increasingly difficult and complex. I liken it to the problem of artificial intelligence. Fifty years ago computer scientists were starting to see Moore’s Law in action, with steady and stunning increasing in computer power. They optimistically extrapolated that increase and predicted that human level computer intelligence would be reached in 20-30 years. We thought nothing in 1969 about the HAL computer in the movie 2001 – sure, in 32 years we will have AI. How cool.
But the AI problem turned out to be much more difficult than anticipated, not just a matter of simple extrapolation from existing advances, and now we still seem to be 30-50 years away from true AI approaching human level intelligence. Maybe there will be unanticipated hurdles still.
Mind reading computers will likely follow the same pattern. We will make incremental advances, but the real futuristic applications may be more challenging than we think. It is difficult to impossible to predict such things, however. While everyone overestimated our ability to develop AI, everyone also underestimated the true power of computers to revolutionize communication.
As an aside, there is some other recent research involving the hippocampus. As I stated, the hippocampus is involved in both memory and learning. Recent research, however, suggests it may be involved more in processing information than just storing it. These researchers ran rats through a maze and then looked at their hippocampus function with fMRI. They found that the pattern of activation in the rats following learning the maze did not match the routes they learned, but rather correlated to other parts of the maze they explored, and even to parts of the maze they never ventured into.
The researchers interpret this study as possibly meaning that the hippocampus was processing the visuo-spatial information of the maze – going over possible routes and reconstructing the maze, even the parts the rats did not directly explore. They conclude that the hippocampus may be more important for information processing than memory.
It is difficult to make too much out of a single study, but the research does suggest the role and function of the hippocampus may be more complex than existing models suggest (which I think we can assume is probably true of all our current brain models). It also reinforces what I have seen from other research – that memory and information processing are closely linked.
One thing is certain – it is an exciting time for neuroscience.
![[del.icio.us]](http://images.del.icio.us/static/img/delicious.small.gif){kind=small}
Print This Post
13 Responses to “Memory and the Hippocampus”
Hi Steve–great post as always. One point I’d like to add is that, when neuroscientists find cells that appear to accomplish a task, the jump to claiming that the cells are DESIGNED for this task is not an entirely logical leap–for this reason, neuroscientists these days oppose the use of the word “place cells” because it’s a bit misleading–it implies that’s all these cells do.
In fact, place cells have been shown to code not only for a specific physical location, but often a more abstract meaning–as related to context. For this reason neuroscientists like to call the locations they code for “place fields” instead–because the cells can do so much more than just plain location! There’s a really interesting study (Kennedy 2004, 2009) that demonstrates that these cells can also code for the contextual retrieval of memory–i.e. thirst versus hunger can direct the location of the place fields coded by these cells.
In addition, a few other really cool studies (Wood et al., 2000; Smith & Mizumori, 2005) showed that many of these place fields are goal-oriented, i.e. they’ll fire in a particular location depending on where you PLAN on going–or even depending on where you came from!
To me, the whole memory system is pretty fascinating, in that we can sort of see how separate areas have evolved somewhat independently, but still interconnectedly. We have areas for habitual motion, areas for emotionally-charged memories and affective states, navigation, path-finding…and then also “recordings” of memories (episodic), abstract working memory, and we can even be tricked into having false memories. The whole system is very labile, and confusing because so many parts of the brain contribute to “memory”–which itself is not super well defined. The recall and interpretation of a memory requires all of the parts of the brain necessary to understand the information it carries–visual cortex lights up when recalling an episode, for example–so memory is really, really hard to study.
Fantastic blog post. It’s pretty amazing what we are finding out about the brain. I enjoy your posts on these kinds of studies very much.
I thought I would offer you a link to an article on fMRI studies that you might want to comment on, or use for your podcast in a detailed discussion – if the topic of fMRI studies comes up.
A couple of days ago, Misha Angrist (from genomeboy.com) posted on Twitter a link to this ScienceNews article from December: http://bit.ly/dg2vPb which indicated that fMRI studies may not be easily replicated, thus inferring that the findings behind some studies may be flawed.
Ah, CW beat me to it. I also find these studies fascinating but am usually left with questions regarding the reliability of fMRI as well as the algorithms used.
Regarding the hippocampus, do the structures in each hemisphere perform the same functions or do they specialize at all? What happens if one or both is damaged?
We’ve studying these grid cells, place fields, etc. in rats for almost a decade (maybe more?) now. This seems like more of a proof of concept that humans work in roughly the same way.
I wonder about the resolution of fMRI in this instance though. The hippocampus isn’t a large brain organ so discerning where these cells and fields are firing could be confounding. Maybe that’s what this new algorithm is accomplishing?
The hippocampus appears to be redundant bilaterally- not lateralized function. However, many people are dominant on one side – one side has more function than the other. We know this primarily from epilepsy surgery studies, such as Wada tests, where we put one side of the brain asleep at a time to see how much function the person loses.
fMRI are certainly very tricky to do, and I suspect there are a lot of crappy fMRI studies out there. I put more weight on studies that can use the fMRI to make predictions, like this study, rather than just generate pretty pictures.
This is indeed a fascinating story that’s unfolding. I remember feeling shocked and almost incredulous reading some of the mouse studies of the hippocampal “place cells” from Susumu Tonegawa’s group. The fact that we can connect the activity of a neuron to some abstract property of the environment such as context is mind-blowing.
However, I agree with Steve that we can’t get too excited about mind-reading technology. I suspect that when it comes to regions of the neocortex responsible for higher-level abstract thinking, the patterns will be far more specific to the individual and less generalizable. The alternate view, that specific ideas or concepts reside in identifiable locations in the brain is essentially a version of the pretty much defunct “grandmother neuron” theory (that you have a neuron that fires every time you think of your grandmother, etc.).
Those are some interesting studies for sure, thanks for posting them.
Their relevancy strikes me sine I am currently reading ” Dragons of Eden,” by Carl Sagan.
Interesting stuff.
The analogy to AI is useful. Of course the point of an AI would be data analysis so if we do get one to a certain level, reading brains wouldn’t so far away (dollhouse, here we come!) So, its almost not an analogy but the same thing
But the history of AI and machine translation is very humbling.
But where are the madelaine neurons?
A coworker with a penchant for fantastical beliefs was maintaining that fMRI studies have proven that we map objects we are using as part of our body, like tools and golf clubs. He then extended this belief to the airplane, saying we treat the entire airplane as if it were part of us. When I had the temerity to doubt this, he became more vehement; they PROVED it, he said. Yet I remember reading somewhere that fMRI studies are rather dubious–they show blood flow, but this may not correlate precisely to brain function. I couldn’t imagine doing a fMRI on someone swinging a golf club. Meditating monks are the ones I’ve read about, done in a search for “god” or serenity in the brain, but that’s easy since the subject is very still.
Are such studies conducted after the activity takes place, or on people laying in an MRI machine thinking about the activity? It just seems that drawing sweeping conclusions from preliminary studies is premature at best. And what are the limits to this? It’s very interesting that the mice seemed to “know” about parts of the maze they hadn’t visited; this sounds magical and impossible to my inherent skepticism. How could this be? And what are the limits to this mapping? If we enter a building, does that now become part of the body map–the whole building, even a large skyscraper–and what are the limits to this?
Interesting studies–
Nit-pick
to ‘make much of’ is ‘to treat as of great importance’. (American Heritage)
The second to the last paragraph seems to imply the opposite of what I’m sure you mean.
Stephen,
“the hippocampus…gets its name from the seahorse, because it looks curled up like the tail of a seahorse”
Actually, it looks like a seahorse.
http://upload.wikimedia.org/wikipedia/commons/thumb/5/5b/Hippocampus_and_seahorse_cropped.JPG
“It is difficult to make too much out of a single study”
I think you meant: it is *easy* to make too much out of a single study.
regards,
BillyJoe
The link didn’t work. Try this:
http://en.wikipedia.org/wiki/Hippocampus