#SfN18 Day 3 Grass Lecture: A Series of Thoughts
On Monday, the annual Grass Lecture took place, highlighting the work of a high-profile researcher chosen by SfN from within its own ranks. This year, the speaker was David Tank, who has co-directed the Princeton Neuroscience Institute since 2006. Before going to Princeton, he did research at Bell Labs, helping to develop and advance fMRI techniques. While at Princeton, his work has generally focused on how the brain internally represents and encodes its own experiences.
His first studies at Princeton focused primarily on how the brain represents location. In 2009, he and his student Chris Harvey invented the first virtual reality environment for mouse experiments, keeping the mouse in place on a rolling ball while IMAX-style screens surrounding the mouse display images giving the impression of motion along a linear track. Using this technique, they were able to not just observe sequences of place cells firing in the hippocampus, but they achieved enough stability to do the first intracellular identification of a place cell while spiking, which helped show that these were actually cells, and not just small regions with similar activity.
A mouse in the virtual reality environment.
He did more work for a while expanding this task from a 1-dimensional line in virtual reality to a 2-dimensional environment, but soon he asked an even more interesting question: are these sequences of cells firing specific to location, or is it just one aspect of a more general phenomenon? To find the answer, they allowed mice to experience sound as well as place, playing a series of tones while the mouse navigated an environment. To their surprise, they found that certain cells in the hippocampus fired in sequence to tone as well as location, but the two populations seemed to sample from each other. In other words, the brain used the same method to track multiple modes of information over time: the map was not just spatial, but possibly cognitive.
Next, they took this idea one step further and fully decoupled the task from physical location, relying solely on sound. Again, a sequence of tones would play, but the mouse would remain stationary at a lever, and had to push it when the tones reached a certain frequency, all the while recording how their brain cells’ activity changed. They found the same thing here as well: their brain also represented the tone as a sequence of activity across brain cells, just like it would represent a physical sequence in space.
The sound-exclusive sequence task.
Interestingly, this general purpose mapping can allow for the reverse to work as well. Sequences in space can actually get reconstructed from neural activity alone in a process known as manifold inference, based on what we know about how these sequences represent the external world. These inferred positions can then be compared to the animal’s real position throughout, and most of the time the two seem pretty similar. However, at certain points the two drift apart. The exact reason for this is unknown, but it may result from our representations coming from both an external and an internal state: when the external state dominates, the two remain similar, but when the internal state predominates, the two could drift apart.
In fact, this drift may specifically represent the difference between our internal representation and external reality. The drift seems especially pronounced when mice get predictions wrong, where our external reality does not match how we represent it internally. Further, across many series of many activities, paths gradually diverge from each other as time goes on, but they generally return to the same point in the end. When an error is made, however, the paths do not return to the same point in the end, but the difference becomes even more pronounced instead.
Overall, the brain seems to represent experiences as sequences of individual cells firing, regardless of what this experience actually is. This paradigm could provide a key to decoding the numerous mysteries of the brain, like how we observe the world around us or how to tell whether our internal representations are accurate. Such possibilities are immensely promising, and hopefully David Tank will have much more to show us for a long time to come.