Animals create a mental map of their environment by leaping, running, swimming, and scurrying through it. This map is used to find food, navigate home, and locate other vital points. For decades, neuroscientists have been trying to figure out how animals do it. Researchers discovered a sophisticated neural code in rats' brains by studying them in laboratory conditions. This is a crucial part of the solution. This landmark discovery won a Nobel Prize in 2014. Many scientists believe that the code may be key to understanding how other abstract information is processed by the brain.
However, lab animals can only navigate in two dimensions using a box with a flat surface. Researchers are finding that applying the lessons learned from that experiment to the real world presents many challenges and pitfalls. Scientists working with rats and bats discovered that 3D spaces are encoded differently by the brain than 2D ones. This was revealed in a pair recently published in Nature Neuroscience and Nature Neuroscience. It is a mechanism they still struggle to understand.
We were expecting something completely different, stated Nachum Ulanovsky (a neurobiologist at Israel's Weizmann Institute of Science) who was the leader of the work in Nature. He has been studying neural representations of 3D spaces for over 10 years. Our thinking had to be re-evaluated.
These findings suggest that neuroscientists may need to reexamine what they believed they knew about the brain's ability to encode natural environments and how animals navigate them. Researchers have been able to see that memory and other cognitive processes might be operating in a different way than previously thought.
Grid Cells go 3D when they are used in grid cells
Many types of neurons are involved in brain navigation. This has been proven over many decades. When an animal moves through a specific area in its environment, place cells in the hippocampal erupt. When an animal's head points in a particular direction, such as north or south, its head direction cells will fire. Border cells light up at certain distances from a boundary.
The most fascinating grid cells are located in the brain region close to the hippocampus. This is called the entorhinal cortex and plays an important role both in spatial navigation as well as memory. These neurons are activated when an animal navigates a two-dimensional space, such as a maze or flat room. Each grid cell's locations are organized like the vertices in a hexagonal lattice floor tile. Different grid cells can fire hexagonal patterns at different spatial scales. Offsets allow them to cover the whole 2D plane.
Grid cells are often viewed as a simple, consistent, and consistent way for animals to track exact distances and directions. Researchers have found evidence that grid cells may be using the hexagonal code to not only represent physical spaces but also abstract cognitive spaces.
However, all the experiments were done in 2D environments so it is not clear how grid cells could represent environments with three dimensions (or higher in the case cognitive spaces).
Kate Jeffery is a University College London behavioral neuroscientist who, like Ulanovsky, has sought to answer this question for over a decade. She and her colleagues created a playground for her rats, adding spiral staircases, climbing walls, and jungle gyms to get them exploring the vertical dimension. Jeffery was able to observe the animals as they explored these contraptions and found clues about how grid cells could extend their regular patterns into 3D space.
If the system were to simply apply its 2D optimal packing to the 3D dimension, researchers would expect grid cells to fire in spherical spots, neatly arranged in an hexagonal 3D lattice structure much like a stack oranges in a grocery shop. There were already signs that something was more. Grid patterns didn't always look perfectly symmetrical in only two dimensions. For example, researchers discovered that changing the room's geometry could cause grids to shift and pull against each other, causing them to lose their rigid periodicity and activity. It was also observed that the grids became warped in areas of significance to rats or where there were rewards.
It was possible, however, that these observations were simply deviations from the hexagonal framework. Researchers were able to record animals using grid cells to navigate 3D spaces. The results became more dramatic. Ulanovsky stated that the findings showed not only deviations but also departures from the framework.
Jeffery and her team spent years perfecting the technology and experimental setup. This included creating a lattice-like climbing frame for rats, setting up wireless recording and three dimensional tracking systems, and finally being able to observe grid cell activity during 3D navigation.
Surprised, researchers found no trace of the hexagonal patterns that governed cells behavior in 2D. Instead, grid cell activity appeared to have been distributed randomly throughout three-dimensional space. Jeffery stated that some properties were retained, but the most striking visual property of grid cells wasn't.
Ulanovsky, on the other hand, was seeing something similar in Egyptian fruit bats while they flew through a large space. It was actually difficult for him and his team to understand what they were seeing when they started recording from grid cells almost 10 years ago. Ulanovsky stated that it took them two to three years to reach the point where they were actually beginning to see the right track.
The bat grid cells of Jefferys rats did not appear to fire in a three-dimensional hexagonal pattern, just like Jefferys rats. Extensive analyses revealed that the cells had no consistent global structure.
However, the firing of grid cells was not random. There was a local order to the firing of grid cells. Although the grid cells were not arranged in a perfectly periodic lattice pattern, the distances between them were too consistent for it to be random. Researchers saw something more than a neat stack of oranges. They were instead seeing something less organized, more like marbles in a box. Ulanovsky stated that they are always stuck in a local minimum. The local distances are fixed because the [marbles] touch their neighbors.
Jeffery stated that the pattern everyone loves and which has inspired many theoretical studies, wasn't there. Perhaps the grid cell's regularity isn't the most important thing, even though it is the most fascinating thing to us.
Be captivated by the Elegance
Loren Frank, a neuroscientist from the University of California San Francisco, stated that it was a beautiful thing to discover the hexagonal periodicity of grid cell grid cells in two dimensions. He wasn't involved in either study. This is what happens in science. When something is beautiful, people attach a lot of importance to it.
He said that the absence of a crystalline structure for grid cells firing in three dimensions forces one to take a step back and ask, "OK, have I been imbuing my particular network with too many capabilities?"
The results show that the brain's spatial map is not as precise as models suggest. This is especially true in natural environments where landmarks, obstacles and other complexities can affect the landscape. Jeffery stated that the mental map may not be used to plot out precise geometric relationships among reference points. Instead, it might create broader connections and a loose metric to be applied to the world. We can then build topological and adjacency relationships. Frank compares it to having a map that shows the subway system in a city. This gives Frank a sense and idea of connectivity, but not necessarily a precise distance.
Scientists will be able to change the way they think about path integration. This is an animal's ability to determine exactly where it is in space relative to its starting point without external cues. It is often attributed grid cells. Path integration is thought to allow animals to find their way home over long distances using novel shortcuts and other abilities. Ulanovsky points out that these hypotheses assume grid cells map landscapes with perfect periodicity. He said that if it isn't perfect, this whole idea crumbles and you can't encode your position robustly any more using current grid cell models.
The new results raise concerns about grid-based distance estimation and path integration mechanisms. Roddy Grieves, who was the first author of their paper and a postdoctoral researcher at Dartmouth College, stated that we need to examine how this works without strict hexagonal periodicity.
Some models that depict animals' navigation across long distances and toward goals might need to be updated. Ila Fiete is a neurobiologist from the McGovern Institute of Brain Research at Massachusetts Institute of Technology. She has done theoretical modeling work about grid codes and speculates that the brain may perceive higher dimensions differently than flat surfaces because there is no global structure in grid cell activity. She suggested that 3D and higher dimensions might not allow for a seamless representation of all space. Perhaps the brain uses a different strategy.
Moving beyond the Grid
Generally, theories about grid cell activity and the associated functions involve continuous attractor models where each grid cell seeks out to activate its neighbor while suppressing theirs. This results in the hexagonal pattern of local excitation and inhibition, which can be seen in 2D. These models don't predict the collapse of periodicity in 3D. Grieves stated that the grid cells are very rigid in terms of how they are connected and what arrangement their firing fields can take. These models are heavily doubted by our 3D data and the flying bats data.
Scientists believe that the models still have the potential to work, but they will need to be modified to accommodate the three-dimensional new observations. Ulanovsky's team has suggested additional dynamics that will guide this adaptation. They are currently working with theoreticians to create a new model where global hexagonal order emerges not in 3D, but in 2D.
Fiete, Mirko Klukas, her postdoc, and their collaborators have developed a new model to save the 2D attractor framework. Fiete believes that some of Jefferys' and Ulanovskys research supports this model. Fiete was skeptical about models that extended 2D attractor states into 3D dimensions even before the results were published. This was because they are biologically expensive. Fiete calculated that to get nice patterns in 3D would require more grid cells than what the entorhinal cortex has. A 3D grid requires different connectivity than a 2D grid.
Klukas and she came up with an alternate idea. It was similar to reconstructing a 3D statue using a variety of photos taken from different vantage points. Some grid cells are used to create a 2-dimensional slice of the 3D space. Others do the same thing but from a different angle. They form several interconnected columns of grid activity in the entorhinal cortex. Other cells combine these responses to produce local, but not global, structure as Ulanovsky's group discovered.
This model retains the benefits of classical theories such as path integration, attractor networks, and path integration. Fiete stated that you can reuse the exact same network and connectivity. It doesn't cost anything to redo anything. There is no overhead cost and you can still represent 3D, or according to the model, 5D, or higher.
Others scientists believe something completely different. Grieves stated that despite this, I am eager to see the next generation models emerge from it.
Jeffery stated that the main point is that the pattern in nature doesn't always crystallize perfectly. It is not a natural state for grid cells, I think.
Although we were captivated by grid cell spatial regularity, it is a minor issue. They are not the most fascinating thing about them.
It is important to have perfect regularity
Scientists have many opinions on what global regularity means. Ulanovsky believes it is the distances between cells that are firing, which his team has observed. Jeffery believes it is the way cells fire. Even though it may not be perfectly periodic, it may still allow the brain's spatial representations to remain separate. Fiete stresses grid cells' ability to combine information about velocity and movement.
Edvard Moser is a neuroscientist from the Norwegian University of Science and Technology and one of Nobel laureate researchers of grid cells. He believes that their global order and periodicity are still what makes them unique. His colleagues and he recently demonstrated that even grid cells which seem to fire in nonperiodic, distorted patterns in 2D, maintain the same relationships with other grid cells in different environments and brain states. This preserves an intrinsic grid.
Moser stated that the internal structure activity of an organism is precise. It's the mapping to the environment, however, that is more difficult.
Lisa Giocomo at Stanford University is a neuroscientist who believes that grid cell activity's breakdown of long-range structure suggests that cells may be encoding other variables than spatial position. For example, visual cues or the position of an animal's eye as it surveys its environment. She said that if you knew the latent variable, you might be able to see more structure.
The results show that grid cells may be encoded with additional non-spatial variables. They might also play a larger role in memory and other non-spatial processes. Memory is usually the domain of the hippocampus. It combines streams of information from different brain regions to create representations of past experiences as well as general knowledge. The entorhinal cortex, and its grid cells, provide one of these streams of information. It may also be responsible for a spatial component of memories. Federico Stella, a neuroscientist from the Donders Institute for Brain, Cognition and Behavior, in the Netherlands, stated that there is a division.
He said that grid cells can be imperfect and could be causing problems. Stella believes that Jefferys and Ulanovsky's findings are flawed. She suggests that grid cells may play a greater role in memory processing, consolidation and memory formation than is commonly believed. He said that the medial-entorhinal cortex can be considered a memory system on its own.
This opens up the possibility of other brain regions processing memories simultaneously. The flow of such information can be complex and may involve other types neurons that havent received as much attention than grid cells. This suggests that it might be necessary to understand other memory processes such as replay and activation in the hippocampus in the context of grid cells and the entorhinal cortex.
It could be a step back from grid cells periodic hexagonality to gain more valuable insights. Fiete stated that while it is true that we could have missed some things if we were too focused on perfect regularity, Fiete believes that periodic hexagonality has allowed us to see more important insights.
Fiete said that it seems like 2D is very special. It seems privileged. This could be because animals tend to stick to 2D space even when they navigate in three-dimensional space. Even fish and bats prefer different heights or depths to move. These 2D discoveries might provide a glimpse into how grid systems evolved and how they were adapted to create other representations.
Stella stated that it was crucial to first discover that these cells can become very perfect. But, it is important to ask: What can these cells actually do, even without such symmetry?
Fiete stated that while 3D work has not made her life easier conceptually, it has helped her find the joy in 3D. Our brain holds so many surprises. This is a system you can understand and it's orderly. But the brain throws curveballs.
Editor's Note: Ila Fiete and Loren Frank received funding from Simons Foundation. Quanta is also funded by the Simons Foundation. Editorial coverage is not affected by funding decisions.
