News FocusNeuroscience

The Map in the Brain: Grid Cells May Help Us Navigate

See allHide authors and affiliations

Science  05 May 2006:
Vol. 312, Issue 5774, pp. 680-681
DOI: 10.1126/science.312.5774.680

A newfound class of neurons enables the brain to perform complex spatial navigation—and may even help form memories

On the grid.

As rodents explore an environment, neurons called grid cells fire in a regular geometric pattern such as this one.


If you're in an unfamiliar city and trying to locate the convention center for yet another conference, a map prepared by the local Visitor's Bureau can come in handy. It's likely to incorporate a navigational aid: an overlaid square grid, often with rows labeled with letters of the alphabet and columns labeled with numbers. An index might tell you that your hotel is in the A2 square, and the convention center can be found in Q22. Last November, scientists who successfully reached the new Washington, D.C., convention center heard Norwegian neuroscientist Edvard Moser address the Society for Neuroscience's annual meeting. In an invited talk, he told the audience that rodents, and presumably people, have their own versions of such navigational grids embedded in their brains. “I swear my jaw just dropped,” recalls computational neuroscientist David Redish of the University of Minnesota, Minneapolis. “It's amazing to see the interaction between theory and experiment that has come together in the Mosers' work.” James Knierim, a neuroscientist at the University of Texas Medical School in Houston, was equally impressed by the talk, declaring, “Moser's work is the most important discovery in our field in over 20 years.”

The jaw-dropping discovery, announced 3 months earlier in a paper in Nature, centers on a brain region called the entorhinal cortex. Moser and his colleagues, including his wife May-Britt Moser and postdoctoral researchers Torkel Hafting and Marianne Fyhn, let rats roam freely in large enclosures while recording the firing of individual neurons from this cortical region. The neurons fired at distinct places: If plotted on a map of the enclosure, the firing locations of each neuron formed a triangular grid. “It's a lattice that's repeated over and over again,” says Edvard Moser. Consequently, he and his team at the Norwegian University of Science and Technology in Trondheim have dubbed these previously unknown neurons “grid cells.”

Fired up.

Edvard and May-Britt Moser analyze the neuronal activity of brain regions involved in spatial navigation.


Notably, the grid is self-generated. The cells fire almost as if there were invisible, overlapping grids painted on the enclosure's floor, with each individual neuron laying out its own “virtual” grid—and the scale of the grids can vary in size from neuron to neuron. “Somehow the system, in this totally abstract way, is saying it's time for this cell to fire,” explains Patricia Sharp of Bowling Green State University in Ohio. “You just can't explain it through external input; … there's no such pattern in the animal's world.” Indeed, the rat doesn't need visual cues once it has been in an enclosure; a grid cell still fires in the same pattern if the animal roams in the dark.

Even Moser and his team doubted their results at first. “We didn't really believe it,” he says. “We had to do some additional analysis to make sure this was biological.” On page 758, the Mosers and their colleagues add to the unfolding story, describing neurons in the entorhinal cortex that encode not only where a rat is, but also how fast it is moving and in what direction. Theoreticians believe that grid cells and their connections to other neurons may finally clarify ideas about how the brain performs spatial navigation. “We're over the moon about this discovery,” says John O'Keefe of University College London (UCL).

Everything in its place

In the 1970s, O'Keefe and his UCL colleagues discovered “place cells” in the hippocampus of the rat. On the simplest level, these neurons fire in response to where an animal is in space. Similar to the Mosers' new work, place cell experiments often feature a rat, with multiple electrodes implanted in its hippocampus, freely moving within an experimental environment; a given place cell only fires when the rat walks in a particular area, dubbed the cell's “place field.” Transfer the rodent to a different environment, and some of the same place cells are used to create a new map of the surroundings, a process called “hippocampal remapping.” This allows place cells to store representations of many different environments.

After O'Keefe's initial discovery, he and colleague Lynn Nadel felt that place cells must form the basis of a “cognitive map” in the hippocampus, but they realized it would take more than just place cells for an animal to navigate its world. Says O'Keefe, “We predicted right from the beginning that there would have to be information about directions and distances to tie together the place cells into something like a map formation.”

In the 1980s, James Ranck of the State University of New York Downstate Medical Center in Brooklyn identified another key group of navigational neurons: head direction cells, which have connections projecting into the entorhinal cortex. Depending on which way the animal is pointing its head, different groups of these cells fire, letting the animal know which way it faces.

Still, the brain needs more to navigate through a complex world. “You have information about place cells—it tells you where you are located—and you have information from head direction cells that tells you what direction you're facing, but how do you then use all that information to update over time your path through an environment?” asks Jeffrey Taube, who was a postdoc with Ranck and continues to study head direction cells at Dartmouth College.

That process of real-time updating is known as “dead reckoning” or “path integration,” and it is now thought that grid cells might be the key to how it works. These neurons' discovery came about through follow-up experiments on place cells. As scientists probed more deeply into the hippocampus, they found place cells with larger and larger place fields. But it's hard to gauge the extent of a large place field, because one can't track neuronal activity in rats running in the wild. In the standard, small experimental enclosure, “the cells either don't fire, because you're not in the place field, or they fire everywhere, because the place field is huge,” explains the Mosers' collaborator Bruce McNaughton of the University of Arizona, Tucson.

In order to better understand larger place fields in the hippocampus, the Mosers had built an enclosure about twice the standard size. After they published a paper in Science in 2004 that reported regularly structured peaks in the firing patterns of entorhinal cortical cells, they conducted follow-up experiments in the larger enclosure. “It seemed almost silly not to do it,” says Edvard Moser. Once they were able to analyze more entorhinal neurons firing over a larger territory, the group realized they were looking at grids.

How do grid cells help rats, and presumably people, find their way? “The fundamental duty of the grid cells is to provide the coordinate system, with which the place cells can then do their association of objects,” says Redish.

Consider the find-the-college-library problem. For most schools, you can get a map that lays the campus out on a grid. At Princeton, the Gothic Firestone Library is at grid square F1. At the University of California, San Diego, the spaceship-like Geisel Library is at grid square E7. The underlying map grid is the same; only the identity and position of individual buildings differ. The hippocampus notes the landmarks (Gothic buildings = Princeton) and maps them onto the entorhinal grid. “The grid cells in the entorhinal cortex provide a coordinate system, but the hippocampus must use those inputs to create a map of the environment,” says Terry Sejnowski of the Salk Institute in San Diego, California.

What lies beneath

That still begs the question of how the brain generates its grids. The leading theories, proposed by several groups, among them McNaughton and his collaborators, and David Touretzky and Mark Fuhs of Carnegie Mellon University in Pittsburgh, Pennsylvania, have to do with “attractor networks,” which can be conceptualized as a sheet of neurons packed closely together like marbles on the surface of a table. In the grid cell attractor network, each neuron in the sheet is a different grid cell. Grid cells adjacent to one another in this sheet are not necessarily next to each other in the brain, but they represent locations that are next to each other in the real world. Thus, if one grid cell fires at a certain location, then the neurons surrounding it in the sheet will fire at nearby locations.

When the rat stands at one location, the grid cell that represents that position will be very active along with neighboring cells in the network. All this activity forms a “bump” on the sheet. Excitatory connections between each cell and itself, and between it and its nearest neighbors, make the bump self-sustaining. The activated neurons also excite a bunch of inhibitory cells that prevent distant neurons from firing, thus the surrounding flatness. As the rat walks through space, the bump moves along the grid to keep track of its location, like Bugs Bunny burrowing in an old cartoon.

“In order for this network to work, the bump has to move with the rat exactly, so if the rat moves 3 feet in real space, the bump has to move however many cells in the brain correspond to 3 feet in real space—and that's not a trivial problem,” says Hugh Blair of the University of California, Los Angeles. “You would need neurons in the attractor network that would encode the speed and direction in which the rat is moving.”

Rat with hat.

Researchers can use a cap of electrodes (upper right) to monitor neuronal activity as a rodent explores a large enclosure (above). The firing pattern of an individual “grid cell” (bottom right) helps the animal find its way, even in the dark.


And that is what the Mosers, postdoc Francesca Sargolini, and their collaborators report finding in the new Science paper. The work reveals “conjunctive cells,” a class of grid cells that give exactly the path integration information—speed of movement and direction—needed to move the bump.

The entorhinal cortex is “a complicated network consisting of four layers of principal cells … with different morphologies and different interconnections,” notes Edvard Moser. As they probed deeper into these layers with electrodes, they found cells with the properties of both head direction cells and grid cells. Additionally, they found other cells whose firing rate expressed the speed of the animal. “You have cells that express position, direction, and speed,” says Moser. “That's what you need to really tell you where you are at any given time as you're walking around. It's the conjunction of those properties that I believe—and I say ‘believe’ because we haven't shown it—could be fed up to the attractor network of the pure grid cells.”

Mapping the future

Because of the connections between the entorhinal cortex and the hippocampus, O'Keefe, UCL's Neil Burgess, and others have begun to explore whether grid cells have something to do with the mysterious oscillatory firing of groups of hippocampal cells. The role of these 6-to-10-Hz “theta” oscillations has fascinated O'Keefe for years. He and Burgess have recently suggested that interference patterns formed by theta waves with slightly differing frequencies guide the firing of grid cells and thus the creation of the grid.

Grid cells could have an even more fundamental purpose than navigation, according to Blair. “Grid cells may be giving us our first real glimpse of the building blocks of hippocampal-dependent memories,” he suggests. In collaboration with Kechen Zhang of Johns Hopkins University in Baltimore, Maryland, Blair's lab is exploring whether grid cells can be used as fundamental components for assembling representations of two-dimensional objects. Blair suggests that grid cells could also construct mental representations of more complex objects, such as visual images of faces and scenes.

“The really remarkable thing is that when you build a two-dimensional memory representation out of grid fields, there is a simple trick you can do to make the memory ‘scale-invariant,'” says Blair. “For example, when you meet a new person, you don't have to store a separate memory of what they look like from close up versus far away. Our work suggests that if you build your memory representation of the person's face out of grid fields, then you can easily represent their face at all possible sizes.” If he's right, grid cells may underlie both how you can find the convention center and how you can find a colleague in the crowd there.


Navigate This Article