Cell action hinges on one-on-one communication









At left: Princeton researchers have found that groups of nerve cells in the retina organize themselves by sharing information in one-on-one conversations with each other, rather than through group discussions in which every cell listens to a leader. Members of the team studying this behavior demonstrate the cell communication model they have developed in this photo illustration. The findings from their research may shed more light on how the brain works. (photo illustration by John Jameson)

From the May 1, 2006, Princeton Weekly Bulletin

Neuron communication

When a group of people tries to decide how to carry out an important task, it is sometimes said that the pivotal discussions do not happen in large, well-attended meetings, but in one-on-one conversations around the water cooler.

It turns out that among individual neurons in our brains, the same may hold true.

Curious about the way networks of nerve cells work together to transmit messages, a research group at Princeton has been exploring the communication style of retinal ganglion cells, the neurons that line the back walls of our eyes and transmit everything we know about the visual world from eye to brain. Generations of scientists have found these cells to be good sources of clues about how the brain works, since the ganglion cells share many characteristics of cells deeper inside the brain but are more accessible for experiments and easily stimulated into action with light.

While observing individual retinal cells has enlightened scientists in many ways regarding the basics of mental processes, watching groups of them has seemingly thrust researchers further into the dark. Though a single ganglion cell is adjacent to several others, it seems at first glance that none of them transmits signals in tandem with its neighbors.

But Princeton’s William Bialek believes that underneath this seemingly random behavior there must be some coordination among the different cells.

“We have these enormous networks of cells in our heads, and we still are just beginning to learn about how they act collectively,” said Bialek, a theoretical physicist whose investigations cross over into neuroscience and mathematical approaches to other areas of biology. “The retina should be simpler than the cells that govern our deepest thoughts, but maybe the retina is a system that we actually can understand. Investigating it might give us a way of thinking about the harder problems of understanding the brain as a whole.”

Conversing in pairs

In order to approach the problem, Bialek, the John Archibald Wheeler/Battelle Professor in Physics, teamed up with Michael Berry and Elad Schneidman. Berry is an experimental physicist who also has turned to neuroscience and is an associate professor of molecular biology at Princeton. Schneidman is a postdoctoral fellow who had been working jointly with Berry and Bialek on theoretical questions concerning retinal ganglion cell behavior. Berry and his group, including postdoctoral fellow Ronen Segev, have developed the technology to listen in not just on one cell at a time in the retina, but on 40 cells at once.

Berry and Segev performed experiments in which they projected a movie directly onto the retina of a salamander and then monitored the signals that many ganglion cells sent out in response. Detailed analyses of single cells and of small groups of three or four of them suggested to the group that such incremental studies were not enough to address large networks, and that new approaches were needed.

“You can think about the electrical pulses streaming out of the ganglion cells as being like a coded form of the original movie,” said Berry, “but because we don’t really know the code, it’s hard to recognize the patterns in all of these data — even though that’s the problem our brain actually solves every split second as we look around the world.”

A first look at the data suggested that the patterns of firing among different cells really are quite random. Even two cells right next to each other seem to be marching to different drummers rather than being coordinated very strongly. But when Schneidman looked not at two cells but at larger groups of 10 or even all 40 cells that they could study in the experiment, there did seem to be strong patterns, where different groups of cells would suddenly act together. Since all four of the scientists were educated as physicists, they set out to find a mathematical way of describing what was going on, a model that could capture both the near random character that one sees looking at two cells and the highly structured behavior of the whole group.

Two possibilities emerged, both compelling to a limited degree. In one, the scientists imagined each cell as belonging to a number of large “teams” whose full membership never appeared at once.

“Just as you might sit down to dinner with whatever members of your team happened to show up at the same time as you on a particular night, we thought the cells might fire in groups that way,” Bialek said. “And just like Princeton students often participate in multiple activities, you might have cells on multiple teams, which could produce random-looking behavior if you couldn’t see the team rosters.”

The other possibility was that no such groupings of retinal cells existed. Rather, the 40 cells behaved somewhat like a cocktail party, or people gathered around an office water cooler: Ideas are bandied about and overheard, snippets of information are passed on as conversation partners shift, and gradually a collective opinion emerges.

“Eventually the second possibility seemed to fit the data better, but with a twist,” Schneidman said. “It seems that cells at the cocktail party talk primarily, perhaps exclusively, in pairs alone. No one belongs to a group, or takes dictation from a leader, but everyone bases their decision on what we might call ‘informed pair conversations.’ They participate in as many of them as they can, listen as much as they can, then decide.”

According to this analogy, the retinal cells would transmit messages based on the information culled from these “conversations.” However, Segev said, as in any party gathering, there are subtleties at work as well.

“Just as you might know from past experience that you tend to sympathize with one party guest quite often, but are put off by another, the opinions you draw from different conversations are not all weighted equally,” he said. “You might nod politely at one person’s argument, while agreeing strongly with another, even though they had both come down on the same side of an issue. Nerve cells seem to react to one another in the same way.”

Seeing the light

All four collaborators were surprised by how far these abstract ideas could be pushed in accounting for the detailed results of the experiments.

“Our model does not exclude the possibility that larger groupings within the 40 cells exist,” Berry said. “What we do find is that by correctly combining all of the interactions between pairs of cells, we can predict very accurately what message the whole group is going to send to the brain.”

A paper describing this work, which was supported in part by the National Institutes of Health, appeared in the April 20 issue of Nature.

Bialek, who often considers the boundary between the living and nonliving worlds, said there were many parallels between the cell signaling pattern and the way groups of inert objects respond to an organizing force.

“When a hunk of iron gets magnetized, all of its individual atoms have to adopt the same spin state. But before that happens, you don’t know which end of the new magnet is going to end up as its north pole,” he said. “How do all those atoms ‘decide’ what to do? To a very good approximation, the atoms are also talking to each other in pairs — the same statistical rules that the ganglion cells are using. In fact, many of the ordered structures that we see in nature — whether they are magnets or crystals or more exotic things — can be understood just by thinking about interactions between pairs of atoms or molecules.”

So what does this tell us about the behavior of our brain cells, which are certainly a bit more complex than a hunk of iron?

“It’s actually an old idea that one could make analogies between magnets and the coordinated activity of cells in the brain,” Berry noted, referring to pioneering work in the early 1980s by John Hopfield, now the Howard A. Prior Professor in the Life Sciences at Princeton. “What is exciting in the new work is that by combining theory and experiment we have been able to move beyond analogies to see an exact mapping, using the same mathematics that describes the patterns of spins in a magnet to describe the patterns of activity and silence from different ganglion cells in the retina.”

The study reveals something previously unknown about collective behavior among neurons, Schneidman explained.

“It’s a great deal more complex than previously thought,” he said, “but it is also comprehensible based on a reasonably simple principle, that you have to keep track of interactions only among pairs.”

Segev and Berry are pushing the technology of their experiments so that they will be able to listen in on up to 200 cells in the near future, partly in the hopes of testing predictions from the pairs model.

“We all know biological systems, from the tiniest cell up to the largest whale, are more than the sum of their parts,” Bialek said. “The individual parts aren’t alive, but when they come together they are. It could be that this is horrendously complicated. But if what we have learned from the retina is more general, perhaps we can understand a lot just by knowing about the interactions between two parts at a time. Then we might be able to use our mathematical tools to understand how all of these pairs can add up to the whole: complex behavior arising from the simple principle of hanging out at the water cooler from time to time.”