A large group of gamers, working with computational neuroscientists, has produced a wiring diagram of the nerve cell connections at the back of the eye, which may have solved the long-standing question of how cells in the retina detect motion.
50 years ago, researchers discovered that retinal ganglion cells, which transmit information from eye to brain via the optic nerve, are sensitive to the direction and speed of moving images, and have been trying explain how ever since. The new diagram, published today in the journal Nature, points to an elegant ‘space-time wiring’ mechanism that makes a certain type of cell sensitive to motion in very specific directions.
Light entering the eye falls on photoreceptors, which convert it into electrical impulses and then transmit the information to bipolar cells. The information then passes to neurons called starburst amacrine cells, and then is transmitted to the retinal ganglion cells, which relay it to the visual cortex at the back of the brain.
The secret of how the retina detects motion seems to lie in the starburst amacrine cells (above), and the connections they form with the bipolar neurons. Starburst amacrine cells are flat with an elaborate array of extremely fine, branched dendrites. Each acts as a computational unit in its own right, and is sensitive to visual stimuli that move away from the centre of the cell out towards the tip of the branch. Earlier research has also shown that some bipolar neurons respond to visual stimuli more slowly than others.
Both starburst amacrine cells and bipolar neurons come in various different types, but the arrangement of connections between them was not know in any great detail. Jinseop Kim of the Massachusetts Institute of Technology and his colleagues reasoned that the pattern of connections between them might reveal something about how the starburst amacrine cells detect motion, and so decided to map them.
The researchers collected a series of electron microscope images of the mouse retina and partly reconstructed them into a large series of ‘cubes,’ each comprising about 17 Megabytes of data, and representing a tiny three-dimensional chunk of retinal tissue about 5 microns (thousandths of a millimetre) across. They then handed the cubes over to members of EyeWire, an online community of more than 100,000 gamers. The information was distributed via CloudFront, Amazon’s content distribution network, which cached the cubes in servers around the world so that they could be delivered to the ‘EyeWirers’ quickly.
“2,183 people from the EyeWire community helped us to reconstruct the starburst amacrine cells,” explains senior author Sebastian Seung, now at Princeton University, adding that only elite members of the online community were involved. “They had to pass the Starburst Challenge. They got a sequence of cubes, and for each one they had to reconstruct part of the branch of an amacrine cell. They had to do a certain number of cubes in a row with a certain degree of accuracy to pass the challenge, and those who qualified could go on to unlock the starburst cells.”
The gamers used their consoles or PCs to add colour to the grey images. “We give them each cube with a seed [of colour],” says Seung, “and from that seed they have to keep on colouring within the same neuron, colouring as much of the same branch as possible.”
The results reveal two distinct neural pathways between the starburst amacrine cells and bipolar neurons. The bipolar neurons that respond to light slowly preferentially connect to the starburst amacrine cell branches nearest the centre, whereas those that respond quickly form their connections further away. This organization, together with what was already known about the properties of these cells, suggests a ‘time-delay’ neural network that functions to make electrical impulses converge in the same place at the same time to signal motion in a certain direction.
Consider a starburst amacrine cell that is sensitive to rightward motion. As an object passes from left to right through the visual field, it will activate first the slow bipolar cells and then the fast ones. Because of the way the connections are arranged, the impulses from both bipolar cells will reach the starburst amacrine cell simultaneously, causing it to send motion-related signals to the retinal ganglion cells. An object moving in any other direction, however, would produce a different pattern of impulses, which do not converge on the starburst cell and therefore do not activate it.
“This is a very interesting study that uses an innovative and exciting approach,” says Rowland Taylor of Oregon Health and Science University. “It demonstrates that a combination large numbers of online human volunteers can accomplish the analysis of a very large data-set that would be otherwise intractable.”
“The connectivity data lead the authors to propose a novel and compelling hypothesis that may finally explain the synaptic basis for direction-selectivity in the retina,” he continues, “[but] further analysis will be required to fully validate the model since it’s unclear whether the contacts they observed represent functional synapses.”
Seung acknowledges that the results are only provisional, and that the proposed mechanism needs to be tested. The study is part of the wider connectome project to map the circuitry of the brain in its entirety, and he wants to focus on mapping the rest of the connections in the retina, while letting physiologists see if they can get evidence for the time-delay network.
“This paper is about the significance of the connectome,” he says. “Critics say that even if we get all the connectome information, it’ll be useless because it doesn’t tell us how the brain functions. This paper demonstrates that a static wiring diagram could be crucial for detection of a moving stimulus.”