Researchers studying how neurons compute--such as tallying the myriad of incoming signals and concluding whether or not to fire--have long focused on the retina. By studying cells known to fire only in response to objects moving in one direction only, they hoped to learn more general lessons about how brain neurons compute. But the studies were handicapped by the fact that no one knew which retinal neurons exactly do the math. Now, an Australian team reports evidence that the computations take place in retinal neurons called ganglion cells (Science, 29 September, p. 2347).

A first sign that these cells might do directional computations came in 1965, when scientists showed that some of them respond only to objects moving in a certain direction. The ganglion cells are third or fourth in a chain of neurons triggered when light strikes the retina; the study suggested that neurons somewhere in this path calculate movement direction from the timed interplay of excitatory and inhibitory neural impulses.

In this scenario, when an object moved in the neuron's preferred direction, excitatory impulses would reach the target neuron first, triggering positively charged sodium ions to flow into the cell--an excitatory current. But when the object moved in the opposite direction, inhibitory and excitatory signals would arrive together. The inhibitory signal would cause chloride ions to enter the cell, their negative charge effectively canceling the excitatory effect. But a key question remained. Do the inhibitory and excitatory impulses converge on the ganglion cells or on earlier cells in the pathway?

To answer that question, a team led by W. Rowland Taylor of Australian National University in Canberra and David Vaney of the University of Queensland in Brisbane used a method called patch clamping, which enables researchers to detect electrical changes in a single cell--in this case, ganglion cells in cultured rabbit retinas. They found, as expected, that movement in a cell's preferred direction caused a greater excitatory current to enter the cells' dendrites, the structures that receive incoming signals. But that didn't pinpoint the site of computation; cells earlier in the pathway might be analyzing motion and delivering a larger excitatory signal to the ganglion cells in response to movement in the preferred direction.

To find out, the researchers shifted the voltage across the dendrite membrane of individual ganglion cells in a way that would favor inhibitory currents over excitatory ones. They found increased inhibitory currents in response to movement in the nonpreferred direction, suggesting that inhibitory inputs play a role in the ganglion cell's response. Next they flooded the interior of the dendrites with chloride to block the inhibitory inward flow of chloride ions; that change abolished directional selectivity.

These results provide "strong evidence" that the computation is going on in the ganglion cell dendrites, says Berkeley neuroscientist Alexander Borst. "This is really important work," he adds--especially because it offers researchers a welcome chance to explore how neurons compute in a well-defined system. But Lyle Borg-Graham, a neuroscientist at the French research agency CNRS in Gif-sur-Yvette, is not convinced. His team recently found that the critical direction-selective computation in turtle retinas occurs earlier in the signaling pathway. "I doubt that the different interpretations may be ascribed to using the turtle as opposed to the rabbit," says Borg-Graham.

--MARCIA BARINAGA