The code of the retina

Seeing is a dynamic and fast process. With quick eye move- ments – several per second – we continuously scan our environ- ment. Within a few milliseconds, the eye must capture an image. “In everyday life, we are not aware of the complex processes going on in our brain in order for us to see,” says Tim Gollisch, scientist at the Munich Bernstein Center for Computational Neuroscience and since 2007 head of an independent research group at the Max Planck Institute for Neurobiology in Martinsried. Every light signal that reaches the retina must be converted into neural signals, which can be processed and interpreted by the brain. Which code is used by the retina for this conversion of image information? And what are the underlying neuronal circuits? These questions are addressed by Tim Gollisch. For his work, he was honored with the Bernard Katz Lecture Award in 2009.

Gollisch studied physics in Heidelberg. “Right from the beginning of my studies, I had this idea of someday using physical methods for working in biology,” Gollisch reveals. Even if this idea had temporarily waned, Gollisch came back to it at the end of his studies – also inspired by a seminar given by Andreas Herz in Berlin. In this seminar, he realized that the knowledge of a physicist can make a major contribution to solving biological problems – especially in neuroscience. Thus, Gollisch had not only found his field of research, but also the adequate laboratory for his doctoral thesis – he later obtained his doctorate under the supervision of Andreas Herz. After that, he worked as a postdoc in the laboratory of Markus Meister at the Harvard University in Boston, USA.

Already during his postdoc period, Gollisch investigated the function of the retina and discovered a central mechanism that explains the enormous speed of neural coding in the retina. Gollisch showed that some ganglion cells in the retina, nerve cells that transmit the signal of the retina to the brain, use a neural code that is exceptionally quick. Each ganglion cell processes a certain section of an image. “We found that certain ganglion cells react to every change in light conditions in the respective image section, no matter whether it gets darker or lighter. The exact onset time of their reaction, however, differs depending on the content of the image section,” says Gollisch. If the image section gets darker, the cells react within 60 milliseconds. If the new image in this section also contains lighter components, they react a little more slowly. The rough outlines of a new image – which reaches the retina after an eye movement, for example – can be captured very quickly by comparing these latency times.

In his group at the Max Planck Institute, Gollisch uses a special technique to examine the neural function of the retina. “I combine a method we developed during my doctorate in the lab of Andreas Herz with the system I worked on as a postdoc,” says Gollisch. When scientists investigate the relation between sensory stimuli and neural response, they have to go through endless repetitions of presenting a sensory stimulus to nerve cells and measuring their respective responses. How the stimulus changes from trial to trial depends on the exact question addressed by the experiment. In collaboration with Jan Benda and Christian Machens in the group of Andreas Herz, then in Berlin, Gollisch automated this procedure. Based on the measurements of neural activity, the sensory stimulus is automatically varied and adapted for the next measurement. At the time, Gollisch used this technology to investigate how grasshoppers react to acoustic signals. Now, the same procedure serves to examine the retina.

In one of his research approaches, Gollisch uses this technology to determine which visual stimuli trigger the same response in a neuron. “We look at ganglion cells that process an image section consisting of two different parts – e.g. dark on the right and light on the left side,” Gollisch explains. He examines how the neuron integrates the right and the left part. Does it plainly sum up the inputs of both halves, such that a darker right part can compensate for a lighter left one? “It’s not as simple as that,” says Gollisch. “Astonishingly, the cells we currently examine are particularly active when the image is as homogeneous as possible – i.e. when the parts of the right and left side are similar. Thus, cells react better when they receive relatively weak signals from all upstream cells, and they are less active when they receive strong signals from some cells and no signals from others.” Gollisch is now interested in understanding the synaptic properties on which this functioning of the cells is based. Also in other experiments, Gollisch starts from a mathematical question (how does a nerve cell compute its input signals?) in order to finally learn something about the biology of the retina, its circuit structures and synaptic properties.

Gollisch not only examines the properties and computations of single cells in the retina, but also how cell groups are coordinated and adjusted to one another. When it is dark and the retina does not receive any visual information, the ganglion cells show a spontaneous activity – a certain background noise. Upon closer examinations of this background noise, scientists noticed that neighboring cells often send out signals simultaneously – they are synchronized. This leads to the conclusion that the cells in the retina adjust their activities to one another. “We want to examine this phenomenon more closely,” says Gollisch. “We would like to know which neural circuits it is based on and what this coordination of cells means for image processing.”

The correlations between neighboring cells seem to be especially relevant for vision when the light conditions are bad and the image is low in contrast. In a way, correlations thus serve to sharpen images. If contrasts are low, noise predominates in the nerve cells’ responses – they react imprecisely, sometimes a little earlier and sometimes slightly later. “We found out that the coordination between cells keeps this imprecision in the cells’ responses relatively low. If one cell responds a little earlier, it can be observed that neighboring cells do the same,” says Gollisch. Thus, coordinated cell behavior counteracts the noise. are particularly active when the image is as homogeneous as possible – i.e. when the parts of the right and left side are similar. Thus, cells react better when they receive relatively weak signals from all upstream cells, and they are less active when they receive strong signals from some cells and no signals from others.” Gollisch is now interested in understanding the synaptic properties on which this functioning of the cells is based. Also in other experiments, Gollisch starts from a mathematical question (how does a nerve cell compute its input signals?) in order to finally learn something about the biology of the retina, its circuit structures and synaptic properties.

“What is fascinating about working on the retina is that it is an ideal system to investigate how a large number of neurons act together in order to convert a visual signal into a neural one,” says Gollisch. The system is ideally suited to analyze which biological properties and neural circuits underlie the reaction of the cells and how an entire neural network produces a code that converts image properties into neural signals.