Harvard University - Department of Molecular & Cellular Biology

A NEURAL CODE BASED ON SPIKE TIMING IN THE RETINA

by Tim Gollisch and Markus Meister

April 2nd, 2008


Co-authors (L to R):
Tim Gollisch and Markus Meister

The process of vision begins in the retina. This neuronal network at the back of the eyeball receives incident light and extracts relevant visual information. Nerve cells in the retina then send on this information to different brain regions in the form of electrical pulses ("spikes"). These spikes must encode all the information needed for the brain to construct an internal model of the visual environment and produce appropriate reactions. Experience tells us that this must be a fast process. Indeed, psychophysical experiments have shown that human subjects require only about a tenth of a second to detect objects in a photograph. Every-day life suggests similar speeds for visual processing; sudden, rapid movements of the eyes or the head ("saccades") occur several times per second. Each saccade brings a new image onto the retina and initiates a short episode of visual processing.

We were interested in how the early visual system encodes and processes information that is available only for a short period. We therefore briefly flashed visual images, ranging from simple patterns of black and white stripes to photographs of natural objects, onto the isolated salamander retina. At the same time, we recorded the spikes of many retinal nerve cells with an array of microscopic wires. Classically, a neuron is thought to convey information by how many spikes it produces per unit time ("rate code"). By contrast, we found that nerve cells of a particular type in the retina respond to very different images with the same number of spikes. What changes, however, is the exact timing of those spikes. Depending on the pattern of illumination, these cells vary the time of their first spike (the "latency") by up to 40 milliseconds. If the same image is presented repeatedly, on the other hand, the response timing is reproducible down to few milliseconds. This means that these neurons can transmit a lot of information in the timing of their spikes.

But how can downstream brain regions read out this information? Clearly, a reference time is needed to make sense of this spike timing information. We found that the spike times from two cells of this neuronal population can function as reference points for each other. In fact, the relative timing between two such cells has intriguing properties: 1) it remains approximately constant even if the visual contrast of an image is reduced and all responses occur later in time, and 2) it can be even more precise than the timing of the individual spikes themselves. The reason for the latter is that noisy fluctuations in spike timing are often correlated between different neurons, so that they partly cancel in the relative timing.

These spike timing relations that we observed in the retina may thus provide a powerful neural code for rapid image processing: they transmit information about visual objects independent of contrast; they are robust to retinal noise; and most importantly, they provide information in the shortest time possible, namely with the first spike of a neural population.

Read paper in Science


The left image shows schematically the responses of the retinal neurons that we have investigated. The response timing depends on the visual image that is presented to the retina. In particular, one neuron may fire earlier than a second neuron for a given image, whereas a different image leads to the reverse order. If the same images are presented at lower contrast, all responses occur later, but the timing relations between neurons are preserved.

 

The right image shows how this timing information can be used to decode a natural scene. If an image (here a photograph of a swimming salamander
larva) appears in front of the eye, the neurons in the retina fire patterns of "spikes". When early responses are converted to black and late responses to white, a faithful reconstruction of the image is obtained. Note that this illustration is based on Figure 4 of our manuscript.

View Markus Meister's Faculty Profile