The Neural Network of the Limulus Retina: From Computer to Behavior

SYNOPSIS. The visual system of the horseshoe crab, Limulus polyphemus,provides an excellent opportunity for studying the neural basisof behavior. Quantitative analysis of the animal's visual behavioris now possible as is theoretical analysis of information processingin its retina. We combine these theoretical and behavioral approachesto investigate the nature of the signals the eye transmits tothe brain for the animal to see. Over the years theoretical studies of the Limulus eye were restrictedby the limited capabilities of single processor digital computers.However, a breakthrough in technology with the advent of parallelcomputers greatly enhances the analysis of large neural networkssuch as that of the retina. We have developed a time-dependentmodel of the Limulus retina on the Connection Machine (ModelCM-2), which is a massively parallel computer containing 32,768processors. The model represents a matrix of 64 x 128 receptorsand simulates interactions among receptors with digital filtersand transduction and adaptation within a receptor by a multistagecascade. Neural response patterns computed with the ConnectionMachine model replicate to a first approximation the patternsof neural activity recorded in the laboratory. Behavioral studies of Limulus vision carried out in the fieldcan be simulated on the Connection Machine. Neural responsesrecorded from behaving animals serve to test the accuracy ofthe model. Thus far we have developed just one model of theretina, but it eventually will have two forms, "daytime" and"nighttime," to account for the known circadian rhythms in retinalfunction. With a combination of field, physiological, and theoreticalstudies, we hope to gain a better understanding of the neuralmechanisms that underlie the animal's visually-guided behavior.