Localized retinal electrical stimulation in blind volunteers results in discrete round visual percepts corresponding to the location of the stimulating electrode. The success of such an approach to provide useful vision depends on elucidating the neuronal target of surface electrical stimulation. To determine if electrodes preferentially stimulate ganglion cells directly below them or passing fibers from distant ganglion cells, we developed a compartmental model for electric field stimulation of the retinal ganglion cell (RGC). In this model a RGC is stimulated by extracellular electrical fields with active channels and realistic cell morphology derived directly from a neuronal tracing. Three membrane models were applied: a linear passive model, a Hodgkin-Huxley model with passive dendrites (HH), and a model composed of all active compartments (FCM) with five nonlinear ion channels. Idealized monopolar point and disk stimulating electrodes were positioned above the cell. For the HH and FCM models, the position of lowest cathodal threshold to propagate an action potential was over the soma. Brief (100 microseconds) cathodic stimuli were 20% (HH with disk electrode) to 73% (FCM with point-source) more effective over the soma than over the axon. In the passive model, the axon is preferentially stimulated versus the soma. Although it may be possible to electrically stimulate RGC's near their cell body at lower thresholds than at their axon, these differences are relatively small. Alternative explanations should be sought to explain the focal perceptions observed in previously reported patient trials.

The somas and dendrites of intact retinal ganglion cells were exposed by enzymatic removal of the overlying endfeet of the Müller glia. Simultaneous whole cell patch recordings were made from a ganglion cell's dendrite and the cell's soma. When a dendrite was stimulated with depolarizing current, impulses often propagated to the soma, where they appeared as a mixture of small depolarizations and action potentials. When the soma was stimulated, action potentials always propagated back through the dendrite. The site of initiation of action potentials, as judged by their timing, could be shifted between soma and dendrite by changing the site of stimulation. Applying QX-314 to the soma could eliminate somatic action potentials while leaving dendritic impulses intact. The absolute amplitudes of the dendritic action potentials varied somewhat at different distances from the soma, and it is not clear whether these variations are real or technical. Nonetheless, the qualitative experiments clearly suggest that the dendrites of retinal ganglion cells generate regenerative Na+ action potentials, at least in response to large direct depolarizations.

The integrative properties of starburst amacrine cells in the rabbit retina were studied with compartmental models and computer-simulation techniques. The anatomical basis for these simulations was provided by computer reconstructions of intracellularly stained starburst amacrine cells and published data on dendritic diameter and biophysical properties. Passive and active membrane properties were included to simulate spiking and nonspiking behavior. Simulated synaptic inputs into one or more compartments consisted of a bipolar-like conductance change with peak and steady-state components provided by the sum of two Gaussian responses. Simulated impulse generation was achieved by using a model of impulse generation that included five nonlinear channels (/ Na , l Ca , 1A, 'K. /K.CO)-The magnitude of the sodium channel conductance change was altered to meet several different types of impulse generation and propagation behaviors. We studied a range of model constraints which included variations in membrane resistance (R m ) from 4,000 ft-cm 2 to 100,000 fl-cm 2 , and dendritic diameter from 0.1 to 0.3 /xm. In a separate series of simulations, we studied the feasibility of voltage-clamping starburst amacrine cells using a soma-applied, single-electrode voltage clamp, based on models with and without dendritic and somatic spiking behavior. Our simulation studies suggest that single dendrites of starburst amacrine cells can behave as independent functional subunits when the R m is high, provided that one or a small number of dendrites is synaptically co-activated. However, as the number of co-activated dendrites increases, the starburst cell behavior becomes more uniform and independent dendritic function is less prevalent. The presence of impulse activity in the dendrites raises new questions about dendritic function. However, dendritic impulses do not necessarily eliminate independent dendritic function, because dendritic impulses commonly fail as they propagate toward the soma, where they contribute EPSP-like responses which summate with conventional synaptic currents.