Responses to pulsatile subretinal electric stimulation: effects of amplitude and duration

Abstract

In working to improve the quality of visual percepts elicited by retinal prosthetics, considerable effort has been made to understand how retinal neurons respond to electric stimulation. Whereas responses arising from direct activation of retinal ganglion cells have been well studied, responses arising through indirect activation (e.g., secondary to activation of bipolar cells) are not as well understood. Here, we used cell-attached, patch-clamp recordings to measure the responses of rabbit ganglion cells in vitro to a wide range of stimulus-pulse parameters (amplitudes: 0–100 μA; durations: 0.1–50 ms), applied to a 400-μm-diameter, subretinal-stimulating electrode. The indirect responses generally consisted of multiple action potentials that were clustered into bursts, although the latency and number of spikes within a burst were highly variable. When different parameter pairs representing identical charge levels were compared, the shortest pulse durations generally elicited the most spikes. In addition, latencies were shortest, and jitter was lowest for short pulses. These findings suggest that short pulses are optimum for activation of presynaptic neurons, and therefore, short pulses are more effective for both direct as well as indirect activation.

retinal prostheses strive to restore vision to those blinded by retinal degenerative diseases, such as macular degeneration and retinitis pigmentosa (Humayun et al. 1996; Rizzo et al. 2003; Zrenner et al. 2009). These diseases destroy the neurons of the outer retina (e.g., photoreceptors and horizontal cells) but largely spare the neurons of the inner retina (bipolar, amacrine, and ganglion cells). Electric stimulation excites one or more of these surviving cell classes, resulting in the generation of action potentials in ganglion cells (the output neurons of the retina). The spike trains transmitted from ganglion cells carry the only visual information conveyed to higher visual centers and therefore, underlie all properties of elicited psychophysical percepts. This suggests that the ability to elicit patterns of neural activity that mimic normal responses is likely to improve clinical outcomes. Therefore, it is desirable to learn how different stimulus parameters influence the resulting spike patterns.

Single pulses of electric stimulation elicit retinal ganglion cell (RGC) activity in two ways: by direct electrical activation of the RGC and/or by activation of one or more of the neurons that delivers synaptic input to the ganglion cell (i.e., bipolar cells) (Fried et al. 2006; Jensen et al. 2003; Jensen et al. 2005b; Margalit et al. 2011; Sekirnjak et al. 2006). Direct activation arises when the stimulus pulse depolarizes voltage-gated sodium channels within the proximal or distal portions of the ganglion cell axon (Behrend et al. 2009; Fried et al. 2009; Jensen et al. 2003; Sekirnjak et al. 2008). This form of activation has been well characterized and typically results in a single action potential that occurs within 0.5 ms of pulse onset. In contrast, the mechanism underlying the response of presynaptic neurons to electric stimulation remains less well understood. Several recent studies argue that stimuli acting on voltage-gated calcium channels, found in the synaptic terminals of bipolar cells and photoreceptors, may result in the activation of these cells (the increased calcium entry leads to an increase in release of synaptic neurotransmitter) (Freeman et al. 2010, 2011). Regardless of the exact mechanism, the RGC activity arising from indirect activation is different from that of direct activation, as it consists of multiple action potentials with wide ranges in latency (Jensen et al. 2005a; Stett et al. 2000).

Although the full extent of the responses that can arise through indirect activation has not been explored, increases to the duration and/or the amplitude of a single stimulus pulse are generally thought to increase the number of spikes elicited. For example, Stett et al. (2000, 2007) found increased spiking when the amplitude of 0.5 ms constant-voltage (single) pulses was increased from 0.6 to 1.8 V. Tsai et al. (2009) similarly found increased levels of spiking when the amplitude of (single) current pulses was increased. Similarly, the level of bipolar cell activation was found to increase when pulse duration was increased from 1 to 3 ms (Fried et al. 2006); this is consistent with increased bipolar cell activation causing the reported increases in RGC spiking.

The neural circuitry of the retina raises the possibility, however, that some increases in pulse strength (amplitude and/or duration) could lead to decreases in the number of elicited spikes. Bipolar cells synapse onto inhibitory amacrine cells (as well as ganglion cells), and therefore, increases in pulse strength are also likely to increase the level of amacrine cell-mediated inhibitory activity delivered to ganglion cells (Fried et al. 2006; Margalit and Thoreson 2006). In addition, some (or many) of the different amacrine cell types may also be activated directly [i.e., not secondary to bipolar cell activation (Margalit and Thoreson 2006)] at stronger stimulus levels. Because there is a wide array of amacrine cell types that can deliver feedback inhibition onto bipolar cells as well as feedforward inhibition to ganglion cells (MacNeil et al. 1999; MacNeil and Masland 1998), the possibility exists that some increases in stimulus strength may actually reduce the number of spikes arising in the ganglion cell. Consistent with this, previous work has shown that periods of increased amacrine cell activity are correlated temporally with periods in which RGC spike activity is silenced (Fried et al. 2006). Surprisingly, however, the interplay between the parameters of stimulation that elicit indirect activation and the corresponding patterns of elicited spiking has not been well established.

Ganglion cell responses to natural stimuli consist of sparse bursts of spikes (Puchalla et al. 2005). The timing of individual spikes within the bursts as well as the overall timing of the bursts are highly consistent from trial to trial, suggesting that such bursts may comprise the basic signaling block used by the retina to transmit information. If so, the ability to elicit bursts of spikes with prosthetic stimulation is likely to be important. Some preliminary evidence suggests that increases in stimulus amplitude reduce the trial-to-trial jitter of elicited spikes (within a burst) (Stett et al. 2000), but a better understanding of how the parameters of stimulation shape jitter and the other timing properties of elicited bursts could lead to more effective stimulation methods.

Higher charge levels of the pulse are associated with increased charge density at the stimulating electrodes; such increases can lead to electrode failure, adverse tissue reactions, and/or increases in the size of the stimulating electrode (to mitigate the increased charge-density levels) (Brummer and Turner 1977; Cogan 2008; Cohen 2007, 2009; McCreery et al. 1990; Sekirnjak et al. 2006). In addition, higher charge levels of the stimulus pulse may necessitate increases of the overall power requirements for the device. Thus although longer and/or higher amplitude pulses may create stronger responses (higher spike count) or perhaps even burst patterns that are more “physiological like,” the creation of such activity must be considered in light of these engineering and safety concerns. Thus a better understanding of how burst responses are shaped by the different parameters of the stimulus pulse will help to resolve the tradeoff between the increased levels of spiking that arise from stronger pulses and the adverse effects of increased charge levels associated with stronger pulses.

Here, we explored the response of RGCs to a wide range of stimulus-pulse durations and amplitudes. The results we report are a first step in characterizing how the RGC response depends on these stimulus parameters and suggest strategies that may optimize the number of spikes elicited, as well as their temporal arrangement.

METHODS

Animal preparation and retina isolation.

The care and use of animals followed all federal and institutional guidelines, and all protocols were approved by the Institutional Animal Care and Use Committee of the Veterans Affairs Boston Healthcare System and/or the Subcommittee of Research Animal Care of the Massachusetts General Hospital. New Zealand white rabbits (∼2.5 kg) were anesthetized with injections of xylazine/ketamine and subsequently euthanized with an intracardial injection of pentobarbital sodium. Immediately after death, the eyes were removed. All procedures following eye removal were performed under dim, red illumination. The front of the eye was removed, and the vitreous was eliminated. The retina was separated from the retinal pigment epithelium and mounted—photoreceptor side down—to a 10-mm square piece of Millipore filter paper (0.45 μm HA membrane filter) that was mounted with vacuum grease to the recording chamber (∼1.0 ml vol). A 2-mm circle in the center of the Millipore paper allowed light from below to be projected onto the photoreceptor layer.

Electrophysiology.

Patch pipettes were used to make small holes in the inner-limiting membrane, and ganglion cells with large somata (diameters ≥20 μm) were targeted. The restriction of soma size helped ensure that displaced amacrine cells were avoided. It also is likely to have restricted the types of ganglion cells targeted to only a subset of the entire population (O'Brien et al. 2002; Rockhill et al. 2002). Furthermore, only those cells whose somas were in the immediate periphery of the stimulating electrode were targeted (Fig. 1D); previous studies suggest that the somas in this region are strongly activated with thresholds that are comparable with those somas that are more centrally located (Behrend et al. 2011). Spiking was recorded with a loose, cell-attached patch electrode (4–8 MΩ), filled with Ames medium. The retina was perfused continuously at 4 ml/min with Ames (pH 7.4) at 36°C, equilibrated with 95% O₂-5% CO₂.

Fig. 1.

Spike patterns vary with stimulus amplitude. A, top: raw response to a 20-μA, 3-ms anodic pulse (top trace); 4 action potentials were elicited, the 1st of which occurs during the stimulus artifact. Bottom: raster responses to 5 presentations of the same pulse. B: same as A, except that the pulse amplitude was 60 μA. C: raster plot of responses to 3-ms anodic pulses with amplitudes ranging from 0 to 100 μA. Five pulses were delivered at each stimulus level; each repeat is offset slightly in amplitude. The stimulus waveform is plotted above the raster plot (dark trace). D: schematic of the stimulating electrode vs. the soma location for ganglion cells targeted in this study. Solid, curved line represents the approximate periphery of the stimulating electrode; curved, broken lines represent the approximate region from which ganglion cells were targeted. Dashed circle indicates the position of 1 soma.

Electric stimulation.

Electric stimulation was delivered via a 15-electrode planar array, consisting of 400 μm-diameter disc electrodes in a 5 × 3 arrangement with 200 μm edge-to-edge spacing (Shire et al. 2009). The array was placed on the floor of the recording chamber with the exposed electrode side facing up. Each electrode site was coated with a reactive direct current-sputtered iridium oxide film of 300 nm thickness. Pulse stimuli were controlled by Multi Channel Systems MCS (Reutlingen, Germany) STG2004 hardware and software. This electric stimulator and the data acquisition hardware were controlled by custom software (D. Freeman), written in LabVIEW (National Instruments, Austin, TX) and Matlab (Mathworks, Natick, MA). Two silver chloride-coated silver wires served as the return; each was positioned ∼8 mm from the targeted cell and ∼6 mm from the other wire (return).

Stimulus waveforms consisted of equal and opposite pulses (anodic first), separated by an interval of 600 ms; the interval between the cathodic phase and the anodic phase of the next waveform was 1 s. These large intervals allowed the response to each phase to be analyzed separately. Two sets of pulse trains were delivered in each experiment. In the first set, amplitudes were stepped from 0 to 30 μA in steps of 2 μA. After pulses at each amplitude had been delivered, the amplitude was “reset” back to 0 μA and the process repeated four additional times with a minimum of 3 s between iterations. The second set of pulse trains was identical to the first, except amplitudes ranged from 30 to 100 μA in steps of 5 μA; the interval between completion of the first set and onset of the second set was always at least 5 s. Because of maximum charge-injection limitations, the maximum amplitude for 10- and 50-ms-duration pulses was limited to 30 and 20 μA, respectively. In all cases, charge density levels were limited to 796 μC/cm² to avoid adverse effects on the electrode surface.

Data analysis.

Raw recordings were collected at a sample rate of 20 kHz and processed with custom software written in Matlab. The timing of the negative peak of each spike (corresponding to the peak of depolarization) was used to approximate the spike latency (relative to stimulus-pulse onset) in raster plots; thus reported latencies throughout this study are ∼0.5 ms greater than actual onsets. To view the variation of spike timing over the whole range of stimulus amplitudes, the five presentations at each stimulus amplitude were plotted at the approximate stimulus amplitude level on the y-axis but each with a slightly different offset to prevent overlapping of responses. For example, the second presentation at 10 μA was positioned at 10.4 μA on the y-axis, the third presentation at 10.8 μA, and additional presentations plotted successively with 0.4-μA offsets (e.g., Fig. 1C). In this manner, the five presentations extended from 10 to 12 μA. A burst was defined as a series of three or more consecutive spikes, for which interspike intervals were ≤4 ms. Previous burst definitions (Reinagel et al. 1999), e.g., no spiking for a minimum of 100 ms and then maximum interspike intervals of 4 ms, were not suitable here, because intervals of 100 ms prior to the onset of additional spiking were rare (i.e., cf. Fig. 2). To evaluate spike-timing precision across all pulse durations, both spike latency and jitter for each pulse duration were plotted as a function of pulse amplitude and fitted with a power-law curve of the form y = a/x^b (Berry et al. 1997; Sekirnjak et al. 2006).

Fig. 2.

Spike patterns vary for different ganglion cells. Rasters on left show 600 ms of response to 3-ms anodal pulses of varying amplitude; right: the time axis expanded to reveal the 1st 100 ms of the response. Five presentations at each stimulus amplitude are offset slightly in amplitude. A–D: responses from a different ganglion cell. Pulse timing shown at the top of each column applies to all traces in the column. A is the same cell shown in Fig. 1. Upward pointing arrows in A (right) and C (right) show decrease of latency with increasing stimulus amplitude. Downward arrows in B (left) indicate multiple bursts. Horizontal arrows in B (right) indicate the 1st bursts at 40 and 90 μA. Downward pointing arrows in B (right) and C (right) indicate increase of latency with increasing stimulus amplitude. Leftward pointing arrowheads in C (right) indicate spikes at 85 and 50 μA to show the reduction in the number of spikes within a burst at increasing stimulus levels. Downward arrowhead in C (right) indicates the suppressed spike at amplitude levels >80 μA. Leftward pointing arrows in C (right) and D (right) indicate the variability in the number of spikes across cells. Downward arrow in D (left) indicates a single burst.

RESULTS

Consistent with previous work on subretinal stimulation (Jensen and Rizzo 2006; Stett et al. 2007), we found that the cathodic phase elicited only weak spiking (e.g., zero or one spike), and therefore, we report only responses to the anodic phase. Responses were obtained from 17 different ganglion cells taken from nine retinas.

Response patterns vary with stimulus amplitude.

A typical RGC response to a 20-μA anodic pulse with a duration of 3 ms consisted of four spikes, the first occurring within 1 ms of pulse onset (Fig. 1A). We extracted the timing of each spike (see methods) to present the response in raster form and show the results of five presentations in Fig. 1A. Responses were generally similar in four of the five presentations: the first spike had a latency of 1–2 ms, and one or more additional spikes had latencies ranging from 15 to 40 ms. When the stimulus amplitude was increased to 60 μA (Fig. 1B), the latency of the early (first) spike and the late (second) spike decreased, and the latency differences between corresponding spikes (e.g., first, second, etc.) from each presentation were reduced. These changes in response for different stimulus amplitudes led us to systematically investigate the response to a wide range of amplitudes.

A typical raster plot of responses to amplitudes ranging from 0 to 100 μA (3-ms pulse duration) is shown in Fig. 1C. The y-axis corresponds to the amplitude of the stimulus pulse, and multiple presentations at each level were offset slightly (see methods). As the plot suggests, several elements of the response patterns were sensitive to stimulus amplitude. For example, more spikes were elicited for stimulus amplitudes >40 μA than for amplitudes <30 μA. In addition, a second, “early” spike, with a latency of ∼3 ms, was initiated by stimulus amplitudes ≥60 μA. The fact that spikes with different latencies had different thresholds suggests that multiple mechanisms of activation may be involved and that each mechanism may have a different threshold for activation. These and other types of response variability (vs. amplitude) were found in all cells that we tested (n = 17) and are analyzed in more detail below.

As suggested by previous studies (Jensen and Rizzo 2008; Tsai et al. 2009), we found considerable variability in the patterns of response across different RGCs (Fig. 2). For example, some cells generated only a single burst (see methods; Fig. 2D), whereas others generated multiple bursts (Fig. 2B). The number of spikes within bursts was also variable; such variability was found within the response of a single cell across stimulus amplitudes (compare the number of spikes within the first burst at 40 μA with 90 μA; Fig. 2B), as well as across cells (i.e., compare 60 μA in Fig. 2C with Fig. 2D). Casual observation of the latency of individual spikes within the first burst suggests that the pattern of spike intervals (i.e., the timing of individual spikes within the burst) becomes more uniform as pulse amplitude increases (see Spike-timing latency and jitter are altered by pulse duration below for further analysis). In RGCs with no baseline spiking (n = 11/17), the duration of the longest bursts observed was close to 78 ms. In the subset of these cells that generated multiple bursts, responses could persist up to 449 ms. In RGCs that exhibited baseline spiking, i.e., the cell of Fig. 2A, the time for baseline spiking to return to prestimulus levels generally extended beyond the duration of the spiking response, although this was not quantified further (but see below).

In addition to the variability associated with the latencies of bursts across cells, the onset latency of a given burst within the same cell could either decrease as stimulus amplitude increased (Fig. 2, A and C) or increase (Fig. 2, B and C). Interestingly, the early bursts, typically those in which the first spike onset latency was <35 ms, were the ones that decreased in latency with increasing stimulus amplitude (n = 17/17). In contrast, the latency of the later bursts—those with an onset latency >35 ms—tended to increase with increasing amplitude (n = 12/14). Because the delays between bursts are thought to arise from amacrine cell activity (Fried et al. 2006; Margalit and Thoreson 2006), they suggest that amacrine cell activation levels may also be sensitive to stimulus amplitude. A reduction in baseline spiking could be observed for all cells that exhibited baseline spiking (n = 6). The reduction for the cell of Fig. 2A persisted for ∼400 ms, suggesting that at least some forms of amacrine cell activation may outlast bipolar cell activation. Yet, another manifestation of possible amacrine cell-mediated inhibition was the reduction in the number of spikes within a burst at increasing stimulus levels (Fig. 2C; compare spikes elicited at 85 μA with that at 50 μA). Taken together, our results suggest the activation of many different types of inhibitory pathways, raising the possibility that multiple subtypes of amacrine cells are activated by electric stimulation. Whereas our data are consistent with the possibility that different types of amacrine cells have different sensitivities to electric stimulation, we did not attempt to further determine the sensitivity of any given type.

Response patterns vary with pulse duration.

We explored the sensitivity of the response patterns to changes in pulse duration. The responses elicited by pulse durations of 0.1, 0.3, 1.0, 3.0, and 10.0 ms are shown in Fig. 3 (all are from the same cell). Several interesting observations arise from a comparison of the responses with different durations. First, the approximate timing of bursts was consistent across pulse durations. For example, one burst of spikes was centered ∼40 ms across all pulse durations. This held true even for the weaker responses elicited by pulse durations of 0.1 and 0.3 ms. Similarly, a smaller burst of spikes with latency of ∼120 ms can be seen in response to pulse durations of 1, 3, and 10 ms.

Fig. 3.

The response pattern varies with pulse duration. Responses to pulse durations of 0.1, 0.3, 1.0, 3.0, and 10.0 ms are shown for a single cell in successive rows (pulse duration indicated at left); left plots show 600 ms of the raster response, whereas right plots expand the 1st 100 ms. Pulse timing is shown at the top of each panel. For each pulse duration, the charge associated with the highest amplitude is given in parentheses in the left panels. The shaded regions at the bottom of the 0.1-ms pulse panels indicate that responses were not measured for pulse amplitudes <20 μA. Note that the y-axes are different for the 10-ms pulse responses (due to the charge limit capacity of the electrode).

Second, the shortest pulse durations, especially 0.1-ms pulses, generated only weak spiking, even at the maximum amplitudes we tested. Unlike the response to longer pulse durations, the number of spikes within weak bursts did not increase as the stimulus pulse was increased for short pulses (0.1 ms: n = 4/4; 0.3 ms: n = 3/4). Third, the longest pulse durations (10 ms) did not elicit the highest levels of spiking. A comparison of the response between 3- and 10-ms pulses revealed that more spikes were elicited by the 3-ms pulse, even when the same amount of charge was delivered with a 10-ms pulse [e.g., compare 30 μA of the 10-ms response (300 nC/phase) with 100 μA for 3-ms pulses (300 nC/phase)]. This raises the possibility that similar to ganglion cells (Jensen and Rizzo 2006; Jensen et al. 2005b; Sekirnjak et al. 2006) and many other spiking neurons (Tehovnik et al. 2006), shorter pulse durations may activate bipolar cells more effectively (with less total charge) than longer pulse durations. These differences are quantified further below (see Figs. 5 and 6). Note that for the 10-ms response, the pattern of spiking that occurs in the range of 300- to 600-ms poststimulus appears different than the spiking patterns observed in the same time periods following shorter pulse durations. This raises the possibility that the effect of the 10-ms duration pulse persists beyond the period for which spike bursts can be observed clearly.

Fig. 5.

Shorter pulse durations generate more spikes for a given charge. A: comparison of raster responses for pulses with constant charge (100 nC/phase) but varying pulse duration. B, top: average number of spikes elicited within the 1st 100 ms of the stimulus onset; averages are reported separately for each pulse duration and each charge level. Bottom: average number of spikes elicited within 600 ms following stimulus onset. Bars represent SE.

Fig. 6.

Increasing pulse duration elicits more spikes. A: each point represents the number of spikes elicited in a typical cell within the 1st 100 ms after stimulus onset for a given pulse duration and pulse amplitude. Lines connect points of constant pulse duration. B: same as A, except for a different cell in which 50-ms pulse durations (but not 100 and 300 μs) were applied. Legend in B also applies to A. Dashed vertical lines in A and B indicate fixed amplitude for comparison of the number of spikes with increasing pulse durations.

Response comparisons for constant charge stimuli.

To further explore the sensitivity of presynaptic neurons to different pulse durations, we overlaid responses to all combinations of pulse duration and amplitude in a single plot (Fig. 4). The responses were overlaid such that the y-value at which each response was plotted corresponds to the calculated charge/phase of the pulse. Similar to previous figures, the five presentations of each stimulus condition were offset slightly (along the y-axis) to prevent overlap. All responses with the same phase duration were plotted with the same color. On the x-axis, t = 0 corresponds to pulse onset for all pulse durations.

Fig. 4.

Constant charge comparisons for different pulse durations. A: overlay of responses to anodal pulses of varying pulse duration and amplitude; pulse durations range from 0.1 to 10.0 ms (0.1 ms: dark blue; 0.3 ms: red; 1 ms: green; 3 ms: purple; 10 ms: light blue), and amplitudes ranged from 0 to 100 μA. In all cases, pulse onset occurred at t = 0. The y value of each response corresponds to the calculated charge/phase of the pulse. Similar to previous figures, the 5 presentations for each stimulus condition are offset slightly (along the y-axis). B: expanded view of the 1st 100 ms of each response. Red horizontal arrows in A and B indicate 3 levels of charge (30, 100, and 300 nC) for comparison of spike patterns among the pulse durations.

Whereas the timing of bursts and nonburst periods was generally similar across pulse durations (Fig. 4A), an expanded view of the first 100 ms of the response (Fig. 4B) revealed some notable differences. For example, when the delivered charge was held constant at 100 nC/phase, the number of spikes occurring within the burst centered around 40 ms was considerably greater for pulses of 1-ms duration than for duration of 3 ms. A similar observation was made for a charge level of 300 nC/phase: the 3-ms pulse duration elicited more spikes than the pulse duration of 10 ms. To probe this further, we isolated the responses to three distinct levels of charge: 30, 100, and 300 nC/phase (Fig. 4, A and B). At each charge level, responses from three different pulse durations were examined.

For a fixed charge level of 100 nC/phase (Fig. 5A), the number of spikes elicited by the 1-ms pulse was larger than that for the 3- or the 10-ms pulses (see below for further analysis of the contribution of early and late spikes to these differences). Similar observations were made at 30 and 300 nC/phase; i.e., the shortest pulse duration elicited the largest number of spikes. To quantify this across the population, we generated plots similar to those in Fig. 5A for all cells and determined the average number of spikes elicited at each duration. These averages (from each cell) were further averaged across all 17 cells, and the results are plotted in Fig. 5B (results for spikes within the first 100 ms following the onset of the stimulus pulse and spikes within the first 600 ms). In all cases, when comparisons were limited to pulse parameters that delivered identical charge/pulse, the shortest-duration pulses elicited the largest number of spikes (ANOVA, P < 0.006).

Response as a function of amplitude.

We also examined the number of spikes elicited as a function of amplitude (constant pulse duration) for fixed pulse durations. Similar to the approach described previously, we counted the total number of elicited spikes within the first 100 ms after stimulus onset for each combination of pulse duration and amplitude. Results from two typical cells are shown in Fig. 6. Unfortunately, we did not test all six pulse durations in any one cell; therefore, plots from two different cells are required to show results from all pulse durations. As expected, when pulse duration was held constant, the number of spikes generally increased with amplitude; this occurred for all pulse durations. Furthermore, for any fixed amplitude (Fig. 6, A and B), the number of elicited spikes increased as pulse duration increased.

Response duration varies across population of cells.

We also examined the duration over which spikes were elicited for different ganglion cells by plotting the number of spikes elicited as a function of time (Fig. 7). The maximum amplitudes were 100, 100, 30, and 20 μA for pulse durations of 1, 3, 10, and 50 ms, respectively; responses at these amplitudes were at or close to the maximum part of their range, i.e., Fig. 6B. As suggested by earlier results (i.e., Fig. 2), the response from some cells (n = 6) was “complete” within a relatively short period of time, i.e., 100 ms (Fig. 7C), whereas the response of other cells (n = 8) persisted for several hundred milliseconds (Fig. 7B) or even longer (Fig. 7A). The duration over which spikes were elicited was generally affected only minimally by pulse duration, e.g., Fig. 7, A and B.

Fig. 7.

Duration of elicited spiking varies across ganglion cells. Each point represents the total number of spikes (see text) elicited as a function of time (poststimulus onset). Lines connect points with constant pulse duration; symbols and colors below C apply to all panels. A–C are from 3 different ganglion cells.

Spike-timing latency and jitter are altered by pulse duration.

The results from several physiological studies suggest that the timing of elicited spikes may play a critical role in the quality of elicited percepts. For example, raster responses to natural scenes reveal that within a given cell, the timing of elicited spikes is highly consistent across trials (Puchalla et al. 2005). Other studies have found that the timing of spikes from neighboring ganglion cells can be highly correlated (DeVries and Baylor 1997; Pillow et al. 2008). Taken together, these and other studies suggest that the ability of a prosthetic to reliably elicit spikes with precise timing may be important. Therefore, we sought to compare the timing of individual spikes (see methods) as a function of pulse parameters (amplitude, duration, and total charge delivered).

Our approach was to examine the latency of the first three spikes by averaging over repeats and then exploring the trends as a function of the different pulse parameters. Because the first spike arises from direct activation of the ganglion cell, and subsequent spikes arise only through activation of presynaptic neurons, we hoped that the latency analysis of the first three spikes might provide insight into the mechanism(s) underlying both forms of activation. There were several challenges associated with this approach, however. The raster plot in Fig. 1C reveals that a single, short-latency spike (<5 ms) was elicited for amplitudes ≤55 μA, whereas for amplitudes ≥60 μA, a second short-latency spike was elicited as well. This means that the latency of the second spike was ∼15 ms for amplitudes ≤55 μA and ∼4 ms for amplitudes ≥60 μA. The reasons for this “jump” in latency are not known, but it seems likely that it arises, because higher stimulation levels use a different mechanism of activation. For example, photoreceptors could be activated at low thresholds (resulting in relatively long latencies), whereas bipolar cells are activated at higher thresholds (resulting in shorter latencies). Regardless of the exact mechanism, the jump suggests that the activation mechanism for the second spike is different at high vs. low amplitudes and therefore, raises questions about whether all “second” spikes should be treated as a single population. The results presented below do not attempt to address the complexities noted above but are simply a first step toward the analysis of spike latency.

The latencies of the first three spikes elicited by each stimulus pulse were each averaged (over five presentations), and plots of latency vs. amplitude were generated for pulse durations of 1, 3, 10, and 50 ms; Fig. 8A plots average latency as a function of pulse amplitude for a typical cell. Consistent with previous work (Sekirnjak et al. 2008; Stett et al. 2000), latency generally decreased as the amplitude of the stimulus pulse increased. Latency values from individual cells were averaged across all cells (n = 9); Fig. 8B plots the latency of the first spike (mean ± SE) vs. amplitude for pulse durations of 3 ms and 50 ms. Mean latencies were fit with a power-law curve (Berry et al. 1997) (see methods) to facilitate comparisons across different stimulus conditions (Fig. 8C). Curve fits were extended along the x-axis to compare latencies across a wider range of amplitudes and pulse durations than could be measured experimentally (due to charge density and maximum amplitude limitations).

Fig. 8.

Spike latency decreases with pulse amplitude. A: average (over 5 repeats) latency of a typical cell's 1st spike, plotted as a function of stimulus-pulse amplitude. B: average [across all cells (n = 9 for 1, 3, and 10 ms, and n = 7 for 50 ms) and all repeats] latency of the 1st spike as a function of pulse amplitude; for clarity, only results from pulse durations of 3 and 50 ms are shown. Bars represent SE. C: power-law curves fit to the data for each pulse duration in B. The vertical, dashed line indicates the maximum amplitude (20 μA) used in the experiments with 50-ms pulses; the dashed curve represents the extension of the power-law curve.

Plots of mean latency of the first, second, and third spike as a function of pulse amplitude for pulse durations of 1, 3, 10, and 50 ms are shown in Fig. 9, A–D, respectively. Figure 9, E–H, shows the same latency data plotted as a function of the total charge delivered. Consistent with our initial observation, latency decreased significantly with increasing amplitude or charge. Incremental increases at low amplitude (or charge) levels resulted in relatively large decreases in latency. Conversely, incremental increases at higher amplitude (or charge) levels resulted in smaller latency decreases.

Fig. 9.

Spike latency decreases are sensitive to pulse duration. A–H: the solid lines represent the power-law curve fit from the average latency data (as in Fig. 8). Separate plots are presented for the 1st (blue)-, 2nd (red)-, and 3rd (green)-elicited spikes. The shaded regions for each spike represent ±1 SE of latency (n = 9 for 1, 3, and 10 ms, and n = 7 for 50 ms). A–D are functions of pulse amplitude, whereas E–H are functions of charge. As in Fig. 8C, the vertical, dashed lines indicate the upper limit at which measurements were made.

To explore the effectiveness of different pulse durations, we compared latencies for parameter sets in which charge levels were constant. Under these conditions, we found that spike latency was always lowest for the shortest-duration pulses; this was true for the first, second, and third spikes (Fig. 10, B–D, respectively). Note that this finding was consistent at relatively low charge levels (i.e., 100 nC/phase), at which all latency values arose from direct measurements, as well as at higher charge levels, at which some latency values arose from predictions.

Fig. 10.

: Shorter pulse durations reduce latency. A: average latency for 1st, 2nd, and 3rd spikes in response to the maximum amplitude at each pulse duration (see text). B–D: latencies of the 1st (B), 2nd (C), and 3rd (D) spike; each bar is the mean latency of a single spike across cells for a fixed pulse duration (color-coded) as well as a fixed charge level (x-axis). Bars represent SE (n = 9 for 1, 3, 10 ms, and n = 7 for 50 ms).

Because shorter latencies suggest that less charge was required for the onset of activation, these results imply that shorter pulse durations are more effective for activating all three spikes. This result is not very surprising for the first spike, because shorter pulse durations are known to be more effective for direct activation of ganglion cells (Fried et al. 2006; Jensen et al. 2003; Jensen et al. 2005b; Sekirnjak et al. 2006; Tsai et al. 2009). However, the analogous findings for the second and third spikes are surprising, given that these two spikes arise from activation of neurons presynaptic to ganglion cells, and previous work suggests that such neurons are targeted preferentially by long pulses (Behrend et al. 2011; Fried et al. 2006; Greenberg 1998).

In addition to exploring the latency of individual spikes, we explored the consistency in the timing with which each spike was elicited (jitter). The plots of Figs. 1–3 provided a general impression that the variability in timing between spikes decreased as amplitude increased. For example, Fig. 11A shows the latency of the first and second spikes in response to a 3-ms pulse with amplitudes of 60 and 100 μA. These data were extracted from Fig. 2B. Casual observation of the variability in spike timing across the repeats suggests that higher amplitudes might be associated with less jitter. To quantify this variability, we calculated the SD for the first, second, and third spikes across the five presentations for each stimulus condition within a given cell. Similar to the latency analysis described previously, SDs for identical stimulus conditions were then averaged across all cells (n = 9), and averages were then fit with a power-law curve to facilitate comparisons (Fig. 11).

Fig. 11.

Jitter is reduced with increasing pulse amplitude and charge. A, inset: expanded raster view of the 1st 2 elicited spikes in response to a 3-ms pulse at 60 (left) or 100 (right) μA. Each line in the main panel represents the power-law curve fit to the average SD (STDEV) of the 1st spike latency. B and C: similar to A, except the power-law fits are based on the average jitter for the 2nd and 3rd spikes as a function of pulse amplitude. D–F: analogous to A–C, except plotted as a function of total charge delivered.

Several general features of spike-timing jitter were revealed by these plots. First, spike-timing variability decreased rapidly with increasing pulse amplitude and/or total charge; some portion of this decrease may arise from the jumps in latency described above, but these were not controlled for in this (initial) analysis. Second, the amplitude levels at which jitter approaches 1 ms correspond approximately to the levels at which spikes are first reliably detected for each duration; e.g., four or five of the pulses elicited spikes. For example, 1-ms pulses reliably produced short-latency spikes when amplitude exceeds 25 μA (Fig. 3), the same approximate level at which jitter falls below 1 ms (Fig. 11A). Similarly, 3-ms pulses begin to reliably elicit short-latency spikes, i.e., <5 ms at ∼16 μA (Fig. 3), the same amplitude at which jitter falls below 1 ms. Third, comparisons at the same level of charge reveal that jitter decreases with pulse duration. In Fig. 11D, at the highest stimulus level used for 1-ms pulse duration (100 nC/phase), latency is ∼0.3 ms, ∼0.6 ms, ∼0.9 ms, and >4 ms for pulse durations of 1 ms, 3 ms, 10 ms, and 50 ms, respectively. In other words, to get the same level of jitter required less total charge with a short pulse vs. a long pulse. The discrepancy was even larger for the third spike (Fig. 11F), in which charge levels of 68, 162, 194, and 531 nC/phase were now needed. This held true for the short-latency first spike (presumably arising through direct activation of the ganglion cell), as well as the second and third spikes (presumably arising secondary to activation of presynaptic bipolar cells), suggesting that both neurons respond more reliably when charge is delivered rapidly.

DISCUSSION

Our study provides several important, new insights about indirect activation and the resulting patterns of spiking in RGCs. First, by applying a wide range of stimulation parameters, we were able to document a diverse range of spike patterns that can be elicited. Importantly, however, this range may not reflect the full extent of ganglion cell responses, because only large somata cells were targeted. Second, analysis of the elicited patterns yielded a much-improved understanding of how the amplitude and the duration of the stimulus pulse shape the response. As part of this analysis, different parameter sets that delivered the same total charge were compared and revealed that the largest number of spikes was elicited when pulse durations were short, and amplitudes were high. These same parameter sets were also associated with the shortest latencies and lowest jitter. Short pulses were known previously to activate ganglion cells optimally, but the results here indicate that they are also more effective for activation of bipolar cells. Each of these findings is discussed below, along with the implications of this work for improving the efficacy of the retinal prosthetic.

Complexity of elicited patterns.

Our measurements suggest that the RGC responses to indirect activation are diverse, even though only those cell types with large somata were targeted (see methods). There was variability in the total number of spikes elicited, the latency of elicited spikes, and the arrangement with which individual spikes clustered into bursts. Both pulse amplitude (Figs. 2 and 6A) and duration (Figs. 3 and 6B) significantly affected the strength of the response, although the sensitivity to changes in these parameters could vary considerably across cells (Fig. 2). Latencies were also sensitive to the amplitude (Figs. 2 and 9) and duration (Figs. 3 and 10) of the pulse. Because all of the responses reported here arose from RGCs whose somas were ≥20 μm and were also in the immediate periphery of the stimulating electrode, it is possible that additional diversity would arise if other types of RGCs and/or other relative locations had also been evaluated.

Tsai et al.(2009) also reported diversity in the responses to indirect activation, and several of the response elements identified in our patterns are consistent with their results. However, our use of a larger stimulating electrode (400 vs. 25 μm) allowed responses to a much wider range of stimulating conditions to be collected, which enabled response patterns across this wider range to be studied. The higher charge levels available with the larger electrodes revealed that the jitter associated with secondary spikes (those with latency >20 ms) was actually reduced at higher stimulus levels.

The effect of stimulus amplitude on spike latency was bimodal (Fig. 2). For example, the latencies of spikes that arose within the first 35 ms decreased with increasing stimulus amplitude, whereas spikes with poststimulus latencies >35 ms were delayed (increased latencies) as amplitude increased. In addition, spike activity in a poststimulus window is sometimes extinguished with increasing stimulus amplitude. For example, the spikes occurring with a poststimulus latency of ∼10 ms in Fig. 2C were suppressed at amplitude levels >80 μA. These types of effects suggest that multiple excitatory and inhibitory sources, each with a different sensitivity to stimulation, contribute to the observed patterns.

In support of this notion, there are ∼10 different types of bipolar cells, each with different morphological and biophysical properties (Euler and Masland 2000) and therefore, each likely to have a different sensitivity to stimulation. Because many types of RGCs are thought to receive excitatory synaptic input from a single type of bipolar cell, it is straightforward to imagine that different types of RGCs receive different levels of excitatory drive in response to a given pulse. Activated bipolar cells also deliver synaptic input to amacrine cells, and therefore, amacrine cells are likely to exhibit different levels of activation as well. Because more than one type of amacrine cell synapses onto each type of ganglion cell, as well as onto each type of bipolar cell, it is not surprising that complex inhibitory effects were observed (Fig. 2). Furthermore, the possibility of direct activation of amacrine cells, i.e., not secondary to bipolar cell activation, may also contribute to and even enhance the response complexity. Unfortunately, neither the specific synaptic connections made by most types of amacrine cells nor their sensitivity to electric stimulation are well understood. Therefore, it is difficult to ascribe the specific response elements observed here to one or more specific type of retinal neuron or circuit.

Response sensitivity to duration and amplitude.

Whereas the direct response of RGCs typically consists of a single spike, responses arising through indirect activation typically consist of multiple spikes. Therefore, we used the number of spikes elicited as a measure of the effectiveness of different combinations of pulse duration and amplitude. Because normal retinal signaling is thought to be comprised of spike bursts, we felt that counting the number of spikes provided a more-relevant measure of stimulus effectiveness than simply determining activation threshold for one of the many elicited spikes, especially since some spikes could be unpredictably suppressed at different levels of stimulation. When stimulus parameter sets, with their product of amplitude and duration (the amount of charge delivered), were held constant, pulses with the shortest durations (and strongest amplitudes) elicited the most indirect spikes (Fig. 5B). Also, under these same conditions, the latencies of the first, second, and third spikes were all lowest with short pulses (Fig. 10). Furthermore, jitter was also lowest for these same parameter sets. Thus for a given amount of charge, our results suggest that short pulses activate bipolar cells more quickly and more strongly than equivalent-charge, longer pulses. Because the amount of charge required to elicit the direct response in RGCs (and other central nervous system neurons) also becomes more efficient as pulse duration decreases, i.e., shorter pulses require less charge (Jensen et al. 2005b; Sekirnjak et al. 2006; Tehovnik et al. 2006), our results further suggest that the rapid delivery of charge may generally optimize activation of all neurons to extracellular stimulation.

Activation of presynaptic neurons has generally been achieved with long pulses (Fried et al. 2006; Greenberg 1998; Jensen et al. 2005b), as well as with other long-duration waveforms (Freeman et al. 2010). Interestingly, however, our findings show that short pulses are most effective for activation of presynaptic neurons when comparing equal charge alternatives—this comparison has not been studied previously, and so, there is no discrepancy between our results and previous studies. Furthermore, if we compare responses from two equal amplitude alternatives (e.g., Fig. 6), we find that similar to previous studies, the longer pulse elicits more spikes.

Work from Freeman et al. (2010) found that low-frequency sinusoids (i.e., 25 Hz) activated bipolar cells strongly and selectively. This could suggest that 20-ms pulses might be most effective for bipolar cell activation (if the difference in waveforms did not lead to response differences), but here, we generally found that durations of 3 ms elicited more spikes than durations of either 10 or 50 ms; this held true even when the 50-ms pulse used considerably more charge (not shown). A potential explanation for the discrepancy (in addition to the waveform differences) arises from the difference in methodology used to measure activation: here, we compared responses to single pulses, whereas Freeman et al. (2010) counted the total spikes from multiple periods as a measure of activation. Accommodative effects are known to reduce responsiveness when multiple stimuli that activate the retinal network are delivered consecutively. Thus an important limitation of the present study is that even though a single, short pulse is optimum for activating bipolar cells, it may not be optimum when multiple stimuli are delivered sequentially. It is unlikely that the passive membrane properties of the bipolar cell play a role, since previous modeling work suggests that such properties are essentially constant at low stimulation rates (Freeman et al. 2011).

The effectiveness of short pulses was also somewhat surprising, given our current understanding of the factors influencing responsiveness (Freeman et al. 2011). Direct activation of ganglion cells is mediated through depolarization of voltage-gated sodium channels, most likely those channels within the axon initial segment, and it is well established that shorter pulses are most effective (initiate spiking with the lowest amount of delivered charge) (Fried et al. 2006; Sekirnjak et al. 2006; Tehovnik et al. 2006; Tsai et al. 2009). However, the presynaptic neurons that respond to indirect activation do not contain the same dense regions of sodium channels as RGCs, and therefore, it is likely that indirect activation of these neurons is mediated through one or more other type(s) of voltage-gated ion channels. In support of this, a recent computational study (Freeman et al. 2011) suggests that voltage-gated calcium channels within the bipolar cell terminal may play a key role. Time constants of voltage-gated calcium channels in bipolar cells are in the range of 1–5 ms (Freeman et al. 2011; Pan 2000; Pan et al. 2001), which suggests that stimuli that depolarize the cell membrane for relatively long periods of time might be most effective. Therefore, our finding here that shorter pulses were more effective is surprising. Unfortunately, the maximum stimulus amplitude that was used in this study (100 μA) limited the amount of charge that could be delivered, especially with the shortest pulse durations (0.1 and 0.3 ms). It will be interesting in future testing to learn whether such (short) durations (e.g., those less than the time constant of calcium channels) are more effective if their charge level is made comparable with that from pulses ∼1 ms (comparable with the time constants of calcium channels). The effectiveness of short pulses may also indicate that the time constants of the calcium channels involved are faster than thought previously, or the mechanism underlying indirect activation involves ion channels other than calcium.

Limitations of this study.

Several elements of our experimental methodology impose limitations on our findings and raise important questions for further study. For example, all experiments in this study were performed on the healthy, in vitro retina. Previous reports have shown that the dendrites of bipolar cells retract in the degenerate retina (Mazzoni et al. 2008) and that the threshold for indirect activation increases (Goo et al. 2011; Jensen and Rizzo 2008). It is not known whether threshold increase is due to the change in dendritic morphology or in some other as-yet-uncharacterized aspect of the bipolar cell morphology or because the synaptic pathway between bipolar and ganglion cells is altered. Excitatory synaptic currents have been measured in ganglion cells of the degenerate retina (Margolis et al. 2008), suggesting that the pathway from bipolar cells remains, at least partially, intact. In the normal retina, photoreceptors may be activated by certain forms of electric stimulation (Margalit et al. 2011); therefore, if photoreceptors have a lower threshold for activation than bipolar cells, then their degeneration would lead to the observed threshold increase. Our current results do not allow us to distinguish between these possibilities, and it will be important to determine whether the spike patterns observed here are similar to those that arise in the degenerate retina. Future studies that identify parameter sets that optimize activation of bipolar cells (over photoreceptors) may help to improve the efficacy of indirect activation in clinical applications.

The fact that the somas of the ganglion cells used in this study were restricted to the immediate periphery of the stimulating electrode (see methods) raises additional questions about the relative contribution of center vs. surround pathways. For example, activation of those bipolar cells within the dendritic field of a given ganglion cell (the center), would likely have an excitatory effect on the cell, whereas activation of bipolar cells beyond the dendritic field would likely have an inhibitory effect on the ganglion cell. However, because targeted somas were within 25 μm of the edge of the stimulating electrode, it is likely that a sizable portion of the central pathways was involved in all cells tested. The fact that we observed robust spiking in all targeted cells suggests that central pathways were dominant, consistent with much previous experimental work on center-surround mechanisms. Thus it is likely that the ganglion cells in the periphery studied here are activated by similar mechanisms to those in the center of the electrode. This is supported by a recent study that showed that ganglion cells within the perimeter of a stimulating electrode, as well as ganglion cells in a large annulus around the electrode, were activated by long-duration pulses with only minor differences in threshold (Behrend et al. 2011).

A second important question arises from the fact that only single pulses were tested in this study. Previous work has shown that a stimulus that activates the retinal network can have a suppressive effect on subsequent stimuli (Freeman and Fried 2011; Jensen and Rizzo 2007). Thus it is possible that the responses measured here would be different for a second pulse that was delivered within a few hundred milliseconds of the first. Because the raster responses that we recorded have clear periods of activation and suppression, it will be interesting to explore how delivery of subsequent pulses in excitatory and inhibitory periods alters sensitivity. This understanding may help to improve consistency for higher rates of stimulation.

GRANTS

Support for this work was provided by the Veterans Administration (MR1I01RX000350-01A1) and by the National Eye Institute (R01-EY019967).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

Author contributions: S.W.L., D.K.E., and S.I.F. conception and design of research; S.W.L. performed experiments; S.W.L., D.K.E., and S.I.F. analyzed data; S.W.L., D.K.E., and S.I.F. interpreted results of experiments; S.W.L. prepared figures; S.W.L. and S.I.F. drafted manuscript; S.W.L., D.K.E., and S.I.F. edited and revised manuscript; S.W.L., D.K.E., and S.I.F. approved final version of manuscript.