The effect of waveform asymmetry on perception with epiretinal prostheses

Abstract

Objective.

Retinal prosthetic implants have helped improve vision in patients blinded by photoreceptor degeneration. Retinal implant users report improvements in light perception and performing visual tasks, but their ability to perceive shapes and letters is limited due to the low precision of retinal activation, which is exacerbated by axonal stimulation and high perceptual thresholds. A previous in vitro study in our lab used calcium imaging to measure the spatial activity of mouse retinal ganglion cells (RGCs) in response to electrical stimulation. Based on this study, symmetric anodic-first (SA) stimulation effectively avoided axonal activation and asymmetric anodic-first stimulation (AA) with duration ratios (ratio of the anodic to cathodic phase) greater than 10 reduced RGC activation thresholds significantly. Applying these novel stimulation strategies in clinic may increase perception precision and improve the overall patient outcomes.

Approach.

We combined human subject testing and computational modeling to further examine the effect of SA and AA stimuli on perception shapes and thresholds for epiretinal stimulation of RGCs.

Main results.

Threshold measurement in three Argus II participants indicated that AA stimulation could increase perception probabilities compared to a standard symmetric cathodic-first (SC) pulse, and this effect can be intensified by addition of an interphae gap (IPG). Our in silico RGC model predicts lower thresholds with AA and asymmetric cathodic-first (AC) stimuli compared to a SC pulse. This effect was more pronounced at shorter pulse widths. The most effective pulse for threshold reduction with short pulse durations (≤ 0.12 ms) was AA stimulation with small duration ratios (≤ 5) and long IPGs (≥ 2 ms). For the 0.5 ms pulse duration, SC stimulation with IPGs longer than 0.5 ms, or asymmetric stimuli with large duration ratios (≥ 20) were most effective in threshold reduction. Phosphene shape analysis did not reveal a significant change in percept elongation with SA stimulation. However, there was a significant increase in percept size (P < 0.01) with AA stimulation compared to the standard pulse in one participant.

Significance:

Including asymmetric waveform capability will provide more flexible options for optimization and personalized fitting of retinal implants.

1. Introduction

Retinitis pigmentosa and age-related macular degeneration are prevalent retinal degenerative diseases that can lead to severe visual impairment or blindness (Hartong, Berson and Dryja, 2006; Jager, Mieler and Miller, 2008). Retinal prosthetic implants have helped improve vision in people blinded by these conditions through electrically stimulating the remaining cells in the inner retina (Humayun et al., 1999; Jones and Marc, 2005; Zrenner, 2013). Patients with these implants report improvements in light perception and performing visual tasks (Weiland and Humayun, 2005; Yanai et al., 2007; Humayun et al., 2009; Ahuja et al., 2011; Finn, Grewal and Vajzovic, 2018), but their ability to perceive shapes and letters is currently limited (Zrenner et al., 2011; Da Cruz et al., 2013). The best visual acuity (reported in peer-reviewed journals) is 20/1260 for epiretinal (Humayun et al., 2012) and 20/460 for subretinal implants (Palanker et al., 2020). These acuity values do not yet demonstrate a level of restored vision that is better than legal blindness (20/200).

Clinical studies have shown that single electrode activation can lead to perception of elongated phosphenes in epiretinal implant users (Nanduri et al., 2012; Beyeler et al., 2019). Unintended activation of off-target retinal ganglion cells (RGC) due to stimulation of axons of passage (Weitz et al., 2015; Chang et al., 2019) is identified as an important contributing factor to elongated responses and therefore lower precision of retinal activation. A study of 30 patients with the Argus II retinal prosthesis demonstrated that about half of the electrodes had perception thresholds above the acute stimulation safety limit (Humayun et al., 2012). To allow safe stimulation, such electrodes are used in unison with neighboring electrodes, effectively creating a larger electrode, which further decreases the perceptual resolution. Investigating stimulation strategies to reduce perception thresholds and create focal phosphenes has the potential to increase the resolution of retinal prostheses and improve the overall outcome for blind patients.

Our previous in vitro calcium imaging study (Chang et al., 2019) showed that asymmetric anodic-first stimulation with duration ratios (ratio of the anodic to cathodic phase) greater than 10 reduces RGC activation thresholds significantly, compared to a standard cathodic-first pulse. We also demonstrated that symmetric anodic-first stimulation with short durations (< 120 μs) results in a more focal response confined to the area near the electrode and presumably avoids stimulating axons of passage; however this type of stimulation also increases activation thresholds significantly.

In this study, we combined human subject testing and computational modeling to further examine the effect of pulse polarity order and asymmetry on perceptual thresholds and phosphene shape characteristics in Argus II patients. Psychophysical testing was done to determine how probability of generating a percept was effected by use of asymmetric anodic-first (AA) stimulation and, interphase gap (IPG), both individually and in combination. We also performed phosphene shape analysis to study the effect of symmetric anodic-first (SA) and AA stimuli on the shape of elicited percepts.

2. Methods

2.1. Human Subject Testing

Four eligible participants with the Argus II retinal prosthesis were recruited from the W.K. Kellogg Eye Center (University of Michigan, Ann Arbor, MI). Informed consent was obtained from each participant, following approval from the University of Michigan’s Institutional Review Board (IRB). The study adhered to the tenets of the Declaration of Helsinki and the national regulations for medical device clinical trials (Clinicaltrials.gov identifier: NCT03635645).

Stimulation was delivered to patient implants using a research Video Processing Unit (VPU), controlled with the Clinician Fitting System (CFS) and the Asymmetric Waveform Research Tool (Second Sight Medical Products, Sylmar, CA, USA). A short pulse width (0.2 ms) was used for stimulation to induce direct activation of RGCs (Fried, 2005; Chang et al., 2019), and to maintain the total pulse duration below 6 ms (Argus II limit) when applying AA stimulation. IPG was included as a variable, since it has been shown to reduce stimulation thresholds by delaying the opposing effect of the anodic phase on neural depolarization (Weitz et al., 2014). We compared its effect on threshold reduction with that of AA stimulation. In all experiments, 15% of the trials were catch trials randomly distributed among stimulus-present trials.

2.1.1. Threshold Measurement

Perceptual thresholds were measured for symmetric cathodic-first (SC) and SA stimuli for 4 electrodes per participant. First, an approximate threshold range was determined for each electrode by applying stimuli with increasing amplitude. Within the approximate threshold range, 10 pulse amplitudes were chosen for more thorough testing. Threshold was measured for all 4 electrodes during one “block” of testing and each electrode had a distinct set of 10 amplitudes for testing. Each trial consisted of one electrode stimulated at one amplitude with 20 identical pulses over 1 second. The participant was cued with an audio tone to indicate (verbally) whether or not they saw a percept. Responses were collected for all 4 electrodes and 10 amplitudes in random order, to avoid desensitzation and order effects. A logistic function was fit to the data and threshold was defined as the pulse amplitude corresponding to 50% probablility on the logistic curve.

Asymmetric pulses use two different amplitudes but must maintain charge balance. The Argus II implant has a limited set of current amplitudes available. Thus, only a limited number of asymmetric pulse durations and amplitudes were available to use. As a result, we could not follow the same protocol for estimating threshold described above, since a full logistic threshold curves could not be generated for asymmetric pulses. Instead, perception probabilities for multiple pulse types were measured and compared at a single amplitude. For asymmetric pulses, amplitude refers to the current in the cathodic phase. For each electrode we selected an amplitude value that satisfied three conditions: 1) A subthreshold amplitude (between 0% and 50% probability) based on the SC threshold curve calculated earlier, 2) availability of that amplitude with AA stimulation, and 3) availability of a duration ratio higher than 10 at that amplitude. Four stimulation types (SC, SA, AA, SC + IPG) were delivered with the aforementioned amplitude and repeated 20 times each in order to calculate the perception probabilities. Each trial consisted of a pulse train of 20 pulses in 1 second. The active electrode and pulse type were chosen randomly in each block of trials. Based on our previous in vitro study (Chang et al., 2019), we hypothesized that AA pulses would have lower thresholds, and thus higher perception probabilities at the same amplitude. The same process was repeated (3 electrodes per participant) to measure perception probabilities in response to SC and AA stimuli with different IPG values (SCI and AAI). We used paired t-tests with Bonferroni correction to compare different groups of perception probabilities.

2.1.2. Phosphene Shape Analysis

Each trial began with a control experiment using tactile targets to calculate the participant’s drawing bias and variability (Nanduri et al., 2012). Participants were asked to feel the tactile shapes and draw them on a touch-screen monitor, considering the shape, size, and orientation of the objects. For retinal stimulation trials, we asked participants to draw their perceptions in response to stimuli consisting of a pulse train delivered at 20 Hz for a total duration of 3 seconds. We used SC, SA, and AA stimuli at a single amplitude. The amplitude value had to satisfy three conditions: 1) A suprathreshold amplitude based on the previously calculated SC threshold curve, 2) availability of that amplitude with AA stimulation, and 3) availability of a duration ratio higher than 10 at that amplitude. We chose a suprathreshold amplitude to ensure a phosphene in the majority of trials. Phosphene shapes were analyzed using two descriptors: area and elongation. Area is defined as the total number of non-zero pixels in the image, and elongation is the ratio of the major to minor axis of the best-fit ellipse. Inclusion of catch trials and randomization of electrodes were done similarly to the threshold measurement task. Paired t-tests with Bonferroni correction were used to compare area and elongation for different pulse types.

2.2. Computational Modeling

A computational model of a RGC with simplified geometry was built in the NEURON simulation environment (Hines and Carnevale, 1997) to study the effect of pulse asymmetry and polarity order on activation thresholds. This allowed us to test a more exhaustive parameter set than what was feasible with human subjects. The geometry and membrane kinetics were based on a previously published model (Schiefer and Grill, 2006) with modifications made to the length of segments. The model consisted of 1065 compartments of 1 μm length constructing an axon, narrow region of axon, initial segment of axon, and soma, with 1 mm, 45 μm, 10 μm, and 10 μm lengths respectively (Fig. 1). The RGC membrane included five nonlinear ion channels: sodium $\left({\overline{g}}_{\text{Na}}\right)$, delayed rectifier potassium $\left({\overline{g}}_{\text{K},\text{dr}}\right)$, inactivating potassium $\left({\overline{g}}_{\text{K},\text{A}}\right)$, calcium activated potassium $\left({\overline{g}}_{\text{K},\text{Ca}}\right)$, and L-type calcium $\left({\overline{g}}_{\text{Ca}}\right)$. The ion channel conductances varied by RGC region, as described in Schiefer et al. These were in parallel with a leakage conductance and a membrane capacitance. A point current source was placed above the axon - axon narrow region junction, at a 50 μm distance from the neuron vertically to simulate the thickness of the nerve fiber layer and inner limiting membrane (Weinreb et al., 1990). The extracellular space was modeled as a purely resistive homogenous volume conductor with 500 Ω-cm resistivity (Heynen and Van Norren, 1985). To study the effect of AA stimulation on thresholds, six different pulse types were used: SC, and AA with 2, 5, 10, 20, and 30 duration ratios. Pulse widths ranged from 0.05 to 0.5 ms, and IPGs ranged from 0.1 to 2 ms. Electrical current pulses were delivered at 20 Hz for a total duration of 1500 ms (30 pulses). This frequency was chosen for consistency with the clinically used parameters. Electrical potential at each point along the neuron was calculated using equation (1).

$$V_e=\frac{{\rho }_{e}I}{4\pi r}$$ (1)

Where V_e = extracellular potential, ρ_e = resistivity of the extracellular space, I = electrode current, r = distance between the center of each compartment to the electrode.

Figure 1.

Computational model of a single retinal ganglion cell. The geometry consists of the RGC soma, axon initial segment, axon narrow region, and axon. A point current source is positioned 50 μm above the axon- axon narrow region junction. Each compartment consists of five ion channels (Na⁺, delayed rectifier K⁺, inactivating K⁺, Ca²⁺ activated K⁺, and L-type Ca²⁺), a leak channel, and a membrane capacitance. (figure is not drawn to scale)

Numerical integration was done using a backward Euler method with a time step of 2 μs. Activation threshold was defined as the minimum amplitude at which the cell fired an action potential in response to at least 50% of the delivered pulses and calculated using a bisection algorithm with a ±0.1 μA error. Based on our in vitro study (Chang et al., 2019) and the human subject testing results, we hypothesized that AA stimulation with a large duration ratio reduced the threshold by the anode break excitation mechanism (Plonsey and Barr, 2007). Differences in time constants for the m, n, and h gating variables are responsible for anode break excitation. After termination of a prolonged hyperpolarizing stimulation, the short time constant for sodium channel activation (m) leads to a sodium influx greater than potassium efflux, thereby increasing the probability of excitation. Adding a cathodic phase after termination of the anodic pulse further increases the excitation probability. Anode break was demonstrated in the squid giant axon, which is modeled by Hodgkin and Huxley using only Na⁺ and K⁺ channels. However, RGCs have at least 5 types of ion channels (Lipton and Tauck, 1987; Lukasiewicz and Werblin, 1988), and the membrane potential follows (2) (Fohlmeister, Coleman and Miller, 1990).

$$C_m\frac{dE}{dt}=-{\overline{g}}_{Na}{m}^{3}h\left(E-E_{Na}\right)-{\overline{g}}_{Ca}{c}^3\left(E-E_{Ca}\right)-{\overline{g}}_{K,dr}{n}^4\left(E-E_K\right)-{\overline{g}}_{K,A}{p}^{3}q\left(E-E_K\right)-g_{K,Ca}\left(E-E_K\right)-{\overline{g}}_L\left(E-E_L\right)+I_{\mathit{\text{stim}}}$$

Where m, h, c, n, p, and q are the gating variables of the voltage-gated ion channels, E_Na, E_Ca, E_K, and E_L are the equilibrium potentials of the ion channels, E is the membrane potential, ${\overline{g}}_{\text{Na}}$, ${\overline{g}}_{\text{K},\text{dr}}$, ${\overline{g}}_{\text{K},\text{A}}$, ${\overline{g}}_{\text{K},\text{Ca}}$, ${\overline{g}}_{\text{Ca}}$ are the ion channel conductances, and c_m is the membrane capacitance. To gain insight into the mechanism of threshold reduction by AA stimulation in a RGC, we analyzed the membrane voltage and the numerical values of gating variables over the course of stimulation.

3. Results

3.1. Human Subject Testing

Perception threshold and probability measurement was performed on four Argus II participants. One participant was removed from the analysis due to a high false positive rate (55–70%). Results from the remaining participants were analyzed separately and are shown in figure (2). The full logistic threshold curve could not be generated for SA stimulation for participant 3, since the perception probability did not reach the threshold despite stimulation with the maximum amplitude allowable within the safety limits. For participants 1 and 2 there was an increase in thresholds with SA compared to SC stimulation, consistent with previous in vitro studies (Boinagrov et al., 2014; Ahn et al., 2015; Chang et al., 2019). This difference was significant for participant 1 (29.66% increase in average threshold, P < 0.05), and not significant for participant 2 (19.58% increase in average threshold). All p-values were adjusted by the Bonferroni correction since multiple hypotheses were being tested in each experiment. We demonstrated that AA stimulation and SC stimulation with an IPG increase perception probability significantly compared to both SC and SA stimuli in participants 1 and 2. Participant 3 showed no difference in perception probabilities (Fig. 2. A3). This might be due to the participant not reporting phosphenes accurately, evident from his high false positive rate on that day (31.67 %). Introducing an IPG to the SC pulse raised the perception probability significantly regardless of the IPG duration in participant 1. Perception probability increased with the duration of IPG in participants 2 and 3, however this change was not significant (Fig. 2. B2 and B3). Adding an IPG to the AA pulse increased the perception probability in participant 1, however not significantly (Fig. 2. C1). This effect was less obvious for the other two participants (Fig. 2. C2 and C3).

Figure 2.

A1-A3) Perception probability comparison for different pulse types in human participants. Each plot represents perception probability values for SA, SC, AA, and SCI stimuli for 4 electrodes. False positive rates: participant 1 = 0 %, participant 2 = 0 %, participant 3 = 31.67 % B1-B3) Perception probability comparison for SC stimulation with different IPG values. Each plot represents perception probability values for SC pulse with different IPG values measured for 3 electrodes. False positive rates: participant 1 = 9 %, participant 2 = 0.8 %, participant 3 = 12 % C1-C3) Perception probability comparison for AA stimulation with different IPG values. Each plot represents perception probability values for AA pulse with different IPG values measured for 3 electrodes. False positive rates: participant 1 = 9 %, participant 2 = 0.8 %, participant 3 = 12 %. (P<0.05*, P<0.01**, P<0.001 ***). In some cases (plot 1B for example), the data points overlap. Black dots represent perception probability values for individual electrodes. Red diamonds represent the average perception probability for each column of data.

Phosphene shape analysis revealed no significant difference between percept elongations in response to SA, SC, and AA stimuli. Average phosphene area was significantly larger with AA stimulation compared to SC (121.67% increase, P < 0.01) and larger compared to SA (64.8% increase) for participant 2. An increase in average phosphene area with AA stimulation was observed in participants 1 (38.64% increase from SC, 8.1% increase from SA) and 3 (2.39% increase from SC, 3.26% increase from SA), but this change was not significant. Perception thresholds were lower with AA stimulation, therefore with the same amplitude it is expected that this pulse creates larger percepts compared to symmetric pulses.

3.2. Computational modeling

A computational model of a RGC (described in 2.2) was used to further investigate the effect of order and duration ratio of pulse polarities on activation thresholds in epiretinal stimulation. The model predicts lower activation thresholds with longer IPGs, consistent with our human subject testing results and with previous research studies (Shepherd and Javel, 1999; McKay and Henshall, 2003; Weitz et al., 2014). Our results reveal that the effect of IPG is more pronounced for shorter pulse widths (Fig. 3). Figure (4) shows the percent change in threshold with AA stimulation of different duration ratios vs. SC stimulation. The model predicts lower activation thresholds with higher duration ratios, consistent with our previous in vitro research study (Chang et al., 2019). The threshold reduction effect is more pronounced with shorter pulse widths.

Figure 3.

Model predictions of percent change in activation threshold with SC stimulation with interphase gap vs. no interphase gap in a RGC. Effect of IPG on threshold reduction is stronger for shorter pulse widths.

Figure 4.

Model predictions of percent change in activation threshold with AA vs. SC stimulation in a RGC. Effect of AA stimulation on threshold reduction is stronger for shorter pulse widths

Figure (5) shows these parameters during an anodic pulse of 2 ms and 5 μA amplitude. In this model, with the electrode positioned nearest the junction of axon - axon narrow region, the action potential initiated in the narrow region. The sodium (Na⁺) and delayed rectifier potassium (K⁺,dr) currents are dominant because the membrane conductances for other ionic currents are negligible in the axon narrow region (Greenberg et al., 1999; Schiefer and Grill, 2006). There is a reduction in the m and n values and an increase in the h value. These changes are more gradual for n and h, and more immediate for m during the anodic pulse due to their different time constants (τ_m << τ_n, τ_h ). After removal of the hyperpolarizing pulse, m rapidly returns to its normal value while h is still elevated and n is depressed, which promotes I_Na > I_K and a high excitation probability. Figure 5.A and B were single pulses extracted from trains of 10 pulses. An anodic only pulse evoked an action potential once out of 10 pulses, while the AA pulse evoked an action potential 10 out of 10 pulses. This model result supports the mechanism of anode break as a means of making cells more sensitive to cathodic stimulation, if not exciting cells directly.

Figure 5.

Membrane response to stimulation in the RGC computational model. A) Top: Membrane voltage in response to an anodic pulse. The cell spikes with some latency after termination of stimulation. Stimulation includes an anodic current pulse of 2.4 ms duration and 5 μA amplitude. Bottom: Membrane gating variables before, after and during stimulation (between the dashed lines). Varying time constants of the gating variables results in an action potential. B) Top: Membrane voltage in response to an AA pulse. The cell spikes in response to the cathodic phase of the stimulus. Stimulation includes an AA current pulse of 2.4 ms anodic and 0.12 ms cathodic phase duration, with 5 μA anodic and 100 μA cathodic phase amplitude. Bottom: membrane gating variables before, after and during stimulation. Dashed lines show the limits of the anodic phase. The anodic only pulse evoked an action potential once out of 10 pulses, while the AA pulse evoked an action potential 10 out of 10 pulses. The differences in the initial values of the gating variables between A and B are due to the cumulative effect of previously applied pulses.

If the action potential is generated in the initial segment of axon (axon hillock), other ionic currents (I_K,_A, I_Ca, and I_K,_Ca) also need to be studied for a comprehensive examination of RGC activation mechanism with AA stimulation. The long hyperpolarizing current causes a drop in m, n, p, and c, and an increase in h and q values. The reduction in p and elevation in q (Fig. 5) results in an overall lower I_K,_A and lower total potassium current according to (2). I_Ca is expected to decrease during the hyperpolarization as c is lowered. However, the Ca²⁺ current has a negligible contribution to action potential generation and due to slower kinetics, only appears after the sodium channel activation has initiated the spike (Fohlmeister and Miller, 1997). I_K,_Ca is very small compared to other ionic currents during an action potential and is only dominant during the early portion of the interspike interval. There is no independent gating variable defined for the calcium activated potassium channel and the change in its conductance is ligand-gated based on the internal Ca²⁺ concentration. The Ca²⁺ concentration is virtually constant before initiation of the spike. However, a small decrease in I_K,_Ca is expected due to the membrane hyperpolarization according to (2). After removal of the hyperpolarization, sodium current begins to recover rapidly while the total potassium current is still lowered due to the slower kinetics of potassium channels. This makes the membrane more excitable.

We used our model to investigate a wider range of stimulus settings than was practical in the human subjects. In particular, we investigated AA stimulation with different IPG values for comparison to human testing results, and asymmetric cathodic-first (AC), which was not tested in human subjects. Figure (6) displays the percent threshold change in response to a range of IPGs (0–2 ms) and pulse types (AC and AA with 2, 5, 10, 20, 30 duration ratios, and SC). The effect of IPG duration on threshold with all pulse types could be described by a decaying logarithmic function. The effects of both IPG and asymmetric pulses on thresholds are less pronounced for longer pulses.

Figure 6.

Model predictions of percent change in activation thresholds with different pulse types and IPGs vs. standard SC stimulation. A1-A4) threshold change for AA vs. SC. The x-axis for each colormap represents pulse types: SC, AA with 2, 5, 10, 20, and 30 duration ratios. B1-B4) threshold change for AC vs. SC. The x-axis for each colormap represents pulse types: SC, AC with 2, 5, 10, 20, and 30 duration ratios. The y-axis represents IPG values. The color bar limits were kept constant to allow comparison between different pulse widths. The tiles with the highest threshold reduction percentage are outlined with a red line for each pulse width. The threshold reduction effects of IPG and asymmetric stimuli are less with longer pulse durations.

The comparison of AA to AC pulses demonstrated that AA is more effective than AC in threshold reduction but only at smaller duration ratios with longer IPGs for 0.05 – 0.12 ms pulse widths. This indicates that the anode break mechanism is more effective at reducing thresholds with a higher amplitude anodic phase and a gap between the anodic and cathodic phases. This is in agreement with the anode break excitation mechanism (Fig. 5) since the gating variable dynamics can result in membrane excitation only with some latency after termination of the hyperpolarizing pulse (Plonsey and Barr, 2007). In absence of a gap or with shorter gaps, AC is more effective than AA. Both reduce the threshold by applying an anodic phase with lower amplitude and longer duration compared to the anodic phase of symmetric pulses, reducing the inhibitory effect of the anodic phase. However, with AC the inhibitory phase comes after the excitatory phase. Thus, multiple factors appear to contribute to achieve the lowest excitation threshold. Our initial hypothesis that the threshold reduction effect of AA stimulation is based on the anode break excitation mechanism is partially supported by the results. AA pulses with small duration ratios (≤ 5) and long IPGs (≥ 2 ms), were most effective in threshold reduction for 0.05 – 0.12 ms pulse widths. SC pulses with IPGs longer than 0.5 ms, and asymmetric pulses with large duration ratios (≥ 20) were most effective in threshold reduction for 0.5 ms pulse widths (Fig. 6).

4. Discussion

Our human subject testing and computational modeling of a single RGC establish that asymmetric anodic-first stimulation is an effective strategy to reduce stimulation thresholds in epiretinal stimulation and that this effect is stronger at higher duration ratios, consistent with our in vitro findings in mice (Chang et al., 2019). Our computational modeling results demonstrate that this threshold reduction effect is more pronounced for shorter pulse widths. Even though the threshold current is lower with longer pulse widths, charge per phase increases with pulse width (Jensen, Ziv and Rizzo, 2005; Sekirnjak et al., 2006; Ahuja et al., 2008). Therefore, using shorter pulses in clinic will likely be advantageous because the safety limit is defined in terms of charge per unit area. Using short AA pulses can lower the threshold currents compared to a standard cathodic-first biphasic pulse, while maintaining the benefit of low charge per phase associated with short pulses.

According to our computational modeling results (Fig. 4) and previous in vitro findings, AA stimulation with smaller duration ratios (≤ 5) is less effective for threshold reduction. However, our model which tested a combination of AA stimulation and IPG (Fig. 6), indicated that AA pulses with small duration ratios (≤ 5) and long IPGs (≥ 2 ms) are amongst the most effective stimuli in threshold reduction. We have previously shown that smaller duration ratios of AA stimulation produce focal responses, meaning the RGC response is confined to the region near the active electrode (Chang et al., 2019). If adding an IPG to small duration ratio AA stimulation maintains a focal response, then this pulse type can offer great potential for threshold reduction and focal phosphenes simultaneously. Experimentation with combinations of AA and IPG will be necessary to suppport this claim. We based our model on a previously published RGC model (Schiefer and Grill, 2006). Although this model has a relatively simple geometry, it has predicted an experimental finding confirmed by our prior work, specifically that pulse widths below 0.1 ms avoid axonal activation (Weitz et al., 2015). However, a number of details have not been included in the model that might affect the threshold values. We investigated only a single electrode-retina distance, though it has been shown that this distance affects excitability (Esler et al., 2018). Different RGC types and neurite geometries respond differently to electrical stimulation (Chichilnisky and Kalmar, 2002; Werginz et al., 2015), and using a disc electrode might affect RGC responses (Greenberg et al., 1999). Thus, the specific predictions of our modeling study (e.g. using a duration ratio less than 5) may not hold when other models are used, and when tested clinically. But more generally, the model predictions are consistent with our clinical data and the results taken together support the claim that asymmetric pulses are an important feature to include in future prosthesis design, to allow optimization of pulse parameters.

The effects of AA stimulation had a similar pattern among our human participants, but variability inherent in human subjects testing made these results less clear. Recruitment is a challenge. We were able to enroll four participants in these experiments. Argus II subjects are typically elderly, and due to their blindness, are dependent on others for transportation. Three of the four participants live several hours away from the University of Michigan, which further limited their participation. The number of electrodes tested was also small because experiments were repetitive and mentally taxing for the participants and their time on campus was limited. We used manual process to measure perception thresholds since the automated process could not deliver anodic-first pulses. This added to the testing duration. In addition to the small sample size, other factors contributed to variability in participants’ responses. Participants’ accuracy in reporting phosphenes can be influenced by the time they take to respond to stimulation and their background visual activity (spontaneous phosphenes). This background activity can vary day-to-day. Intersubject differences that might be introducing variability to our data include the position of the implant on the retina, the electrode-retina distance, and the amount of retinal degeneration under the implant. The comparison of perception probabilities in response to different pulse types and IPGs was done at a single amplitude value due to the following reasons: 1) The software limitation prevented us from directly measuring the thresholds and 2) Our amplitude selection criteria and the tedious nature of the experiments did not allow for testing with more settings. The pattern of change in perception probability can be a representation of change in perception threshold. However the variability of location of the chosen amplitude on the logistic curve (based on the three criteria mentioned in the methods section), introduces some noise to the perception probability values.

Phosphene shape analysis revealed no significant difference among percept elongations with different pulse types. We expected less elongated phosphenes with short SA stimulation based on our in vitro findings. We have demonstrated that this pulse type activates cell bodies preferentially and avoids axonal stimulation, resulting in a more focal response (Chang et al., 2019). In the current study, phosphene elongation was compared for different pulse types at only one amplitude. Since pulse amplitude has been shown to effect the elongation of activation area with SA stimulation (Chang et al., 2019), this comparison needs to be done at different amplitudes for a more valid comparison with the in vitro study, but was not possible with the limitations on AA pulse amplitude. We observed a significant increase in phosphene size with AA stimulation compared to SC and SA in participant 2. Larger phosphenes with AA stimulation may be due to the lower activation thresholds with this pulse type, but this was noted in only one participant. Overall, there were not reliable trends relating pulse type to phosphene shape.

Our results suggest that incorporating asymmetric stimulation as an option can improve retinal prosthesis function. Lower perceptual threshold provides a wider dynamic range for stimulation amplitudes and phosphene sizes, which results in more flexibility for retinal stimulation. In addition, it allows for individual usage of electrodes that were previously grouped with neighboring electrodes due to high thresholds, and potentially improving the perception resolution. Lower perceptual thresholds will also result in lower power consumption. Our conclusions are limited by the small sample size and must be validated in a larger number of participants. On the whole, our data demonstrates the need for more flexible programming options in retinal prostheses. Optimization of the stimulation protocol for each patient will be necessary for the best outcomes. Systems that can only produce symmetric, biphasic pulses will limit optimization.