Spatiotemporal Interactions in the Visual Cortex Following Paired Electrical Stimulation of the Retina

Abstract

Purpose.: Retinal prostheses use spatiotemporal patterns of electrical stimulation across multiple electrodes to provide visual percepts to blind patients. It is generally assumed that percepts produced by individual electrodes are independent of one another, which may not be the case. In this study, we aimed to quantify interactions between pairs of electrical stimuli delivered to the retina.

Methods.: Normally sighted cats were implanted with a suprachoroidal electrode array. The retina was stimulated with a paired-pulse paradigm that consisted of a conditioning stimulus followed by a test stimulus, while recording multiunit activity in the visual cortex. Conditioning current, and spatial and temporal separation between the conditioning and test stimuli were varied. Cortical interactions were quantified by changes in multiunit activity elicited by stimulation with the paired-pulse paradigm, compared to stimulation of the test stimulus alone (control).

Results.: Interactions varied as a function of conditioning current and temporal separation between the two stimulating pulses. Cortical activity increased compared to the control condition at an interstimulus delay of 1.025 ms and was significantly suppressed for delays between 20 and 90 ms, returning to near control levels for longer delays. The level of interactions increased when the conditioning current was increased. Interactions were found to be similar for electrode separations up to 3 mm.

Conclusions.: Interactions between sequential stimulation of pairs of electrodes in a suprachoroidal retinal prosthesis occur for delays up to 100 ms and electrode separations of several millimeters. Knowledge of these spatiotemporal interactions is essential for developing effective patterns of stimulation for retinal prostheses.

Introduction

Retinal prostheses are designed for patients suffering from degenerative retinal disorders, such as retinitis pigmentosa, and provide sensations of light, termed “phosphenes,” by electrical stimulation of surviving retinal neurons. To provide functional vision to the blind to perform tasks such as navigating a room, or reading, the activation of potentially hundreds of electrodes may be required.^1,2 Therefore, retinal prostheses will need to use complex spatial and temporal patterns of electrical stimulation. By stimulating appropriate electrodes in a defined spatiotemporal sequence, one can take advantage of the visuotopic organization of the retina³ to convey the desired pattern to higher visual centers. Clinical trials have shown that retinal prostheses can provide useful visual cues, with patients able to perform tasks in a controlled clinical setting, such as identifying orientation of lines, reading letters, and identifying simple objects.^4–7 However, testing in humans has shown that electrode interactions can significantly influence the resultant percept (Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458).^8,9 Therefore, a thorough understanding of the interactions that occur between repetitive electrical stimuli applied to a single retinal electrode, as well as the spatiotemporal interactions that occur between retinal electrodes, is integral to the development of efficacious stimulation strategies. 

The timing between successive electrical stimuli on a single retinal electrode has been shown to affect retinal ganglion cell (RGC) activity. Studies performing in vitro electrical stimulation of the retina suggest that direct stimulation of RGCs may reliably be able to evoke stimulus phase-locked spike activity at stimulating rates of several hundred hertz,^10–12 similar to that seen when using visual stimuli.^13,14 Other studies, however, report conflicting results and have shown that the number of spikes evoked in RGCs decreases with increasing stimulation rate beyond 50 Hz.¹⁵ With indirect activation of RGCs via depolarization of the neurons in the inner retina, the number of spikes evoked is greatly reduced when stimulation rates in excess of 10 Hz are used.¹⁰ The desensitization of RGC spiking activity to fast rates of electrical stimulation is also evident when the retina is activated with only two electrical pulses¹⁶ as opposed to a train of stimulating pulses.^10,15 Upon presentation of two pulses of equal strength, a slight reduction occurs in the spikes evoked by the second pulse, compared to those evoked after the first pulse.¹⁶ The reduction in spiking activity following the second pulse becomes greater when the two pulses are closer in time, such that almost no spikes are evoked by the second pulse with temporal separations less than 25 ms (equivalent to 40 Hz). As larger numbers of stimulus pulses are presented to the retina in the form of a pulse train, further reductions in spike numbers occur after each subsequent pulse for rates above 4 Hz.^16,17 These findings may possibly play an important role in the fading of phosphenes observed by patients upon continuous electrical stimulation of the retina,^7,18 whereby phosphenes can disappear into background levels of illumination. However, phosphene fading is complex with great intersubject variability.¹⁸ Initially upon beginning stimulation, the appearance of the phosphene changes notably. Aside from losing brightness, the phosphene can be marked by changes to its size, shape, and even its color, as well as sometimes reappearing as a bright flash when stimulation ceases.¹⁸ These varying sensations may influence a person's ability to interpret a scene presented by a retinal prosthesis. 

Retinal prostheses used for clinical trials use one of two stimulation strategies to present visual patterns to the retina. The electrodes that comprise the pattern may be simultaneously stimulated.^19,20 Alternatively, the same group of electrodes can be electrically stimulated in a sequential fashion (Blamey P, et al. IOVS 2013;54:ARVO E-Abstract 1044).^20,21 Although implementation of a simultaneous strategy is straightforward, activation of neural prostheses in this manner can result in an uncontrollable percept due to electric field interactions between electrodes. The use of simultaneous stimulation in a cochlear implant (CI) has demonstrated the existence of interactions between each of the activated electrodes due to overlapping electric fields. It is thought that neurons located between stimulated electrodes receive a summation of stimuli from each of the active electrodes, resulting in unexpected variations to the threshold,²² loudness,²³ and pitch²⁴ of the resultant auditory percept. Therefore, sequential activation of electrodes on an array has been used in CIs to minimize electrode interactions,²⁵ as it is expected that interactions arising from electric field summation should be eliminated when stimulation of neighboring electrodes are separated in time. Sequential stimulation has been shown to produce much weaker interactions than simultaneous stimulation,^22,24 and the degree of these interactions is found to be a function of both spatial and temporal separation between electrodes. In the animal model, changes to the threshold and spatial profile of recordings from the auditory cortex reflect the spatiotemporal interactions that have been demonstrated to occur with CI recipients.²⁶ 

Psychophysical testing of retinal prostheses have shown that patients are able to discern simple lines with simultaneous electrode activation.^19,20 However, patients experience difficulties discerning multifaceted patterns such as letters, possibly due to interactions between the many, concurrently active electrodes.²⁰ Multifaceted pattern recognition has been successful in patients with retinal prostheses using sequential electrical stimulation.²⁰ Furthermore, simultaneous stimulation of a group of retinal electrodes elicits a brighter percept compared to the sequential stimulation of the same group of electrodes.⁹ However, stimulation of pairs of retinal electrodes results in more complex changes in perceived brightness, with both increases and decreases in brightness possible.⁸ Others have reported that the ability to detect individual phosphenes from adjacent electrodes is compromised if stimulation of the two electrodes occurs closely in both space and time (Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458). The extent to which these interactions occur varies widely between the different electrode placements (epiretinal⁸ versus subretinal [Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458]) and vastly different electrode array designs. It is difficult to predict the spatial and temporal range over which these interactions may occur, based on studies by others. This is especially true of the spatial range in which electrode interactions can occur, as each electrode placement and design vary in the area of the retina that is stimulated. Therefore, spatiotemporal interactions need to be evaluated by electrode array design. 

In the present study, we explored spatiotemporal interactions arising between electrical stimuli presented to electrodes on a clinical-grade suprachoroidal prosthesis that is currently in trial in three human patients (www.clinicaltrials.gov; ClinicalTrials.gov number, NCT01603576). We used a paired-pulse paradigm¹⁶ to study interactions at the level of the visual cortex in an animal model. This allowed us to systematically investigate in detail how a range of stimulating parameters (current level, spatial shift, and temporal shift) influences spatiotemporal interactions, which is not necessarily possible in the limited time frame available with patients during clinical trial. 

Methods

Anesthesia and Surgery

All experiments were performed with approval from the Royal Victorian Eye and Ear Hospital Animal Ethics Committee and conformed with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. Anesthesia before surgery was induced in eight normally sighted adult cats with ketamine (intramuscular [i.m.] 20 mg kg⁻¹) and xylazil (subcutaneous [s.c.] 2 mg kg⁻¹). Anesthesia was sustained with an infusion of sodium pentobarbitone (60 mg kg⁻¹ h⁻¹) and vital signs (respiration rate, end-tidal CO₂, and temperature) were monitored throughout the experiment. Sodium lactate (Hartmann's solution 2 mL kg⁻¹ h⁻¹) was infused intravenously, and injections of dexamethasone (i.m. 0.1 mg kg⁻¹) to minimize brain swelling and clavulox (s.c. 10 mg kg⁻¹) as an antibiotic were given regularly over the 3- to 4-day experimental period. 

Surgical implantation involved inserting the suprachoroidal array approximately 15 to 17 mm into a pocket created between the sclera and choroid,²⁷ such that the tip of the electrode array was located beneath or near the area centralis. Suturing of the array onto the sclera facilitated wound closure and ensured stable positioning of the array throughout the experiment. Following suprachoroidal surgery, the animal was placed in a stereotaxic frame (David Kopf Instruments, Tujunga, CA, USA) and a craniotomy overlying the visual cortex was performed contralaterally to the implanted eye. To determine the optimal placement for a planar 60-channel recording array (Blackrock Microsytems, Foxborough, MA, USA), an evoked potentials mapping procedure was used as previously described.²⁸ Briefly, a platinum ball electrode was used to measure thresholds of evoked potentials on the surface of the contralateral cortex at various points in response to electrical stimulation of a group of retinal electrodes. The point on the cortex with the lowest evoked potential threshold was chosen for implantation of the recording array, which corresponded well to the retinotopic positioning of the suprachoroidal array in the visual space.^28–32 Recording electrodes were typically inserted contralaterally to the stimulated eye, in the lateral gyrus in the visual cortex, corresponding to the macular region of the retina.³³ Recording areas included both cortical areas 17 and 18. 

Suprachoroidal Electrode Array

All experiments used a clinical-grade suprachoroidal retinal prosthesis, manufactured in-house. The prosthesis (Fig. 1A) was made of a silicon elastomer substrate that was contoured to conform to the eye. It consisted of an electrode array with 21 platinum disk electrodes (600-μm diameter) plus an additional two return electrodes (2-mm diameter) placed on the proximal end of the array to complete the current return path for electrical stimulation. Electrodes on the array were arranged in one of two ways. In the first design (4/8 eyes), electrodes were organized into seven rows by three columns. For the second design, which has recently been implanted into three patients as part of a clinical trial (www.clinicaltrials.gov; ClinicalTrials.gov number, NCT01603576) (4/8 eyes), electrodes were organized in a pattern such that the 21 stimulating electrodes were surrounded by an additional return made up of the electrodes on the outer edge shorted together. In both designs, electrodes were spaced 1 mm, center-to-center, equating to a separation of approximately 4° visual angle in the cat.³⁴ 

Figure 1

 

(A) Photograph of one of the clinical-grade suprachoroidal retinal prostheses used during the experiments (design 1). The leadwire (*) allows access to individual electrodes. (B) The paired-pulse protocol. The electrode for the test stimulus (P2) was kept constant, while surrounding electrodes of varying separations were chosen to apply the conditioning stimulus (P1). The conditioning stimulus always preceded the test stimulus.

Electrical Stimulation Protocol

The focus of this study was to investigate how repeated stimulation of the retina affects neural responses in the visual cortex. In particular, we aimed to characterize the effect of varying delay, intensity, and spatial location of electrical stimuli on multiunit spiking activity (MUA) in the visual cortex. An electrode on the retinal array was chosen to apply a fixed-amplitude current pulse (test pulse; P2), while a conditioning current pulse (P1) was applied to either the same or surrounding electrode as the test stimulus (Fig. 1B). Stimuli used were biphasic, cathodic leading current pulses (500 μs per phase, 25 μs interphase gap). The conditioning current varied from zero to a maximum stimulating level of 1.5 mA to ensure that charge densities were below 300 μC cm⁻².³⁵ The delay or stimulus onset asynchrony (SOA; i.e., onset of the conditioning stimulus to onset of the test stimulus) was either fixed to 1.025 ms (the fastest possible stimulation rate between electrical stimuli for the given stimulating pulse parameters) to assess the combined effect of the two stimuli, or varied in nonlinear increments between 20 and 500 ms to distinguish the response elicited by one stimulus from the other. Stimulus onset asynchronies between 1.025 and 20 ms were not chosen for the paradigm, as MUA arising from retinal stimulation usually occurs within a 3- to 20-ms window post stimulus.^28,36 Thus, the use of SOAs between 1.025 and 20 ms would not allow us to distinguish if the spikes recorded following the paired-pulse protocol were due to the conditioning stimulus, or the test stimulus, or the combination of the two stimuli. Each SOA and conditioning current combination was presented at random with a repetition rate of 1 Hz; all possible combinations were repeated 10 times. The situation whereby the amplitude of the conditioning stimulus was 0 μA (i.e., only test stimulus present) was considered as the control. Most of the electrodes used for stimulation were located on either the tip or one of the lateral edges of the stimulating array. While this was the case, the arrays themselves were specifically designed and shaped to conform to the curvature of the eye in the suprachoroidal space and relied on the active pressure in the eye to keep all electrodes flush against the choroid.³⁷ Therefore, we did not expect any differences in response properties between different stimulating electrodes according to the location of the electrode on the suprachoroidal array. 

To choose a suitable current for the test stimulus in the paired-pulse protocol, each retinal electrode was electrically stimulated by using single biphasic current pulses (0–1.5 mA, Δ50 μA, repetition rate of 1 Hz), and a spike rate versus stimulating current curve (Fig. 2A) was obtained for each recording site. From these curves, we obtained the current required to evoke 50% of the maximum neural activity, chosen to be the midpoint of the fitted sigmoid curve (P50).²⁸ The mean P50 current across all recording sites for the given stimulating electrode was taken to be the amplitude of the test stimulus in the paired-pulse protocol. 

Figure 2

 

Example of electrically evoked multiunit (MUA) in the visual cortex inclusion criteria. (A) Spike rate versus stimulating current curve. Symbols denote average spike counts for each current level, in the window 3 to 20 ms poststimulus onset. The test stimulating current, the P50 value, was derived from the fitted sigmoid curve. The shaded area denotes the 25% to 75% current range. (B) Poststimulus time histogram of spiking activity for electrical stimulation of a single retinal electrode across all current levels (0–1.5 mA). Although there is a second phase of spiking centered at 50 ms, this response was classified as a single phase response as per the criterion.

Data Analysis

Multiunit spiking activity was recorded continuously at a sampling frequency of 30 kHz with the Cerebus data acquisition system (Blackrock Microsystems, Salt Lake City, UT, USA). Spiking events collected at the time of the experiment with the Cerebus GUI software were used to determine the test electrode stimulating current. Additional offline processing with custom scripts written in Igor Pro (Wavemetrics, Lake Oswego, OR, USA) were used to identify spikes for subsequent data analysis. Briefly, the methods described by Heffer and Fallon³⁸ were used to remove signal artifacts due to electrical stimulation, after which the signal was band-passed filtered (Butterworth filter; frequency: 0.3–5 kHz, order: 3). An estimate of the root mean square (RMS) activity was computed over a 60-s moving window and spikes were detected if the signal exceeded four times the RMS value for the given window. 

To quantify the degree of stimulus interactions, spike counts on each individual recording site following the test stimulus in the paired-pulse protocol were compared to those obtained in the control condition, that is, 0-μA conditioning stimulus. For each conditioning current–SOA combination, spikes were counted in the time window 3 to 20 ms immediately after the test stimulus. Spike counts were then normalized to the average spike rate obtained for the control conditions across all SOAs, giving a spike ratio that defined the degree of interaction between the conditioning and test stimuli. A spike ratio of 1 indicated no interactions, with conditioning and test stimuli completely independent of each other. A spike ratio greater than 1 indicated facilitation or summation caused by presence of both the conditioning and test stimulus, while a spike ratio less than 1 indicated suppression of neural activity following the test stimulus due to the presence of the conditioning stimulus. The results of such analyses are presented with the conditioning current normalized with respect to the P50 current level obtained for a particular recording site, that is, Conditioning Current (dB) Display Formula . Normalization in this manner better reflects the strength of the conditioning stimulus with respect to its dynamic range. This has previously been shown to be approximately 10 dB with monopolar stimulation of a single retinal electrode,²⁸ equivalent to ±5 dB from the P50 current level; therefore, conditioning currents were restricted to the range of ±5 dB.  

Cortical Site Inclusion Criteria

A given cortical recording site was included for analysis provided the following criteria were met. 

Test Stimulating Current Was Within 25% to 75% of the Activation Current.

The current of the test stimulus was obtained from the average P50 level across cortical recording sites. Therefore, one would expect to see a wide range in the spike counts obtained for the control condition (i.e., no conditioning stimulus) according to where the test stimulating current falls along an individual cortical site's response curve. A recording site was only included in the analyses if the chosen test stimulating current was within the current range that evoked between 25% to 75% of the maximum activity on that recording site (Fig. 2A). It was important to apply this criterion as test pulse stimulating currents much below the P50 current were not large enough to produce a robust spiking response under the control condition, which may confound our results through the inability to observe a drop in spiking activity. Similarly, test pulse currents much larger than the P50 current resulted in cortical sites spiking near to its maximum capacity under the control condition, in which case, we may not observe an increase in spiking activity. In cases where we may still observe small levels of spiking, particularly for test stimulating currents that fall toward the lower end of this range, an additional minimum spike rate criterion for the control condition of 0.5 spikes/pulse needed to be met for a cortical site to be included for analysis. 

P1 Electrode Exhibits a Single Phase Spiking Response.

Electrical stimulation of the retina can exhibit a multiphased spiking response in the visual cortex.^28,36 It is believed that the first phase of spiking is due to the direct and indirect electrical stimulation of RGCs.¹⁷ Later spiking phases may be due to the late-stage spiking events shown to occur in vitro,¹⁷ possibly from other retinal circuits or cortical feedback mechanisms.³⁶ In this study, analysis was restricted to the first phase occurring 3 to 20 ms after the stimulus presentation. However, in cases where the conditioning stimulus evoked a multiphase response, it was not possible to distinguish spikes evoked after presentation of the test stimulus with those evoked as part of the multiphase response to the conditioner. These recording sites were removed from analysis to ensure that any interactions seen were due to what is presumed to be direct electrical stimulation of RGCs. To ensure unbiased classification of a multiphased response, a cortical site was classified as multiphased if the peak amplitude of the second phase of spiking exceeded 10% of the maximum spiking amplitude of the first phase of spiking (Fig. 2B). 

Statistical Analyses

All statistical analyses were performed in Igor Pro, whereby one-sample t-tests were used to compare computed spike ratios with expected population means, performed at a significance level of 0.05. The conditioning and test pulses were considered to be independent of one another if spike ratios were within 10% of 1, that is, interactions were considered to be absent for spike ratios between 0.9 and 1.1. Therefore, interactions were deemed facilitatory in nature if average spike ratios exceed a value of 1.1 and suppressive for average spike ratios below 0.9. Furthermore, spike ratios less than 0.5 were defined as heavily suppressed and between 0.5 and 0.9, moderately suppressed. Given that several hypotheses were tested for each classification, Bonferroni adjustment of the P value to control for family-wise error was too conservative a method; therefore, post hoc adjustments of P values were performed with the Benjamini and Hochberg correction method.³⁹ 

Results

Across the eight animals implanted, 132 cortical recording sites from the stimulation of 16/23 retinal electrodes passed all cortical site inclusion criteria and were analyzed when the paired-pulse protocol was applied to the same retinal electrode, that is, 0-mm separation. In addition, analysis of interactions was performed from 55 cortical sites, obtained from 15/21 electrode pairs for a separation of 1 mm (center-to-center distance, rounded to the nearest millimeter), and 58 cortical sites, obtained from 8 of 18 electrode pairs for a separation of 2 mm. A breakdown of the number of cortical sites that passed each of the inclusion criteria is provided in the Table. 

Table.

 

Details of the Number of Cortical Sites Available for Analysis

Electrode Spacing, mm	No. Cortical Sites*	Passed Test Current Criterion†	Passed Minimum Spike Rate Criterion‡	Passed Single Phase Criterion
0	480	194	178	132
1	239	100	62	55
2	170	89	63	58

*  Denotes cortical recording sites in which both stimulation of the test and conditioning electrode alone elicited a spike response.

†  Test stimulating current was within 25% to 75% of the activation current.

‡  Minimum spike rate of 0.5 spikes/pulse.

Cortical Responses to Paired-Pulse Stimulation

Electrical stimulation of the retina with the paired-pulse protocol was found to influence MUA in the visual cortex, compared to when the retina was electrically stimulated with the test stimulus alone (Fig. 3). Both SOA and conditioning current were found to influence spikes elicited by the test stimulus in the paired-pulse condition. 

Figure 3

 

Poststimulus time histogram demonstrating the effect of the paired-pulse protocol on MUA. In each example, the conditioning and test stimuli were applied to the same retinal electrode. (A) Fixed conditioning current, varying SOA. Control level of spike activity recorded after the test stimulus with the conditioning current equal to 0 μA (left). Spiking activity observed after the test stimulus for SOAs between 1.025 and 500 ms with conditioning current fixed to 0 dB (right). (B) Varying conditioning current and fixed SOA of 80 ms, whereby the arrows on the horizontal axis indicate the timing of the conditioning and test stimulus. Each row represents an increasing conditioning current amplitude. Spiking activity recorded up to 20 ms post stimulus is attributed to the conditioning stimulus and between 80 and 100 ms post stimulus, the test stimulus.

Figure 3A demonstrates the effect of SOA on MUA recorded after the test stimulus when the conditioning current was equal to the P50 level for each recording site. Despite a robust pattern of spikes obtained for the control condition when the retina was stimulated with only the test stimulus (Fig. 3A, left column), this spike pattern varied with the introduction of the conditioning stimulus and was dependent on the delay between the two stimuli (Fig. 3A, right column). At the shortest SOA (1.025 ms), an increase in the number of spikes was observed as compared to the control condition, indicating facilitation. However, the next SOA shown in the figure (30 ms) resulted in a strong suppression of MUA following the test stimulus where almost no spikes were recorded. As the SOA was increased, the suppression following the test stimulus was found to reduce, with near-normal activity, compared to the control condition, returned at an SOA of 500 ms. 

The amplitude of the conditioning stimulus was found to influence the strength of interactions (Fig. 3B). As expected, the number of spikes recorded immediately after the conditioning stimulus increased as the amplitude of the conditioning current was increased. However, as the conditioning current was increased, the number of spikes recorded after the test stimulus kept decreasing as compared to the control condition when the conditioning stimulus was absent. In fact, at the largest conditioning current (Fig. 3B, bottom row), almost all spikes following the test stimulus were eliminated. Thus, larger currents applied to the conditioning stimulus resulted in greater facilitation (SOA = 1.025 ms only) or suppression of neural activities following the test stimulus. Interestingly, even conditioning currents corresponding to the lower end of the dynamic range for each recording site (−5 dB), and eliciting only a weak response, resulted in suppression of activity elicited by the test stimulus. 

Facilitatory Neural Interactions

Facilitatory interactions occurred between paired pulses separated by 0 (Fig. 4A), 1 (Fig. 4B), and 2 mm (Fig. 4C) in the retina, whereby stimulation with the conditioning stimulus resulted in an increase in MUA that followed stimulation of the test stimulus, compared to stimulation with the test stimulus alone. These facilitatory interactions were apparent only for a SOA of 1.025 ms, whereby stimulation of the test stimulus immediately followed cessation of the conditioning stimulus. Facilitatory interactions measured in the cortex steadily increased with increasing conditioning current with an approximate 2-fold increase in spike ratios for larger conditioning currents. Conditioning currents below the P50 current level (i.e., less than 0 dB) were also able to elicit facilitatory interactions (spike ratio > 1.1). One-sample t-tests using the Benjamini and Hochberg correction method showed that the minimum conditioning current required to elicit significant facilitatory interactions was found to be −3 dB for 0- and 1-mm electrode separations and −1 dB for the 2-mm electrode separation. 

Figure 4

 

Spike ratios (mean ± SE) versus conditioning current obtained when SOA between conditioning and test pulse was 1.025 ms. Data for each current bin were obtained from (A) 97 to 130 of 132, (B) 37 to 54 of 55, and (C) 40 to 57 of 58 cortical recording sites. The number of cortical sites for each conditioning current is indicated in Figure 5. The + sign indicates those conditioning currents that yielded spike ratios that exhibited facilitation (i.e., spike ratios significantly above 1.1).

Figure 5

 

Color scale image representing the mean spike ratios obtained for SOAs 20 to 500 ms versus conditioning current (A) 0 mm, (B) 1 mm and (C) 2 mm. The black dashed line outlines the conditioning current–SOA combinations that yield heavily suppressed interactions and the gray dashed line, moderately suppressed interactions based on statistical analyses of spike ratios to population means. To the right of the moderately suppressed line are the conditioning current–SOA combinations where pulse pairs were found to be independent of one another. The number of cortical sites for each conditioning current is indicated on the right axes. A total of 97 to 130 (0 mm), 37 to 54 (1 mm), and 40 to 57 (2 mm) cortical recording sites were included for analysis for each current bin.

Suppressive Neural Interactions

Stimulus onset asynchronies ranging from 20 to 500 ms exhibited suppressive neural interactions (Fig. 5). These interactions were dependent on both conditioning current (abscissa) and SOA (ordinate) and were very similar for each of the electrode separations tested. Suppressive interactions were strongest for SOAs between 20 to approximately 100 ms. In this region, cortical spikes following the test stimulus could be suppressed by as much as 90% compared to the control, particularly for large conditioning currents and small SOA combinations. For conditioning currents below the P50 current level (i.e., below 0 dB), a 40% to 50% suppression following the test stimulus was observed. Stimulus onset asynchronies beyond 100 ms marked the beginning of the recovery period for spike activity, with notably higher spike ratios than those obtained with SOAs less than 100 ms, although a drop in spike rate by 10% to 20% was still apparent, even for long SOAs. 

One-sample t-tests using the Benjamini and Hochberg correction method were used to compare spike ratio data for each conditioning current and SOA combination to expected population means in order to test for suppression. As there was a varying degree of suppression found, these interactions were further classified as heavily suppressed (spike ratio < 0.5, black dashed line), moderately suppressed (0.5 ≤ spike ratio < 0.9, gray dashed line), or absent (0.9 ≤ spike ratio ≤ 1.1). Interactions were found to mostly show heavy suppression for SOAs up to 90 ms (0-mm electrode separation; Fig. 5A) and 80 ms (1- and 2-mm electrode separation; Figs. 5B, 5C, respectively) and moderate suppression beyond this critical SOA. For SOAs of 400 and 500 ms, the two stimuli in the paired-pulse paradigm were found to be independent of one another only if the current of the conditioning stimulus was below the P50 level. 

Differences in Interactions With Increasing Electrode Separation

It is evident from Figures 4 and 5 that the conditioning current and SOA combinations that resulted in facilitation, heavy suppression, and moderate suppression are comparable for the three electrode spacings that were investigated in this study. Despite these similarities, we may expect that the degree of facilitation and suppression is greater between pulse pairs presented on the same retinal electrode as opposed to two different electrodes. To confirm if this hypothesis was true, differences in spike ratios obtained with single- and paired-electrode stimulation were compared. Where data were available for both single-electrode and paired-electrode stimulation, spike ratios obtained with single-electrode stimulation were subtracted from those with paired-electrode stimulation to determine differences in the magnitude of interactions (Fig. 6). A difference of zero indicated that interactions were not affected by the location of the conditioning stimulus relative to the test stimulus. A two-tailed t-test with Benjamini and Hochberg post hoc correction of P values was used to determine the significance of these differences. 

Figure 6

 

Differences in mean spike ratios obtained when conditioning and test stimuli were presented on the same electrode compared to 1-mm and 2-mm electrode pairs. (A, C) Differences in facilitatory interactions (mean ± SE) obtained at an SOA of 1.025 ms. (B, D) Differences in suppressive interactions obtained with SOAs between 20 and 500 ms. The number of cortical sites for each conditioning current is indicated on the right axes (these numbers also apply to [A, C]). Data for the 1-mm electrode pair separation were obtained from 20 to 37 of 40 cortical recording sites and for the 2-mm electrode pair, 18 to 33 of 34 cortical recording sites. The + sign indicates mean differences in spike ratio that were found to be significantly different from zero.

Spike ratio differences between paired-electrode (1-mm separation) and single-electrode stimulation are shown in Figures 6A and 6B. For an SOA of 1.025 ms (Fig. 6A), the spike ratio difference did not deviate significantly from zero. Spike ratios exhibited a trend of positive differences for SOAs between 20 and 500 ms (Fig. 6B). This suggests that the suppressive interactions between 1-mm paired electrodes tended to be weaker than those obtained with single-electrode stimulation; however, these differences were not significant. Therefore, there was little difference in neural interactions resulting between two stimuli separated by 0 or 1 mm in the retina. 

Figures 6C and 6D present the analyses of the difference in mean spike ratio for 2-mm–separated paired-electrode and single-electrode stimulation. Negative differences for facilitatory interactions (Fig. 6C) were obtained with conditioning currents of 4 to 5 dB, which is indicative of weaker facilitation occurring when the conditioning and test stimuli were applied between the 2-mm–spaced electrode pairs. Generally, suppression of MUA occurring at SOAs longer than 50 to 60 ms was greater when the conditioning and test stimuli were applied to the same site in the retina (Fig. 6D). Unexpectedly, for SOAs between 20 and 60 ms, suppression was stronger when the two stimuli were separated in the retina by 2 mm than that seen with single-electrode stimulation. Statistical analyses revealed, however, that very few neural interactions between the 2-mm–separated paired-electrode stimulation were different from those obtained with single-electrode stimulation. 

Interactions Between Paired-Electrode Stimulation Separated by 3 mm in the Retina

In some instances, there was the opportunity to apply the paired-pulse paradigm to two retinal electrodes spaced 3 mm apart. The results of such stimulation are shown in Figure 7 (six cortical sites obtained from three electrode pairs). It is difficult to make inference from such few cortical sites; however, the results appear to show that facilitation may still be possible between two stimuli separated by 3 mm (Fig. 7A). Conditioning currents of 0 dB and greater resulted in the doubling of spikes evoked by the test stimulus when presented as part of a pair. However, statistical analyses to qualify these interactions revealed no significant facilitation. 

Figure 7

 

Interactions obtained between conditioning and test stimuli with a 3-mm center-to-center electrode spacing. (A) Facilitatory interactions. (B) Suppressive interactions. The black dashed line outlines the conditioning current–SOA combinations that yield heavily suppressed interactions and the gray dashed line, moderately suppressed interactions based on statistical analyses of spike ratios to population means. To the right of the moderately suppressed line are the conditioning current–SOA combinations where pulse pairs were found to be independent of one another. (C) Mean difference in facilitatory spike ratios between 0-mm and 3-mm electrode pairs. (D) Mean difference in suppressive spike ratios between 0-mm and 3-mm electrode pairs. The + sign indicates mean differences in spike ratio that were found to be significantly different from zero.

In addition, as was seen previously with the 1-mm and 2-mm paired-electrode stimulation, spikes evoked after electrical stimulation with the test stimulus can be more than halved if the stimulation occurs within 80 ms of the conditioning stimulus (Fig. 7B). What is notable however is that spiking activity for SOAs greater than 80 ms typically ranges between 80% and 100% of the number of spikes evoked by electrical stimulation with the test stimulus only. Classifications of neural interactions revealed that for SOAs of 80 ms or shorter, interactions between the two stimuli primarily exhibited moderate suppression. When an SOA of 90 ms or more separated the two stimuli, the conditioning and test stimuli were found to be independent of one another, exhibiting no interactions. 

Following from these results, the differences in spike ratios used to quantify interactions between the 3-mm–separated paired-electrode and single-electrode stimulation indicated that stronger facilitation and suppression occurred when the pulse-pairs were applied to the same retinal site (Figs. 7C, 7D). In particular for the suppressive interactions, the magnitude of the difference in spike ratios between 0-mm and 3-mm electrode pairings (Fig. 7D) were greater than those obtained with the 1-mm (Fig. 6B) and 2-mm (Fig. 6D) electrode pairs. However, statistical analyses did not show a significant difference in spike ratios between single-electrode and 3-mm paired stimulation, which may be a consequence of the small number of cortical recording sites available for analysis (average statistical power = 0.36). 

Discussion

The aim of this study was to evaluate, by recording from the visual cortex, interactions arising from electrical stimulation of the retina with a pair of current pulses presented on a suprachoroidal electrode array. Pairs of stimuli separated in the retina by 0 to 3 mm were found to interact. The resultant interactions showed both facilitation and suppression, dependent on the current of the conditioning stimulus and the delay between the onsets of the conditioning and test stimuli, but not on the spatial distance separating the two stimuli. Interactions were found to begin to abate with wider electrode separations and long delays. 

Comparison to Other Electrophysiological Data

Stimulation of the paired-pulse paradigm at the fastest rate possible for the given biphasic pulse parameters resulted in facilitation of MUA. Similar facilitatory responses have been reported with paired-pulse electrical stimulation of auditory nerve fibres,^40,41 whereby the presence of the conditioning stimulus is found to reduce threshold of spiking activity elicited by the test stimulus. The facilitation is a consequence of the charge-storing properties of neural membranes⁴⁰; electrical stimuli depolarize the neuron⁴² altering the charge on its membrane, which takes time to dissipate.⁴⁰ As a result of the altered membrane potential, the neuron is more susceptible to subsequent stimuli presented in quick succession after the first, summing the resultant charge from each pulse, and so, increasing the likelihood of a spiking event. High rates of stimulation, in which we can take advantage of the summation of electrical charge on the neural membrane, could potentially be used as a strategy to reduce thresholds or elicit brighter phosphenes. While phosphene thresholds have been reported to increase⁴³ and perceptual brightness decrease⁴⁴ with increasing stimulation rate in normally sighted individuals following extraocular stimulation, in patients with retinitis pigmentosa, perceptual thresholds have been shown to decrease when electrical stimulation rates are increased,^45,46 and the phosphenes evoked in such situations have been described as brighter.^45,47,48 However, these perceptual observations may not necessarily be solely due to the facilitation described here, but may also reflect the ability of RGCs to follow fast rates of electrical stimulation,⁴⁷ as has been demonstrated in vitro. 

In contrast, most SOAs resulted in suppressive interactions occurring between pulse pairs. In particular, SOAs between 20 and 90 ms, equivalent to a stimulating pulse rate of approximately between 10 and 50 Hz, resulted in the greatest amount of spike suppression with a reduction in MUA by 50% or greater. The similarities in our results to those obtained in vitro^15–17 suggest that the mechanisms behind the heavily suppressed response may be due to the abolishment of the responses from RGCs that occurs with indirect stimulation at fast rates of electrical stimulation (termed as “RGC desensitization”). There are several mechanisms that have been presented in the literature behind RGC desensitization including the role of the neural network¹⁶ as well as amacrine cell inhibition.¹⁷ It must be noted however that RGC desensitization is known to persist in the absence of amacrine cell inhibition as shown in a study by Freeman and Fried.¹⁷ The same study suggested that desensitization is likely to occur upstream of the spike generator, further supporting the role of the neural network, including photoreceptors, bipolar cells, and horizontal cells, in contributing to RGC desensitization. Furthermore, only a single region of suppression was evident in our study. In an early study examining RGC activity with paired-pulse stimulation, several regions of suppression are evident.⁴⁹ In the study by Crapper and Noell,⁴⁹ a single electrical stimulus elicits distinct phases of excitation and inhibition, leading to several distinct bursts of spiking activity in the RGCs. To paired-pulse stimulation, the resultant activity measured in the RGCs represented the algebraic sum of the spiking activity from each of the individual pulses in the pair, leading to several regions of suppression.⁴⁹ Cortical MUA recorded in response to retinal stimulation also elicits several phases of spiking. Incorporating these latter phases of spiking into our analyses may possibly reveal additional periods of suppression in our results, as has been observed by Crapper and Noell.⁴⁹ However, as the origin of the late phase of spiking is unknown (i.e., possible retinal or cortical mechanisms), our results were conducted on the analysis of cortical recording sites that exhibited only a single phase of spiking. 

Stimulus onset asynchronies in excess of 90 ms between the two stimulating pulses also resulted in a reduction in spiking activity to the test stimulus, with spike rates dropping by as much as 10% to 50% compared to when an electrode was stimulated alone. It has been reported that suppressive interactions between pairs of stimulating pulses separated by as much as 400 ms can occur, which are only abolished when the pulses are separated by an interval of 650 ms,¹⁶ a delay outside the range used in this study. Excitatory current produced by inner retina cells to electrical stimuli decreases at pulse repetition rates as low as 2 Hz,¹² presumably resulting in a reduction in the subsequent excitation of RGCs. These findings could explain why even with SOAs as long as 500 ms, spike counts in the cortex after the test stimulus did not match the number of spikes evoked in the control condition. 

Although prior studies delving into temporal interactions between electrical stimuli have been conducted in the retina,^15–17 evidence of inhibition is also apparent in the electrically evoked potential (EEP), recorded from the visual cortex, whereby the amplitude of the EEP is found to decrease with increasing stimulation frequency.^50–52 Temporal transfer properties are known to differ between retinal and cortical neurons,⁵³ as well as between lateral geniculate nucleus (LGN) neurons and cortical neurons.⁵⁴ Therefore, neural processes occurring in either the LGN or visual cortex may also have contributed to the inhibition observed in our study, as well as the known retinal desensitization that occurs with repeated electrical stimuli. 

Interactions Between Pulses Exist Over Several Millimeters in the Retina

Our results are in accordance with studies performed in CIs²⁶ and retinal prosthesis simulations,⁵⁵ which have shown that interactions between stimulus pulses become weaker as a function of the spatial separation between the two stimulating electrodes. The reduced interactions are due to the reduction in the overlap of activated neural populations from each of the electrodes. Electrical stimulation of two suprachoroidal electrodes with the paired-pulse paradigm, even separated by up to 2 mm in the retina, resulted in comparable cortical interactions to stimulation on a single electrode. Using methods developed from an earlier study,²⁸ we estimate that the retinal spread of activation from the 600-μm diameter electrode used in this study spans a diameter of 2.9 mm. Furthermore, a mathematical modeling study assessing the effects of electrode spacing on cross-talk⁵⁵ found that an array placed at a distance of 400 μm from the surface of the retina required a center-to-center spacing of at least 2.5 mm to eliminate cross-talk, which is similar to limit of retinal spread of activation found in our study. This suggests that owing to spread of current in the retina, a similar population of retinal cells may have been excited from the electrical stimulation of each of the electrodes in the 1-mm– and 2-mm–separated pairs. One way in which these spatiotemporal interactions may potentially be reduced is with the appropriate choice of electrode return configuration, which is known to influence the electric field emanating from a stimulating electrode.^28,56–58 Electrode return configurations that produce focused currents in the retina are expected to exhibit weaker spatiotemporal interactions as current spread between electrodes is minimized. Indeed, with CIs, bipolar and tripolar stimulation have been shown to be effective in reducing electrode interactions between pairs of cochlear electrodes that have been activated simultaneously and with a temporal delay, compared to monopolar stimulation.^26,59 Similarly, simulations in the retina have demonstrated that hexagonal stimulation can reduce “cross-talk” between retinal electrodes.⁵⁵ 

For the 3-mm electrode spacing, where the distance between the conditioning and test stimuli exceeded the electric receptive field of a cortical site, electrical stimulation with the pulse pairs were found to be more independent of one another. Our current spread data suggest that stimulation of two electrodes spaced 3 mm or greater apart should result in two independent phosphenes whose shape and brightness should not be influenced by the stimulation of the other electrode, thus providing an alternative way to reduce spatiotemporal interactions. Preliminary testing with the suprachoroidal electrode array used in this study in human subjects has shown that simultaneous stimulation of two electrodes separated by 3 mm can appear as two, nonoverlapping phosphenes⁶⁰ and supports the findings of this study. However, the absence of a statistically significant interaction was apparent only if the SOA between the stimulation of each electrode was at least 90 ms. Suppressive interactions still occurred after paired-pulse stimulation with SOAs smaller than 90 ms, albeit this suppression tended to be less than the 0-, 1-, and 2-mm electrode separations. Therefore, spreading electrodes further apart to abate interactions may not necessarily hold true when several electrodes are stimulated sequentially to form a pattern. Unlike the 1-mm– and 2-mm–separated electrode pairs, the suppression of neural activity between the 3-mm–separated electrode pairs is not thought to have arisen from current spread between neighboring electrodes. Rather, the interactions are likely to arise from the wide-field amacrine cells that span several millimeters in the retina,^61,62 causing inhibition through the release of the neurotransmitter γ-aminobutyric acid to postsynaptic cells.⁶³ Electrical stimulation of the retina has been shown to excite amacrine cells,^12,64 releasing inhibitory currents that last approximately 100 ms.¹² This is of a similar range to the SOAs that resulted in moderate suppression in the 3-mm–separated electrode pair. We must also consider that in addition to current spread, amacrine cell activation may also have played a role in the suppression of neural activity in the 1-mm– and 2-mm–separated electrode pairs, also possibly explaining the stronger suppressive interactions observed with the 2-mm electrode pair compared to single-electrode stimulation occurring for SOAs between 20 and 50 ms. However, the present work was performed in a normal feline retina, and it is unknown how neurite sprouting and rewiring of retinal cells that occur in conditions such as retinitis pigmentosa^65,66 will affect such spatiotemporal interactions. 

Comparison to Human Psychophysics Data

Recently, Horsager et al.⁸ have reported a facilitatory interaction in humans between electrode pairs upon epiretinal stimulation. This results in a percept that is brighter than expected, based on the brightness of the phosphene evoked by the stimulation of each electrode alone. The facilitatory effect described by Horsager et al.⁸ is strongest for temporally overlapping stimuli and declines as the delay between the stimulation of each electrode increases. In our study, facilitation was only observed with an SOA of 1.025 ms, while for the next SOA used, 20 ms, significant suppression was observed. The longest SOA used in the study by Horsager et al.⁸ is 9 ms. Therefore, we can assume that facilitatory interactions between two suprachoroidal electrodes will occur with SOAs longer than 1.025 ms, and that interactions may transition from facilitation to suppression at an SOA less than 20 ms. A similar such trend has been observed in humans with paired-pulse paradigms in the form of twin- or two-flash experiments that explore interactions between successive light flash visual stimuli.^67–71 

In contrast to the results obtained in this study where both facilitatory and suppressive interactions were seen, only suppressive interactions have been observed between electrode pairs in a study with subretinal stimulation in humans (Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458). The suppressive interactions reduced the ability to detect the phosphene evoked by the stimulation of the second electrode in the pair. Irrespective of the spatial separation between the two electrodes, in the study by Wilke et al., interactions have been observed when a delay of 155 ms or greater is applied between the stimulation of each electrode. In our study, neural suppression was reduced when an SOA of 100 ms (0-mm electrode separation) or 90 ms (1-mm, 2-mm, and 3-mm electrode separation) was applied between the stimulation of the conditioning and test electrode, similar to the range of 155 ms reported with subretinal stimulation. Therefore, to avoid suppression, it would seem preferable to use stimulating frequencies below approximately 10 Hz where cortical suppression was found to be minimal. In practice however, stimulating frequencies greater than 10 Hz are used so that the patient does not perceive flickering of the phosphene. Testing in patients also shows that it is difficult to achieve persistent phosphenes,¹⁸ even with stimulating frequencies lower than the 10 Hz required to overcome suppression of MUA. Phosphenes initially appear bright and fade to nothing long before electrical stimulation with the pulse train ceases.¹⁸ Similar phenomena have been observed with transcorneal electrical stimulation of normally sighted and blind individuals, although these phosphenes still maintain their visibility⁷²; therefore, phosphene fading with electrical stimulation of the retina may be unavoidable. It must be noted that our study was performed by using a paired-pulse paradigm, while retinal prostheses mostly use pulse trains for stimulation. In such cases, interactions will likely not only occur between the first two pulses in the train but also extend over subsequent pulses as well. For example, while facilitation may occur between the first two pulses, the first pulse may have a suppressive effect on the third or subsequent pulses. 

The spatial dynamics of interactions occurring between two suprachoroidal stimuli vary compared to those reported previously in human subjects with epiretinal⁸ and subretinal (Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458) stimulation, but this may not be surprising owing to species differences and the fact that the human studies were conducted in blind individuals, or the different electrode designs. Interactions between epiretinally placed electrodes can occur between two electrodes separated by up to 2.4 mm,⁸ while for subretinally placed electrodes, interactions between electrodes can span a distance of approximately 0.8 mm (Wilke RG, et al. IOVS 2011;52:ARVO E-Abstract 458). From the results of our study, interactions are likely to exist between two suprachoroidal electrodes spaced anywhere between 2 to 3 mm apart. Aside from differences in the models used to investigate electrode interactions, the differences in the size of the stimulating electrodes may also account for the different results among these studies owing to differences in the area of the retina that is stimulated. We also need to consider the placement of the retinal prosthesis on the spatial aspects of interactions. Suprachoroidally placed electrodes may activate larger areas of the retina owing to a more widely dispersed electric field resulting from an increased distance from the retina, compared to epiretinal and subretinal electrodes.^73,74 Therefore, for comparably sized electrodes, spatial interactions are expected to occur over a wider distance in the retina with suprachoroidal stimulation compared to epiretinal and subretinal stimulation. 

Summary

Interactions between sequential stimulation of a pair of suprachoroidal retinal electrodes span several millimeters in the retina. These interactions were dependent on conditioning current and the spatial and temporal shifts between paired-pulse stimulation. Interactions between stimuli were comparable for 0-mm–, 1-mm–, and 2-mm–spaced electrode pairs: successive stimulation with a pair of pulses separated by 1.025 ms resulted in an increase in MUA evoked by the test stimulus, a significant reduction in MUA evoked by the test stimulus when separated by less than approximately 80 to 90 ms, and slight changes to MUA when paired stimuli were separated by greater than 100 ms. Interactions between stimuli were abolished when presented on pairs of electrodes spaced 3 mm apart only when combined with temporal shifts of 90 ms or greater. Knowledge of these spatiotemporal interactions is essential for developing effective patterns of spatiotemporal stimulation for retinal prostheses. 

Acknowledgments

The authors thank Felix Aplin, Sam John, and Ronald Leung for assistance with data collection; Michelle McPhedran and Alexia Saunders for technical assistance; Penelope Allen and Jonathan Yeoh who performed all surgeries; Owen Burns, Helen Feng, and Vanessa Maxim for electrode array fabrication; and Robert Shepherd for reviewing earlier versions of this manuscript. This study was conducted at the Bionics Institute at St. Vincent's Hospital and the Biological Research Centre at the Royal Victorian Eye and Ear Hospital. 

Supported by the Australian Research Council through its Special Research Initiative in Bionic Vision Science and Technology awarded to Bionic Vision Australia and by the Bertalli Family Foundation to the Bionics Institute. The Bionics Institute acknowledges the support received from the Victorian Government through its Operational Infrastructure Program. 

Disclosure: R. Cicione, None; J.B. Fallon, None; G.D. Rathbone, None; C.E. Williams, P; M.N. Shivdasani, P