Survey of electrically evoked responses in the retina - stimulus preferences and oscillation among neurons

Electrical stimulation is an important tool in neuroscience research and clinically. In the retina, extensive work has revealed how the retinal ganglion cells respond to extracellular electrical stimulation. But little is known about the responses of other neuronal types, and more generally, how the network responds to stimulation. We conducted a survey of electrically evoked responses, over a range of pulse amplitudes and pulse widths, for 21 cell types spanning the inner two layers of the rabbit retina. It revealed: (i) the evoked responses of some neurons were charge insensitive; (ii) pulse-width sensitivity varied between cell types, allowing preferential recruitment of cell types; and (iii) 10–20 Hz damped oscillations across retinal layers. These oscillations were generated by reciprocal excitatory / inhibitory synapses, at locations as early as the cone-horizontal-cell synapses. These results illustrate at cellular resolution how a network responds to extracellular stimulation, and could inform the development of bioelectronic implants for treating blindness.


Results
A library of electrically evoked responses. We assembled a library of morphology, light evoked responses and electrically evoked responses for 21 cell types across the inner two layers of the rabbit retina, encompassing all major interneuron types, including horizontal cells, bipolar cells, amacrine cells, as well as the retinal ganglion cells (RGCs). The isolated rabbit retinas were placed photoreceptor-side down (Fig. 1a) on a multielectrode array (MEA) (Fig. 1b). Each neuron was characterized by intracellular recording and by morphology ( Supplementary Figs 1-4). Light responses were evoked by flashing a spot over the neuron.
For extracellular electrical stimulation, one of the MEA electrodes was chosen as the stimulating electrode and eight distant electrodes around the MEA perimeter served as the stimulus return. For consistency, we only recorded from neurons ≤ 100 µm from the stimulation electrode center. For each neuron, we tested 0.1, 0.2 and 0.5 ms charge-balanced, cathodic-first, biphasic pulses, over a range of stimulus amplitudes. This permitted us to examine how the evoked responses varied as a function of pulse width and pulse amplitude.

Horizontal cells and cone bipolar cells.
We began with the A-type horizontal cells (Fig. 1c) in the distal retina. Consistent with previous findings 26,27 , these neurons have a depolarized resting potential in darkness (−48 mV in Fig. 1d), and exhibited sustained hyperpolarization in response to the stationary light spot (Fig. 1d). Following a single extracellular stimulus, horizontal cells exhibited an initial transient depolarization of ~20 ms duration, followed by a series of damped oscillations, which decayed over several hundred milliseconds. This is illustrated in Fig. 1e as overlaid grey traces, for 0.1 ms and 0.5 ms biphasic pulses. We note that these long-duration, damped oscillations are of biological origin, rather than electronic artifacts, because they could be suppressed by pharmacological blockers (see section "Damped oscillations arose through synaptic interactions") and their temporal kinetics varied between cell types (see also Figs 2c, 3c and 4c). To illustrate how the membrane potential varied over time, following stimulation, as a function of stimulus amplitude, we calculated the average deviation from resting potential over ten trials ( ∆V mean m ) and plotted the data as color charts. The size of the initial transient response and the subsequent damped oscillation increased with stronger stimulus strength (Fig. 1e). These observations were consistent across all horizontal cells. This is summarized for all recorded horizontal cells in Fig. 1f, where we computed the standard deviation of the change in membrane potential, averaged over all recorded neurons, in the 390 ms analysis period (s V mean m ∆ ). Next we examined the cone bipolar cells (Fig. 1c). The Off and On cone bipolar cells hyperpolarized and depolarized, respectively, in response to the light spot (Fig. 1d). Following extracellular stimulation, these neurons also responded with an initial transient depolarization, followed by damped oscillation (Fig. 1e), akin to the horizontal cells. However, given the same stimulus strength, cone bipolar cells' responses were considerably larger than those of horizontal cells (Fig. 1f).
Amacrine cells. The amacrine cells are the most diverse class of retinal neurons. We organized these neurons into three categories according to their dendritic field size 28 : narrow (<125 µm), medium (125-400 µm) and wide (>400 µm). Whenever possible, we matched our recorded neurons to documented amacrine cell types in the rabbit retina 28-34 based on morphology and, if reported previously, also by light responses. The previously given names are shown between parentheses in Figs 2a, 3a and 4a.
The light-evoked responses (Fig. 2b) and electrically evoked responses (Fig. 2c) of narrow-field amacrine cells were extremely diverse. Similar diversity also held for the medium-field (Fig. 3) and wide-field ( Fig. 4) amacrine cells. Nevertheless, several general observations could be made. First, a single extracellular stimulus elicited responses of several hundred milliseconds in most amacrine cells. Second, increasing the stimulus amplitude gradually increased the size of the evoked response. Third, electrical stimulation evoked oscillatory membrane potentials in most amacrine cells.
To provide a reference for comparison against the existing studies, we examined the On-Off direction selective RGCs. These neurons respond to both the onset and the offset of light stimuli with spikes 39 (Fig. 4b). Extracellular stimulation pulses of sufficient amplitude elicited a short latency spike (Fig. 4c), through direct activation of these neurons 18 . Subthreshold responses were observed across a range of stimulus amplitudes. Long latency spikes were often seen above the subthreshold responses (Fig. 4c). Increasing the stimulus charge increased the overall response magnitude over the 390-ms analysis time window (Fig. 4d).
Wide-field neurons & horizontal cells are charge insensitive. The library we have assembled offers unique opportunities for within and between cell-type analyses. We asked if increasing the stimulus charge consistently resulted in larger responses, that is, larger membrane potential variations ( ∆ s V mean m ) following pulse delivery. For each type of neuron, we compared the s V mean m ∆ as the charge increased (repeated measure 2-way ANOVA, Fig. 5a). All but the horizontal cells and two wide-field amacrine cells exhibited significant increases in SCiENTifiC REPORTS | 7: 13802 | DOI:10.1038/s41598-017-14357-1 response amplitude with increasing charge (p < 0.05, marked by stars in Fig. 5a). Morphologically, these wide-field amacrine cells have dendrites spanning several hundred µm or more across the retina. The A-type horizontal cells also have broad-coverage neurites, and are connected to neighboring horizontal cells of the same type via gap junctions, forming a syncytium 40 . These neurons' response amplitude (    Fig. 5a). Approximately half of the cell types showed significant differences in response amplitude between pulse widths (p < 0.05, marked by stars in Fig. 5a).
Changing the most strongly driven cell type with pulse width. The prevalence of pulse width sensitivity among retinal neurons (Fig. 5a, charge fixed at 20 nC) raises the question of whether pulse width could be used to preferentially recruit neuronal types. We performed pairwise comparisons of electrically evoked  Fig. 5b-d). Given the same stimulus, some cell types responded significantly more strongly than the rest of the population (2-sample t-test, *denotes p < 0.05 in Fig. 5b-d).
Using 0.1 ms pulses (Fig. 5b), the NF5 amacrine cells responded most strongly among the neuronal types. The response magnitudes of MF3, MF5, SACa (A-type starburst amacrine cell) and WF1 were also statistically higher than the rest of the population. As the pulse width was increased to 0.2 ms (Fig. 5c), then to 0.5 ms (Fig. 5d), the most responsive cell type shifted from NF5 to MF3. Concomitantly, the responses of MF5, SACa and WF1 amacrine cells decreased with increasing pulse width. Thus, pulse width variations changed the most strongly driven cell types, offering opportunities for preferentially activating certain cell types.

Electrical stimulation evokes damped oscillation.
A brief extracellular stimulus elicited dramatic, damped oscillations in the majority of cell types we examined, lasting several hundred milliseconds (Figs 1e, 2c and 4c). To precisely quantify these oscillations, we computed the difference in response power spectral density (PSD) with and without electrical stimulation (PSD stim − PSD baseline ). We observed an increase in power over the 10~20 Hz range for most neuronal types (Fig. 6a). Notably, the damped oscillations were widespread throughout the network. As a comparison to rare cell types with little oscillation, the change in PSD of NF1 amacrines was minimal following electrical stimulation (Fig. 6b).

Damped oscillations arose through synaptic interactions.
How did the widespread, damped oscillation arise? Oscillations could be generated by: (i) spontaneously oscillating neurons 41 ; (ii) neurons that resonate in response to a stimulus through ionic channel mechanisms 42 ; and/or (iii) network interaction of neurons, which in isolation are incapable of oscillation 43 . A thorough analysis on the origin(s) of the damped oscillation is hampered by the lack of network wiring information for the majority of amacrine cell types. To address this question, we therefore focused on the direction encoding pathway in the retina, where the connectivity has been explored extensively 39 . This pathway (Fig. 7a) consists of photoreceptors, horizontal cells, cone bipolar cells, starburst amacrine cells and direction selective RGCs.
None of these neurons are known to oscillate spontaneously. Consistent with this notion, baseline recordings without stimulus revealed stable membrane potentials (green traces in Fig. 7b-e). Therefore, mechanism-(i) cannot be responsible for the damped oscillation.
We next recorded from the same neurons with Ames' Medium containing glutamatergic, GABAa/c and glycinergic blockers. The neurons lost the ability to oscillate when they were synaptically isolated. Besides  The effect of pulse width (0.1-0.5 ms, with charge fixed at 20 nC) on response amplitude is only statistically distinguishable in some cell types (repeated measure 2-way ANOVA). (b-d) We pair-wise compared the response amplitude of every cell type against every other cell type (s mean ∆Va − s mean ∆Vb ). In each colored comparison plot the vertical and horizontal axis represents cell type 'a' and cell type 'b' , respectively. At each pulse width, some neurons were more strongly driven (2-sample t-test) than the population. Insets: the response amplitudes (s mean ∆V ) for each cell type.
the displaced starburst amacrine cell, all neurons also continued to exhibit an initial depolarizing transient in response to the extracellular stimulus. These observations ruled out mechanism-(ii) -intrinsic ionic channel properties. Importantly, these results also suggest that synapses -mechanism-(iii) -are crucial for the damped oscillations.
Cone-horizontal-cell reciprocal synapses generate damped oscillation. Knowing the putative mechanism responsible for the damped oscillation, we next searched for the locations within the direction encoding network where it first arises. Systematic recordings across the retina (Fig. 7a) indicated that the damped oscillation was already present at the horizontal cells (red traces, Fig. 7b). Horizontal cells receive excitatory inputs from the photoreceptors and provide inhibitory feedback onto the photoreceptors. Disrupting this reciprocal connectivity with glutamatergic blockers alone was sufficient to eliminate the damped oscillation (Fig. 7f), further supporting the notion that synaptic interactions underlie the oscillatory responses.
Horizontal cell syncytium modulates damped oscillation. Cones and horizontal cells are interconnected to adjacent cones and horizontal cells, respectively, with gap junctions. In particular, the horizontal cell gap junctional conductance is modulated by exogenous factors 40,44 . What roles does this conductance play in the electrically evoked responses of horizontal cells? To address this question, we constructed a computational model consisting of the cone and horizontal cell layer (Fig. 8a). Both layers were internally connected by gap junctions (Fig. 8b). The cones and horizontal cells were connected by excitatory and inhibitory reciprocal synapses. We then placed an extracellular stimulation electrode above the photoreceptor layer, over the center of the grid, and observed the activities of a row of horizontal cells.
Following an extracellular stimulation pulse, the horizontal cells responded with damped oscillation (Fig. 8c) spanning the 10-20 Hz range (Fig. 8d), similar to those of intracellular recordings (Fig. 7b). Cells in the grid center, directly beneath the stimulation electrode, had larger evoked responses than those at the periphery. Increasing the stimulus increased the horizontal cells' response amplitude, again consistent with intracellular recordings (Fig. 1e).
The oscillation amplitude of horizontal cells decreased with increasing gap junctional resistance (Fig. 8e), approaching an asymptotic limit beyond 250 MΩ. Therefore, a tightly coupled horizontal cell syncytium, through increased gap junctional conductance, gave rise to damped oscillations of larger amplitude. The junctional conductance had little effect on the spectral range of oscillation, but did increase the power of the oscillation with decreasing junctional resistance (Fig. 8f).
We also simulated a network containing only horizontal cells, interconnected with gap junctions (Fig. 8g). In this network, without reciprocal synapses, we never observed oscillatory behavior over a variety of junctional resistances (Fig. 8h), illustrating the inability of such a network to generate damped oscillations.

Discussion
In this study, we began by surveying the electrically evoked responses for neurons in the inner two layers of the retina, over a variety of stimulus configurations. Analyses of these data revealed that: (i) the response amplitude of wide-field amacrine cells and gap-junction-coupled horizontal cells were charge insensitive; (ii) some neurons were sensitive to stimulus pulse width, resulting in a change of the most strongly driven cell types with changing pulse width (at a fixed charge of 20 nC); and (iii) damped oscillations, arising from reciprocal excitatory and inhibitory synapses between the cones and horizontal cells, were widespread throughout the retina following a single stimulus pulse. Finally, computational analyses suggested that the oscillatory amplitude can be modulated by gap junctional conductance.
There are three major electrode placement locations for a retinal prosthesis: epiretinal, subretinal and suprachoroidal. The electrodes are placed adjacent to the RGC layer, adjacent to the photoreceptor layer, or in the suprachoroidal space, respectively. By virtue of its close proximity to the RGCs, epiretinal stimulation is able to directly stimulate these neurons while avoiding retinal network activation 45 . The spatiotemporally precise RGC spikes thus elicited could match, or even exceed, the spike timing precision of visually evoked responses in the same neurons 20 . It is not known whether the other placement strategies are able to replicate the neural code of natural vision with such fidelity. We found that a single subretinal pulse generated responses lasting several hundred milliseconds, across multiple layers of the retina. Oscillatory or phasic RGC spiking responses have been observed previously 35,[45][46][47] . Though pharmacological manipulations, these responses were broadly attributed to source(s) presynaptic to the RGCs. Our observations are consistent with these studies, and in particular, we also identified a network substrate responsible for these responses. Given the architectural similarities between the primate and rabbit retina, and more importantly, the simplicity of the network substrate required to create such damped oscillations, we anticipate similar responses to occur also in the human retina following subretinal and suprachoroidal stimulation. Therefore, clinical stimulation protocols should take these long-duration responses into account, when a sequence of well-timed RGC spikes are needed to encode specific visual features.
The effects of stimulus charge on the RGCs have been extensively investigated 35,45,[48][49][50] . Much like the prior RGC findings, we found that stimulus charge variations generally had the same effects on neurons in the inner retina -a positive correlation between response amplitude and charge. The exceptions were two wide-field amacrine cells and the horizontal cells. All three types are spatially extensive, either due to dendritic branching or through high-conductance, gap-junction-mediated syncytium. These anatomical features may have influenced the charge sensitivity of these neurons. In the retina, visual information is conveyed by several parallel information streams 51 , each represented by different cell types. There has been a long-standing interest in preferential activation of retinal neurons with electrical stimulation. Recent work by Twyford et al. 52 offered the possibility of preferentially recruiting RGC types using high-frequency pulse trains. Analysing a wide variety of retinal neurons, across a range of stimulation configurations, we found here that certain neuronal types in the inner retina could be preferentially activated through pulse-width variations. By increasing the repertoire of preferential stimulation techniques, it may eventually be possible to target specific cell types, across multiple retinal layers, thereby improving the clinical efficacy of retinal prosthesis through better device-to-neuron interfacing.
Based on RGC spike train analyses, it has been suggested that short-duration pulses are effective at eliciting direct RGC responses, and that long-duration pulses are more effective for evoking responses from the interneurons 35,45,49 . We observed a shift in the most strongly driven cell type as the pulse width increased, thus supporting the notion that some neurons may be preferentially targeted with specific choices of pulse width. A similar strategy has been used recently by Weitz and colleagues 53 to achieve spatially focused excitation with 25-ms pulses, which was postulated to preferentially stimulate bipolar cells.
Two factors influenced our sample size in this study. First, there is great diversity in retinal cell types. Many of these are thought to be rare 28 , limiting the probability of encounter. Second, an inherent challenge of interneuron recordings in the intact retina is difficult access. While sharp microelectrodes permitted intracellular recordings without slicing the retina or delaminating the photoreceptor layer, it tends to be less stable than whole-cell recordings, thus limiting the recording duration. Recent developments in genetically labelled retinal neurons 54 and the increasing availability of these mouse lines should dramatically improve experimental yield by permitting targeted recordings with fluorescent labels.
The retinal network undergoes progressive remodelling in neurodegenerative diseases 24 . We anticipate the results here to serve as a useful benchmark for electrically evoked responses in a healthy neural network, against which subsequent studies in diseased retinas could be compared. Persistent oscillations have been observed in the degenerate retina. The mechanisms underlying these responses have been the focus of recent studies [55][56][57][58] . Due to the presence of reciprocal synaptic connections in the degenerate retina, for instance, between the bipolar cells and the amacrine cells, the electrically evoked oscillations we observed here are likely to also occur in the degenerate retina. Furthermore, similar to previous findings in the rd1 and rd10 mice 55,57,58 , the damped oscillations are enhanced by gap junctions. In future studies, it will be important to conduct similar surveys in the diseased retina, at different degeneration stages, to understand the extent to which disease progression could impact vision restoration through electrical stimulation.

Methods
Electrophysiology. Adult New Zealand White rabbits weighing 2.0~3.0 kg, were anesthetized with ketamine (70 mg/kg) + xylazine (10 mg/kg). After enucleating both eyes, the animals were euthanized with an overdose of sodium pentobarbital. Inferior retina with the sclera attached was dissected and kept under darkness in Ames' Medium supplemented with 1% Penicillin/Streptomycin (Invitrogen). After 1~10 hours, small pieces of the retina were isolated and transferred RGC-side up into an imaging chamber. All experimental procedures were approved and monitored by the University of New South Wales (UNSW) Animal Care and Ethics Committee. All animal-related experimental procedures were performed in accordance with the guidelines and regulations defined in the UNSW Animal Research Ethics Procedure document.
The retinal tissue was perfused with Ames' Medium at 34~35 °C during electrophysiological recordings. Data were low-pass filtered at 10 kHz and digitized at 50 kHz. We recorded the responses of RGCs and displaced starburst amacrine cells with whole-cell patch clamp. The electrodes were filled with (mM): 120 KMeSO 4 , 10 KCl, 0.008 CaCl 2 , 0.5 EGTA, 1 MgCl 2 , 10 HEPES, 4 ATP-Na 2 , 0.5 GTP-Na 3 and 0.075 Alexa Fluor 488, pH 7.2. The electrode resistance ranged 3.0~5.5 MΩ with the above solution. A calculated and measured 5 mV liquid junction potential has been applied to all results. We conducted intracellular recording of neurons in the inner nuclear layer (INL) using sharp microelectrodes filled with: 1 M KMeSO 4 , 10 mM KCl, 2 mM EGTA and 4% Neurobiotin-Cl. The electrode resistance ranged from 100~150 MΩ. All chemicals, except Alexa Fluor 488 (Life Technologies) and Neurobiotin-Cl (Vector Laboratories), were supplied by Sigma Aldrich.

Morphological reconstruction and cell identification. For patch clamp recordings, where cells
were filled with Alexa Fluor 488, dye loading was achieved by passive diffusion, followed by epi-fluorescence microscopy. For sharp microelectrode recordings, Neurobiotin was loaded by passing +1 nA square pulses at 2 Hz for ≥120 s, followed by 30 minutes of incubation. The retinas were fixed with 4% paraformaldehyde (Sigma Aldrich) in phosphate buffer, incubated in PBS with 0.5% Triton-X and 1% BSA (Sigma Aldrich), reacted against Streptavidin -Alexa Fluor 488 conjugate (Life Technologies), counter stained with DAPI (Life Technologies), then mounted with Pro-long Gold (Life Technologies). The retinas were imaged with a confocal microscope (Olympus FV1000). Amacrine cells were first broadly categorized by dendritic field diameter 28 , namely: narrow field (<125 µm), medium field (125~400 µm) and wide field (>400 µm). Further subdivisions were made on the basis of dendritic stratification, branching patterns, presence of varicosities and dye coupling. When matching our amacrine cells to previously reported types, we used the naming schemes of MacNeil and colleagues 28,29 . The exceptional cells were: MF1, which was called A18 by Kolb 30 and a TH-IR cell by Casini et al. 31 ; MF2, which was called A13 by Kolb and Nelson 32 ; and WF2, which was called A17 by Nelson and Kolb 33 and an indoleamine S-2 cell by Vaney 34 . The bipolar cells were classified into the On or the Off subtypes, based on light responses, then confirmed morphologically by dendritic stratification. We identified the A-type horizontal cells by their light-induced hyperpolarization, morphology and extensive gap-junction coupling. The On-off direction selective RGCs (DSGCs) were identified by their responses to both light onset and offset.
Electrical stimulation. The wholemount retinas were stimulated at the photoreceptor side (subretinal stimulation) with a 40-µm diameter platinum electrode in a multielectrode array (MEA; Qwane Biosciences) embedded in the base of the imaging chamber. Sixty evenly spaced electrodes were arranged in a square grid within the MEA. The electrode center-to-center distance was 200 µm. The cells were stimulated with a monopolar return configuration. Eight distant and evenly spaced electrodes connected in parallel at the MEA perimeter served as the stimulus return. These distant return electrodes were always at least 400 µm away from the stimulation working electrode. The electrical stimuli were constant-current, charge-balanced, cathodic-first, rectangular biphasic pulses without inter-phase delay. We repeated all stimulus configurations at least 10 times, with 1 s between presentations.
Light stimulation. Recordings were made at constant mesopic lighting (~7 cd.sr/m 2 ). To study the light evoked responses, we projected a 160 µm diameter circular light spot onto the photoreceptor layer and centered over the soma of the recorded neuron. The light stimulus was white and 4 s in duration, consisting of the sequence On-Off-On-Off, with 1 s for each segment.
Analyses. All results are in the mean ± SEM format. Statistical tests with p < 0.05 (2-tailed) were considered significant. In the stimulus-time plots (e.g. Fig. 1e), the changes in membrane potential (∆V m ), averaged over 10 trials, are denoted by color variations. To show the responses in conventional format, we also overlaid in grey a representative single trial response at the highest stimulus amplitude.
To evaluate the effect of stimulus pulse width, we first expressed the electric pulse as the total charge delivered (pulse amplitude × pulse width). Comparisons were then made between pulse widths at identical charge levels, thereby eliminating the effect of pulse amplitude.
The retina contains both spiking and graded responses. Both types were analyzed by the same method. To characterize the perturbation caused by the stimulus independently of the time evolution of the response, we used the following conceptual treatment. The trials were combined to form an average time-waveform. The time-coordinate of the waveform was collapsed by projecting all the ∆V m (t) points onto the common vertical (∆V m ) coordinate axis. The assemblage of V m values was treated as samples drawn from a hypothetical probability density function to yield its standard deviation (s mean ∆Vm ). The latter is the measure of perturbation. In practice, we repeated the stimulus trial 10 times, to compute the average time-waveform. The pre-stimulus data and the stimulus artifacts were excluded from analysis.
We used multi-tapered spectral analysis 59 to characterize the frequency domain structure of the evoked responses. The spectra were computed from a 390 ms window using three Slepian data tapers at 10 Hz.
Computational model. The computational network model was implemented in NEURON 60 , using strategies similar to our previous work 61 . It had a three-dimensional topology containing multi-compartmental neurons with conductance-based description of ion channel dynamics. The network contained cone photoreceptors and horizontal cells. We used the cone photoreceptor model developed in Publio et al. 62 . The horizontal cell was based on the model of Aoyama and colleagues 63 , modified into three-dimensional form, to contain dendrites. It reproduced all behaviors of the original model. The network contained 13 × 13 cones and 13 × 13 horizontal cells in two planar grids separated by 15 µm. The dendritic fields of neighboring horizontal cells were spatially overlapping to mimic experimental observations. Every cone and horizontal cell was connected to four neighbors by gap junctions. In practice, a many-to-many connectivity exists for the cones and horizontal cells. For computational efficiency, we used a one-to-one relationship. Each cone and horizontal cell pair was connected by an excitatory feed-forward synapse and an inhibitory feedback synapse.
Gap junctions were modeled as resistors connecting two compartments. Chemical synapses followed the general scheme of Destexhe et al. 64 , with a stationary relationship between the presynaptic membrane voltage and the post-synaptic membrane current: where V pre is the presynaptic voltage and I syn is the postsynaptic current. The parameters g syn , v λ and v slope determine the maximum synaptic conductance, half-activation voltage and activation gradient, respectively. When the cones and horizontal cells were connected, their membrane potentials converged to an equilibrium of approximately −49 and −50 mV, respectively. These values were close to the experimental observations. The network was stimulated above its center position with an extracellular current pulse in a homogeneous medium. The extracellular potential at each point v(x, y, z) was determined by a three-dimensional form of the equation in McIntyre and Grill 65 : where I stim and ρ ext are the extracellular current amplitude and extracellular resistivity, respectively. The stimulus induced trans-membrane current could then be computed for each segment per McIntyre and Grill 65 . Supplementary Table 1 contains further details of the computational model. The electrically evoked responses in the model were comparable to the experimental results.
Data availability. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.