Morphological Factors that Underlie Neural Sensitivity to Stimulation in the Retina

Abstract

Retinal prostheses are a promising therapeutic intervention for patients afflicted by outer retinal degenerative diseases like retinitis pigmentosa and age-related macular degeneration. While significant advances in the development of retinal implants have been made, the quality of vision elicited by these devices remains largely sub-optimal. The variability in the responses produced by retinal devices is most likely due to the differences between the natural cell type-specific signaling that occur in the healthy retina vs. the non-specific activation of multiple cell types arising from artificial stimulation. In order to replicate these natural signaling patterns, stimulation strategies must be capable of preferentially activating specific RGC types. To design more selective stimulation strategies, a better understanding of the morphological factors that underlie the sensitivity to prosthetic stimulation must be developed. This review will focus on the role that different anatomical components play in driving the direct activation of RGCs by extracellular stimulation. Briefly, it will (1) characterize the variability in morphological properties of α-RGCs, (2) detail the influence of morphology on the direct activation of RGCs by electric stimulation, and (3) describe some of the potential biophysical mechanisms that could explain differences in activation thresholds and electrically evoked responses between RGC types.

Graphical Abstract

To design improved stimulation strategies that achieve selective activation of specific RGC types, a better understanding of the morphological and biophysical factors that underlie the sensitivity and response of RGCs to prosthetic stimulation needs to be developed. This review focuses on studies that have investigated the role that somatodendritic morphology and the axon initial segment play in driving the direct activation of RGCs by extracellular stimulation.

1. Introduction

Retinitis pigmentosa (RP) and age-related macular degeneration (AMD) are among the leading causes of retinal-degeneration associated blindness, with about 1 in 4,000 people diagnosed with RP in the U.S. every year ^[1] and 11 million people afflicted with some form of AMD ^[2]. Both diseases involve the gradual deterioration of photoreceptors, the outer retinal neurons that are responsible for initiating the biochemical cascade that converts light into the downstream electrical signals that are interpreted by the brain (Figure 1a). This results in vision loss that can worsen over the course of degeneration. Existing treatment strategies for the disease are limited, and few patients see improvement in their condition over time in response to treatments ^[3,4]. The use of retinal prostheses as a therapeutic intervention for individuals afflicted by these outer retinal degenerative diseases is currently an active area of research by many groups (reviewed in ^[5]). Devices are usually implanted directly above the retinal surface and in contact with the retinal ganglion cell (RGC) layer (epiretinal) or below the RGC layer in the space previously occupied by the photoreceptors and the remaining retinal neurons (subretinal) (Figure 1b). Subretinal devices function by stimulating the bipolar cells which are presynaptic to the RGC layer (referred to as indirect activation), whereas epiretinal devices function by stimulating RGCs directly. Retinal implants typically consist of a multi-electrode array (MEA) in which each electrode is intended to a create localized region of neural activation that is perceived as a light percept. In this review, we will focus on studies relating to the direct activation of RGCs. We will pay particular attention to the morphological properties of RGCs that play an important role in shaping the sensitivity and overall response to electric stimulation.

Figure 1. Fundamentals of retinal anatomy and implants.

a) Schematic illustrating the basic anatomy of the retina. Light enters the eye through the cornea and is initially processed by photoreceptors (PRs) which transduce light into electrical signals that are fed into bipolar cells (BCs). BCs generate electric signals which are transmitted to retinal ganglion cells (RGCs). The axons of RGCs generate action potentials and form an input at the optic nerve to downstream visual circuits. Blue arrow indicates input stream (light), and red arrow indicates output stream (neuronal signals). b) Retinal implants can be distinguished by their location. Epiretinal devices (e.g., Argus II, Second Sight Medical Products) are situated on the surface of the RGC layer and directly stimulate the RGCs whereas subretinal devices (e.g., Alpha AMS, Retina Implant AG; PRIMA System, Pixium Vision) are placed in the space previously occupied by photoreceptors and indirectly activate RGCs by stimulation of the BCs. Blue and red arrows indicate input (electric stimulation) and output streams (neuronal signals), respectively.

Despite the significant progress in prosthetic design/development, the quality of vision elicited by these implants has been limited (reviewed in ^[6]). The limited quality of the responses produced by retinal devices is most likely due to the differences between the natural, synchronized patterns of cell type-specific signaling that occur in the healthy retina vs. the non-specific, simultaneous activation of multiple cell types that occurs with a MEA. This has led to a variability in visual percepts reported across different patients ^[7–9]. While the development of stimulation strategies that better replicate the patterns of neural signaling that arise naturally in the retina is highly desirable, it has proven difficult to preferentially target specific RGC types while avoiding the activation of others, e.g., activation of ON cells without simultaneous activation of OFF cells (but see ^[10–13] for novel approaches for preferential targeting). The sensitivity and response characteristics of individual RGCs to extracellular stimulation has been found to be dependent upon intrinsic morphological differences that exist within and across cell types ^[14–16]. Unfortunately, the extent of these differences is only partly understood. In addition, a much better understanding of the relationship between specific morphological features and the sensitivity to electric stimulation will need to be developed before stimulation strategies can be customized in a way that allows specific types to be targeted. The goal of this review is to (1) summarize the variability in morphological features that exists across RGC types in both the mouse as well as the primate/human retina, (2) analyze the studies which have investigated the influence of morphology on the direct activation of RGCs by prosthetic stimulation, and (3) describe what is known about the biophysical mechanisms that underlie differences in activation thresholds and electrically evoked responses between RGC types.

In the first section, we present a summary of various aspects of neuronal morphology within RGCs that have been shown to influence their response to epiretinal electric stimulation. In addition to the variations in somatodendritic morphology across RGCs, we will pay specific attention to the properties of the axon initial segment (AIS). The AIS is a specialized region of the proximal axon of central nervous system (CNS) neurons which is responsible for spike initiation ^[17,18], and which has been found to be the site of greatest sensitivity for extracellular electric stimulation in RGCs ^[14,19,20]. Following a characterization of the morphological variability present in RGCs within the healthy retina, we will present a brief overview of studies that have investigated the impact of retinal degeneration on the anatomical structure of RGCs. In the second section of the review, we will summarize the results of different physiological studies that have investigated the direct activation of RGCs via extracellular electric stimulation. In particular, we will focus on (1) how somatodendritic morphology and the properties of the AIS may underlie activation thresholds, (2) physiological differences between the healthy and degenerate retina, and (3) sophisticated stimulation strategies to achieve preferential activation of specific RGC types.

In the third section of the review, we will examine computational studies that aim to better understand the biophysical mechanisms underlying the response of RGCs to extracellular electric stimulation. Briefly, we will review how RGC spiking is affected by the morphology of somatodendritic compartments as well as the role of different AIS features in modulating stimulation thresholds for different RGC types. Developing a better understanding of the biological mechanisms underlying the activation and responses of various ganglion cell types will help facilitate the translational development of preferential stimulation paradigms to be applied in next generation devices.

2. Review of somatodendritic morphology and the properties of the axon initial segment across major RGC types

RGCs extract different features from the visual world and use unique patterns of action potentials to convey information to higher visual centers. Individual RGCs are classified as ON or OFF based upon their characteristic spiking responses to either the onset (ON cells) or offset (OFF cells) of a light stimulus. In addition, the response patterns of ON and OFF cells are further classified as being either sustained or transient. This results in 4 major types of RGCs which are characterized based upon their responses to light stimuli: ON Sustained, OFF Sustained, ON Transient, and OFF Transient. Furthermore, ON and OFF RGC types can be anatomically identified based on the stratification depth of their terminal dendrites within the inner plexiform layer of the retina; ON cells stratify in the inner portion of the inner plexiform layer (IPL) (closest to the RGC layer) whereas OFF cells stratify in the outer portion (Figure 2a, bottom) ^[21]. In addition to variability in their physiological responses, there is also a considerable degree of morphological variability within an individual RGC type, as well as across multiple cell types ^[22,23] (Figure 2a, top). These differences include somatodendritic properties, such as the size of the soma and dendritic field (Figure 2c), dendritic complexity, dendritic length, etc. More recently, differences in axonal properties such as the length and location of the axon initial segment, a specialized region in the proximal axon of CNS neurons that is the site of spike initiation, have also been identified within and across RGC types ^[20,24,25] (Figure 2b,c). Much evidence suggests that both the somatodendritic and axonal properties of a given cell can shape its responses to electric stimulation, and therefore, it is important to develop a detailed understanding of the anatomical variability present within and across types as this may lead to stimulation strategies that enable specific types to be selectively activated. As the natural patterns of retinal signaling do not involve the simultaneous activation of multiple cell types, selective activation of specific RGC types is an important requirement for a retinal prosthesis that aims to restore functional vison. In this section, we will examine the anatomical variability present in mouse alpha RGCs (α-RGC), focusing specifically on their somatodendritic morphology and AIS properties. α-RGCs represent an ideal population for further study as they have homologs in most mammalian species including the primate and human retina. Next, we will briefly compare the findings in mouse retina to morphological studies performed in the primate and human retinas. Finally, we will look at studies which have examined the impact of retinal degeneration on the morphology of RGCs.

Figure 2. Comparison of anatomical properties between RGCs.

a) (top) Flat mount views of traced dendritic fields in mouse RGCs display variability in size, density, and underlying structure. Dendritic field diameter of an example cell is estimated by calculating the diameter of the circle surrounding the edges of an individual cell’s peripheral dendrites (see methods in ^[24]). (bottom) RGCs are designated into ON and OFF sub-types based on the stratification of their terminal dendrites relative to the ON and OFF choline acetlytransferase (ChAT) bands, respectively. Reproduced with permission.^[23] 2015, Annual Reviews (US). b) In addition to the dendritic field diameter, the morphology of individual RGC can be quantified by measuring the properties of both the soma (soma diameter) and axon initial segment (AIS length (L) and distance from the soma (D)). Soma diameter is estimated by calculating the diameter of the circle surrounding the cell body dendrites (see methods in ^[24]). The AIS of an individual RGC can be identified through the colocalization of the cytoskeletal marker Ankyrin_G (blue), and proximal axon (green). Na_v1.6 sodium channels (red) make up much of the relative sodium channel composition of the AIS and can be labeled using anti-Na_v1.6/SCN8 antibodies. Reproduced with permission.^[24] 2019, Frontiers Media SA. c) Comparison of soma diameter, dendritic field diameter, distance from the soma to the AIS, AIS length, and Na_v1.6/AIS ratio between the 4 mouse α-RGC types. Reproduced with permission.^[20] 2020, IOP Publishing.

2.1. Morphological differences across α-RGCs

We will focus our review of RGC morphology primarily on the α-RGC type for several reasons. Firstly, as one of the most well-characterized and studied RGC types of the mouse retina ^[21,24–28], α-RGCs provide a wealth of information from which to compare morphological features. Second, α-RGCs are a major subset of the output neurons in the mouse retina which project to both the superior colliculus and core of the visual thalamus, and form a major input into the downstream visual processing circuits ^[29]. As a result, the ability to preferentially activate α-RGCs would provide important insight as to how to activate an RGC type that forms a major input into the visual cortex and plays an important role in regulating functional vison. Third, as α-RGCs are thought to be homologs of the parasol ganglion cells found in the primate and human retinas ^[30,31], developing stimulation strategies to be able selectively target them may provide insight as to how to target the parasol ganglion cells in human subjects. Fourth, α-RGCs display a systematic gradient in size both as a function of cell type and retinal eccentricity. This provides a natural platform from which to study the relative influence of different components of cellular anatomy, such as soma and dendritic field area, on the activation of RGCs by prosthetic stimulation.

Prior studies in the mouse retina have found that ON and OFF α-sustained (α-S) RGCs display a gradient in both soma and dendritic field size along the nasal-temporal axis of the visual field ^[27], with ON α-S RGCs being slightly larger than OFF α-S RGCs at any given retinal eccentricity ^[24]. OFF α-transient (α-T) RGCs display a gradient in only dendritic field size along the dorsal-ventral axis ^[32]. In terms of soma size, ON α-S RGCs had the largest soma diameter ^[20,28,32] while ON α-transient (α-T) RGCs had the smallest soma diameter as compared to the other α-cells (Figure 2c). In terms of dendritic field size, ON α-T RGCs had the smallest dendritic field diameter as compared to the other α-RGCs. All four sub-types displayed a strong correlation between their soma and dendritic field diameters ^[20], similar to the trend observed within midget and parasol cells of the primate and human retinas ^[30,33]. In addition to the α-RGCs, melanopsin 1 (M1), and melanopsin 4 (M4) sub-types of intrinsically photosensitive RGCs (ipRGCs) ^[34,35] also display a gradient in soma as well as dendritic field size. In summary, there exists a large degree of somatodendritic variability both within a specific RGC type and across multiple types. While some cell types of exhibit gradients in both soma and dendritic field size, others exhibit gradients in either one or neither of these parameters as well. These gradients in cell size have also been found to be a function of retinal location.

2.2. AIS properties vary systematically across α-RGCs

The AIS is a specialized region of the proximal axon of CNS neurons that is comprised of a dense band of ion channels and cytoskeletal proteins and is the region where spikes are initiated. The length, location, and composition of the AIS have been found to be specifically customized in various CNS populations to support varying degrees of functional output ^[36,37]. For example, neurons of the chick nucleus laminaris that are sensitive to low auditory frequencies have AISs that are shorter and further from the soma than the AISs of high-frequency neurons; differences in the length and location are thought to help maximize sensitivity to the interaural time differences associated with each frequency range ^[36]. Similarly, variability in the AIS properties of different RGC types has also been found. For example, brisk-transient (BT) cells of the rabbit retina (thought to be homologs of the α-RGCs in the mouse) were found to have longer AISs when compared to the directionally selective (DS) population (~40μm vs. 26μm); the distance of the AIS to the soma, however, did not vary between the populations ^[14].

Recent studies in α-RGCs of the mouse retina have also found significant variability in their AIS properties. Within the α-S RGCs, Raghuram et al. ^[24] found that the AISs of ON α-S cells were longer and located further from the soma (AIS distance) than OFF α-S cells; both the AIS length and its distance from the soma were significantly correlated to the size of the soma as well as the diameter of the dendritic field. The scaling of the AIS properties with cell size was found to help maintain physiological response consistency across the entire population. In OFF α-T RGCs, the AISs of cells located in the dorsal region were longer and located further from the soma than cells in the ventral region. Differences in the AIS properties between dorsal and ventral cells were found to correlate to differences in their intrinsic physiological properties ^[25]. For example, the larger AIS lengths and distances from the soma in OFF α-T RGCs located in the dorsal retina were able to support stronger mean firing rates as compared to cells in the ventral retina with shorter AISs located closer to the soma. Like the α-S RGC population, the length and location of the AIS in these cells were also correlated to dendritic field size. Overall, AIS length was found to be longest in ON α-S cells, while AIS distance was largest in OFF α-T cells (Figure 2c). In addition to AIS length and location, the percentage composition of Na_v1.6 sodium channels within the AIS has also been found to have important functional implications for the generation of high-frequency action potentials ^[38]. Recent studies have found a difference in the sodium channel 1.6 (Na_v1.6) composition (%) between the α-S and α-T RGC types; ~90% in ON/OFF α-S RGCs vs. 80% in OFF α-T RGCs ^[20,24,25] (Figure 2c). In terms of the relationship between AIS composition and somatodendritic properties, however, no correlation between cell size and Na_v1.6 composition was observed within ON and OFF α-S RGCs ^[24].

Prior studies in the rabbit retina have found that the AIS is site of lowest threshold for electric stimulation due to the high density of sodium channels, and that variability in extracellular stimulation thresholds across rabbit cell types could be attributed to differences in AIS properties (Figure 3d) ^[14]. As the properties of the AIS have been found to vary systematically across populations and retinal location, future efforts studying the response of RGCs to electric stimulation should consider not only consider cell type but also the relative location of the RGC in the mouse visual field.

Figure 3. The site of lowest threshold for extracellular stimulation is colocalized with the AIS and varies between different cell types.

a) Schematic of experimental technique used to map electric stimulation thresholds around the anatomy of a single rabbit RGC. Using patch-clamp electrophysiology, an individual cell (white outline indicates soma) was classified into a known RGC type through its response to light stimuli. Once classified, a series of stimulus pulses, ranging from low to high amplitudes, was delivered using a microelectrode (black) positioned 25μm above the RGC cell body. Threshold was defined as the minimum current amplitude that elicited a 67% response rate (i.e., 2/3 pulses produced a response). b) The location of the stimulating electrode was then systematically stepped along the region surrounding the cell body and proximal axon, and threshold measured at each location; each colored square corresponds to a separate threshold measurement (color bar at top: blue and red are low and high thresholds, respectively). c) The overlay of the threshold map with the traced cell reveals that the region of lowest threshold occurs over the proximal axon. Subsequent immunochemical analysis (not shown here but see Fig. 2b) showed the region of lowest threshold corresponded to the dense band of voltage-gated sodium channels in the axon initial segment. d) Comparison of stimulus thresholds between the brisk-transient (BT), directionally selective (DS), and local edge detector (LED) populations. Reproduced with permission.^[14] 2009, American Physiological Society (US).

2.3. Morphological properties of primate and human RGCs

Similar to the observations in the mouse retina ^[24,25,32], work within the major RGC types of the primate/human retina (i.e., the midget and parasol ganglion cells) has found (1) a linear correlation between soma and dendritic field areas, (2) a gradient in cell size as a function of retinal eccentricity, and (3) that ON cells are larger on average than OFF at any given location ^[30,33,39]. Scaling of the somatodendritic compartments as function of retinal eccentricity, however, is not ubiquitous amongst all human RGC types, as melanopsin + RGCs display a gradient in dendritic field size but not in soma diameter between the central and peripheral retina ^[40]. Nevertheless, melanopsin + RGCs exhibit significant variability in their anatomical properties as a function of sub-type ^[41].

It remains an open question whether the properties of the AIS are also functionally customized within each cell type of the primate or human retina and whether they vary as a function of retinal eccentricity. A recent study in the primate retina showed that differences in the presynaptic network of midget ganglion cells in the foveal, central, and peripheral network contributed to differences in their intrinsic spiking response patterns ^[42]. The analogy to variations of α RGCs across the mouse retina raises the question of whether the properties of the AIS may also be similarly tailored in each retinal location to support differences in neuronal output. Immunochemical reconstructions of the soma and axonal trajectories from primate parasol RGCs from MEA recordings suggested that the spike initiation zone colocalized with AIS staining and may be different in ON vs. OFF cells ^[43]. The variability in the AIS properties of ON/OFF midget and parasol ganglion cells could also explain the differences observed in their extracellular stimulation thresholds ^[16,44].

2.4. Changes in RGC morphology within rodent models of retinal degeneration

Given that morphological properties of neurons contribute to their sensitivity to prosthetic stimulation, it is important to understand whether the changes occur over the course of disease. Prior studies using mouse models of retinal degeneration have found that alterations in the morphology of α-RGCs over the course of photoreceptor-mediated degeneration, i.e., in diseases such as retinitis pigmentosa, are strongly linked to the time course of the degenerative process ^[45,46]. Two of the most widely studied mouse models of retinal degeneration are the C57BL/6J-Pde6b^rd10/J (referred to as rd10) and C57BL/6J-Pde6b^rd1/J (referred to as rd1) ^[45,47–49]. The major difference between the models is the later onset and milder degree of degeneration in the rd10 mouse (more reminiscent of the time course of degeneration in humans with retinitis pigmentosa) as compared to the rd1which has a more aggressive rate of degeneration that starts during the mouse’s critical period of development. A survey of non-specific RGCs within the rd10 mutant mouse line found that they largely retained their dendritic morphology well into the final stages of degeneration ^[45]. In addition, the number of RGCs in the ganglion cell layer also did not change with degeneration. This study, however, did not report on measurements of soma diameter, AIS properties, and relative location of these RGCs within the mouse visual field. In contrast to the rd10 model, work in the rd1 mouse found a reduction in the total number of RGCs in the ganglion cell layer of rd1 mice vs. wild-type (wt) mice at 3 months ^[50] as well as when comparing mice aged 4.5 months to mice at 12 months ^[45]. Similarly, work in S334ter-line-3 rats (a well-studied rat model of retinal degeneration analogous to rd1 mice) found a reduced number of SMI-32 positive RGCs (an immunochemical marker for α-cells) in p500 and p800 animals at the temporal, dorsal, and ventral quadrants ^[51]. Looking further at whether the degree of degeneration is related to retinal location, a recent study in the retina of Royal College of Surgeons (another well-studied rat model of retinal degeneration) rats found that photoreceptor degeneration was significantly greater in the ventral vs. dorsal retinal locations ^[52]. Since prior studies have found a systematic difference in the morphology and physiology of α-RGCs along the nasal-temporal ^[24,27] and dorsal-ventral axis ^[25,32] of the rodent retina, it would be interesting in future studies to explore whether changes to specific cell types during degeneration are location specific. Mazzoni et al. ^[45] also reported a significant decrease in the dendritic field area of B3 outer, B3 inner, C2 outer, and C2 inner RGCs (classified as per ^[53]) at both the 3 month and 8 month time points of degeneration. The soma diameters of C2 inner, C2 outer and Melanopsin + cells were also found to be significantly reduced as compared to the wt population. In addition to anatomical changes to the somatodendritic compartments, work by Saha et al. ^[50] found cell type-dependent changes in the density of excitatory and inhibitory puncta on the dendrites of α-RGCs within the rd1 mouse. ON α-S RGCs were found to have a reduction in the density of excitatory (RIBEYE) puncta in their dendrites within 3 months of degeneration but no change in inhibitory (Gephyrin) puncta at the same time-point. OFF α-S RGCs displayed no significant changes in excitatory puncta but did display a significant reduction in inhibitory puncta. These results suggest a differential influence of retinal degeneration on the morphology ON vs. OFF α-S RGCs. Subsequent studies are needed to better understand how these degeneration-induced morphological changes vary across other RGC types.

Studies that have investigated the properties of the AIS in RGCs of the degenerate retina have reported conflicting results. Damiani et al. ^[46] found no change in AIS length at the 1 month and 1 year time points, but a slight reduction in distance between the AIS and the soma at the 1 year time point in rd1 mice when compared to the wild-type mouse. Work in the heterozygous S334ter-line-3 rats found no change in the distance between the AIS and the soma but did observe a significant reduction in AIS length when compared to Long Evans rats (wild-type) ^[54]. In contrast, a more recent study ^[55] found that AIS length increased as a result of light deprivation during the critical period, but did not report on changes in the distance between the AIS and the soma. As the AIS properties of RGCs have been found to vary as function of both cell type and retinal eccentricity ^[24,25], it remains unclear as to (1) how degeneration influences the AIS properties across different RGC types, and (2) whether these changes are location specific. Both these parameters should be considered when studying the effects of retinal degeneration on extracellular stimulation thresholds.

3. Physiological studies exploring the direct activation of RGCs by prosthetic stimulation

Electric stimulation can activate RGCs through both direct and indirect mechanisms (reviewed in ^[56]). Direct activation refers to the direct depolarization of the ganglion cell membrane by the electric stimulus. Depolarization leads to the opening of sodium channels ^[57,58] that, if strong enough, triggers an action potential (AP). Indirect activation refers to the activation of excitatory neurons presynaptic to RGCs (i.e., bipolar cells and/ or photoreceptors) probably by depolarizing their axon terminals ^[56]; the excitatory synaptic input arriving at the RGC induces the membrane depolarization that triggers spiking. Prior studies in the rabbit retina have found that the response to direct activation of RGCs typically consists of a single action potential which is phase-locked to the stimulus pulse with a latency less than 1 millisecond. In contrast, indirect activation of the network typically results in a train of APs that are clustered into bursts, some of which have latencies of tens or even hundreds of milliseconds, depending in part on the properties of the stimulus ^[59,60]. All long-latency responses are eliminated in the presence of synaptic blockers ^[57,58,61]. Furthermore, the thresholds for the direct activation of individual RGCs were found to be significantly lower than the thresholds needed for indirect activation when using short duration pulses (<0.1ms) whereas no significant difference in direct vs. indirect activation thresholds was observed when using long duration pulses (10–20ms), effectively indicating that short duration pulses allow for preferential targeting of RGCs without simultaneously activating the rest of the retinal network ^[57,61,62]. Because activation of the network typically results in long lasting inhibitory signals, likely the result of either direct or indirect activation of amacrine cells, that can reduce sensitivity to subsequent stimuli, direct activation allows for the creation of much higher response rates.

In this section, we will focus on studies related to the direct activation of RGCs. Much recent work has begun to unravel the intrinsic morphological and physiological properties of RGCs that underlie sensitivity to electric stimulation ^[14,16,20] and we will summarize the work here.

Physiological measurements revealed that threshold for direct activation varied for different regions of individual rabbit ^[63] and primate RGCs ^[64], with activation thresholds that were lower when the electrode was slightly offset from the soma. Systematic mapping of threshold around single RGCs revealed that the region of lowest threshold for electric stimulation was indeed along the proximal axon. Immunochemical staining of mapped cells revealed that the region of highest sensitivity corresponded to a region of high density sodium channels in the AIS ^[14] (Figure 3a–d). The size of this region, as well as the distance between this region and the soma, varied both within and across different cell types. The size of this region was found to be largest amongst the rabbit BT cells, while distance from this region to the soma was found to be similar between BT cells and directionally selective (DS) cells.

3.1. Activation thresholds in primate and human RGCs

Similar to Fried et al. ^[14] who found that that RGC activation thresholds can vary across cell type, work in the primate and human retina found also variability in thresholds between ON/OFF midget and parasol cells ^[16,44]. While no morphological correlate to activation threshold was investigated in the study by Jepson et al. ^[16], variability in both soma and dendritic field size ^[30] as well the AIS properties ^[43] across the major population of primate RGCs have been readily identified, and are likely to contribute to the observed differences in threshold.

3.2. Somatodendritic morphology influences the RGC response to electric stimulation

Many anatomical features of RGCs vary between cell types and even within a single type, there can be significant variation. For example, the size of both the soma and the dendritic field vary systematically with eccentricity in midget and parasol ganglion cells of the primate retina ^[30,33]. Analogously, the size of the soma and dendritic field vary systematically along the nasal-temporal axis in ON and OFF α-S RGCs in the mouse retina ^[24,27]. Recent efforts have begun to explore the role of somatodendritic morphology in shaping the direct response of RGCs to electric stimulation. Cho et al. ^[65,66] in the mouse retina found that the extracellular stimulus threshold for single-cell activation across an unspecified group of RGCs was inversely correlated to soma diameter (Figure 4a). These results could potentially be explained by the fact that larger soma size is directly correlated with longer AIS length in some RGC types ^[20,24] which, in turn, has been found to underlie lower stimulation thresholds ^[14]. Within A-2 type RGCs of the rat retina (α-S RGCs in the mouse), Hadjinicoloau et al. ^[67] examined the influence of both soma and dendritic field diameter on a cell’s ability to follow a train of electric stimulus pulses. When comparing the efficacy by which cells could follow a train of 30 pulses, they found that cells with smaller soma and smaller dendritic field diameters were able to keep up with stimuli at all frequencies (5–200 pulses/sec) over the first 10 pulses. In contrast, cells with larger soma and dendritic field diameter displayed decreasing stimulus response efficacy as a function of frequency (Figure 4b). While their work suggests that small A-2 RGCs are able to better sustain their response to stimuli than larger cells at higher stimulus frequencies, this is contradictory to results which have found that larger cells have longer AISs ^[24,25] which are more capable of generating a higher frequency of APs ^[68]. This discrepancy may be due to their use of larger stimulating electrodes (200 × 200 μm²) which have been found to be less sensitive to changes in AIS properties ^[20].

Figure 4. Somatodendritic morphology is related to RGC physiology.

a) Each point is a plot of the activation threshold (during extracellular stimulation) versus soma diameter for an unspecified mouse RGC. Reproduced with permission.^[65] 2011, IEEE. b) (left) Plot correlating the efficacy by which small (grey) and large (black) soma diameter cells can follow a train of electric pulses at a range of stimulus frequencies. (right) Similar to (left) but cells with a small and large dendritic field diameter. Reproduced with permission.^[67] 2015, IEEE.

3.3. Physiological studies in the degenerate retina

Several anatomical changes to RGCs have been identified in mouse models of retinal degeneration. These changes have been linked to the time course of the degenerative process and include alterations in soma size, dendritic field size, dendritic puncta, as well as AIS properties ^{[45,46,50,54]}. In addition to changes in ganglion cell morphology, physiological studies in both the rd1/rd10 retina have reported a number of differences in underlying cell function as well as their sensitivity to electric stimulation. Elevated levels of spontaneous activity have been reported in both ON and OFF RGC populations as early as eye opening (P7) and increasing progressively to P120 where they plateau ^[47,49,69]. Concurrent with increasing spiking, there is a steady increase in oscillatory activity, e.g., spike bursts occurring at 10Hz ^[69]. Responses to light stimuli are also altered during this time period although the specifics may vary somewhat by both species as well as the specific model of degeneration. In rd1 mice, the spiking rates of ON cells to light stimuli were reduced and response latency increased, while the responses of OFF cells were largely conserved ^[47]. In contrast, both ON and OFF cells in the rd10 model showed similar decreases in light-evoked responses ^[49]. Despite a decrease in light evoked activity, the presence of all major RGC types (ON/OFF, sustained and transient) were identified in the adult rd10 mouse ^[49]. Additionally, work by Sekirnjak et al. ^[70] with the P23H degenerate (P23H) rat found that the size of the visual receptive fields decreased in both ON and OFF RGCs as function of age. This physiological finding is consistent with subsequent studies that found decreases in dendritic field size in rd1 mice ^[46,71]. Sekirnjak et al. ^[70] found that the application of synaptic blockers reduced the spontaneous firing of OFF cells by only ~50%, suggesting that the increased rate of spontaneous spiking in P23H rat RGCs does not arise solely from increased excitatory input but instead that changes intrinsic to the RGC may be contributing as well ^[70]. It is not yet clear what underlies the intrinsic changes in firing rate but given the importance of the AIS in spike initiation, it seems a likely candidate. Consistent with this, work outside the retina in nigral dopaminergic neurons ^[72] has found that the length of the AIS is strongly correlated to spontaneous firing rate, raising the question of whether changes in AIS length could play a similar role in modulating the spontaneous firing activating of degenerate RGCs. Prior studies on the AIS in the degenerate retina, have reported conflicting results. Damiani et al. ^[46] found a slight reduction in distance between the AIS and the soma in rd1 mice, while studies in the S334ter-line-3 rats observed a significant reduction in AIS length ^[54]. Given that the properties of the AIS can underlie the sensitivity to prosthetic stimulation, understanding how AIS properties change in the degenerate retina will have important implications for how RGC activation thresholds can vary during degeneration.

Several studies have identified changes to the somatodendritic morphology and intrinsic spiking properties of RGCs within models of retinal degeneration that resemble human disease pathologies ^{[45,50,70,73]}. In addition to changes in the anatomy and physiology of RGCs, several studies have reported on a number of differences in the sensitivity of RGCs to electric stimulation. While initial studies by Jensen et al. ^[74] found that the median thresholds for activation of RGCs were significantly greater in the rd1 retina, they utilized relatively long pulse durations (1ms), i.e., capable of activating RGCs both directly as well as through the network. As a result, it is not clear whether the differences they observed reflect intrinsic changes in the RGC or differences within the network. Larger thresholds to achieve direct activation were observed among the SMI-32 positive cells of the degenerate rat retina (homologs to mouse α-RGCs) and were positively correlated to the age of the animal/degree of degeneration ^[51]. In contrast, Sekirnjak et al. ^[70,75] found the activation thresholds for electric stimulation of RGCs in the degenerate (P23H) rat retina did not change. Cho et al. ^[66] also found an increase in activation thresholds for direct activation of RGCs in the rd10 mouse. Interestingly, however, they found no correlation between thresholds and soma diameter as were observed in wild-type animals ^[65]. These results suggest the possibility that morphological factors other than cell size e.g., AIS properties, could be contributing to the observed changes in activation thresholds.

3.4. High-frequency stimulation leads to differential responses across RGC types

The stimulation of RGCs using high-frequency pulse-trains has been found to elicit differential spiking responses across multiple RGC types and, as a result, been the subject of active investigation to preferentially target individual cell types ^[76–78]. Preliminary studies found that the elicited spiking response of RGCs increased monotonically with stimulus amplitude and could be fit with a sigmoidal function ^[79]. Cai et al. ^[76] questioned how the shape of an individual cell’s amplitude-response curve varied as a function of stimulation frequency (100–700 pulses/sec) and found that the efficiency by which RGCs were able to follow high-frequency trains of pulses was variable across cell types. While OFF-BT cells were able to follow up to 700 pulses/sec, the maximum elicited response by the local edge detector (LED) and ON-OFF-DS population was only ~300 spikes/sec ^[76]. Twyford et al. ^[77] used a 2kHz stimulus waveform and found that (1) increasing stimulation amplitude resulted in a non-monotonic spiking response in ON vs. OFF-BT cells (Figure 5a), i.e., that the number of spikes elicited at a given amplitude increased up to a certain extent after which it proceeded to decline until very minimal to no spiking was observed, and (2) that amplitude modulation could increase the spiking rate of one rabbit RGC type while significantly suppressing the spiking response in the other (Figure 5b). This response pattern persisted in the presence of synaptic blockers, and thus is mostly an effect of direct RGC activation. The amplitude which elicited this peak response was also found to be different in ON vs. OFF-BT cells as well as in the ON vs. OFF-delta population of the rabbit ^[80], further suggesting that ON and OFF cells may be able to be preferentially activated at select amplitudes using high-frequency pulse trains. Recently, Muralidharan et al. ^[78] systematically studied the effects of a range of stimulus frequencies and amplitudes on the activation of α–RGCs in the mouse retina and found that the 3 of the 4 α-RGC types could be preferentially targeted at specific parameter combinations (Figure 5c). Specifically, OFF α-S cells were activated at 20–100μA across all frequencies, while OFF α-T’s exhibited activation between 150–240μA at 1kHz and ON α-T’s between 180–240μA between 4–6kHz. The ability to preferentially target ON vs. OFF RGCs using amplitude modulated high-frequency stimuli also persisted within rd1 mice in the presence of synaptic blockers ^[81], suggesting the potential efficacy of this strategy for use in models of retinal degeneration. As these studies did not take into account the specific location of these cells along the nasal-temporal/dorsal-ventral axis of the mouse visual field, it remains an open question whether preferential stimulation parameters vary across regions of the retina where the morphology of RGCs within a single type has been shown to vary ^[24,25].

Figure 5. High-frequency stimulation elicits differential spike rates in functionally distinct RGC types.

a) OFF and ON brisk-transient (BT) rabbit RGCs have non-monotonic spiking response patterns to high-frequency pulse trains (2kHz) delivered at various amplitudes. The amplitude which elicited the peak spiking response (dashed vertical lines) was found to be different in OFF (blue) and ON BT (orange) cells. b) A transient increase in amplitude during a pulse train delivered at 2000 pulses per second elicits different responses in ON versus OFF cells. ON BT cells (middle panel) displayed an increase in spiking response but OFF BT cells (bottom) displayed a suppression in spiking response. Reproduced with permission.^[77] 2014, IOP Publishing. c) Map summarizing the different stimulus frequency and amplitude combinations which produced preferential spiking response in each of the 4 mouse α–RGC types versus the other types. The specific pairing of frequency and amplitude that elicited the maximal differential spiking response at the population level was found to vary across the 4 cell types. Reproduced with permission from. ^[78] 2020, IOP Publishing

Preliminary investigations into the morphological properties which may be linked to these activation differences have found an inverse correlation between soma size and the amplitude that elicits a peak response to high-frequency stimulation ^[80]. As the ability of RGCs to support high-frequency firing has been attributed to the presence of Na_v1.6 channels in the AIS ^[20,38], the differential response characteristics to high-frequency stimulation exhibited across RGC types could be linked to differences in the length, location, and/or ion-channel composition of the AIS. As natural physiological signaling involves the specific activation of certain RGC types, while not activating others’, knowledge of how the stimulation thresholds and firing properties vary between RGC sub-types during high-frequency stimulation may provide insights into approaches that can achieve selective stimulation.

4. Theory and review of modeling studies involving the influence of somatodendritic morphology and AIS properties on extracellular stimulation

This section of the review will focus on computational studies that have investigated the physiological mechanisms by which the AIS and other anatomical properties shape the response of RGCs to epiretinal stimulation. During extracellular electric stimulation, electrode locations close to the AIS were found to be the site of lowest threshold ^[14] for direct neuronal activation due to the presence of a high density of sodium channels. Sodium channels have been found to be most sensitive to the short duration pulses used in the direct activation of RGCs as the time constant for their activation is in the sub-millisecond range ^[82]. The rapid opening of sodium channels within the AIS in response to small levels of depolarization (~5mV) allows the generation of single action potentials at a high temporal precision within RGCs that are located close to the stimulating electrode ^[58,83]. In summary, the anatomical features of the AIS (i.e., location proximal to the target RGC) in combination with its unique biophysical properties (i.e., low threshold, precise spike timing) make it an ideal stimulation target to achieve RGC activation at the high spatial and temporal resolution which is required for a retinal prosthesis. The following sections will describe the modeling approach which is widely used to simulate the behavior of RGCs in response to electric stimulation.

4.1. Modeling the response of RGCs to electric stimulation: A methodological summary

The response of RGCs to extracellular electric stimulation can be studied in-silico by using detailed morphological reconstructions of an individual RGC’s anatomical structure and biophysical properties. In order to better study the mechanisms of direct activation, the models can be constructed so as to exclude the influence of presynaptic inputs from bipolar and amacrine cells. As a result, the models considered here consist of single RGCs in isolation of the retinal network and their spiking responses depend only on their anatomical and biophysical properties and their interaction with the applied electric fields.

Computational models use the morphological details of traced neurons in conjunction with compartmental modeling ^[84] to simulate current flow along the intracellular space as well as the transversal (ionic and capacitive) currents across the neural membrane. The cellular processes are divided into discrete compartments which allows for numerically solving the arising differential equations. The interaction of spatiotemporally modulated electric fields and each compartment of the model is described by the activating function ^[85,86]. The activating function can be approximated by the second derivative of the extracellular voltage along the membrane, with compartments having a positive activating function initially being depolarized and those having negative values being hyperpolarized. Electric fields can be computed analytically for simple electrode geometries (e.g., point source or disk electrodes) in homogeneous medium or by finite element modeling if more detail of the surrounding tissue is required. Ionic currents are typically computed by Hodgkin-Huxley style models ^[87] which are fitted to experimental data to replicate spiking properties of RGCs. For RGCs, the most widely used family of models are based on the experimental data obtained from tiger Salamander, cat and rat ^[82,88–90].

4.2. Modeling the influence of RGC anatomy during extracellular stimulation

While early computational studies ^[91,92] laid the initial framework for better understanding the response of RGCs to epiretinal stimulation, their cellular models did not take into account the detailed morphology and biophysics of the AIS as revealed by more recent studies. For example, AIS length as well as distance from the soma were recently shown to vary across RGC types as well as across the retinal surface for given RGC types (see section 1). As a result, initial studies obtained conflicting results regarding the region of lowest threshold for extracellular stimulation. For example, in Greenberg et al. ^[91] electrode locations above the soma were found to be the site of lowest threshold whereas electrode locations above the axon and dendrites resulted in higher thresholds. The geometric simplifications made by ^[92] led them to conclude that the most sensitive region for electric stimulation was at the 90° bend of the proximal axon within the ganglion cell layer. While this region was also located in the proximal axon as the AIS, it did not depend on the specific biophysical properties of the AIS but on axon geometry. While both studies ^[91,92] used an increased sodium channel density at the proximal axon as suggested by previous studies ^[82], the maximum sodium channel density within the AIS was lower by a factor of 2–3 as compared to more recent studies ^[19,20].

The first study which focused on the AIS and its impact on the thresholds to extracellular stimulation ^[19] was motivated by findings of an experimental study by Fried et al. which showed that thresholds were lowest when the stimulating electrode was positioned above the AIS ^[14]. Jeng et al. ^[19] incorporated the AIS as a region of the proximal axon with a 5–40-fold increase in sodium channel conductance in comparison to the soma. A realistic range of 30–60μm for both AIS length and distance from soma was sufficient to qualitatively replicate experimental results i.e., minimum activation thresholds were found for electrode locations close to the AIS (Figure 6a, left). Subsequent studies extended this work to develop a more detailed understanding of how the morphology and biophysical properties of the AIS affected stimulation thresholds and response latency. For example, Tsai et al. ^[15] found that RGC response latency during epiretinal stimulation can vary depending on the relative location between target cell and electrode. Werginz et al. ^[93] examined the effect of the AIS on the site of spike initiation for varying electrode locations and found an effect even when the electrode was located fairly distant to the AIS. In order to study the influence of cellular anatomy on thresholds during electric stimulation, Werginz et al. ^[20] developed a library of various morphological parameters obtained from detailed anatomical studies of mouse RGCs ^[24,25] to construct morphologically and biophysically accurate model cells. Briefly, realistic models of 40 mouse α-RGCs were reconstructed using detailed information about AIS length, distance to soma as well as soma and dendritic field diameters. The results from computational experiments confirmed previous experimental data by showing that electrode locations close to the AIS had the lowest thresholds (Figure 6a, right). Additionally, thresholds were also found to be dependent on the placement of the stimulating electrode. For example, when the stimulating electrode was located above the soma, the threshold needed for activation was found to be directly proportional to soma diameter whereas thresholds for electrode locations above the distal AIS were negatively correlated to AIS length (Figure 6b).

Figure 6. Modeled thresholds in response to extracellular stimulation of RGCs.

a) (left) Threshold when the electrode is moved along the axon in a constant distance of 25μm. (right) 2-D threshold map for an individual RGC. The stimulating electrode was systematically stepped (5μm increments) at a constant height of 30μm above the soma. Thresholds were further interpolated to a 1×1μm grid. b) Minimum threshold when the electrode is located above the distal AIS is plotted vs. AIS length. Solid black line indicates the best-fit linear regression (p<0.001). Reproduced with permission.^[19,20] 2011, 2020, IOP Publishing.

The use of computational modeling in conjunction with detailed cellular reconstructions RGC enables the ability (1) to incorporate a large degree of morphological variability into simulations (section 2), as well as (2) to isolate the contribution of individual anatomical parameters on the response to prosthetic stimulation. This is beneficial for developing a link between experimental results, and the mechanisms involved in regulating the response to stimulation. With a better knowledge of the effect of anatomical features of RGCs on threshold levels the observed anatomical differences between RGC types could potentially be exploited to achieve cell type-preferential activation. For example, a previous in-vitro study found that dendritic field diameter played an important role in regulating the responses to repetitive electric stimulation ^[67]. Modeling studies ^[20], however, reported that the dendritic tree size (with all other parameters kept constant) does not significantly influence threshold levels, mainly because of the larger distance between stimulating electrode and dendritic tree during epiretinal stimulation. Results from in-silico studies have the advantage that, after model validation, many additional stimulation paradigms can be tested without performing time-consuming in-vitro experiments. For example, modeling studies have explored the effect of pulse duration and pulse polarity as well as the influence of electrode size on activation of the AIS ^[19,20,92]. In general, short pulses <0.5ms applied by microelectrodes in the range of tens of micrometers in diameter were shown to exploit the low-threshold region of the AIS most effectively. Depending on AIS length, distance from soma, diameter, as well as sodium channel conductance, the ratio between thresholds at the distal (passing) axon and minimum threshold was found to be in the range of 1.5–5 ^[15,19,20]. Small differences in thresholds were shown to be sufficient to achieve local single cell activation in the peripheral primate retina ^[16,83], however, the activation of passing axons from cells located far distant of the stimulating electrode was reported to be a major challenge ^[94].

In addition to work in in-vitro animal models, studies in RP and AMD patients have reported on differences in the thresholds to elicit visual percepts during epiretinal stimulation as a function of retinal eccentricity, with lower thresholds being observed closer to the macula ^[95]. The source of these threshold differences is still an active area of research. Recent efforts have begun to develop 3-D patient specific computational models to predict the neuronal response to prosthetic stimulation as well as the inter-patient variability in stimulation thresholds needed to elicit a visual percept ^[96]. As a result, these studies warrant the need for morphologically and biophysically accurate reconstructions of human RGCs. These models will need to consider (1) how the AIS properties of midget and parasol cells vary as function of retinal eccentricity (e.g., in the fovea vs. periphery), (2) whether these differences correspond to differences in stimulation thresholds, and (3) how these properties change as a function of retinal degeneration and influence the patient-observed visual percepts.

5. Novel materials approaches toward retinal protheses

Conventional MEAs such as the ones described here generally consist of inert metallic electrodes that are individually addressed by metallic interconnects. While these devices have been widely studied and also commercialized, the need for interconnects and cable feedlines limits device number and density while low conformability and mechanical mismatch between electrodes (Elastic modulus E=150–170 GPa for silicon)^[97] and retinal tissue (E=200–400 kPa)^[98] may hinder long term device integration. Recent advances in materials science present new avenues that could address these challenges.

5.1. Photovoltaics for optical modulation

Neurons can be modulated using photovoltaic (PV) devices which transduce optical inputs into electrical stimuli. PVs eliminate the need for electrical interconnects and can be fabricated on a wide range of substrates, enabling large area, pixel-like device configurations that are difficult to achieve by other means. The Ghezzi group recently demonstrated POLYRETINA, an epiretinal implant consisting of up to 10,498 functionally independent PV pixels consisting of polymer anode and semiconductor layers and Ti cathode layer (Figure 7a). These devices were fabricated on a flexible and conformable poly(dimethyl siloxane) substrate and with 120 μm pitch, allowing for the direct and spatially selective activation of RGCs with single-pixel resolution ^[99–101].

Figure 7. Photovoltaics for retinal implants.

a) Optical image of high-density POLYRETINA prosthesis with 10,498 PV pixels fabricated on PDMS substrate. (inset) Magnified view showing individual PV pixels with 80-μm diameter and 120-μm pitch. b) Schematic of a photocurrent current produced by a PIN-SiNW at a neuronal cell membrane, inducing action potential generation in the neuron via membrane depolarization. Solid blue and orange lines represent movement of electrons and holes towards n-type and p-type Si, respectively, on light stimulation. c) (left) HAADF STEM image of PIN-SiNW with p-type core outlined by white dotted line. (right) TEM image of PIN-SiNW. d) TEM image of Au–TiO₂ NW structure. (inset) Corresponding schematic showing (green) TiO₂ NW and (red) spherical TiO₂ nanocrystals. e) Illustration of an eye highlighting a section of retina (blue dotted box). f) Illustration of (left) healthy retina and (right) blind retina, where necrotic photoreceptors were replaced by biomimetic Au-TiO₂ artificial photoreceptors. g) Schematic of syringe-injectable electronics. h) Schematic showing intravitreal injection of mesh electronics onto the RGC layer. i) In vivo through-lens images of a mouse eye on days 0 and 14 after injection of mesh electronics; note electrode indexing in the day 14 image. Reproduced with permission from a) ^[99], 2021, Nature Publishing Group; b,c) ^[102], 2018, Nature Publishing Group; d,e,f) ^[108], 2018, Nature Publishing Group. g) ^[109], 2015, Nature Publishing Group; h,i) ^[110], 2018, American Association for the Advancement of Science.

Recent advances in nanoscience have enabled new routes toward PVs that offer single-cell or subcellular resolution, tunable absorption properties and stable cellular integration. Silicon nanowires (SiNWs) are one illustrative class of nanomaterials that are especially advantageous because their geometry and doping profiles—and therefore optoelectronic properties—can be precisely controlled during growth. For example, coaxial p-type/intrinsic/n-type (PIN) SiNWs consisting of p-doped cores and intrinsic and n-type shells have been achieved. These structures produced photocurrents when illuminated with 532 nm laser light, and the photocurrents elicited action potentials in primary rat dorsal root ganglia neurons (Figure 7b) ^[102]. SiNW PVs offer several advantages over their microscale counterparts. First, they exhibit nanoscale surface roughness. These features tend to enhance focal adhesion formation, which induces tight junctions between cells and devices (Figure 7c). Second, through rational control at the time of synthesis, PIN SiNWs could be generated with either capacitive or Faradic photocurrents dominant ^[103]. While both mechanisms have been shown to successfully activate CNS neurons, future experiments are needed to better compare the relative merits of capacitive and Faradic stimulation methods in terms of biocompatibility (see ^[104]). Finally, SiNWs could be assembled on nearly any type of substrate, including those that are stretchable and flexible. This property enabled large- area implants that were soft and conformal, as recently demonstrated for electrical modulation in cardiac tissue ^[105]. Alternately, freestanding SiNWs could be internalized by cells. Examples include SiNWs coated with a cell penetrating peptide which that delivered into primary hippocampal and dorsal root ganglia neurons ^[106]; as well as unmodified SiNWs that were internalized by multiple cell lines via phagocytosis ^[107]. Individual glial cells with internalized PIN SiNWs could be remotely controlled, exhibiting a fast calcium release upon irradiation. That wave propagated to neighboring gial and neural cells ^[103].

PV nanowires (NWs) and similar nanostructures may also be exploited for their biomimicry, opening new avenues for tissue-device hybrids. The Zheng group explored this concept by developing a subretinal implant consisting of Au-TiO₂ nanowire (NW) arrays (Figure 7d), where the NWs functioned as artificial rod and cone photoreceptors (Figure 7e,f) ^[108]. Significantly, these structures exhibited high absorption coefficients (0.5 – 1.5 μm⁻¹) that rivaled those of natural photoreceptors in UV, blue and green regimes. To demonstrate a hybrid retina the authors studied rd1/cDTA (blind) mice which had degenerated rod and cone photoreceptors but intact bipolar and RGC. Photocurrents produced by NWs activated nearby bipolar cells which in turn activated synaptically-connected RGCs. Notably, RGCs could be activated with irradiation as little as 0.5 μW/mm², while ca. 80% of the RGCs were activated at 133 μW/mm². In a similar in vivo study with blind mice, NW array implants restored light-invoked spiking activity in the contralateral primary visual cortex for up to 5 months post-surgery, and pupil restriction for 4–8 weeks post-surgery ^[108].

5.2. Injectable Electronics

Injectable “reagent-like” electronics would preclude the need for surgery, which would be especially useful for delicate or inaccessible systems such as the brain or retina. The Lieber laboratory demonstrated a flexible bioelectronic mesh composed of SU-8, a bioinert polymer, that could be administered through a needle with diameter as small as 100 μm (Figure 7g). The meshes supported up to 16 recording elements, interconnects, and input/output pads that could be connected to recording electronics. Despite the small bore of the needle, after administration the meshes unfurled to cover centimeter-scale areas. Electronics administered into a rat brain recorded brain activity with >80% yield and moreover demonstrated little chronic immunogenicity ^[109]. A similar system was administered onto the retina of live rodents by intravitreal injection (Figure 7h,i),. These meshes interfaced with RGCs and recorded responses to visual stimuli for at least two weeks ^[110]. While these studies focused on recording, bioelectronic meshes could also be enabled with stimulating elements including gold pads (as demonstrated in cardiac studies ^[111,112]) or PV NWs as described in the previous section.

Other materials innovations could further enhance the function of bioelectronic meshes. Built-in strain causes meshes to unfurl into rationally-designed shapes ^[113], and so these structures could be designed to conform to the shape of the retina. To further enhance long-term stability SU-8 might be replaced with a bioactive or natural material such as photo-crosslinkable silk fibroin, which exhibited similar bioelectronic performance but also demonstrated enhanced binding affinity toward neurons ^[114]. Alternatively, both meshes and electronics could be designed with fully bioresorbable components should a transient intervention be desired ^[115]. Efforts to tune the interface between devices and cell are also an active area of research, and these have been reviewed in detail elsewhere ^[116].

6. Conclusion

In this review, we have (1) summarized the variability in somatodendritic morphology and AIS properties that exist across α-RGCs in the mouse retina, (2) analyzed studies which have investigated the influence of morphology on the direct activation of RGCs by extracellular electric stimulation, (3) and described what is known about the biophysical mechanisms that underlie differences in stimulation thresholds and electrically evoked responses in RGCs. By better understanding the biological mechanisms underlying the activation and responses of various ganglion cell types, in both healthy and diseased models, it will facilitate the development of stimulation techniques to achieve preferential activation of specific types. In addition, a plethora of novel stimulation techniques are currently under development ^{[99,117–123]}, many of which may allow for improved targeting and control of specific neurons.

Biographies

Vineeth Raghuram received his B.S. in Biomedical Engineering from the Georgia Institute of Technology, M.S in Biomedical Engineering from the Rensselaer Polytechnic Institute, and his Ph.D. in Biomedical Engineering from Tufts University. His research interests lie in at the intersection of visual neuroscience and bioelectronics, with a focus on better understanding the biophysical factors that underlie the neuronal response to artificial stimulation in the retina/cortex and developing the next generation of flexible electronic devices for neural stimulation.

Paul Werginz is currently an independent postdoctoral fellow at the Institute for Analysis and Scientific Computing at Vienna University of Technology and also holds a joint academic affiliation with Shelley Fried’s lab at Massachusetts General Hospital / Harvard Medical School. He received his B.A., M.Sc. and Ph.D. from Vienna University of Technology in 2010, 2012 and 2016, respectively. Paul’s research interests lie in the combination of computational and experimental neuroscience to improve the understanding of the mechanisms involved in the electrically stimulated retina; this knowledge will help to develop more sophisticated stimulating strategies for future retinal implants. Furthermore, he is interested in the anatomical and biophysical factors that shape the intrinsic properties of retinal ganglion cells.

Shelley I. Fried was born in Brooklyn, NY, USA, in 1961. He received the B.E. degree in mechanical engineering from Cooper Union, New York, NY, USA, in 1982, the M.S. degree in biomedical engineering from the Pennsylvania State University, State College, PA, USA, in 1986, and the Ph.D. degree in vision science from the University of California, Berkeley, CA, USA, in 2004. From 2004 to 2006, he was a Postdoctoral Fellow in the Molecular and Cell Biology at Berkeley and from 2006 to 2007 he was a Research Fellow at the Massachusetts General Hospital, Department of Neurosurgery. He currently heads up the Neural Prosthetic Research Laboratory at Massachusetts General Hospital and is an Associate Professor in the Department of Neurosurgery at Harvard Medical School as well as a Research Scientist at the Boston VA Healthcare System. His research focuses on understanding the fundamental mechanisms by which CNS neurons respond to artificial stimulation and using that information to develop new, more effective stimulation strategies and devices.

Brian P. Timko is an Assistant Professor in the Department of Biomedical Engineering at Tufts University. He earned B.S. degrees in chemistry and chemical engineering from Lehigh University in 2002 and a Ph.D. in chemistry from Harvard University in 2009. He was an NIH F32 postdoctoral fellow and subsequently Instructor at Massachusetts Institute of Technology and Boston Children’s Hospital. Throughout his training he studied solid-state nanomaterials and their applications in nanoelectronics, tissue engineering and drug delivery. His current research interests focus on novel materials for bioelectronics that offer recording and modulation at the cell, tissue and organ levels; routes toward stable bioelectronics-embedded tissue hybrids; and nanomaterials-based approaches to direct cellular assembly and maturation in engineered tissues.