Electrical activation of degenerated photoreceptors in blind mouse retina elicited network-mediated responses in different types of ganglion cells

Electrical (e-) stimulation is explored in schemes to rescue the vision of blind people, e.g. those affected by Retinitis Pigmentosa (RP). We e-activated subretinally the surviving degenerated photoreceptors (d-Phrs) of the rd1 mouse (RP model) and evoked visual responses in the blind retina. The e-stimulation was applied with a single platinum/iridium electrode. The d-Phrs (calcium-imaging) and ganglion cells (GC) activity (MEA-recording) were recorded in simultaneous multilayer recordings. The findings of this study confirm that the d-Phrs responded to e-stimulation and modulated the retinal network-activity. The application of blockers revealed that the synaptic interactions were dependent on voltage-gated calcium channels and mediated by the transmitters glutamate and GABA. Moreover, the gap junctions coupled networks promoted the lateral-spread of the e-evoked activity in the outer (~60 µm) and inner (~120 µm) retina. The activated GCs were identified as subtypes of the ON, OFF and ON-OFF classes. In conclusion, d-Phrs are the ideal interface partners for implants to elicit enhanced visual responses at higher temporal and spatial resolution. Furthermore, the retina’s intact circuity at the onset of complete blindness makes it a tempting target when considering the implantation of implants into young patients to provide a seamless transition from blinding to chip-aided vision.


Results
The d-Phrs of the rd1 mouse retina in subretinal configuration were e-stimulated by a single platinum/iridium electrode, and both the e-activated d-Phrs's calcium responses and the network-mediated correlated spike responses of the GCs were simultaneously recorded (Fig. 1). The compositions of the network mediated retinal signalling were identified as glutamatergic and GABAergic for vertical retina signalling while the gap junctional (GJ) cell coupling was responsible for the lateral retinal signalling in the outer and inner retina, which was revealed by antagonising each signalling pathway with respective pharmaceutical drugs.
The d-Phrs are activated by e-stimulation. The rd1 retina's light-insensitive d-Phrs 17,18 are responsive to e-stimulation, applied by a single electrode (Fig. 1a,b1). The d-Phrs were e-stimulated subretinally using an increasing voltage ramp of 0.3-1.2 V (Fig. 1c1 and c2) and two d-Phrs response parameters were evaluated: the number of responsive d-Phrs and their distance from the stimulation electrode. Control experiments were used to ensure that the recorded rd1 retina-patch (P20) was blind and unresponsive to UV light illuminated during the calcium-imaging recordings, as shown previously for white-light stimulation 17,19 . The application of UV light flashes at 380 nm and 3 sec duration yielded no retinal responses. D-Phrs were responsive to e-stimulation at 0.3 V, and according to the incremental voltage step applied, the number of e-activated d-Phrs increased on average by 8.13 ± 5.25 cells (Fig. 2a). The increase in the number of e-activated d-Phrs from one voltage step to the next, was not significant ( The number of e-activated d-Phrs declined the spatial distance from the stimulation electrode increased, in a half-sigmoid curve (at all voltage steps, Fig. 2a). The highest number of d-Phrs are e-activated inside the first binning range of 20 µm, i.e. the closest distance to the stimulation electrode. With increasing distance, the number of e-activated d-Phrs drops by 1.32 ± 0.29-fold at 40 µm binning range and by 2.37 ± 0.41-fold at 60 µm binning range. The strongest cell response −95.50% of the total e-activated d-Phrs -occurred within the 60 µm binning range (44.90%, 35.45% and 15.15% cells at 20, 40 and 60 µm, respectively). The few remaining e-activated d-Phrs (4.5%) are distributed across the distant binning ranges of 80-100-120 µm, with strong standard deviations (1.65 ± 0.51, 1.4 ± 0.22 and 0.2 ± 0.13 cells), indicating that the exterior binning area is far enough beyond the influence of e-stimulation to cross the threshold. In conclusion, d-Phrs are responsive to e-stimulation of 0.3-1.2 V and the e-evoked activity extends spatially up to 60 µm from the surrounding boundary of the stimulation electrode.

Ganglion cell responses elicited by means of subretinal e-activation of the d-Phrs. Correlated
GC spike activity was detected in the MEA-recordings during e-activation of the d-Phrs, (Fig. 1b2,c3). The voltage value (0.3-1.2 voltage ramp) activating d-Phrs and promoting robust network mediated GC activity, was determined in the recording field at the shortest distance to (beneath) the stimulation electrode (Fig. 1b2). When the voltage ramp was applied, the first GC activity was detected at 0.5 V, on a few MEA-electrodes (15 ± 11.18%, Fig. 2b). At the increased stimulation voltage (0.7), significantly more MEA-electrodes detected GC activity (85 ± 6.8%). At higher stimulation voltages of 0.9-1.2 V, up to 90% of the MEA-electrodes detected GC activity. This data indicates that no significant increase in GC activity occurs between 0.7 to 1.2 V. Thus, the voltage value of 0.7 V was established as the minimum required for sufficient e-activation of the retinal d-Phrs, to induce network mediated GC activity. On the other hand, the number of e-activated GCs dropped with increasing distance from the stimulating electrode: Around 85% of the MEA-electrodes detecting GC activity were within the distances 80 µm (88 ± 6.8%, 93 ± 8.0% and 86 ± 9.6% at binning of 0, 40 and 80 µm, respectively; average over the 0.7-1.2 V steps). At a more distant binning of 120 µm, the number of MEA-electrodes detecting GC activity dropped to 63.7 ± 28.5% with larger standard deviation. At 160 µm, the number of MEA-electrodes detecting GC activity dropped significantly, with a high standard deviation (19.14 ± 10.83%), indicating that the exterior binning area is beyond sufficient e-stimulation influence. Thus, the distance study of this research suggests that the d-Phrs-induced network activity in the GC layer spreads up to 120 µm from the stimulation electrode. The e-evoked d-Phr responses and network activity rely on VGCC. So far, it has been established that the subretinal e-stimulation of the rd1 retina evoked correlated responses in d-Phrs (calcium-imaging) and GC (MEA-recordings, Fig. 1), although with different spread range. To determine whether the GC activity is network mediated, the synaptic signal transmission was antagonised by the application of the voltage-gated calcium channel (VGCC) blocker verapamil 22,23 . The e-evoked calcium responses of the d-Phrs were reduced in the presence of verapamil, but the GC activity reduced significantly (Fig. 3a, average GC response spike activity at ctr: 10.14 ± 4.15, drug: 3.3 ± 2.41 (***) and wash: 2.05 ± 1.43 (***)). These results suggest that the correlated GC responses due to e-stimulation of d-Phrs, are network mediated and rely on the neuronal synapses VGCC.
E-activated d-Phrs drive the postsynaptic bipolar cell activity by glutamate release. E-activated d-Phrs drive retinal network activity by releasing glutamate at the first synapse. The role of glutamate was investigated by applying various receptor agonist and antagonists and blocking the glutamatergic input of bipolar cells from d-Phrs. Identified ON bipolar cells (see section Network mediated activation of different ganglion cells types) were agonised by LAP-4 (mGluR6 agonist), leading to the suppression of ON-GC responses (Fig. 3b, average GC response spike activity at ctr: 9.35 ± 3.66, L-AP4: 1.35 ± 1.18 (***) and wash: 3.39 ± 1.63 (**)). The response-recovery was significant, but the initial response strength was not completely recovered (~40%). Next, the glutamate reception at the first synapse was completely prevented with a solution containing L-AP4, L-TBOA and NBQX. The ON bipolar cells were agonised by L-AP4 (mGluR6) and additionally antagonised by L-TBOA (EAAT5 receptors). In the rd1 retina, the rod bipolar cells are contacted by d-Phrs via functional ectopic synapses, and promote excitatory rod BC (EAAT5 receptors) signalling by glutamate 18,24 . The OFF system was antagonised by NBQX (iGluR). The application of these drug combinations significantly reduced the GC post stream network dependent activity (Fig. 3c, average GC response spike activity at ctr: 8.39 ± 3.63, drug: 1.53 ± 1.89 (***) and wash: 0.93 ± 1.09 (**)). After the wash, no recovery of network mediated GC signals was observed, instead the blockage effect strengthened over time by further reduction of the GC activity. Reduction of the 120 µm activity boundary to 40 µm in the presence of CBX (29.17%). At larger distances, binning up to 80 µm, the GC activity dropped steeply with a large standard deviation (13.54 ± 13.45, 7.62 ± 9.9 and 1.4 ± 3.9%, respectively), indicating that the exterior binning area was significantly (***) beyond the range of adequate e-stimulation influence. No GC activity recovery effect from the CBX dependent GJ network blockage was detected after the 20-minute wash period. The blockage even strengthened in the inner retina over time, narrowing the activity spread to the vicinity of the stimulation electrode, covering almost 65.9% of the overall activity. In the outer retina, the range of the activated d-Phrs remained at ~40 µm covering 96.2% of the e-activated d-Phrs. Thus, the data obtained in this study suggest that the GJ coupled networks facilitate the lateral spread of the e-evoked activity in the outer and inner retina.

Network mediated activation of different ganglion cells types. The basic GC types -ON, OFF
and ON-OFF -were identified in the recordings of the subretinally e-stimulated blind retina (only GC within 80 µm distance from the e-stim electrode, stimulation voltage 0.7). In the healthy mouse retina, these three GC types are identified by their responses correlating to the state of the light stimulation (ON or OFF). However, for the GC type determination of the blind retina, we employed two different approaches. First, the cells were sorted by cluster analysis into two major types, with mono or biphasic response shape (Fig. 5) responses with larger latencies (Fig. 5a, peak at ~0.2 s) were suppressed, suggesting this signalling to be of the ON-type. Consequently, responses with shorter latency were assigned as OFF-type. Regarding the cells with the biphasic response shape, it is to note, even though the first response peak is of the OFF-type and the second of the ON-type, these cells are labelled as "ON-OFF" cells. Established this, the results of the classification analysis yielded six distinctive GC subtypes (Fig. 5b), four cluster types with mono-modality and two with multi-modality features. The cell type with the mono-modality feature were three OFF-types, including two OFF-sustained subtypes (OFF 1 and OFF 2 ) and one transient type (OFF 3 ). The sustained sub-cell types OFF 1 and OFF 2 had response peaks at 16.7 and 20.9 ms after e-stimulation but differed in their sustained response durations of 93.7 and 155.6 ms (at 75% drop from response peaking time). With a small delay, the activity peak of the second cell OFF-type (OFF 3 , transient) was observed at 66.1 ms. The fourth cell cluster is of the ON-cell type (ON 1 ) with an activity peak delay at 188.2 ms after application of the e-stimulus. The fifth and sixth cell type clusters were from the multi-modality feature group and represented two ON-OFF GC subtypes. The first ON-OFF 1 cell type exhibited two peaks at 60.3 and 121.5 ms after e-stimulation, whereas the second type, ON-OFF 2 , featured three peaks: The first two close together at 18.7 and 49.7 ms and the third delayed at 125 ms after e-stimulation. Overall, the selective elimination of the ON-channel signalling (Figs 3b and 5a) suggests, that the GC activation is network mediated at the retina's first synapse (d-Phrs => ON bipolar cells => GC). Therefore, the activation of different major GC types (ON, OFF and ON-OFF) proves that the network integrity of the RP-affected blind rd1 mouse retina is preserved at a level which might provide enhanced visual sensations by means of e-activation of d-Phrs at the onset of RP-dependent, complete blindness.

Discussion
In the rd1 mouse model of human RP, rod-dystrophy leads to secondary cone degeneration 16 . During the initial disease phase, the photoreceptor cell layers degenerate, and only a few rows of segmentless cone cell bodies and their terminals survive in the blind retina 20,27 . In this study, the potential of these d-Phrs for restoring visual responses at the onset of complete blindness by means of subretinal e-stimulation, was examined.
The subretinal e-stimulation of rd1 mouse d-Phrs by a single electrode elicited visual responses of the ON, OFF and ON-OFF type. Results of pharmacological blockers revealed that the e-stimulus correlated GC spike activity is network-mediated (d-Phr => BC => GC) by interneuronal synaptic interactions and not directly by e-stimulation. In addition, the activation of the OFF and subsequently the ON-channel demonstrates that the retinal network-activity originates from the d-Phrs: E-stim of the d-Phrs induced a depolarisation of the d-Phrs, followed by their subsequent hyperpolarisation. The dynamic of the relatively fast OFF-response caused by electrically induced depolarisation of d-Phrs is consistent with the relatively fast rise of the calcium influx (=>glutamate release). The long latency of the ON-signal (peak at ~0.2 s) is also consistent with the very slow recovery of the calcium levels (slow hyperpolarisation => less glutamate release). In turn, glutamate released by the d-Phrs drives the activity of the cells contacting into the outer plexiform layer: Horizontal cells (iGluR), ON (mGluR6) and OFF (iGluR), but also the rod bipolar cells (mGluR6 and EAAT5). It should be noted that the d-Phrs and rod bipolar cells of the rd1 mouse established a miss-wired connection, which is sign-conserving and regulated by the glutamate transporter EAAT5 18 . In addition, the GABAergic inhibition modulates the synaptic input of the bipolar cells in the rd1 outer retina 18 . In the presence of the GABA inhibitor, the subretinal e-activation of the d-Phrs leads to increased GC spike responses. This result agrees with previous findings, suggesting a vigorous, activity modulating, role of horizontal cells in the degenerated rd1 retina 18 and GABAergic inhibition in the inner retina of the P23H rat model of RP 28 . Overall, the present glutamatergic and GABAerig signalling as well as the basic visual responses of the ON, OFF and ON-OFF type (~40 mouse GC types 29 ) are strong indications for an intact vertical signal transduction pathway of the blind mouse retina at the onset of the complete blindness. Along the morphological changes, the emergence of spontaneous, light independent activity in the inner and outer rd1 retina has also been reported 18,26 . The data obtained in this study do not support the expected negative interference from the auto-activity emerging from the rd1 retina on the retinal activity evoked by subretinal e-stimulation of the d-Phrs. The spontaneous activity described is not stationary; rather it emerges locally in an epochal manner in "generator" cells and travels in a wave-like manner through the GJ coupled cells of the outer and inner retina 17,18,26 . Once the wave has disappeared 17 , the e-stimulation area becomes silent. Thus, we were able to e-activate the remaining degenerated retinal network in a noise-free condition and analyse the underlying mechanism of the functional circuity by the application of pharmacological agents. Furthermore, in this study's calcium-imaging recordings, the responses in d-Phrs evoked by e-stimulation were much stronger than the calcium amplitudes generated by spontaneous activity in the same cell (Fig. 1c calcium trace). This finding establishes the dominance of e-evoked calcium responses over the spontaneous activity in the outer retina, since like the e-evoked activity of the d-Phrs, the spontaneous activity of the d-Phrs also relies on the VGCC 18 .
Furthermore, when subretinal e-stimulation of the d-Phrs is applied, the induced activity spreads horizontally from the stimulation electrode: In the outer (~60 µm) and inner (~120 µm) retina (ratio 1:2). This outcome suggests that the spread of activity around the stimulation electrode is based on horizontal cell signalling, conducted by GJ cell coupling. The GJ distance study data (control: Enlarged GJ coupling -CBX: Reduced GJ coupling) support the findings describing the GJ connectivity during visual adaptation (night: Enlarged GJ coupling -day: Reduced GJ coupling 25,30 ). Due to the loss of light-sensitive photoreceptor segments, the rd1 retina experiences the dark condition continuously, hence presumably the GJ coupling of the d-Phrs is adapted to the dark condition. The e-induced d-Phrs activity extends over a radius of 60 µm via the GJ coupled d-Phr-network, which is a 60% match to the GJ-coupled connectivity range of the photoreceptors (100 µm) in healthy mouse and rabbit retina in darkness 31 . Similarly, increased GJ connectivity during light adaptation, has been previously reported for turtle cone-cone coupling 32 , and rod-cone coupling in salamanders 33 . In contrast, the photoreceptor coupling is restricted to few neighbouring cells in the presence of light, corresponding to the CBX effect observed in this study's experiments: CBX restricts the lateral spread of the e-evoked activity in the GJ coupled network of d-Phrs to few cells. Moreover, the cone-driven horizontal cell coupling is modulated during the light adaption processes [34][35][36] . In the rd1 retina, the GJ coupled networks of the horizontal cells and d-Phrs, are capable of signalling over a larger distance and performing synchronised cell activity 18,37 . Hence, it is conceivable that the horizontal cells also contribute to the lateral spread of the e-evoked activity in the outer retina. In the inner retina, the e-evoked activity in the GC layer extends up to 120 µm from the stimulation electrode. It is very likely that in the rd1 retina, both bipolar cell pathways (d-Phrs => rod & cone bipolar cell 18,38 ) feed the e-evoked activity of d-Phrs into the inner retina's two GJ coupled networks: The amacrine cell network and GC network 25 . These two networks are known to generate synchronised spike activity in neighbouring retinal GCs 39 , and therefore it is very likely that they are responsible for the twice as large activity spread in the inner retina in comparison to the outer retina. After the 20-minute CBX wash, the GJ blockage did not diminish but became stronger. A similar outcome had previously been observed in rd1 retina recordings 18 and discussed as a retinal degeneration-effect leading to increased cell vulnerability. It is important to note that a further effect of CBX on VGCC has also been reported, affecting the cone and horizontal cell signalling 37 . We took this potential side effect into account in the data evaluation, by comparing the GJ blocker data (CBX) and VGCC blocker data (Verapamil): CBX evidently affected the lateral spread of the activity by restricting it to the vicinity of the stimulation electrode, whereas verapamil completely blocked the synaptic transmission, silencing the network-mediated GC-responses. A second GJ blocker, meclofenamic acid (MFA) has previously been reported to eliminate spontaneous activity in rd10 mouse recordings and preserve retinal light responses 40 . However, in this study's experiments, MFA had a very strong impact on the rd1 retina, silencing the retinal activity in calcium-imaging and MEA-recordings. It is conceivable that the activity-diminishing effect of MFA on the rd1 retina recordings, is due to the degeneration dependent vulnerability of the rd1 retinal circuity, when compared to rd10 recordings at the same age.
Overall, the e-activation of d-Phrs is of great relevance for blind RP patients. In the initial phase of RP, the d-Phrs in rd1 mouse 20 and human retinas sprout 41 , even arranging a functional ectopic synapse between cones and rod bipolar cells 1,18,24 , indicating that these connections represent an effort by the bipolar cells to renew input activity. To restore vision by d-Phrs, they were resensitised to light in an optogenetic approach using channelrhodopsin 1 . The operational range of channelrhodopsin is at unnaturally high levels of light intensity and therefore not yet useful for human therapy without complex and extensive light amplification devices. The method employed in this study, e-activation of the d-Phrs using low and safe e-stimulation values, was successful in restoring network-mediated visual responses in different types of GCs. Thus, d-Phrs are a promising target for e-stimulation to substitute retinal afferent input in human therapy using established subretinal e-implants. Along the same lines, some patients have reported experiencing an outstanding visual perception with the e-chip implanted at the foveal area, e.g. recognising letters 6,10 or facial expressions 42 . Thus, it is conceivable that the outstanding e-chip performance was due to the presence of the d-Phr in the foveal region providing intact glutamate transmission to the outer plexiform layer. In terms of the sustainability of therapeutic effects, the genetically introduced channel rhodopsin could not stop RP, so that the benefit is lost with progressing retinal degeneration 1 while vision provided from subretinal implants 43 with a technically overhauled implant Alpha AMS, remains stable over a long period. Additionally, e-stimulation has been reported to be beneficial in the treatment of human RP 44,45 and neuroprotective in RP rat models 46 . Therefore, implantation of the subretinal implant at an early stage of RP-induced blindness while the d-Phrs are present is suggested, to deliver afferent input and neuroprotection at the same time and provide a seamless transition from natural seeing towards e-chip aided vision with a long-term perspective. Moreover, the lower e-stimulation values for d-Phrs activation allow smaller electrodes and therefore denser electrode arrays, enhancing the implants' spatial resolution.

Materials and Methods
Animals. In this study, the crossbred mouse line HR2.1:TN-XL × rd1 (rd1 TN-XL ) 18,47 homozygous for the rd1 allele 15 was used, to identify the fluorescently labelled degenerated cones (d-Phrs). The mouse line is referred to as "rd1". The animals were housed under standard, white, cyclic lighting, had free access to food and water and were used irrespective of gender. Animal protocols, compliant with §4 of the German law on animal protection, were reviewed and approved by the Tübingen University committee on animal protection (Einrichtung für Tierschutz, Tierärztlichen Dienst und Labortierkunde; Registration No.: 21/01/2015). All the experiments were performed in accordance with the ARVO statement for the use of animals in ophthalmic and visual research. The animals were sacrificed in a carbon dioxide atmosphere following cervical dislocation. Experiments were performed at postnatal days (P) 20-50 (n = 17).
Tissue preparation. The eyes of the rd1 mice were enucleated and the retina was isolated in an extracellular solution containing (in mM): 125 NaCl, 26 NaHCO 3 , 2.5 KCl, 2 CaCl 2 , 1 MgCl 2 , 1.25 NaH 2 PO 4 , and 20 glucose, and maintained at pH 7.4 using carboxygen perfusion (95% CO 2 /5% O 2 ). All the chemicals were obtained from Sigma-Aldrich (Germany). Retinal degeneration in rd1 mice progresses with age and retinal eccentricity, therefore the recording fields were consistently chosen within an ~800 µm radius from the optic disc. The retinas were incubated in an extracellular solution containing 0.27 µM Fura-2-AM and 0.1% Pluronic acid (Invitrogen, Eugene, USA) for 35 minutes at room temperature, washed and transferred to the MEA-recording chamber (GC side down), where the tissue was perfused with carboxygenated medium at 32 °C. The chamber perfusion rate was adjusted to 1 ml/min. E-stimulation and recordings. In addition to subretinal e-stimulation, multilayer recordings in a "sandwich" configuration were performed simultaneously, by applying calcium-imaging and MEA recordings (Fig. 1a). See supplementary information for details. All data generated or analysed during this study are included in this published article (and its Supplementary Information files).

Data analyses.
Custom developed scripts (MATLAB, The MathWorks, Germany) were used for the data analysis of calcium-imaging, MEA-recordings and spatial activity spread, unless otherwise indicated. For GC-type analysis online available tools [48][49][50][51][52][53] were modified and utilized. See supplementary information for details. The Wilcoxon signed-rank test and the two-tailed Student's t-test were used for the statistical evaluation of drug effects; statistical significance is indicated as * for p ≤ 0.05, ** for p ≤ 0.01 and *** for p ≤ 0.001; and all parameters are given as mean ± SEM.