Electronic photoreceptors enable prosthetic visual acuity matching the natural resolution in rats

Retinal degenerative diseases such as age-related macular degeneration and retinitis pigmentosa lead to photoreceptor loss, while much of the inner retina remains viable for electrical stimulation. Subretinal photovoltaic implants (e.g., PRIMA) have demonstrated that prosthetic letter acuity can closely match pixel size, but achieving clinically useful acuity (<20/200) requires pixel sizes of 50 µm or smaller, and ideally 25 µm for 20/100. Conventional bipolar pixel designs with local return electrodes limit lateral crosstalk but constrain field penetration, causing stimulation thresholds to rise sharply as pixel size decreases, making pixels below ~55 µm inefficient in rodents and ~75 µm in humans. The research question is whether dynamic, optically controlled confinement of the electric field in a planar monopolar subretinal array—by turning selected pixels into transient return electrodes via forward-biased diode conduction—can minimize crosstalk, preserve high contrast, and thereby enable prosthetic visual acuity at or beyond the pixel sampling limit, potentially matching natural acuity in rats with 20 µm pixels.

Previous clinical and preclinical work established that subretinal photovoltaic implants can restore patterned vision with acuity limited by pixel pitch (e.g., PRIMA with 100 µm pixels yielded letter acuity ~1.17±0.13 pixels). However, shrinking bipolar pixels exacerbates threshold due to limited field penetration depth set by active–return spacing. A 3D honeycomb design with elevated returns can orient fields vertically and reduce thresholds, but fabrication and functional neuronal migration into wells remain under development. Epiretinal stimulation requires complex encoding of ganglion cell spiking, whereas subretinal stimulation leverages extant retinal processing (center–surround, temporal filtering). Prior modeling and experiments showed that full-field activation of monopolar arrays yields nearly vertical fields with much lower thresholds (~0.06 mW/mm² for 10 ms pulses) than 40 µm bipolar pixels (~1.8 mW/mm²), but crosstalk compromises contrast without confinement. This work builds on models of capacitive electrode–electrolyte interfaces, current steering, and ocular field propagation to introduce and validate transient, optically conditioned local returns for field confinement.

- Device and implant: Planar monopolar subretinal photovoltaic arrays, 1.5 mm diameter, 30 µm thick, with hexagonal pixels of 40 µm or 20 µm pitch. Each pixel had a SIROF-coated active electrode (18 µm diameter for 40 µm pixels) and a photosensitive diode; the return was a SIROF-coated common ring at the periphery. Imaging was projected with 880 nm light via a DMD-based system. - Computational modeling: Finite element method (COMSOL 5.6) modeled elementary electric fields for individual pixel activation in a degenerate rat retina. Boundary conditions: active electrodes treated as uniform current density (UCD) and the common return as equipotential (EP), consistent with electrode RC time constants (active ~0.24 ms, return ~40 ms) relative to 4–10 ms pulses. Inter-pixel coupling was quantified by a cross-resistance matrix R; nearest-neighbor cross-resistance R1,2 was 15–30 kΩ for 40 µm pixels. A multi-dimensional nonlinear circuit model of all pixels, incorporating diode I–V, electrode capacitances, photocurrent, and spatial coupling, was solved with a custom adaptive-step gradient descent (MATLAB) to obtain time-varying electrode potentials and currents. Corneal potentials were modeled using an anatomically realistic rat eye and head geometry adapted for a degenerate retina and subretinal implant, then simplified to improve meshing stability while preserving ocular/orbital tissues. - In vitro measurements: Arrays were immersed in DPBS diluted to 10% to approximate retinal resistivity. 10 ms, 880 nm pulses at 40 Hz (8 mW/mm²) were applied. A 5 µm-tip micropipette measured potential 20 µm above selected pixels vs. Ag/AgCl reference. A field stop was rapidly closed between pulses to switch from full-field to ~1 mm octagonal aperture, enabling observation of transient return behavior in darkened pixels. - In vivo experiments: Royal College of Surgeons (RCS) rats (degenerate) received subretinal implants (n=9 total: 40 µm pixels n=4; 20 µm pixels n=5) placed temporal-dorsal ~1 mm from optic nerve. Long Evans (LE) rats served as normally sighted controls for natural acuity. Anesthesia: ketamine/xylazine; OCT used to confirm ONL loss and implant position. Transcranial screw electrodes were placed over visual cortices with a reference anterior to bregma. Corneal ERG electrodes recorded implant-generated signals. - Prosthetic visual acuity: NIR gratings were pulsed at 64 Hz with 4 ms pulses; pattern reversal every 500 ms (2 Hz). Bar widths on the retina: 13–157 µm. Peak irradiance: 1.2 mW/mm². VEPs were sampled at 2 kHz, averaged over 250 trials; corneal signal served as a template to remove stimulus artifacts and for high-frequency filtering. - Natural visual acuity: Black–white gratings (0.12–9.6 cpd) under continuous illumination with 2 Hz reversal; VEPs averaged over 100–150 cycles. Retinal angular-to-linear conversion was calibrated by retinal photocoagulation (577 nm) at known angular separations; conversion factor 64.3±2.9 µm/deg. - Data analysis: VEP amplitude defined as peak-to-peak within first 100 ms post-reversal. Noise floor measured with static gratings at matched intensity and carrier frequency. Acuity defined as the intersection of a logarithmic fit of VEP amplitude vs. spatial frequency (1/µm) with the noise mean; uncertainty from fit covariance and noise variance (95% CI via delta method).

- Dynamic field confinement via transient return pixels: Preconditioning dark pixels to ~0.5 V forward bias made them conductive during the subsequent frame, sinking current and confining fields. Modeling showed near 100% contrast for alternating 40 µm line gratings with preconditioning, versus <30% without (0.2 V bias), with contrast decreasing with distance from implant and with smaller pixels. - Inter-pixel coupling: FEM estimated nearest-neighbor cross-resistance R1,2 of 15–30 kΩ for 40 µm pixels. Typical per-pixel stimulation currents were 0.1–1 µA; summed neighbor potentials can drive forward bias near the silicon diode turn-on (~0.5 V), enabling transient return behavior. - Threshold advantage of monopolar arrays: With full-field activation, monopolar arrays exhibited a stimulation threshold of ~0.06 mW/mm² with 10 ms pulses, ~30× lower than 40 µm bipolar pixels (1.8 mW/mm²). - In vitro validation: Upon closing the field stop between pulses, pixels transitioning to darkness exhibited a fast negative dip at the onset of the next pulse, indicating current sinking by discharging dark pixels, followed by slower negative swings due to current flow to the large transient return—consistent with model predictions. - Corneal potentials (in vivo) and model agreement: Full-field 4 ms pulses at 1.2 mW/mm² produced corneal potentials of ~2–3.5 mV (varied by animal and electrode position). Temporal waveforms and frequency dependence (1–125 Hz) closely matched model predictions. VEP amplitudes decreased faster with frequency, approaching noise near ~60 Hz (flicker fusion). - Prosthetic grating acuity (RCS rats): Using 64 Hz pulsing and 2 Hz reversal at 1.2 mW/mm², prosthetic acuity with 40 µm pixels matched the pixel pitch; with 20 µm pixels, acuity reached the rat’s natural resolution limit (~28 µm). Sample sizes: 20 µm implants n=5; 40 µm implants n=4. Natural acuity was measured in LE rats (n=6) and used for comparison and calibration. - Current distribution in monopolar configuration: In simulations with adjacent monopolar pixels, ~75% of anodal current was collected by a neighboring cathodal pixel, with the remainder flowing to the global return, demonstrating effective local field confinement. - Safety: With photodiode responsivity 0.51 A/W and photosensitive area about twice the active electrode area, maximum charge injection at 1.2 mW/mm² with 10 ms pulses was ~1.2 mC/cm², corresponding to ~0.2 V electrode polarization given SIROF capacitance (~6 mF/cm²), staying within the −0.6 to +0.8 V water window. The single-diode monopolar pixel’s maximum voltage (~0.6 V) is lower than that of two-diode bipolar pixels (~1.2 V), indicating a favorable safety margin.

The study addresses the central challenge of achieving high-acuity prosthetic vision with small pixels by mitigating crosstalk without sacrificing field penetration. By dynamically converting selected pixels into transient returns via forward-biased diode conduction, the system optically steers current to confine fields laterally and axially. Modeling and in vitro/in vivo measurements demonstrate that this strategy restores high contrast in fine gratings and enables acuity limited by pixel pitch, allowing 40 µm pixels to reach their sampling limit and 20 µm pixels to match the rat’s natural acuity (~28 µm). The approach leverages existing retinal processing by stimulating bipolar cells, preserving center–surround and temporal dynamics. The configurability of field confinement allows tailoring stimulation depth to individual retinal thickness and implant–retina proximity, potentially avoiding direct activation of amacrine or ganglion cells. Practical deployment could use optimization-based control to assign preconditioning and stimulation phases within each 30–33 ms frame; predictive tracking (e.g., Kalman filters) may compensate for eye movements. Sequential activation strategies may further reduce crosstalk but must account for intra-frame eye motion. Overall, dynamic optical current steering provides a scalable path to higher prosthetic acuity in conditions such as atrophic AMD.

A planar monopolar photovoltaic array with optically controlled transient return pixels enables dynamic confinement of the electric field, substantially reducing crosstalk and permitting high-contrast stimulation with small pixels. In rats, this design achieved prosthetic grating acuity matching the pixel pitch for 40 µm pixels and reaching the natural visual resolution limit with 20 µm pixels. Computational models of retinal and ocular fields accurately predicted in vitro and in vivo measurements, supporting the mechanism of field confinement via preconditioned diodes. This technology allows customizable field shaping to match patient-specific retinal anatomy and may enable prosthetic acuity exceeding 20/100 in AMD. Future work should optimize real-time current steering algorithms, integrate eye tracking to mitigate eye-movement-induced misalignment, evaluate performance in larger eyes with subretinal debris, and advance clinical translation.

- Dependence on precise spatiotemporal modulation: Misalignment between projected patterns and pixels (e.g., due to eye movements) can reduce contrast; this is more pronounced with smaller pixels and during sequential activation. - Biological and anatomical constraints: Human AMD often presents subretinal debris separating inner nuclear layer from the implant by ~30–40 µm, potentially limiting achievable resolution versus rodents and altering optimal field penetration depth. - Contrast and distance effects: Without sufficient preconditioning, contrast drops markedly due to crosstalk, and contrast decreases with distance from the implant and with smaller pixel sizes. - Modeling assumptions: FEM and circuit models rely on tissue conductivities, simplified boundary conditions, and geometry approximations (e.g., simplified head model), which, while validated, may not capture all in vivo variabilities. - Operational constraints: VEP responses diminish above ~60 Hz, constraining pulsing rates; intra-frame eye motion introduces uncertainty in transient return placement unless mitigated by eye tracking or higher frame rates. - Generalizability: Results in degenerate rat retinas may not fully translate to human retinal architecture and disease variability; long-term functional performance and perceptual outcomes in primates/humans remain to be established.