Electronic photoreceptors restore vision in blind rats to natural-resolution level

Electronic photoreceptors restore vision in blind rats to natural-resolution level

Retinal degenerative diseases, such as age-related macular degeneration (AMD), are the leading cause of visual impairment and blindness, affecting millions of patients globally, including my own grandfather. At the age of 96, he has been struggling with AMD for over a decade - losing his ability to read, to watch TV, and to recognize faces has drastically diminished his quality of life. The advanced stage of atrophic AMD has no available treatment due to the irreversible loss of the photoreceptors, photosensitive cells which convert light into visual signals in the retina. Currently, retinal prosthetics remain the most promising approach to restoring sight in these conditions. Leveraging the remaining intact neural circuits in the retina and in the brain, we can electrically stimulate the secondary retinal neurons, primarily bipolar cells, to elicit visual percepts and thereby provide prosthetic vision. Figure 1 illustrates the anatomical structure of a healthy retina and a degenerated retina interfaced with our electronic photoreceptors.

Figure 1. Diagrams of the healthy and degenerate retinas. a) Diagram of the cellular layers in healthy retina, including the Choroid (Ch), Retinal Pigmented Epithelium (RPE), Photoreceptors (PR), Horizontal cells (HC), Bipolar cells (BC), Amacrine cells (AC), Inner Plexiform Layer (IPL), and Ganglion cells (GC), whose axons comprise the Nerve Fiber layer (NFL). b) Degenerate retina with “electronic photoreceptors”: a subretinal prosthesis composed of 20 μm wide and 30 μm deep photovoltaic pixels, which convert incoming light into electric current flowing through the tissue and polarizing the nearby secondary neurons.

AMD patients suffer from the loss of high-acuity central vision, while retaining their blurrier peripheral vision (Figure 2a). With the subretinal PRIMA implants (Pixium Vision, Paris, France), patients in clinical trials demonstrated a prosthetic letter acuity closely matching the implant pixel size of 100 μm, corresponding to the Snellen range of 20/438–20/565. Furthermore, patients report simultaneous perception of central prosthetic vision and peripheral natural vision1,2. Despite such inspiring proof-of-concept, prosthetic visual acuity still needs significant advancement from state-of-the-art prototypes for wider adoption by patients and greater improvements in quality of life.

As with natural vision, prosthetic visual acuity is fundamentally limited not only by the spatial resolution of the stimulation patterns (i.e. pixel size), but also by contrast, which is affected by crosstalk between the neighboring pixels. To achieve prosthetic visual acuity of 20/200 (the US legal blindness limit and the top row in the eye chart), the pixel size needs to be 50 μm at most, while 20/80 acuity requires 20 μm pixels. Figure 2 demonstrates what an atrophic AMD patient would see with and without the central prosthetic vision provided by our 20 μm pixels.

Figure 2. Simulated visuals for an atrophic AMD patient. a) What atrophic AMD patients see with lost central high-acuity vision and remaining low-acuity peripheral vision. b) Simulation of what atrophic AMD patients would see with natural peripheral vision and prosthetic central vision, enabled by our 20 μm pixels implanted subretinally.

Scaling down pixel size is a nontrivial challenge, however, as the existing “bipolar” pixel design with circumferential return electrodes in each pixel over-constrains the penetration depth of electric field into the retina for small pixels (Figure 3a). Our previous studies have shown that the amount of charge necessary to stimulate retinal neurons increases rapidly with decreasing pixel size, exceeding the safety limit of charge injection even for sputtered Iridium oxide film (SIROF), one of the best electrode materials. The maximum stimulation we could safely apply becomes insufficient for cortical measurements of acuity when the pixel size falls below 55 μm in rats3, and for perceptual measurements of acuity below 75 μm in humans4.

Here, we present a novel design of a photovoltaic pixel array that overcomes such limitations and enables high-resolution prosthetic vision. Our new “monopolar” pixels have no circumferential local return electrodes, but a global return ring on the edge of our 1.5 mm wide implant5. To reduce crosstalk and provide good contrast, the lateral field confinement is achieved through spatial-temporal modulation of the light stimulus, which allows a subset of pixels to become transient returns (Figure 3b). Leveraging the adjustable conductivity of photodiodes, we can pre-condition selected pixels with weak light (to generate sub-threshold current for neural stimulation), so that they become sufficiently conductive to serve as transient current sinks in the actual stimulation frame.

Figure 3. Electric field with various implant geometries overlaid on retinal histology. Electric field above the photovoltaic pixels (40 μm in width) of various geometries, with a potential relative to the middle of IPL (57 μm). a) The traditional (bipolar) pixel design containing a central active and a circumferential return electrode. (b) Two new (monopolar) pixels without local returns, where the active electrode of the left pixel acts as an anode and the one on the right – as a cathode, i.e. a transient return. Bottom panels show the corresponding simulated electric fields overlaid on a histology image of a degenerate rat retina. Diagram of a bipolar cell with its axonal terminals in the middle of IPL in the left panel illustrates its position and size with respect to the field penetration depth. The arrows represent current magnitude on a log scale.

We established computational models to simulate the circuit dynamics of our photovoltaic pixel array and derive the electric field distribution in the retinal tissue. The modeling predictions of this optically-controlled current steering method were validated through in-vitro and in-vivo experiments. In vitro, we verified that the contrast was indeed improved significantly with these pre-conditioned pixels acting as transient local returns, comparing with the case where these pixels are idle. We then implanted our photovoltaic “monopolar” pixels of various sizes subretinally in Royal College of Surgeons (RCS) rats, an animal model with retinal degeneration, as well as in Long Evans (LE) rats, the healthy control group. We projected alternating gratings into the rats’ eyes and measured the cortical responses via transcranial electrodes above visual cortices: visually evoked potentials (VEPs). We demonstrated that the grating acuity with 40 μm pixels matches the pixel pitch, while with 20 μm pixels, it reaches the 28 μm limit of natural visual resolution in rats (Figure 4).

Our results indicate that prosthetic visual acuity with 20 μm pixels in rats is no longer limited by sensor size, but rather by their natural limit of retinal and cortical network integration. In humans, these 20 μm pixels correspond to 20/80 visual acuity. Our method also allows customized field shaping based on individual retinal thickness and the distance between the implant and targeted neurons, paving the way to higher visual acuity provided by retinal prosthetics. For the first time, prosthetic vision has surpassed the US legal blindness limit in any live subject. If successfully translated to clinical use, this technology would benefit many more patients suffering from atrophic AMD.

Figure 4. Prosthetic and natural visual acuity in rats. a) Example VEP waveforms in response to alternating gratings in diseased RCS rats (blue) with implants and in healthy LE rats (red). Grating bar width in mm is shown next to each waveform. Asterisk points at the N1 peak. B-D) VEP amplitude as a function of the grating density (in units of the inverse bar width, 1/μm) for prosthetic vision with 20 μm pixels (b), 40 μm pixels (c), and for natural vision (d). The error bars represent the standard deviations centered at mean values, which are derived from n = 5 (b), n = 4 (c), and n = 6 (d) biologically independent animals, respectively. The vertical error bars indicate the standard deviations across animals, while the horizontal error bars in (d) represent the uncertainty in conversion from angular units of resolution (cpd) to linear bar width on the retina (μm). The black dashed lines represent the mean noise level, while the gray bands indicate the noise spread among animals. Red line is a logarithmic fit, which defines acuity as a crossing point with the noise level, indicated by the red arrow. Red band around the fit line represents the 95% confidence interval.

  1. Palanker, D., Le Mer, Y., Mohand-Said, S., Muqit, M. & Sahel, J. A. Photovoltaic Restoration of Central Vision in Atrophic Age-Related Macular Degeneration. Ophthalmology 127, 1097–1104 (2020).
  2. Palanker, D., Le Mer, Y., Mohand-Said, S. & Sahel, J. A. Simultaneous perception of prosthetic and natural vision in AMD patients. Nat. Commun. 13, (2022).
  3. Flores, T. et al. Honeycomb-shaped electro-neural interface enables cellular-scale pixels in subretinal prosthesis. Sci. Rep. 9, 10657 (2019).
  4. Wang, B.-Y., Chen, Z. C., Bhuckory, M., Goldstein, A. K. & Palanker, D. Pixel size limit of the PRIMA implants: from humans to rodents and back. J. Neural Eng. 19, 055003 (2022).
  5. Huang, T. W. et al. Vertical-junction photodiodes for smaller pixels in retinal prostheses. J Neural Eng 17 (2021).

Please sign in or register for FREE

If you are a registered user on Nature Portfolio Bioengineering Community, please sign in