Optogenetic therapy: high spatiotemporal resolution and pattern discrimination compatible with vision restoration in non-human primates

Vision restoration is an ideal medical application for optogenetics, because the eye provides direct optical access to the retina for stimulation. Optogenetic therapy could be used for diseases involving photoreceptor degeneration, such as retinitis pigmentosa or age-related macular degeneration. We describe here the selection, in non-human primates, of a specific optogenetic construct currently tested in a clinical trial. We used the microbial opsin ChrimsonR, and showed that the AAV2.7m8 vector had a higher transfection efficiency than AAV2 in retinal ganglion cells (RGCs) and that ChrimsonR fused to tdTomato (ChR-tdT) was expressed more efficiently than ChrimsonR. Light at 600 nm activated RGCs transfected with AAV2.7m8 ChR-tdT, from an irradiance of 1015 photons.cm−2.s−1. Vector doses of 5 × 1010 and 5 × 1011 vg/eye transfected up to 7000 RGCs/mm2 in the perifovea, with no significant immune reaction. We recorded RGC responses from a stimulus duration of 1 ms upwards. When using the recorded activity to decode stimulus information, we obtained an estimated visual acuity of 20/249, above the level of legal blindness (20/400). These results lay the groundwork for the ongoing clinical trial with the AAV2.7m8 - ChR-tdT vector for vision restoration in patients with retinitis pigmentosa.


Supplementary Figures
. Information about the animals included in the study.

AAV production
ChrimsonR and ChrimsonR-tdTomato were inserted into an AAV backbone plasmid. The constructs all included WPRE and bovine growth hormone polyA. Recombinant AAVs were produced by the plasmid cotransfection method 1 . The resulting lysates were purified by iodixanol gradient ultracentrifugation, as previously described. Briefly, the 40% iodixanol fraction was concentrated and subjected to buffer exchange with Amicon Ultra-15 centrifugal 3 filter units. The DNase-resistant vector genome titers of vector stocks were then determined by real-time PCR relative to a standard 2 .

Gene delivery, retina isolation and preservation of the primate retina
Primates were anesthetized with a 10 mg/1 kg mixture of ketamine/xylazine. We injected 100 µL of viral vector solution into the vitreous. An ophthalmic steroid and antibiotic ointment was then applied to the cornea. A lethal dose of pentobarbital was administered two months (± 5 days) or 6 months (± 9 days) after AAV injection. Eyeballs were removed and placed in sealed bags for transport with CO 2 -independent medium (Thermo Fisher Scientific), after puncture of the eye with a sterile 20-gauge needle. Retinas were then isolated and conserved as retinal explants in an incubator for 12 to 36 hours before recording. Hemifoveal retinal fragments were transferred onto polycarbonate Transwell inserts (Corning), which were placed on Neurobasal + B27 medium for conservation in the cell culture incubator. In these conditions, natural photoreceptor responses were abolished, and did not recover. We applied pharmacological blockers to check that all natural responses were suppressed (see below).

Two-photon live imaging and single-cell electrophysiological recordings
A custom-made two-photon microscope equipped with a 25x water immersion objective (XLPLN25xWMP, NA: 1.05, Olympus) and a pulsed femto-second laser (InSight™ DeepSee™ -Newport Corporation) was used for imaging ChR-tdT-positive retinal ganglion cells. AAVtreated macaque retinas were imaged in oxygenized (95% O 2 , 5% CO 2 ) Ames medium (Sigma-Aldrich). For live two-photon imaging, whole-mount retinas were placed in the recording chamber of the microscope (ganglion cell layer side up), and images and z-stacks were acquired 4 with an excitation laser used at a wavelength of 1050 nm. Images were processed offline with ImageJ software.
We used an Axon Multiclamp 700B amplifier for whole-cell patch-clamp and cell-attached recordings. Patch electrodes were made from borosilicate glass (BF100-50-10, Sutter Instruments) pulled to 6-9 MΩ. Pipettes were filled with 112.5 mM CsMeSO 4 , 1 mM Mg SO 4 , 7.8 × 10 -3 mM CaCl 2 , 0.5 mM BAPTA, 10 mM HEPES, 4 mM ATP-Na 2 , 0.5 mM GTP-Na 3 , 5 mM lidocaine N-ethyl bromide (QX314-Br) (pH 7.2). We clamped the cells at a potential of -60 mV (the reversal potential of Cl -), to isolate excitatory currents. Recordings were also made in the cell-attached configuration, with pipettes filled with Ames medium. Retinas were darkadapted by incubation for at least 30 minutes in the dark, in the recording chamber, before recordings. All the recorded responses were ON responses. We checked that these ON responses  applied in the bath, via the perfusion system, for 10 minutes before recordings. Action potentials were detected on the filtered electrode signal (2 nd order high-pass Butterworth, cut-off frequency 200 Hz) with a threshold of 4 x the SD of the signal. Spike density function, calculated as previously described 3,4 , was averaged over repeat simulations and used to measure maximal firing rate in a window corresponding to stimulus duration plus 50 ms. When comparing responses for different vector constructs at different light intensities, we computed, for each electrode, the additional firing rate as the maximal firing rate minus the spontaneous firing rate for this electrode, which is calculated as the mean firing rate in the 2 seconds preceding stimulation.

Confocal imaging and quantification
After MEA experiments, the tissue was recovered and fixed by incubation for 30 minutes at We estimated the position of the center of the fovea from the center of a disk drawn to represent the shape of the fovea. Our retina explants were cut to obtain hemifoveae on two explants, but the cut was not at 180°. For each explant, we estimated the angle covered by the retina explants, using the calculated center of the fovea, and used it to estimate the total density of ChR-tdTexpressing RGCs relative to eccentricity.

Descriptive statistics for population responses on multielectrode arrays
In analyses of response patterns for stimuli of increasing duration (1 ms to 2000 m, Fig. 3) we calculated the time to first spike and Fano factor. Time to first spike was determined as the time between the onset of stimulation and the next spike on the electrode. For each electrode, this value was averaged over all repeats for a specific duration. We then constructed a box-plot for each duration, for all electrodes in all experiments. The Fano Factor, F, was calculated for each electrode as: Where is the mean spike variance, and is divided by , the mean spike count.

Letter discrimination based on the population response
Three different shapes (X, circle and square) were presented over the retina, moving in a centerout configuration, in eight directions. The shapes were created with Snellen chart ratios, such that the ratio of width of the individual edges of the shape to the total size was 1:5. Widths of 11 µm to 66 µm were used, creating shapes of 55 µm to 330 µm. Each condition was presented over The spike train S can be considered as a population signature representing each presented stimulus. These spike patterns S were used for pattern discrimination resembling an algorithm 70 developed for event-trains from neuromorphic sensors. The intercell spike intervals were used to create temporal features. For each set S, a feature vector F ∈ R N (N = Number of cells) was created by averaging the temporal feature vectors f for each spike S k (Eq 1).
where t k is the spike time for S k and t i is the last spike at every other cell i and τ is the decay constant dictating the effect of the temporal history of the last spike of the cell (set to 5 ms in this case). Thus, each trial T of a shape is represented by the vector F. Finally, a k-means-based linear discriminator, trained over half the trials, was used to discriminate between the three shapes in the remaining half of the trials. The performance of the feature and k-means method was quantified as the information common to the presented and predicted stimulus shapes, I(L; L p ), (Eq 3) 5 where P(L,L p ) represents the joint probabilities for a presented shape L predicted by the decoder as L p .
10 This measurement is more useful than average discrimination accuracy, as it also incorporates cross-stimulus errors. Indeed, it quantifies both the closeness of the features within the same class cluster and their distance from the features of all other classes. This calculation was performed with the matlab information breakdown toolbox 62 .

Moving bar stimulation and direction estimation
The retina was stimulated with moving bars at two speeds (2.      23 Only the results for the two highest light intensities are given. No significant differences were observed for the other three light intensities. One of the four retinas treated with a medium dose displayed no response to light, two had significantly weaker responses than 5 and 6 (of 8) of the retinas receiving a high dose and illuminated at the highest light intensity. Finally, only one single medium-dose retina ((5E10).BB161i_OG) had a significantly stronger response than three of the eight high-dose retinas, for the highest light intensity only, giving a positive mean difference (bold red). This retina also had a significantly weaker response than four of the highdose retinas.
Supplementary References