Restoration of patterned vision with an engineered photoactivatable G protein-coupled receptor

Retinitis pigmentosa results in blindness due to degeneration of photoreceptors, but spares other retinal cells, leading to the hope that expression of light-activated signaling proteins in the surviving cells could restore vision. We used a retinal G protein-coupled receptor, mGluR2, which we chemically engineered to respond to light. In retinal ganglion cells (RGCs) of blind rd1 mice, photoswitch-charged mGluR2 (“SNAG-mGluR2”) evoked robust OFF responses to light, but not in wild-type retinas, revealing selectivity for RGCs that have lost photoreceptor input. SNAG-mGluR2 enabled animals to discriminate parallel from perpendicular lines and parallel lines at varying spacing. Simultaneous viral delivery of the inhibitory SNAG-mGluR2 and excitatory light-activated ionotropic glutamate receptor LiGluR yielded a distribution of expression ratios, restoration of ON, OFF and ON-OFF light responses and improved visual acuity. Thus, SNAG-mGluR2 restores patterned vision and combinatorial light response diversity provides a new logic for enhanced-acuity retinal prosthetics.

The authors present solid evidence for a photo-engineered G protein coupled receptor with and without a LiGluR that restores light evoked spiking activity to retinal ganglion cells in rd1 mice. The novelty of this approach is the new GPCR that is used that demonstrates superior performance in both sensitivity, timing and duration of effect.
The results are clearly presented with for the most part all controls in place. There are some minor issues with the statistics that are used. Finally, because it has been shown that some of these channels only function in "blind" retinas, it would be useful to know if these receptors function in the WT retina in which synaptic function has been eliminated pharmacologically. Line 48 -of retina is redundant Line 54 -this dual system improve close-line discrimination. What is the comparison? Line 98 "that would turn it on in response to light" -phrase is odd and it is not clear what it means Line 144 -145 localized to the somata and dendrites of both ON and OFF RGCs. -There is no data that definitively shows dendritic label, what the authors show is label throughout the IPL. Line 157 - Figure 1h would be even more convincing if raw data were shown for the condition where LY341495 were included. Figure 1g -The solid bars obscure the response decrease. An outline of the bar would be more appropriate. Figure 1h -how many cells in the number of retinas described? Why is there no statistical comparison? If one is made, the authors need to remember that here as in many of the figures, it is not appropriate to perform a parametric statistic (t-test, ANOVA) on data that are computed as a proportion/percentage, as these data are not normally distributed. Line 166 -loses should be lost Line 180 -500x of what? Figure 2f -please provide statistical comparison Figure 2g, h -The most appropriate analysis here would be a fast fourier transform to examine the power at the frequency of the stimulus. Then a statistical comparison should be made.
In addition, the authors should keep in mind when they plot the inhibitory response that there is a floor effect when using spiking responses, there could be much more inhibition in terms of the cell's membrane potential, but spiking can only go to 0sp/sec. Figure 3 b Stat should be a repeated measure ANOVA.
Figure 3e -f Stat should be a nonparametric ANOVA Figure 4K -Stat should be a nonparametric ANOVA Line 289 -shouldn't the index be a cross correlation index? Figure 5i --Stat should be a nonparametric ANOVA Discussion -It would be useful if the authors discussed how they think that this mGluR2 receptor creates the change in the cell polarization. Could they speculate on what they think might be the downstream target of the receptor and what GPCR cascade is being used.
We thank the reviewers for their helpful comments. We have addressed all of the questions that were raised, as described below. To do so, we added new experiments, analysis and expanded our explanation and interpretation of the results (highlighted yellow in the text and supplement).

Reviewer #1
In their study, Berry et al. deliver a modified GPCR (SNAP-mGluR2), which this group of investigators recently generated and characterized in HEK293 cells ( We thank the reviewer for the positive overview and address the specific issues raised below.

Specific comments 1) RGCs in mice with retinal degeneration (incl. rd1) are spontaneously hyperactive. It is possible that OFF responses generated by SNAG-mGluR2 depend on this increase in baseline firing, which is suppressed during the ON phase of the stimulus with a subsequent rebound at light OFF. The authors should test the effect of SNAG-mGluR2 in wild-type mice when transmission of photoreceptor signals to bipolar cells is blocked by L-APB and ACET.
We have performed the suggested experiment. It appears that SNAG-mGluR2 light response has specificity for the degenerated retina.
SNAP-mGluR2 was virally delivered to RGCs of wt animals as we did in rd1 animals. These retinas displayed a pan-retinal expression pattern that was restricted to the RGC layer ( Supplementary Fig. 4a-c), as seen in the rd1 retina (Fig. 1c,d). Retinas mounted on the MEA and conjugated to BGAG 12,460 had low basal activity in the dark and normal photoreceptor mediated ON and OFF responses when flashed with dim white light that is bright enough to elicit a response from the wt untreated retina, but below the threshold of SNAG-mGluR2 (Fig. 1j,k &  Supplementary Fig. 3j). Administration of 50µM of L-AP4 (agonist of mGluR6 1 ), and 1µM of ACET (blocker of kainate receptors 2 ), which mediate synaptic transmission from photoreceptor cells to ON & OFF bipolar cells, abolished this photoreceptor-driven light response (Fig. 1j,k & Supplementary Fig. 3k). Strikingly, illumination at the an intensity (20 mW/cm 2 ) and wavelength (445 nm) that produces a large SNAG-mGluR2 light response in the RGCs of the rd1 retina elicited little to no light response in the synaptic transmission-blocked wt retina (Fig. 1j,k). When light intensity was increased by 2.5-fold a small light response was detected that was ~7x smaller than that induced in rd1 animals expressing SNAG-mGluR2 and labeled with BGAG ( Fig. 1k & Supplementary Fig. 3l). The mechanism for this selectivity in response may depend, as reviewer #1 and reviewer #2 suggested, on the increase in baseline firing seen in rd1 animals 3,4 . Additionally, the reported increase in Kv7 channel activity and up-regulation of HCN channels in the degenerating retina may also contribute to selective function in the disease model [5][6][7] . Regardless, these results suggests that the signaling of SNAG-mGluR2 occurs selectively (or far more robustly) in the degenerating retina, and that, in early stage of the disease, when only some of the photoreceptors have been lost, this therapy may selectively animate RGCs whose photoreceptor cells have degenerated, with minimal effect on areas whose photoreceptor cells remain intact.

2) It is unclear what conductances are activated by SNAG-mGluR2 in RGCs. This question could be addressed by patch clamp recordings from RGCs and/or pharmacology.
We used pharmacology on the multi-electrode array in effort to identify the downstream target of mGluR2 in RGCs of the rd1 retina. "As expected for the Gi-coupled mGluR2 8 , the light response was absent in retina injected with 150 uM pertussis toxin for 24 hrs (Fig. 1h & Supplementary Fig. 3c). We considered that the inhibitory response could be mediated by activation of a potassium channel. We first tested G protein-coupled inwardly-rectifying potassium (GIRK) channels, since these channels are expressed in RGCs 9 and are activated by Gβγ as a result of activation of mGluR2 10 . However, neither 300 nM Tertiapin-Q nor 1mM barium, potent blockers of GIRK channels 11 , altered the SNAG-mGluR2 light response in the rd1 retina (Fig. 1i & Supplementary Fig. 3e-g). We next considered Kv7 channels, since one of these, Kv7.4, is also activated by Gβγ 12,13 and is expressed in the retina 14,15 . We found that 500 nM linopirdine, a potent and, at this low concentration, selective blocker of Kv7 channels 16 , blocked the SNAG-mGluR2 light response in the rd1 retina (Fig. 1j & Supplementary Fig. 3h). This suggests that the SNAG-mGluR2 inhibitory light response is mediated by activation of the Kv7.4 channel. Lastly, we tested the idea that SNAG-mGluR2 OFF response could be due to hyperpolarization-activated cyclic nucleotide-gated HCN channels, since they are expressed in RGCs 17 and have been established in the mechanism of hyperpolarizationinduced rebound excitation within RGCs 18 . Indeed, we found that 50 µM ivabradine, a blocker of HCN channels 19 , eliminated the excitatory OFF response while maintaining the inhibitory component when the light was on (Fig. 1j  & Supplementary Fig. 3i). Together, these results suggest that SNAG-mGluR2 activates Kv7.4 channels through Gilinked Gβγ coupling to hyperpolarize RGCs and that deactivation of the channels when the light is turned off triggers rebound excitation due to activation of HCN channels.

This is a very well-conceived set of experiments in a manuscript that is a pleasure to read. The authors present solid evidence for a photo-engineered G protein coupled receptor with and without a LiGluR that restores light evoked spiking activity to retinal ganglion cells in rd1 mice. The novelty of this approach is the new GPCR that is used that demonstrates superior performance in both sensitivity, timing and duration of effect.
We thank the reviewer for this positive assessment.
The results are clearly presented with for the most part all controls in place. There are some minor issues with the statistics that are used. Finally, because it has been shown that some of these channels only function in "blind" retinas, it would be useful to know if these receptors function in the WT retina in which synaptic function has been eliminated pharmacologically.
We have performed the suggested experiment and it appears that SNAG-mGluR2 OFF response provides some specificity for degenerative retina. Please see the response to specific comment #1 by Reviewer #1, who posed the same question.

Line 48 -of retina is redundant
Removed.

Line 54 -this dual system improve close-line discrimination. What is the comparison?
Changed to: This dual system improves close-line discrimination over either on its own.

Line 98 "that would turn it on in response to light" -phrase is odd and it is not clear what it means
Changed to: To obtain a native retinal GPCR that endows a rapid light response onto RGCs, we engineered metabotropic glutamate receptor 2 (mGluR2) to contain a nanoscopic chemical photoswitch that activates mGluR2 in response to light.

Line 144 -145 localized to the somata and dendrites of both ON and OFF RGCs. -There is no data that definitively shows dendritic label, what the authors show is label throughout the IPL.
Changed to: and covering the area where the somata and dendrites of both ON-and OFF-RGCs are located.
Line 157 - Figure 1h would be even more convincing if raw data were shown for the condition where LY341495 were included.
We have followed the reviewer's suggestion and added Supplementary Fig. 3a,b which displays a representative MEA recording from rd1 mouse retina expressing SNAG-mGluR2 in RGCs before (a) and after (b) addition of 5 µM LY341495 (a,b: n=71) We have also included traces of MEA recordings as a supplement to the additional experiments suggested by the reviewer (Supplementary Fig. 3). Changed to small bar above denoting duration of illumination.

Figure 1h -how many cells in the number of retinas described? Why is there no statistical comparison? If one is made, the authors need to remember that here as in many of the figures, it is not appropriate to perform a parametric statistic (t-test, ANOVA) on data that are computed as a proportion/percentage, as these data are not normally distributed.
Number of cells has been added to 1h. Additionally we have addressed the reviewer's comments regarding statistical comparisons. All statistical comparisons have been recalculated using nonparametric tests (Supplementary Tables 1  and 4). Figure 1h: To determine significance between drug treatments in cells before and after exposure, a Mann-Whitney U test was applied to the data.

Figure 2f -please provide statistical comparison
The statistical comparison was between the light induced inhibition and OFF response of rd1 SNAG-mGluR2 retina to that of rd1 untreated retina.

Figure 2g, h -The most appropriate analysis here would be a fast fourier transform to examine the power at the frequency of the stimulus. Then a statistical comparison should be made.
In addition, the authors should keep in mind when they plot the inhibitory response that there is a floor effect when using spiking responses, there could be much more inhibition in terms of the cell's membrane potential, but spiking can only go to 0sp/sec.
While an FFT analysis, as for example performed recently to determine the temporal sensitivity of cone and rod mediated responses in wt mouse RGCs 20 , would provide an interesting comparison between the natural photoreceptor-mediated responses and responses mediated by SNAG-mGluR2 expressed in the RGCs of rd1 mice, it is beyond the scope of this study. Here we did not set out to determine peak temporal sensitivity, but instead only endeavored to estimate following frequency, and so our experiments with trains of light pulses simultaneously varied both light pulse frequency and duration.
We agree with the reviewer that there will be a floor effect when spike responses hit zero, while the cell membrane potential may be further hyper-polarized. Indeed, we see spike trains from individual units in several of the figures in which early quiescence during the start of the light pulse relaxes to a low rate of firing, as the average measured spiking rate hits a minimum early and relaxes up. This means that we underestimate the degree of inhibition to some degree under conditions where the inhibition is the strongest, distorting the relationship between pulse duration and inhibition. We now address this in Methods.

Figure 3 b Stat should be a repeated measure ANOVA.
We have followed the reviewer's suggestion and performed a repeated measure ANOVA. Statistical significance calculations for slow release BGAG delivery for behavioral light avoidance (Fig. 3b) was analyzed by repeatedmeasures ANOVA (rANOVA) using "time point" as the within-subject variable and "group" (control untreated, treated) as the between-subject variable. Within-subject effects were analyzed by one-way ANOVA using "time point" as the independent variable. Where sphericity was violated, as assessed by Maulchy's test of sphericity, the Greenhouse-Geisser correction was applied (Supplementary Table 2). So the statistical difference was not significant within subjects across multiple days but was statistically significant between the two groups. This suggests that treated mice show an avoidance advantage over rd1 control animals and display no change in performance over 42 days. This significance was lost at day 48 where treated animals performed no differently from control rd1 animals. Corrected. See Supplementary Fig. 5b,d-g The behavioral performance of all 156 mice that participated in learned avoidance behavior (Fig. 3e-h & Fig. 5i) was reassessed as proportion of success on the behavioral task; an approach commonly employed in nonparametric calculations of clinical trials or behavioral experiments where achieving a normal distribution is unrealistic. For learned dark avoidance behavior Fig. 3e and the learned pattern discrimination behaviors (Fig. 3f-h & 5i), a standard deviation was computed. A success was defined as greater than the sum of the control group average and one S.D, and a failure was any value that did not achieve this criteria. Success ratios were then calculated for each condition ( Supplementary Fig. 5b,d-g). To determine significance in differences between conditions a pairwise contingency table was then constructed, and a Two-Sided Pearson's Chi-Square Test was initially conducted (Supplementary  Table 3). To correct for conditions with a small n, a One-Sided Fisher's Exact Test was also conducted (Supplementary Table 3). In addition to these tests, we request that the behavioral performance analysis on Fig. 3e-h using parametric (ANOVA/t-test) and corrections (Bonferroni) also remain in order to maintain consistency with the behavioral measures of other vision restoration papers.  Corrected -To determine significance between cross correlation values, a Mann-Whitney U test was applied to the data.

Line 289 -shouldn't the index be a cross correlation index?
Correct, changed to cross-correlation index

Discussion -It would be useful if the authors discussed how they think that this mGluR2 receptor creates the change in the cell polarization. Could they speculate on what they think might be the downstream target of the receptor and what GPCR cascade is being used.
Our results suggest that SNAG-mGluR2 activates Kv7.4 channels through Gi-linked Gβγ coupling to hyperpolarize RGCs and that deactivation of the channels when the light is turned off triggers rebound excitation due to activation of HCN channels. Please see the response to specific comment #2 by Reviewer #1, who posed the same question.