Robust cone-mediated signaling persists late into rod photoreceptor degeneration

Article activity feed

1. 

   Version published to 10.7554/elife.80271 on eLife

   Aug 30, 2022
2. 

   eLife

   Aug 8, 2022

   Author Response

   Reviewer #1 (Public Review):

   In this study, Scalabrino et al. show persistent cone-mediated RGC signaling despite changes in cone morphology and density with rod degeneration in CNGB1 mouse model of retinitis pigmentosa. The authors use a linear-nonlinear receptive field model to measure functional changes (spatial and temporal filters and gain) across the RGC populations with space-time separable receptive fields. At mesopic and photopic conditions, receptive field changes were minor until rod death exceeded 50%; while response gain decreased with photoreceptor degeneration. Using information theory, the authors evaluated the fidelity of RGC signaling demonstrated that mutual information decreased with rod loss, but cone-mediated RGC signaling was relatively stable and was more robust for natural movies than …

   Author Response

   Reviewer #1 (Public Review):

   In this study, Scalabrino et al. show persistent cone-mediated RGC signaling despite changes in cone morphology and density with rod degeneration in CNGB1 mouse model of retinitis pigmentosa. The authors use a linear-nonlinear receptive field model to measure functional changes (spatial and temporal filters and gain) across the RGC populations with space-time separable receptive fields. At mesopic and photopic conditions, receptive field changes were minor until rod death exceeded 50%; while response gain decreased with photoreceptor degeneration. Using information theory, the authors evaluated the fidelity of RGC signaling demonstrated that mutual information decreased with rod loss, but cone-mediated RGC signaling was relatively stable and was more robust for natural movies than artificial stimulus. This work reveals the preservation of cone function and a robustness in encoding natural movies across degeneration. This manuscript is the first demonstration of using information theory to evaluate the effects of neural degeneration on sensory coding. The study uses a systematic evaluation of rod and cone function in this model of rod degeneration to make the following findings: (1) cone function persists for 5-7 months, (2) spatial and temporal changes to the ganglion cell receptive fields were not monotonic with time, (3) mutual information between spikes and photopic stimuli remained relatively constant up to 3-5 months, and (4) information rates were higher for natural movies than for checkerboard noise stimuli.

   The strengths of this paper include the following:

   A systemic evaluation of potentially confusing data. The authors do an excellent job of organizing the results in terms of light levels and time points. The results themselves are confusing and difficult to draw across metrics, but the data are presented as clearly as possible. The work is especially well executed and presented.

   The insight that cone responses remain relatively stable despite rod loss. The study clearly demonstrates that despite cone loss and morphological changes, cone-mediated responses remain robust and functional.

   The application of information theory to degeneration is the first of its kind and the study clearly shows the utility of the metric.

   The results are thoughtfully interpreted.

   We thank the reviewer for these comments.

   The weaknesses of this study include the following:

   The inability to follow the same ganglion cell types over time is a major weakness that could confound the interpretation in terms of whether the changes are happening from artifacts of the recording method or from dynamic changes in the pooled population of ganglion cells. Is there even a single cell class, for example the ON-OFF direction-selective ganglion cells, that this group has so well quantified on the MEA, that the study could track over time, in addition to examining the pooled population changes over time? Tracking a single cell type for each of the metrics would make the population data more convincing or could clearly show that not all ganglion cells follow the population trend.

   As suggested by the reviewer, we have added a cell type that is tracked through all the analyses: ON brisk sustained RGCs. Example receptive field mosaics, temporal receptive fields, and spike train autocorrelation functions for WT and 4M Cngb1neo/neo animals are shown in Figure 2-figure supplement 1E-F. These RGCs follow the trends displayed by the larger populations of RGCs in each analysis. We chose this cell type because they are readily identified by their spike train autocorrelation functions compared to other RGC types and they have approximately space-time separable receptive fields (RFs). There are many text changes associated with adding an analysis of the ON Brisk sustained RGCs (see lines 202-207; 227-229; 264-267, etc).

   We chose not to focus on direction selective RGCs because we are analyzing the spatial and temporal RFs of RGCs in Figures 3-5 and direction-selective RGCs do not have space-time separable RFs (see example in Figure 2C-D). Thus, those cells could not be used to track those receptive field properties across degeneration. Also, we did not collect responses to drifting gratings or bar responses across a range of speeds or contrasts, so we are unable to reliably distinguish the different types of direction-selective RGCs (e.g., ON vs ON-OFF) from these data.

   While the non-monotonic changes are interesting, they are also difficult to make sense of. Can the authors speculate in the Discussion what could be underlying mechanisms that give rise to non-monotonic changes. In the absence of potential mechanisms, the concern of recording artifacts arises.

   Thank you for raising this point. We have added some speculation for the cause of these non-monotonic changes in the Discussion (lines 455-462). “While we do not know why non-monotonic changes are occurring for some RF properties, they largely occurred in the 3-5M range. During this time, there is a transient decrease in the rate of rod death (4-5M) and cone death begins (Figure 1). Consequently, there may be complex changes to retinal circuitry as the retina reacts to a temporary stabilization in rod numbers and an acceleration in cone death. Intracellular studies of the light-driven synaptic currents impinging onto bipolar cells and RGCs during this time will be important for understanding the origin of these non-monotonic changes in RF properties.”

   The mutual information calculation seems to be correlated with the spike rate despite the argument made in Fig 10E-F. Can the authors show this directly by calculating the bits per spike in Figures 8 and 9? Of all the metrics, the gain function and the mutual information seem to be more consistent with each other. Can the authors demonstrate or refute a connection between the spike rate and information rates?

   We added a supplementary figure to each of the information figures (see for Figures 8-10 figure supplement 1) showing the trends hold after dividing the information rate by the spike rate. Certainly, changing spike rates are contributing, but there are also clear changes in the bits/spike plots (Figure 8-figure supplement 1D; Figure 9-figure supplement 1D, Figure 10-figure supplement 1D).

   Can the authors provide an explanation for why the mutual information calculation remains stable despite lower SNR and lower gain, especially after the contributions of oscillations have been ruled out?

   The mutual information depends more strongly on the precision of spiking (both in terms of time and spike number within a small time bin) than the mean spike rate (averaged over the stimulus). Diminishing the total number of spikes (because of reduced gain) will have a relatively small effect on the information rate if the spike trains continue to exhibit low variability (high precision). Indeed, spike generation by RGCs is distinctly sub-Poisson (Berry, Warland, and Meister 1997), indicating it can exhibit relatively high information rates even when spike rates are relatively low. We clarified this in Results at lines 493-496.

   Lack of age-matched WT controls to accompany the different time points. It is known that photoreceptor degeneration can occur naturally in WT mice. Though the authors have used controls pooled from across the ages used in the CNGB1 mutants, it would be informative to know if there are age-dependent changes in any of the metrics for WT mice.

   WT recordings were pooled from retinas from littermate control mice between 2 and 7 months of age (n=3 2M, n=1 each 4M, 6M, 7M). We have added data points from individual retinal recordings to the figure supplements for Figure 2-6 and 8-10 to illustrate the consistency between these recordings, which allowed us to confidently pool the results.

   Can the authors elaborate on why cone function persists despite the rod loss and morphological changes? This is unique for other models of rod loss and is worth extra discussion.

   This is something we are also very interested in, but outside the scope of this study. The Sampath Lab (co-author and collaborator) has data from single cell recordings in late stage rd10 retinas that show abnormal cone signaling (and structure similar to the 7M Cngb1neo/neo cones), yet relatively normal cone bipolar cell and horizontal cell responses. Thus, somehow there is either compensation or a high level of redundancy in the transmission of signals from cones to 2nd-order neurons that makes the responses of the 2nd-order neurons robust to deteriorating cone function. These results suggest our observations in Cngb1neo/neo mice are not unique to this model of RP. Future experiments are needed to understand how this compensation is occurring.

   Reviewer #2 (Public Review):

   In this study, the authors assess the decline of retinal function in a mouse model of slow photoreceptor degeneration - the Cngb1neo/neo. Rod loss occurs between 1-7 months and complete cone loss occurs by 8-9 months. The authors characterize cone loss in the first 7 months and find that 70% of cones are still there at 7 months, though their outer segments are highly degraded. They then use MEA recordings to characterize retinal function using a variety of measures. First, they use spike-triggered averaging to determine the spatial and temporal receptive fields, restricting this analysis to RGCs that have separable spatial and temporal receptive fields. They find that both rod and cone receptive fields are surprisingly intact over the first 5 months, identifying primarily a reduction in contrast response functions (and a reduction in the number of rods that are light responsive-though this is not quantified). Second, they show that oscillatory activity does not appear until after photoreceptors are completely deteriorated-in sharp contrast to other PR degeneration models (e.g. rd10) in which oscillatory activity appears while there are still light-evoked responses. Third, they use information theory to assess the reliability of signaling. When examining the 10% of RGCs with the highest information rates they see a significant decrease at mesoscopic light levels, while information rates were mostly stable at photopic light levels. Finally, they showed that at photopic light levels, the mutant retinas conveyed more information about natural movies than a repeating checkerboard, and this was maintained across light levels.

   My primary question is whether this represents a significant advance. There have been many studies regarding the changing retinal circuits in various rodent models of photoreceptor degeneration. The authors make a few arguments regarding the uniqueness of this study.

   One is that this is a novel analysis that is not limited to particular cell types but rather characterized the retinal as a "whole". But in this point is also its weakness. First, one cannot speak to the retinal as a "whole" since they state that there is a reduction in the number of light-responsive cells across degeneration - yet they do not quantify it. This seems incredibly important to know because even presuming the remaining cells have perfect receptive field structure if only 10% of cells are left, assessing the receptive fields of only the remaining cells is clearly not a characterization of the retention of visual function.

   We never claim that we have assessed the “retina as a whole”. We do state that we are measuring certain features of RGC signaling that reflect the “net changes” induced by photoreceptor degeneration (e.g., changes in photoreceptor function, retinal rewiring, homeostatic mechanisms, etc.) on those features. In fact, we are explicit that we are only measuring certain RF properties in certain RGC types, such as the linear spatial and temporal RFs in cells with space-time separable RFs: Figure 2 makes this point explicitly. We do not measure changes in direction-selectivity, object motion sensitivity, orientation selectivity, edge detection, looming detection, luminance encoding, chromatic opponency, contrast adaptation, motion reversal signaling, etc., because doing so would produce a manuscript with at least one figure for every RGC type (e.g., 45 figures). This would clearly be an unreasonable amount for a single study.

   We agree with the Reviewer that explicitly quantifying the number of light responsive RGCs is important, and we now include this information as a function of degeneration time point in Figure 2-figure supplement 1. Under photopic conditions, this fraction is quite stable until 5M and then begins to deteriorate. We also observe a decrease in the number of RGCs with space-time separable RFs at 5M (Figure 2F), suggesting (but not proving) that these RGCs are representative of changes across all RGCs. We also described these results in the Results (lines 167-174).

   Second, it is hard to assess whether this mouse model is better than existing models for human disease. Their phenotype is different than the rat model of this same disease. It also shows a lack of oscillatory activity that is apparent in rd models.

   We are not making the claim that this model is better than other models. Each model has value. However, because the degeneration in this model is relatively slow, it may be more representative of changes that occur in slower forms of human retinal degeneration (emphasis on “may be”). This is a discussion point, not something that we are aiming to prove. We also believe the utility of a model depends on the questions being asked. In this case, we aimed to track changes over time during photoreceptor loss to better understand the extent to which retinal output is impaired.

   Also, retinitis pigmentosa is a heterogenous disease with a spectrum of phenotypes that may or may not be genotype specific. A patient with a PDE6B mutation presents with differing phenotypes than a patient with CNGB1 mutation, despite both having an RP diagnosis. It is fallacy to assume a mouse is the exact same as a human, just as it is incorrect to assume clinical presentations are identical for all patients for one broad disease that is known to have a diverse set of underlying causes. Studying a range of models is thus essential to understanding the disease. Given that mutations causing RP have different impacts on retinal signaling, we believe it is important to contextualize findings to their mutation. We make this point in Discussion: Comparison to previous studies of RGC signaling in retinitis pigmentosa (beginning on line 436).

   Finally, the model we study does not lack oscillatory activity, it simply arises later than in rd1 or rd10 mice and does so only after all the photoreceptors have died (Figure 7). To our knowledge, it is not clear when or even if RGCs exhibit oscillations in human patients with RP. We discuss why oscillation might arise at different time points in different genetic models of RP in lines 555-570.

   Reviewer #3 (Public Review):

   In the manuscript by Scalabrino et al. a rigorous characterization of the functionality of retinal ganglion cells in a mouse model of rod photoreceptor degeneration is presented. The authors analyzed the degeneration of cone photoreceptors, which is known to be linked to rod degeneration. Based on the time course of cone degeneration they investigated the functional properties of retinal ganglion cells aged between 1 month and seven months.

   The most interesting finding is robust preservation of functional properties, as reflected in little changes of the receptive fields (spatial and temporal characteristics) or signaling fidelity/information rate. In contrast to other mouse models, the present one shows no oscillatory activity until a complete loss of cone photoreceptors occurred at an age of nine months.

   Although the receptive fields of retinal ganglion cells remain nearly intact, the number of ganglion cells with identifiable receptive fields decreases significantly with age (Fig.2F). Could the authors comment, if this might imply a "patchy" vision?

   Visual field loss is a predominant clinical observation in patients with retinitis pigmentosa, including those with Cngb1 mutations. We connect to this observation in the Discussion at lines 521-529: “At the latest stages of photoreceptor degeneration in the Cngb1neo/neo mice (5-7M), we did observe a decrease in the fraction of RGCs with spike rates that were strongly modulated by checkerboard noise (Supplemental Figure 2). It is possible these RGCs were losing their light response completely, or that changes in their light response properties made them relatively unresponsive to checkerboard noise. If the former, it is possible that light responsive RGCs are becoming sparser at the later stages of degeneration which may result in inhomogeneous, or “patchy”, visual sensitivity described by RP patients (see reviews by Hull et al., 2017; Nassisi et al., 2021).”

   Reviewer #4 (Public Review):

   Scalabrino et al. report the remarkable persistence of cone-driven retinal ganglion cell responses in a mouse model of retinitis pigmentosa (i.e., Cngb1 KO mice). The authors first map the time course of primary rod and secondary cone degeneration in Cngb1 KO mice. Approximately 30% of rods are gone at one month (1M), and all rods are lost by 7M in Cngb1 KO retinas. The cone morphology changes progressively as rods degenerate, cone outer segments shrink and are largely absent by 5M. Cones die between 8-9M. Scalabrino et al. next perform multielectrode array recordings from wild-type and Cngb1 KO retinas from 1M to 5M in mesopic and photopic stimulus conditions. They find that spatiotemporal receptive fields remain relatively stable in the face of photoreceptor degeneration, whereas contrast gain gradually decreases. Oscillatory spontaneous ganglion cell activity emerges late (~9M) in Cngb1 KO mice compared to other retinal degeneration models. Finally, the authors analyze mutual information between stimuli (white noise and naturalistic movies) and ganglion cell spikes trains and find that the encoding of the most informative ganglion cells is preserved relatively late into photoreceptor degeneration and that information rates decline less in photopic vs. mesopic conditions and for naturalistic movies vs. white noise stimuli.

   Overall, this is an exciting study that shows remarkable preservation of cone-driven ganglion cell light responses in advanced stages of a retinitis pigmentosa model when most rods have died, and cone morphologies are dramatically altered. The results are presented clearly in the text and figures and are scholarly discussed. Nonetheless, the authors should address a few specific comments to clarify and better support some of the conclusions they draw.

   Specific comments:

   1. In describing the results on information encoding, the authors write and show data (panels A of Figures 8-10) that suggest that most ganglion cells, even in recordings from wild-type retinas, respond unreliably to white noise stimuli and naturalistic movies. Why does such a large fraction of cells have such low repeat reliability? Does this reflect unreliable spike detection and sorting, poor cell or tissue health, or true variability in the responses of healthy retinal ganglion cells. The latter does not seem to align with results from patch-clamp recordings targeted to specific ganglion cell types. The limited repeat reliability also raises questions about how well the linear-nonlinear model, which the authors use to compare responses between wild-type and Cngb1 KO mice of different ages, predicts the responses of these cells. Comparing model parameters (receptive field size, temporal filtering, and contrast sensitivity) between genotypes and ages only makes sense if the model is a good description in the acquired datasets.

   We agree with the reviewer that this is an important point to be clear about. In Figures 8-10 some RGCs exhibit high repeatability, some exhibit low repeatability as quantified by their information rates. The reviewer is concerned about those cells with low repeatability and the ability of capturing their responses with an LN model. This is a valid concern, but to be clear, we are not fitting an LN model to cells with low information rates. In Figures 3-6, where an LN model is being used to estimate the spatial and temporal components of the RFs, we are fitting a subset of all the RGCs: those with space-time separable RFs (see Figure 2). Those particular cells exhibit high information rates and highly reproducible responses, and an LN model captures ~60% of the explainable variance in the spike rate (see Figure 2-figure supplement 1A-B; also see lines 157-151). This is typical for LN models that approximately predict the responses of RGCs to checkerboard noise. Thus, we think the LN model reasonably captures the responses of cells for which we use the LN model. The information rate estimates include these cells as well as other cells that are not well described by an LN model. Note, the LN model is not used to calculate the mutual information rates. We have added text in the Results (lines 324-327) to clarify this.

   In addition, the information rates we estimated in mouse are consistent with past studies from guinea pig (Koch et al, 2004 and Koch et al, 2006). We think cells with very low repeatability are not well driven by checkerboard noise or the particular 10s natural movies we showed. We have updated the example neurons to better reflect the reliability of the cells near the median of the MI distributions in Figures 8-10.

   1. The authors should, maybe in figure supplements and parts of the main figures, break results down by recordings. Inter-experimental variability has been well documented (e.g., Shah et al. Neuron 2022, Zhao et al Sci Rep 2020), and it would be reassuring to see that the conclusions drawn by the authors are supported by statistics in which n = number of recordings (e.g., there is a somewhat difficult to explain broadening of temporal filters in 4M Cngb1 KO retinas that recover by 5M).

   We agree that inter-experiment variability can be large and is important to control for. We now show all the analyses broken down by experiment in Supplemental Figures (2, 3, 4, 5, 6, 8, 9, and 10) for each analysis. None of the trends we describe or highlight in the manuscript were driven by inter-experiment variability.

   1. At different points in their manuscript, the authors conclude that their results "suggest that homeostatic mechanisms in the retina serve to compensate for deteriorating photoreceptors" (or similar). I think that this may well be the case. However, in its present form, the study provides no evidence that retinal circuits in Cngb1 KO mice change to preserve function compared to the alternative that the observed stability is evidence for functional redundancy or resilience in retinal circuits (as they are) without the need for adjustments. Distinguishing between these alternatives would be conceptually important. For example, Care et al. Cell Rep 2019 and Care et al. Cell Rep 2020 used partial stimulation to activate fewer photoreceptors and compare light responses in downstream neurons to those in retinas with fewer photoreceptors. Other studies have directly observed changes in circuit wiring in models of retinal degeneration. If the authors cannot provide experimental evidence for homeostatic changes, it would be good to reflect this in the interpretation and discussion.

   The reviewer raises a terrific point and potential alternative interpretation. We agree. We have not been able to identify an equivalent analysis to that in Care et al. 2019 that we can run that will cleanly distinguish between these two possibilities, without doing many more experiments across timepoints of degeneration. We have thus rewritten portions of the Introduction and the Discussion to recognize the potential of this alternative interpretation.

   Introduction (lines 39-44): Alternatively, homeostatic plasticity or redundancy in retinal circuitry may compensate for photoreceptor loss (Care et al., 2020; Lee et al., 2021; Shen et al., 2020). Such mechanisms could facilitate reliable signaling at the level of retinal output, despite deterioration in photoreceptor function. Identifying the extent to which changes in photoreceptor morphology impact retinal output will inform treatment timepoints for gene therapies aimed at halting rod loss to preserve cone-mediated vision.

   Discussion (lines 514-520): There are two potential classes of mechanisms for this compensation. First, homeostatic plasticity has been documented in models of photoreceptor loss in which the retina remodels to preserve signal transmission (Care et al., 2019; Keck et al., 2013, 2011, 2008; Leinonen et al., 2020; Shen et al., 2020). Alternatively, functional redundancy within the circuit could explain how robust retinal signaling is retained longer than the changes in cone morphology would suggest (Care et al., 2020). This study did not distinguish between the two compensation models.

   1. The authors do not attempt to classify retinal ganglion cells into functional types as functional changes from degeneration may confound such classifications. However, it would be beneficial to separate some categorical response types (direction-selective ON-OFF and ON ganglion cells, maybe orientation-selective [horizontal, vertical, ON, OFF] ganglion cells) and compare how their responsiveness, reliability, and information encoding change with degeneration. This would provide additional insights and address concerns that changes caused by degeneration may be obscured by the differences between ganglion cell types in the present analysis.

   We agree. We now track ON brisk sustained RGCs across degeneration time points for the RF analyses and mutual information analyses. These RGCs are likely the ON sustained alpha cells because they generate very large spikes on the MEA as would be expected for cells with large somata. Example receptive field mosaics, temporal receptive fields, and spike train autocorrelation functions for WT and 4M Cngb1neo/neo animals are shown in Figure 2-figure supplement 1E-F. These RGCs follow the trends displayed by the larger populations of RGCs in each analysis. We chose this cell type because they are readily identified by their spike train autocorrelation functions compared to other RGC types and they have approximately space-time separable receptive fields (RFs). There are many text changes associated with adding an analysis of the ON Brisk sustained RGCs (see lines 202-207; 227-229; 264-267, etc).

   We chose not to focus on direction selective RGCs because we are analyzing the spatial and temporal RFs of RGCs in Figures 3-5 and direction-selective RGCs do not have space-time separable RFs (see example in Figure 2C-D). Thus, those cells could not be used to track those receptive field properties across degeneration. Also, we did not collect responses to drifting gratings or bar responses across a range of speeds or contrasts, so we are unable to reliably distinguish the different types of direction-selective RGCs (e.g., ON vs ON-OFF) from these data.

   Read the original source
3. 

   eLife

   Aug 8, 2022

   Evaluation Summary:

   In this study, the authors assess the decline of retinal function in a mouse model of slow photoreceptor degeneration. The authors use a linear-nonlinear receptive field model to characterize functional changes across some RGC populations and information theory to assess the fidelity of RGC signaling. They show remarkable preservation of cone-driven ganglion cell light responses in advanced stages of a retinitis pigmentosa model when most rods have died, and cone morphologies are dramatically altered. The results are presented clearly in the text and figures and are scholarly discussed. However, there are several technical and conceptual concerns with the conclusions that can be drawn.

   (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private …

   Evaluation Summary:

   In this study, the authors assess the decline of retinal function in a mouse model of slow photoreceptor degeneration. The authors use a linear-nonlinear receptive field model to characterize functional changes across some RGC populations and information theory to assess the fidelity of RGC signaling. They show remarkable preservation of cone-driven ganglion cell light responses in advanced stages of a retinitis pigmentosa model when most rods have died, and cone morphologies are dramatically altered. The results are presented clearly in the text and figures and are scholarly discussed. However, there are several technical and conceptual concerns with the conclusions that can be drawn.

   (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

   Read the original source
4. 

   eLife

   Aug 8, 2022

   Reviewer #1 (Public Review):

   In this study, Scalabrino et al. show persistent cone-mediated RGC signaling despite changes in cone morphology and density with rod degeneration in CNGB1 mouse model of retinitis pigmentosa. The authors use a linear-nonlinear receptive field model to measure functional changes (spatial and temporal filters and gain) across the RGC populations with space-time separable receptive fields. At mesopic and photopic conditions, receptive field changes were minor until rod death exceeded 50%; while response gain decreased with photoreceptor degeneration. Using information theory, the authors evaluated the fidelity of RGC signaling demonstrated that mutual information decreased with rod loss, but cone-mediated RGC signaling was relatively stable and was more robust for natural movies than artificial stimulus. This …

   Reviewer #1 (Public Review):

   In this study, Scalabrino et al. show persistent cone-mediated RGC signaling despite changes in cone morphology and density with rod degeneration in CNGB1 mouse model of retinitis pigmentosa. The authors use a linear-nonlinear receptive field model to measure functional changes (spatial and temporal filters and gain) across the RGC populations with space-time separable receptive fields. At mesopic and photopic conditions, receptive field changes were minor until rod death exceeded 50%; while response gain decreased with photoreceptor degeneration. Using information theory, the authors evaluated the fidelity of RGC signaling demonstrated that mutual information decreased with rod loss, but cone-mediated RGC signaling was relatively stable and was more robust for natural movies than artificial stimulus. This work reveals the preservation of cone function and a robustness in encoding natural movies across degeneration. This manuscript is the first demonstration of using information theory to evaluate the effects of neural degeneration on sensory coding. The study uses a systematic evaluation of rod and cone function in this model of rod degeneration to make the following findings: (1) cone function persists for 5-7 months, (2) spatial and temporal changes to the ganglion cell receptive fields were not monotonic with time, (3) mutual information between spikes and photopic stimuli remained relatively constant up to 3-5 months, and (4) information rates were higher for natural movies than for checkerboard noise stimuli.

   The strengths of this paper include the following:

   A systemic evaluation of potentially confusing data. The authors do an excellent job of organizing the results in terms of light levels and time points. The results themselves are confusing and difficult to draw across metrics, but the data are presented as clearly as possible. The work is especially well executed and presented.

   The insight that cone responses remain relatively stable despite rod loss. The study clearly demonstrates that despite cone loss and morphological changes, cone-mediated responses remain robust and functional.

   The application of information theory to degeneration is the first of its kind and the study clearly shows the utility of the metric.

   The results are thoughtfully interpreted.

   The weaknesses of this study include the following:

   The inability to follow the same ganglion cell types over time is a major weakness that could confound the interpretation in terms of whether the changes are happening from artifacts of the recording method or from dynamic changes in the pooled population of ganglion cells. Is there even a single cell class, for example the ON-OFF direction-selective ganglion cells, that this group has so well quantified on the MEA, that the study could track over time, in addition to examining the pooled population changes over time? Tracking a single cell type for each of the metrics would make the population data more convincing or could clearly show that not all ganglion cells follow the population trend.

   While the non-monotonic changes are interesting, they are also difficult to make sense of. Can the authors speculate in the Discussion what could be underlying mechanisms that give rise to non-monotonic changes. In the absence of potential mechanisms, the concern of recording artifacts arises.

   The mutual information calculation seems to be correlated with the spike rate despite the argument made in Fig 10E-F. Can the authors show this directly by calculating the bits per spike in Figures 8 and 9? Of all the metrics, the gain function and the mutual information seem to be more consistent with each other. Can the authors demonstrate or refute a connection between the spike rate and information rates?

   Can the authors provide an explanation for why the mutual information calculation remains stable despite lower SNR and lower gain, especially after the contributions of oscillations have been ruled out?

   Lack of age-matched WT controls to accompany the different time points. It is known that photoreceptor degeneration can occur naturally in WT mice. Though the authors have used controls pooled from across the ages used in the CNGB1 mutants, it would be informative to know if there are age-dependent changes in any of the metrics for WT mice.

   Can the authors elaborate on why cone function persists despite the rod loss and morphological changes? This is unique for other models of rod loss and is worth extra discussion.

   Read the original source
5. 

   eLife

   Aug 8, 2022

   Reviewer #2 (Public Review):

   In this study, the authors assess the decline of retinal function in a mouse model of slow photoreceptor degeneration - the Cngb1neo/neo. Rod loss occurs between 1-7 months and complete cone loss occurs by 8-9 months. The authors characterize cone loss in the first 7 months and find that 70% of cones are still there at 7 months, though their outer segments are highly degraded. They then use MEA recordings to characterize retinal function using a variety of measures. First, they use spike-triggered averaging to determine the spatial and temporal receptive fields, restricting this analysis to RGCs that have separable spatial and temporal receptive fields. They find that both rod and cone receptive fields are surprisingly intact over the first 5 months, identifying primarily a reduction in contrast response …

   Reviewer #2 (Public Review):

   In this study, the authors assess the decline of retinal function in a mouse model of slow photoreceptor degeneration - the Cngb1neo/neo. Rod loss occurs between 1-7 months and complete cone loss occurs by 8-9 months. The authors characterize cone loss in the first 7 months and find that 70% of cones are still there at 7 months, though their outer segments are highly degraded. They then use MEA recordings to characterize retinal function using a variety of measures. First, they use spike-triggered averaging to determine the spatial and temporal receptive fields, restricting this analysis to RGCs that have separable spatial and temporal receptive fields. They find that both rod and cone receptive fields are surprisingly intact over the first 5 months, identifying primarily a reduction in contrast response functions (and a reduction in the number of rods that are light responsive-though this is not quantified). Second, they show that oscillatory activity does not appear until after photoreceptors are completely deteriorated-in sharp contrast to other PR degeneration models (e.g. rd10) in which oscillatory activity appears while there are still light-evoked responses. Third, they use information theory to assess the reliability of signaling. When examining the 10% of RGCs with the highest information rates they see a significant decrease at mesoscopic light levels, while information rates were mostly stable at photopic light levels. Finally, they showed that at photopic light levels, the mutant retinas conveyed more information about natural movies than a repeating checkerboard, and this was maintained across light levels.

   My primary question is whether this represents a significant advance. There have been many studies regarding the changing retinal circuits in various rodent models of photoreceptor degeneration. The authors make a few arguments regarding the uniqueness of this study.

   One is that this is a novel analysis that is not limited to particular cell types but rather characterized the retinal as a "whole". But in this point is also its weakness. First, one cannot speak to the retinal as a "whole" since they state that there is a reduction in the number of light-responsive cells across degeneration - yet they do not quantify it. This seems incredibly important to know because even presuming the remaining cells have perfect receptive field structure if only 10% of cells are left, assessing the receptive fields of only the remaining cells is clearly not a characterization of the retention of visual function.

   Second, it is hard to assess whether this mouse model is better than existing models for human disease. Their phenotype is different than the rat model of this same disease. It also shows a lack of oscillatory activity that is apparent in rd models.

   Read the original source
6. 

   eLife

   Aug 8, 2022

   Reviewer #3 (Public Review):

   In the manuscript by Scalabrino et al. a rigorous characterization of the functionality of retinal ganglion cells in a mouse model of rod photoreceptor degeneration is presented. The authors analyzed the degeneration of cone photoreceptors, which is known to be linked to rod degeneration. Based on the time course of cone degeneration they investigated the functional properties of retinal ganglion cells aged between 1 month and seven months.

   The most interesting finding is robust preservation of functional properties, as reflected in little changes of the receptive fields (spatial and temporal characteristics) or signaling fidelity/information rate. In contrast to other mouse models, the present one shows no oscillatory activity until a complete loss of cone photoreceptors occurred at an age of nine months.

   Al…

   Reviewer #3 (Public Review):

   In the manuscript by Scalabrino et al. a rigorous characterization of the functionality of retinal ganglion cells in a mouse model of rod photoreceptor degeneration is presented. The authors analyzed the degeneration of cone photoreceptors, which is known to be linked to rod degeneration. Based on the time course of cone degeneration they investigated the functional properties of retinal ganglion cells aged between 1 month and seven months.

   The most interesting finding is robust preservation of functional properties, as reflected in little changes of the receptive fields (spatial and temporal characteristics) or signaling fidelity/information rate. In contrast to other mouse models, the present one shows no oscillatory activity until a complete loss of cone photoreceptors occurred at an age of nine months.

   Although the receptive fields of retinal ganglion cells remain nearly intact, the number of ganglion cells with identifiable receptive fields decreases significantly with age (Fig.2F). Could the authors comment, if this might imply a "patchy" vision?

   Read the original source
7. 

   eLife

   Aug 8, 2022

   Reviewer #4 (Public Review):

   Scalabrino et al. report the remarkable persistence of cone-driven retinal ganglion cell responses in a mouse model of retinitis pigmentosa (i.e., Cngb1 KO mice). The authors first map the time course of primary rod and secondary cone degeneration in Cngb1 KO mice. Approximately 30% of rods are gone at one month (1M), and all rods are lost by 7M in Cngb1 KO retinas. The cone morphology changes progressively as rods degenerate, cone outer segments shrink and are largely absent by 5M. Cones die between 8-9M. Scalabrino et al. next perform multielectrode array recordings from wild-type and Cngb1 KO retinas from 1M to 5M in mesopic and photopic stimulus conditions. They find that spatiotemporal receptive fields remain relatively stable in the face of photoreceptor degeneration, whereas contrast gain gradually …

   Reviewer #4 (Public Review):

   Scalabrino et al. report the remarkable persistence of cone-driven retinal ganglion cell responses in a mouse model of retinitis pigmentosa (i.e., Cngb1 KO mice). The authors first map the time course of primary rod and secondary cone degeneration in Cngb1 KO mice. Approximately 30% of rods are gone at one month (1M), and all rods are lost by 7M in Cngb1 KO retinas. The cone morphology changes progressively as rods degenerate, cone outer segments shrink and are largely absent by 5M. Cones die between 8-9M. Scalabrino et al. next perform multielectrode array recordings from wild-type and Cngb1 KO retinas from 1M to 5M in mesopic and photopic stimulus conditions. They find that spatiotemporal receptive fields remain relatively stable in the face of photoreceptor degeneration, whereas contrast gain gradually decreases. Oscillatory spontaneous ganglion cell activity emerges late (~9M) in Cngb1 KO mice compared to other retinal degeneration models. Finally, the authors analyze mutual information between stimuli (white noise and naturalistic movies) and ganglion cell spikes trains and find that the encoding of the most informative ganglion cells is preserved relatively late into photoreceptor degeneration and that information rates decline less in photopic vs. mesopic conditions and for naturalistic movies vs. white noise stimuli.

   Overall, this is an exciting study that shows remarkable preservation of cone-driven ganglion cell light responses in advanced stages of a retinitis pigmentosa model when most rods have died, and cone morphologies are dramatically altered. The results are presented clearly in the text and figures and are scholarly discussed. Nonetheless, the authors should address a few specific comments to clarify and better support some of the conclusions they draw.

   Specific comments:

   1. In describing the results on information encoding, the authors write and show data (panels A of Figures 8-10) that suggest that most ganglion cells, even in recordings from wild-type retinas, respond unreliably to white noise stimuli and naturalistic movies. Why does such a large fraction of cells have such low repeat reliability? Does this reflect unreliable spike detection and sorting, poor cell or tissue health, or true variability in the responses of healthy retinal ganglion cells. The latter does not seem to align with results from patch-clamp recordings targeted to specific ganglion cell types. The limited repeat reliability also raises questions about how well the linear-nonlinear model, which the authors use to compare responses between wild-type and Cngb1 KO mice of different ages, predicts the responses of these cells. Comparing model parameters (receptive field size, temporal filtering, and contrast sensitivity) between genotypes and ages only makes sense if the model is a good description in the acquired datasets.

   2. The authors should, maybe in figure supplements and parts of the main figures, break results down by recordings. Inter-experimental variability has been well documented (e.g., Shah et al. Neuron 2022, Zhao et al Sci Rep 2020), and it would be reassuring to see that the conclusions drawn by the authors are supported by statistics in which n = number of recordings (e.g., there is a somewhat difficult to explain broadening of temporal filters in 4M Cngb1 KO retinas that recover by 5M).

   3. At different points in their manuscript, the authors conclude that their results "suggest that homeostatic mechanisms in the retina serve to compensate for deteriorating photoreceptors" (or similar). I think that this may well be the case. However, in its present form, the study provides no evidence that retinal circuits in Cngb1 KO mice change to preserve function compared to the alternative that the observed stability is evidence for functional redundancy or resilience in retinal circuits (as they are) without the need for adjustments. Distinguishing between these alternatives would be conceptually important. For example, Care et al. Cell Rep 2019 and Care et al. Cell Rep 2020 used partial stimulation to activate fewer photoreceptors and compare light responses in downstream neurons to those in retinas with fewer photoreceptors. Other studies have directly observed changes in circuit wiring in models of retinal degeneration. If the authors cannot provide experimental evidence for homeostatic changes, it would be good to reflect this in the interpretation and discussion.

   4. The authors do not attempt to classify retinal ganglion cells into functional types as functional changes from degeneration may confound such classifications. However, it would be beneficial to separate some categorical response types (direction-selective ON-OFF and ON ganglion cells, maybe orientation-selective [horizontal, vertical, ON, OFF] ganglion cells) and compare how their responsiveness, reliability, and information encoding change with degeneration. This would provide additional insights and address concerns that changes caused by degeneration may be obscured by the differences between ganglion cell types in the present analysis.

   Read the original source
8. 

   Version published to 10.1101/2022.04.27.489758v2 on bioRxiv

   May 12, 2022
9. 

   Version published to 10.1101/2022.04.27.489758v1 on bioRxiv

   Apr 28, 2022