Differences in gene and protein levels in long-term versus short-term light damage to vision cells in adult zebrafish

All experiments utilized adult albino (alb) zebrafish (6–9 months of age). Zebrafish were fed three times daily a combination of brine shrimp and flake food. The fish were maintained under a closed water circulation system with a standard circadian light cycle which consists of 14 h of light (∼250 lux) and 10 h of dark at 28.5°C (Westerfield, 2000). The Institutional Animal Care and Use Committee at Wayne State University approved all procedures used in this study (Protocol # 22-02-4412).

The acute light treatment model used for these experiments was exactly as described previously (Kramer et al., 2021). For the chronic low light treatment protocol, fish were only dark adapted for 24-h prior to initiation of the light exposure (vs. 5 days for the acute light treatment), so as to only partially sensitize the animals to the subsequent light. The “0 h” sample set was harvested immediately after dark adaptation. The remaining fish were subsequently exposed to 28 days of continuous exposure to two 250 W halogen bulbs (Woods Home Products, Carrollton, Georgia; Figure 1B). Two rounds of light treatments were conducted with a total of n = 3 animals per tissue collection time-point, resulting in an overall n = 6. The tissue collection time-points were as follows: a non-light treated control group (0 h), 24 h post light onset (hpl), 36 hpl, 72 hpl, 5 days post light onset (dpl), 10 dpl, 14 dpl, and 28 dpl.

Right eyes were removed from euthanized animals then transferred into a 9:1 ethanolic formaldehyde solution overnight at 4°C. Next, eyes were washed in 5% sucrose/1X Phosphate Buffered Saline (PBS) at room temperature (RT) for 1–2 h, and then transferred to 30% sucrose/1XPBS overnight at 4°C. The following day, eyes were transferred to a 1:1 solution of 30% sucrose/1XPBS and Tissue Freezing Medium (General Data, Cincinnati, OH) overnight at 4°C. Next, eyes were embedded in 100% TFM and stored at −80°C, prior to cryosectioning at a thickness of 16 μm. Sections were transferred to Superfrost Plus glass microscope slides (Fisher Scientific, Waltham, MA) and dried on a slider warmer at 55°C for 1–2 h before being stored at −80°C.

Stored slides were re-warmed for 20 min on a 55°C slide warmer. Tissue was rehydrated with 1xPBS for 20 min, and then incubated in blocking solution (2% normal goat serum, 0.2% Triton X-100, and 1% dimethyl sulfoxide (DMSO) in 1XPBS) at RT for 1 h. Slides were incubated overnight at RT with primary antibodies diluted in blocking solution. The next day, sections were washed 3% × 0.05% Tween-20 in 1XPBS (1XPBT). Secondary antibodies along with the nuclear stain TO-PRO-3 (Life Technologies, Grand Island, NY) were diluted 1:500 in 1XPBT. The tissue was washed with 1XPBT after incubation and mounted using Prolong Gold Antifade Reagent (Molecular Probes, Eugene, OR).

Primary antibodies used included: rabbit anti-Rhodopsin antisera (gift from David Hyde; 1:5,000), mouse anti-PCNA (Sigma; 1:1,000), rabbit anti-Blue Opsin (gift from David Hyde; 1:500), rabbit anti-UV Opsin (gift from David Hyde; 1:1,000), rabbit anti-Green Opsin (gift from David Hyde; 1:500), mouse anti-zpr-1 (Zebrafish International Resource Center; 1:200) rabbit anti-GFAP (DakoCytomation; 1:500), mouse anti-4C4 (gift from Peter Hitchcock; 1:250), mouse anti-HuC/D (Life Technologies, Grand Island, NY; 1:50). Secondary antibodies used were all AlexaFluor-conjugated 488 and 594 anti-primary (1:500, Life Technologies, Grand Island, NY).

A terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) assay was performed to detect DNA damage. Slides containing sectioned tissue were removed from −80°C and warmed for 20 min on a 55°C slide warmer. Sections were rehydrated for 20 min in 1XPBS. Slides were then incubated with ApoTag^® Equillibration Buffer (EMD Millipore, Darmstadt, Germany) for 10 min at 37°C. The TUNEL reaction mix was prepared using the TdT dNTP mix (R&D Systems, Minneapolis, MN) and TdT enzyme (Takara Bio, Japan, Korea) according to manufacturer protocol, and slides were incubated for 2 h at 37°C. Slides were then washed in 2X SSC buffer for 15 min at RT, followed by 2 washes with 1XPBS. Slides were then labeled with an AlexaFlour 488 Streptavidin secondary antibody (1:200; Life Technologies, Grand Island, NY) for detection of incorporated dNTPs and TO-PRO-3 (1:500) in 0.1% Tween-20 in 1XPBS for 1 h at RT. Slides were then washed 3 x in 0.1% Tween-20 and a final wash in 1XPBS before being mounted and cover-slipped with Prolong Gold prior to imaging.

A Lecia TCS SP8 confocal microscope was used to acquire all images. Images were taken of the central dorsal retina on a singular plane using the same exposure and gain. Next, images were processed using ImageJ to isolate the fluorescence channel of interest, and the mean fluorescence intensity within a given ROI was quantified (n = 5-6 images quantified per marker). Quantified fluorescent intensities were normalized to the 0 h baseline intensity. PR outer segment lengths were also quantified using ImageJ (Schneider et al., 2012). Three measurements were taken at different areas and averaged for each image quantified. The length measurements were taken in arbitrary units and normalized to the 0 h baseline. For the TUNEL assay, 4c4 staining of microglia, PCNA staining, and cone cell quantification, cells were counted by hand over a 300 µm linear distance in the central dorsal retina. For ONL nuclear counts, TO-PRO-3 positive cells in the ONL proper were quantified (i.e., within the outer limiting membrane). Given that the vast majority of the ONL nuclei are rod PR nuclei, this is commonly used as a proxy for rod PR numbers; however, it should be noted that the ONL also contains UV cone nuclei (Raymond and Barthel, 2004; Lagman et al., 2015). Nuclei for the rest of the cone PRs lie outside of the outer limiting membrane (Raymond and Barthel, 2004; Lagman et al., 2015). For all quantification methods, a one-way ANOVA was used to determine statistical significance across the time-course, followed by a post hoc Tukey test to determine any statistical differences between any time-points.

RGB confocal images were split into the respective channel image components using ImageJ 1.53k (Schneider et al., 2012). An automated image processing pipeline was created using CellProfiler 4.2.4 (CellProfiler Image Analysis Software, RRID:SCR_007358) (Carpenter et al., 2006; Jones et al., 2008; Jones et al., 2009; Kamentsky et al., 2011). A region of interest was manually defined in each confocal image based on TO-PRO-3 signal. This region was then used to mask channel images used for subsequent analyses. Green channel images were converted to grayscale, and the 4c4 fluorescence signal was enhanced using the EnhanceorSuppressFeatures module with neurites feature type and line structures enhancement method. Cell bodies were then identified based on these enhanced 4c4 grayscale images as primary objects, using global otsu two-class thresholding strategy. Overlapping objects were distinguished based on intensity, and dividing lines drawn by propagation. The area of objects was measured in pixels using the MeasureObjectSizeShape module, and reported on an exported spreadsheet. Subsequent statistical analyses were performed using GraphPad Prism 9.

Retinas from left eyes of animals were dissected and processed for RNA isolation as individual biological replicates for each of the eight tissue collection time-points as previously described (Kramer et al., 2021). Notably, with the retinal harvest technique, there is a possibility of some RPE and vitreous inclusion in the sample, however the optic nerve is excised. A set of non-dark adapted 0 h control age-matched adult albino control retinas were also harvested to use as a naïve control group to compare dark adaptations in the AL and CLL models. This purified RNA was submitted to the Genome Sciences Core at Wayne State University for RNA QC, cDNA library preparation, and sequencing. Sequencing libraries were generated from 250 ng of total RNA using Lexogen’s Quantseq 3′mRNA-Seq Library Prep Kit FWD. Sequencing was performed on a NovaSeq (minimum of 5Mreads per sample). Reads were aligned to the zebrafish genome (Build dR10) (Dobin et al., 2013) and tabulated for each gene region (Anders et al., 2015).

Differential gene expression analysis was used to compare transcriptome changes between conditions (6 replicates per condition across two separate experiments) (Robinson et al., 2010). In addition, separate pairwise comparisons were also performed for the 0–5 dpl samples in the CLL dataset and the 5–10 dpl samples in the AL data set for subsequent gene ontology analysis and comparison was performed with the open-source GOrilla platform (Eden et al., 2007; Eden et al., 2009) using a fold change cutoff of greater than 2-fold increase or decrease in gene expression with an FDR of <0.05. Redundant terms were consolidated and visualized in the “Reduce and Visualize Gene Ontology” (REVIGO) platform as an output (Supek et al., 2011). From the REVIGO output, the. tsv file was recovered from the “tree map view,” and formatted for input into the 2D pie chart gene ontology visualization software CirGO to visually represent the GO categories (Kuznetsova et al., 2019). The 13 overlapping genes identified in Figure 6 were input into the STRING platform to asses literature-based functional protein association network relationships (Szklarczyk et al., 2015). Mapped reads for the entire dataset can be found in Supplementary Table S1. The datasets presented in this study can be found in the NCBI Gene Expression Omnibus, accession no: GSE233896↗, and the associated metadata for this submission can be found in Supplementary Table S2.

To characterize and compare the response of the zebrafish retina under chronic low light (CLL) exposure to the classic PR ablative acute light (AL) damage paradigm (Vihtelic and Hyde, 2000; Thummel et al., 2008a; Kramer et al., 2021), we subjected adult albino zebrafish to either 3 days of intense AL followed by 25 days of recovery, or 28-day of CLL (Figures 1A, B). Eye tissue was harvested at the same 8 time-points throughout both light treatment time-courses, with whole right eyes from each animal processed for immunohistochemistry, and dissected retinas from left eyes harvested for single biological replicate transcriptomic studies. The 8 time-points chosen for investigation were largely governed by our previous detailed characterization of the AL damage model (Kramer et al., 2021), which allowed us to directly compare the CLL and AL models in this follow-up work.

Principal component analysis (PCA) was performed on the top 200 genes which maximized separation of the dark-adapted 0 h baseline controls for both the chronic and acute light datasets, such that the distance between each pair of samples represented the root-mean-square deviation for these top genes. This analysis revealed that AL treatment resulted in a counterclockwise trajectory that followed a chronological progression, with early damage time-points (24–72 hpl) diverging most significantly from 0 h baseline controls, followed by a gradual return to near the 0 h baseline at later time-points (10–28 dpl; Figure 1C, grey arrows), which is consistent with the full recovery observed in this model. In the CLL model, 1-day dark-adapted transcriptomes of the samples clustered more closely with the grouping of 10–28 dpl samples in the AL treatment, rather than the 5-day dark-adapted samples in that dataset (Figure 1C), suggesting that the transcriptomes of the CLL-dark adapted retinas align more closely to that of newly regenerated retinas than to 5-day dark-adapted retinas. In addition, after the initiation of light, we observed a considerably higher variability in the time-point clustering in the CLL than in the AL. Interestingly, the CLL time-points were arranged in a linear fashion within the surrounding AL time-points, with the majority of early time-points in the CLL clustered toward the bottom and the majority of the later time-points near the top (Figure 1C).

We next performed a morphological analysis of cone PRs in response to CLL (Figures 2A–H) and AL (Figures 2I–P). In the CLL model, green (medium wavelength, mw) cone outer segments (mwCOS) exhibited a ∼50% truncation in length within the first 24 h of light exposure, which was maintained through the remainder of the time-course (p < 0.0001, Figures 2B–H, Q). Sparse opsin-positive debris appeared posterior to the cone OS boundary at 24 hpl, but subsided by 72 hpl (Figures 2B–H, Q). UV and blue cones exhibited a similar pattern of OS truncation (Supplementary Figure S1), although short-wavelength (blue) cones appeared to be the least impacted by the CLL damage paradigm, as evidenced by a minimal debris field at 24 hpl, and less severe OS truncation (60%–80% normal lengths; Supplementary Figures S1M, T). In contrast, consistent with our previous findings (Kramer et al., 2021), in the AL model, green cones were rapidly ablated by 36 hpl (hours post light onset; Figure 2K). The resultant debris field was largely cleared by 72 hpl (Figure 2L), and new cones emerged by 10 dpl (days post light onset; Figure 2N).

Analysis for red/green double cone somas in the CLL model using the anti-Arrestin3 (Arr3) antibody, zpr-1, confirmed that double cones remained intact throughout the CLL time-course (). In addition, quantification of UV, blue, and green opsin outer segments confirmed that these cones also remained intact during the CLL time-course (). Nevertheless, we observed a translocation of Arr3 from the cone soma to both the cone OS and pedicles between 24–36 hpl (), and the pedicle expression was sustained throughout the CLL time-course (). Finally, the inner retina appeared unaffected in the CLL model based on morphology and HuC/D staining of amacrine and ganglion cells (). Supplementary Figure S2 Supplementary Figure S3 Supplementary Figure S2A–C Supplementary Figure S2H Supplementary Figure S4

Next, we compared the transcripts of the cone PR opsins in the CLL and AL models. We previously demonstrated that the 5-day dark adaptation utilized in the AL model, which sensitizes PRs to light, decreased all PR opsin transcript levels (Kramer et al., 2021). Here we show that the 1-day dark adaptation used in the CLL model also decreased most PR opsin levels, but not to the degree observed in the AL model (Figures 2R, S). In the AL model, transcripts of the dominant Green Opsin paralog opn1mw2 were reduced to zero by 24 hpl, due to the ablation of these PRs (Figure 2S). Transcript levels increased from 5 to 10 dpl as progenitors differentiated into new double cones, and increased even further by 28 dpl, exceeding the expression level of the 0 h control retinas (Figure 2S). Surprisingly, in the CLL model, opn1mw2 transcripts were also reduced to near zero by 24 hpl, even though these PRs remained intact (Figures 2B, S. From this point, opn1mw2 transcripts slowly rose throughout the time-course, but never reached even 50% of the baseline level of the 0 h control retinas (Figure 2S).

It should also be noted that Green Opsin has four paralogues in zebrafish (opn1mw1, opn1mw2, opn1mw3, opn1mw4). In both damage models, all four paralogues were highly expressed in naïve retinas, with opn1mw2 exhibiting the highest expression levels at the dark adapted 0 h baseline (Figures 2R, S; black). However, the different dark adaptation durations of 1 day in the CLL model, and 5 days in the AL model, revealed surprising transcriptional consequences. In the CLL model, the least expressed opsin at naïve control baseline levels, opn1mw1, exhibited an increase in transcript levels of nearly 10-fold during the 1-day dark adaptation (Figure 2R, green), the only opsin across both models to increase with dark adaptation. Next, following an initial decrease in expression at 24 hpl, we observed that opn1mw1 transcripts increased significantly, surpassing the naïve control baseline levels to match the levels of the dominantly expressed paralogue opn1mw2 (Figure 2R, green), which remained well below its baseline level, along with the other two paralogues, opn1mw3 and opn1mw4. The transcriptional kinetics of these paralogues follows a completely distinct trajectory in the AL model as compared to the CLL model. In the AL model, all 4 paralogues showed a significant decrease in transcript levels during the 5-day dark adaptation (Figure 2S). Interestingly, opn1mw3 and opn1mw4 returned to near-naïve levels by 28 dpl, whereas the dominant opsin, opn1mw2, only returned to dark-adapted control levels (Figure 2S). In the AL model, the opn1mw1 paralogue, which was the most active from baseline in the CLL model, had a very minor response, just reaching its low naïve control baseline level. These data suggest regulated differential control of opsin expression in the two damage paradigms.

Next, we compared the response of rod PRs to CLL and AL exposure by labeling the Rhodopsin protein with the zpr-3 monoclonal antibody, which normally labels rod outer segments (ROS; Figure 3). In contrast to the early, maintained phenotypes for cones in the CLL model, rod PRs showed a gradual damage and loss under CLL conditions. By 72 hpl, we observed the beginning of a progressive truncation of ROS which continued through the end of the time-course, at which point the remaining OS were less than 20% of their normal length (Figures 3D–G, Q; p < 0.0001). By 5 dpl we also began to observe a slow loss of nuclei in the ONL over the remainder of the time-course (Figures 3E–H) and reduced rhodopsin gene expression (Figure 3S). From 36 hpl forward, 1-4 TUNEL-positive cells were detected within a 300 µm linear distance in the dorsal retina at each time-point, consistent with a slow loss of nuclei on a daily basis (Supplementary Figure S5) (Vihtelic and Hyde, 2000). By 28 days, a ∼55% reduction in ONL nuclei was observed (0 h baseline: 287 nuc ±18.7, 28 dpl: 126.8 nuc ±26.2). We also observed additional morphological consequences of CLL exposure, beginning with faint zpr-3 expression in the rod inner segments (RIS) starting at 24–36 hpl (Figures 3C, D). This RIS misexpression continued to progressively localize to the inner compartments of rod PRs, resulting in robust perinuclear localization of the zpr-3 signal by 10 dpl (Figure 3F). This finding is consistent with aberrancies seen in various mouse models of retinal degeneration and our previous observations (Kong et al., 2006; Li et al., 2020; Turkalj et al., 2021).

All of these observations were in stark contrast to the AL model, where it is well-established that rod PRs die within the first 24–36 hpl (Vihtelic and Hyde, 2000; Nelson et al., 2013; Kramer et al., 2021). This resulted in extreme hypertrophy of the ROS and disorganization of the ONL nuclei at 24–36 hpl (Figures 3J, K) and a dramatic drop in rhodopsin transcript levels (Figure 3S). A large opsin-positive debris field was still present at 72 hpl (Figure 3L) but was largely cleared by 5 dpl (Figure 3M). Finally, rhodopsin transcript levels increased dramatically from 5 to 10 dpl (Figure 3S), at which point new rod PRs were evident (Figure 3N). Maturation and ROS length increased from 10 to 28 dpl, but never returned to baseline control levels (Figure 3O–Q).

Müller glia/macroglia function together with microglia in a number of developmental, disease, and regeneration contexts (White et al., 2017; Mitchell et al., 2019; Issaka Salia and Mitchell, 2020; Var and Byrd-Jacobs, 2020). In 0 h CLL control retinas, microglia appeared to exhibit two distinct morphologies (Figure 4A; Supplementary Figure S6A). Within the inner and outer plexiform layers, microglia exhibited a ramified morphology with thin projections (Figure 4A, arrowhead). This morphology is associated with microglia in a resting state. In contrast, microglia that resided at the RPE/ROS margin exhibited a more rounded morphology with fewer projections (Figure 4A, white chevron). This morphology is associated with microglia in an intermediate/activated state, but not actively phagocytosing debris. We quantified microglia area in 0 h CLL control retinas and compared microglia with resting morphologies in the inner retina to microglia with rounded morphologies in the RPE/ROS margin and found no statistical difference (Supplementary Figure S6A, p = 0.0524).

In the CLL model, microglia remained at near baseline levels within the first 36 hpl (Figures 4B, Q). Starting at 72 hpl, we observed an increase in the presence of microglia with a rounded morphology at the outer margin of the degenerating ROS (Figures 4D–H, Q). Compared with baseline controls, this increase in number was moderate (p = 0.011 at 10 dpl), but sustained throughout the time-course, and by 14 dpl, microglia with a rounded morphology formed a single line at the RPE/ROS interface (Figure 4G). In contrast, in the AL model, large numbers of microglia with ameboid morphology appeared at 36 hpl (Figure 4K, arrowhead), and the peak of microglia accumulation occurred at 72 hpl (Figures 4L, Q). The morphology of the microglia in the outer retina in the AL model was nearly all ameboid, and the response was nearly 4-fold greater at peak in the AL model vs. the CLL model (Figure 4Q, p < 0.0001). This ameboid morphology is characteristic of active phagocytosis, which is consistent with these microglia clearing the debris field of dead PRs at these time-points. We quantified microglia area at 72 hpl in the CLL and AL models and found that microglia in the AL model with amoeboid morphology had a significantly greater area than microglia with a rounded morphology in the CLL model (Supplementary Figure S6B, p < 0.0001). These data are consistent with a greater phagocytosing activity in the microglia with an amoeboid morphology in the AL model. Finally, at later time-points, microglia slowly decreased in number in the AL model, but did not return to baseline control levels by 28 dpl (Figure 4Q).

As an initial comparison of the transcriptional kinetics of the microglial infiltration in the AL and CLL models, we plotted the gene expression data for two microglial associated genes, mpeg1.1 and apoeb (Butovsky et al., 2014; Ferrero et al., 2020; Thiel et al., 2022). In the CLL model, these genes showed a slow, prolonged wave of elevation by ∼72 hpl, a peak at 10 dpl, and then a slow decline through 28 days, trending towards baseline expression levels (Figures 4R, S; blue line). This expression profile correlated with the peak of activated microglia at 10 dpl in the CLL, and the subsequent plateau in these numbers through 28 dpl. This was in contrast to the AL model, which exhibited a rapid immune response as the PRs were destroyed, with mpeg1.1 peaking at 24 hpl and apoeb at 72hpl, followed by a sharp decrease in expression of both genes towards baseline levels (Figures 4R, S; gray line). In both models, we observed an apparent transcriptional resolution of the activated microglial-associated genes, despite the removal of stimulus in the CLL model. Interestingly, this transcriptional resolution was somewhat in contrast to the moderately elevated numbers of microglia still present at the conclusion of the 28-day time-course.

Using a limited number of time-points, our previous work suggested that MG do not undergo gliosis in the CLL model (Turkalj et al., 2021). Consistent with this report, we observed that MG in CLL did not significantly upregulate the expression of the intermediate filament GFAP, which is a hallmark of reactive gliosis (Figures 5B–H, Q) (Thomas et al., 2016; Ranski et al., 2018). In addition, our transcriptome studies confirmed no significant changes in gfap expression in the CLL model (Figure 5S). These findings were both in contrast to the AL model (Figures 5I–P), wherein MG first exhibited a gliotic response with a consolidation of GFAP in the ONL at 24–36 hpl (Figures 5J, K). This was followed by a more global MG gliosis at 5 dpl, with GFAP expression extending beyond the end feet into the inner retina (Figure 5M). These two waves were also represented transcriptionally, with a significant increase in gfap expression at 24 hpl, followed by a dramatic rise to a larger peak at 72 hpl (Figure 5S).

Next, we compared the proliferation responses in the CLL and AL models. In the CLL model, we confirmed the absence of a robust MG proliferation response in the INL throughout the time-course (Figures 5A–H). Instead, we observed a marked and significant increase in a subset of cells within the ONL that expressed the proliferative marker PCNA at 14 dpl, and continued at 28 dpl (Figures 5G, H; p < 0.0001). Likewise, pcna gene expression showed a mild, but steady increase starting at 3 dpl, which continued through the end of the time-course (Figure 5R). Based on morphology and location, it is highly likely that these PCNA + cells in the ONL represented rod precursors. During constant neurogenesis of the adult zebrafish retina, rod PRs are added to the retina in an injury-independent manner (Stenkamp, 2011). As clusters of MG-derived progenitor cells (MGPCs) were never observed in this model, we hypothesized that the CLL model triggered an upregulation of this injury-independent neurogenesis mechanism. In support of this, we found evidence of both rod precursor proliferation within the ONL and migration of single PCNA + nuclei from INL to the ONL at 14 dpl in the CLL model (Supplementary Figure S7). In contrast, in the AL model, MG reentered the cell cycle at 36 hpl (Figure 5K) and generated a daughter cell that amplified to become a pool of multipotent MGPCs. This robust response was observed transcriptionally with a dramatic increase in pcna expression at 24–36 hpl (Figure 5R). The MGPCs then migrated to the ONL at 72 hpl (Figures 5L, Q), and from 5 to 10 dpl differentiated into new PRs (Figure 2N, 3N) and PCNA positivity decreased throughout the remainder of the time-course (Figures 5N–Q). Collectively, these data suggest very distinct responses from MG in that AL and CL models.

Using transcriptomic data for both AL and CLL models, we probed for genes corresponding early PR degeneration in CLL (0–5 dpl) and PR regeneration in AL (5–10 dpl). We hypothesized that critical PR genes that were significantly downregulated during sustained degeneration in the CLL damage paradigm, would overlap with a subset of genes that were significantly upregulated during regeneration in the AL model. Pairwise analyses between the 0–5 dpl CLL samples and the 5–10 dpl AL samples was performed, and genes with ≥2-fold decreased expression in 0 vs. 5 CLL (FDR<0.05) and ≥2-fold increased expression in 5 vs. 10 AL were categorized by gene ontology (GO) analysis using the GOrilla platform (Eden et al., 2007; Eden et al., 2009). REVIGO was used to consolidate redundant terms, and CirGO software was used to visualize the output in 2D pie chart format (Figures 6A, B) (Supek et al., 2011; Kuznetsova et al., 2019). The majority of differentially expressed genes (DEGs) in both the 0–5 dpl CLL and 5–10 dpl AL datasets were related to phototransduction (light purple GO1 in Figures 6A, B), with categories such as “response to light stimulus” and “visual perception.” The next largest GO term, related to a cellular response to an external stimulus, was also shared between the datasets (slate blue GO2 in Figures 6A, B). Next, when we directly compared the genes within the AL upregulated group (grey) and CLL downregulated group (blue), we found 196 genes in the former category and 22 in the later (Figure 6C). Remarkably, 70% of the CLL downregulated genes were also in the AL upregulated gene set (Figure 6C), confirming our original hypothesis. The 13 genes that overlapped in these datasets included the dominant cone cell opsin, opn1mw2, genes known to be involved in rod phototransduction, such as gnat1, and pde6a (Sun et al., 2018), and a transcription factor known to play a role in retinal development, rx1 (Figures 6C, D). Interestingly, STRING analysis of the 13 overlapping genes pointed to a relational cross-talk between rod and cone PRs in these models (Figure 6D) (Szklarczyk et al., 2015). We found that a hub of rod-specific phototransduction genes (gnat1, pde6a, gc2, and gngt1) have an association with the medium wave-length cone opsins (opn1mw1 and opn1mw2) (Figure 6E), potentially indicating a connection between rod and cones in these time windows.

Next, we probed the CLL dataset for genes with a positive fold-change ≥2 (FDR<0.05) from 0 to 5 dpl. We hypothesized that these genes represent pro-survival networks for rod and cone PRs. The top two GO categories for this group of genes included classifications of “response to stimulus,” “response to stress,” and “regulation of synapse structure” (beige and dark purple GOs, Figure 7A). Various other GO categories related to cell surface signaling and immune system processes (Figure 7A). We also found significant changes in select genes known to encode proteins that play roles in visual perception, DNA repair, and eye development (Figure 7B). For example, the arrestin3b gene (arr3b), encodes a G-protein coupled receptor binding protein required for the termination of cone phototransduction. Similar to GNAT1 in rods, ARR3 translocates from the outer segment to the inner segment of cones following bright light stimulation (Zhang et al., 2003), supporting a role for these proteins in light-dark adaptation. The ddb2 gene encodes a p53-associated protein involved in global DNA repair following UV light damage (Hwang et al., 1999). The sox11b transcription factor is critical in central nervous system development, including retina and lens formation (Wurm et al., 2008; Jiang et al., 2013; Anand and Lachke, 2017) where it plays a role to maintain proper levels of Hedgehog signaling during ocular morphogenesis (Pillai-Kastoori et al., 2014). Finally, rcvrn3 (visinin) encodes a cone-specific recoverin that modulates the recovery phase of the PR response to light stimulation, leading to their prolonged life. This visinin gene (rcvrn3), however, is unique to zebrafish and some species of birds and amphibians and we found that it had an inverse expression pattern in the CLL model relative to its rod counterpart, rcvrna (recoverin) (Figure 7B) (Lamb and Hunt, 2018). Collectively, this dataset offers the field a new set of genetic networks associated with chronic PR degeneration and survival.

At 10 dpl in the CLL model both rod and cone OS were truncated by 40%–60%; however, the cone cells were still present, whereas a loss of some rod PRs was evident based on the thinning ONL (Figures 2, 3). This predilection for damaging rods over cones is similar to what is observed in some human retinal degenerative diseases, such as age-related macular degeneration (Curcio et al., 1996). Unlike the disease state, which is not reversible, the CLL model can be discontinued at any time, allowing us to test whether this chronic stress on PRs is reversible. Specifically, we asked whether the discontinuation of the CLL stimulus would allow for recovery of truncated rod and cone OS and the restoration of rod nuclei. To test this, we subjected the animals to CLL for 10 days, and then returned the remaining unharvested animals back to their normal light:dark housing conditions for 2 weeks (Figures 1, 8). Eyes were harvested for immunohistochemical analysis at 0 h, 10 dpl, and after a 2-week recovery period at 24 dpl (Figure 8).

We observed a significant recovery of both rod and cone OS (Figures 8C, G). Green cone outer segments recovered to ∼75% their original length (Figure 8D), but lacked the classic cone-shaped morphology, with the tips of the outer segments having a frayed appearance (Figure 8C). Rod outer segments also recovered to ∼75% of their length; however, zpr-3 staining was still evident in the RIS. ONL nuclei, which are primarily rod PRs, also recovered to ∼73% of normal from the reduction at 10 dpl (Figure 8H). Finally, despite removal of the light stimulus, we observed a significant number of activated microglia present in the subretinal space near the RPE/ROS boundary compared with undamaged control retinas (Figures 8I, L; p < 0.01). Of note, microglia in the inner retina remained in a resting/ramified morphological state, similar to control retinas (Figures 8I, K). Future studies will be aimed at determining the role of these microglia in CLL damage conditions and recovery.