The present work was performed with adult fish between 12 and 24 months old. Wild-type fish were AB (Oregon) and transgenic lines were careg:EGFP, originally named as Tg(ctgfa:EGFP)^zf620 (also named cnn2a:EGFP; ZFIN Database ID: ZDB-ALT-160802-2) (Chiou et al., 2006), Tg(careg:dmKO)^fri1 (ZFIN Database ID: ZDB-ALT-180626-1) (Pfefferli and Jaźwińska, 2017), Tg(–3.5ubb:loxP-EGFP-loxP-mCherry)^cz1701, referred to as ubi:EGFP (also named ubi:Switch; ZFIN Database ID: ZDB-ALT-110124-1) (Mosimann et al., 2011), and TgBAC(gfap-GFP)^zf167 (ZFIN Database ID: ZDB-ALT-100308-3) (Lam et al., 2009).

The retina injury was induced by MNU treatment and it was performed following the previously established method (Maurer et al., 2014). Briefly, fish were treated for 1 h in 1 L of system water containing 150 mg of MNU. Following the treatments and procedures, fish were kept in 1 L tank apart from the system for 1 day, then reintegrated to the system during the period of regeneration. For scRNA-seq experiment, control fish were treated with inactivated MNU heated at 90°C for 30 min. For chemical inhibition experiments, the animals were pre-treated for 2 days with water containing 1 μM rapamycin (Selleckchem) before MNU-treatment, followed by 2, 7, or 22 days after MNU treatment. A total of 0.1% DMSO was used as control, as solvent for preparation of stock concentration of the drug.

Animal experimentation was performed in accordance with Swiss regulations and approved by the Cantonal Veterinary Office of Fribourg, Switzerland.

At the end of experiments, the eyes were collected in daylight and fixed overnight at 4°C in 4% paraformaldehyde. They were then rinsed in PBS and equilibrated in 30% sucrose for a minimum of 3 h. Lenses were removed before embedding in Tissue-Tek OCT compound (Sakura Finetek Europe B.V.) and cryo-sectioned at a thickness of 16 μm. The immunofluorescence procedures were performed as previously described (Chablais et al., 2011). The following primary antibodies were used: mouse anti glutamine synthetase (GS) at 1:200 (MAB302; Millipore, Billerica, MA, USA), (provided by Enzmann V. Group, University of Bern), mouse anti-proliferating cell nuclear antigen (PCNA) at 1:200 (Clone PC10; M0879; DAKO), rabbit anti-MCM5 at 1:500 (kindly provided by Soojin Ryu, Heidelberg); mouse IgG ZRF-1/GFAP 1:100 (ZIRC, University of Oregon), mouse IgG ZPR-1 1:100 (ZIRC, University of Oregon), mouse IgM Xap-1 (Clone 3D2) at 1:10 (DSHB), mouse IgG3 Xap-2 (Clone 5B9) at 1:10 (DSHB), mouse anti Vimentin at 1:50 (40E-C, DSHB) Chick anti-GFP antibody at 1:500 (GFP-1020; Aves labs); Rabbit anti KO2 at 1:200 (PM051M, MBL International Corporation). The Alexa-Fluor conjugated secondary antibodies (Jackson Immunoresearch) were used at 1:500, and DAPI was used at 1:2,000.

Proliferating cell nuclear antigen staining required heat-induced epitope retrieval in 10 mM citric buffer, pH 6.0, 120°C for 3 min, whereas all other staining were performed without this step, according to standard immunofluorescence staining protocol (Bise et al., 2020).

The fish were dark-adapted for 1 day prior to retina collection. Zebrafish eyeballs were collected on ice, in a darkroom under red light in brown Eppendorf 1.5 ml containing 1 ml ice-cold PBS1x. Under a binocular microscope in ice-cold PBS, each retina was directly extracted from eyeball, minced into small pieces, and placed in 500 μl of ice-cold RNAse-free PBS within 2 min following eyeball collection and for a maximum of 30 min (to avoid RPE sticking on the neuroepithelium).

To dissociate retinal cells, PBS was replaced with a solution of liberase 2.5 mg/ml (Merck, Liberase DH) at 35°C and incubated at 28°C for 20 min. To ensure maximal tissue dissociation, 3–4 rounds of trituration were done using wide bore pipet tips. The enzymatic reaction was stopped by adding 100 μl of 10% BSA.

The resulting cell suspension was then filtered using 40 μm Cell strainer (Corning) and centrifuged at 800 g for 5 min at 4°C. Supernatant was discarded and the pellet resuspended in 500 μl of PBS containing 0.5% BSA. Cell concentration and viability were determined using both MACSQuant flow cytometer and hemocytometer (improved Neubauer). Cell viability was determined using Propidium Iodide (PI) with both methods.

Libraries for scRNA-seq were prepared using the Chromium Single Cell 3′ Library and Gel Bead Kit v3 (10X Genomics), according to the manufacturer’s protocol (User Guide). Briefly, wells were loaded after calculating single-cell suspension concentration of each sample equal to 1,200 cells/μl. Targeted recovery rate was approximately 10,000 cells. 10X Chromium Chip performed GEMs generation, reverse transcription, and cDNA amplification. Illumina NovaSeq 6000 S2 flow cell was used for deep sequencing generating paired-end reads. Different sequencing cycles were performed for the different reads, R1 and R2. R1, contained 10X barcodes and UMIs, in addition to an Illumina i7 index. R2, contained the transcript-specific sequences.

Illumina BCLs data were demultiplexed into FASTQ files using Cellranger mkfastq pipelines v3.0.2 according to the 10X genomic support. To allow EGFP transcripts detection, referenced and annotated genome GRCz11 from NCBI were manually modified to add EGFP sequence. Reads were aligned on referenced genome using Cellranger Count pipeline.

Using Seurat package from Satijas lab in R (v4.1.0) (Stuart et al., 2019), biological duplicates were merged into single conditions and filtered according to metadata. Genes with UMIs <250 and <200 were trimmed. The Scater tool was used to identify low-quality cells based on experiment-specific aspects of the data (McCarthy et al., 2017). This includes cells with a low library size, low number of expressed genes, high proportion of mitochondrial reads.

All datasets were then associated and merged into a single object for all further analysis. The data was normalized using a SCTransform normalization (Hafemeister and Satija, 2019), which builds regularized negative binomial models of gene expression in order to account for technical artifacts while preserving biological variance. During the normalization, we also removed confounding sources of variation (mitochondrial and ribosomal mapping percentage). The quality control report is included (Supplementary Data 1).

An integration analysis (dim = 30) corrected for confounders and batch effects (technical and biological variability) between samples. We identified single-cell clusters through k-nearest neighbors and a shared nearest neighbor modularity optimization. Clusters were determined with a resolution of 0.2 and visualized through UMAP dimensional reduction. Cell clusters were identified by specific marker genes according to previous published studies (Hoang et al., 2020; Ogawa and Corbo, 2021). Subsets of careg:EGFP expressing cells were obtained using the subset function in Seurat v4.1.0.

Genes upregulated in each cluster were identified using the FindAllMarkers function in Seurat v4.1.0 and are provided in Supplementary Table 1. Genes differentially expressed in specific cell types between post-MNU time-points (3, 7, and 10 dpMNU merged datasets) and uninjured control were obtained using the FindMarkers function in Seurat v4.1.0 using non-parametric Wilcoxon rank sum test. The DoHeatmap function of the Seurat package was used to create heatmaps of scaled gene expressions. GO term enrichment analysis were done using the topGO package (Alexa and Rahnenfuhrer, 2022), applying the Fisher’s exact test.

Additionally, all cells from clusters 1 and 9 were subset and reanalyzed following the approach above. Cell clusters were identified by specific marker genes and the new cluster 0–5 were merged into one cluster.

Fluorescent images were taken with a Leica SPE II confocal microscope, and ImageJ 1.53f51 software was used for subsequent measurements. A minimum of 3 retinas were used per experiment and 2–3 representative images were taken per individual (n). From these, one representative image was selected for the relevant figure. Images of the same specimens were taken for area quantification. The area of careg:EGFP or p-rpS6 positive signal was measured using thresholds in ImageJ and was compared with the total area of the analyzed retina fragment.

Quantification of PCNA-positive nuclei was performed using the plugins colocalization and ITCN counter of ImageJ, and was normalized to the number of DAPI-positive nuclei. Error bars correspond to standard error of the mean (SEM). Significance was calculated using one-way ANOVA with Sidák multiple comparison test. Statistical analyses were performed with GraphPad Prism. All results are expressed as the mean ± SEM.

To identify a transgenic tool for monitoring retina regeneration, we investigated the expression of a cis-regulatory DNA element, called careg, which we have previously characterized in the fin blastema and the peri-injured myocardium (Figure 1A; Pfefferli and Jaźwińska, 2017). Bioinformatic analysis of the careg sequence predicted various transcription factor-binding sites, many of which are known to regulate retina regeneration or development, such as pax6, ascl1a, meis1, and oct4 (Bessa et al., 2008; Thummel et al., 2010; Gorsuch et al., 2017; Sharma et al., 2019; Figure 1B). Thus, the expression of this reporter might be regulated in the retina. Firstly, we aimed to investigate whether careg is expressed in the zebrafish eye during development. Between 2 and 5 days post-fertilization (dpf), careg:EGFP was detected in the optic nerve, whereas at 6 and 7 dpf, it was also observed in the retinal pigmented epithelium (RPE) and the corneal stroma (Supplementary Figure 1A). In adult zebrafish, the transgene expression expanded into the iris and certain blood vessels irrigating the eye. Importantly, no expression was observed in the adult retina (Supplementary Figure 1B). These analyses demonstrate that careg:EGFP is not expressed in retinal neurons, the germinal zone and glia cells at any stage.

The retina consists of several layers of specialized cells (Raymond and Barthel, 2004; Figure 1C). To assess the reactivity of the careg:EGFP reporter during retina regeneration, we aimed to damage the rod outer nuclear layer (rONL), which also comprises nuclei of UV cones. We applied a pulse treatment with MNU that causes disintegration of this layer after 5–8 days (Tappeiner et al., 2013). Our analysis of retina sections at 7 days post-MNU treatment (dpMNU) revealed that careg:EGFP was activated in elongated cells spanning all retinal layers, that resemble MG (Figure 1D). To identify these cells, we analyzed three MG markers, glutamine synthetase (GS), the glial fibrillary acidic protein (GFAP, visualized with ZRF-1 antibody), which is confined to a thin segment of MG, and the intermediate filament vimentin (Luna et al., 2010; Gao et al., 2021). We found that 62 ± 2% (n = 4) of careg:EGFP-positive area overlapped with GS staining and displayed a positional association with ZRF1- and vimentin-immunoreactive segments (Figure 1D). We concluded that at 7 dpMNU, the majority of careg-expressing cells are MG, which are known to have a regenerative potential in the zebrafish retina (Fischer and Reh, 2001).

To test the specificity of careg activation after retina injury, we used another transgenic line containing a fluorescent protein with a rapid turnover. Specifically, we selected destabilized monomeric Kusabira-Orange 2 (dmKO2), which contains a C-terminal PEST domain targeted for rapid degradation via ubiquitination (Collery and Link, 2011). Double transgenic fish, careg:EGFP;careg:dmKO2 displayed a partially overlapping pattern of both reporters, whereby nearly all dmKO2-positive cells were also EGFP-positive (Supplementary Figures 2A, B). Furthermore, we used the fish strain TgBAC(gfap-GFP)^zf167, which detects a subpopulation of MG (Lam et al., 2009) and generated double transgenic fish by crossing with careg:dmKO2 (Supplementary Figure 2C). Consistent with previous studies (Conedera et al., 2017), gfap:GFP^zf167 was detected mostly in the thin segment of MG often in close proximity to careg:dmKO2-positive cell (Supplementary Figure 2D). Taken together, we concluded that a population of careg-expressing cells belong to MG, which are activated during retina regeneration.

Activated MG undergo an asymmetric cell division to give rise to a pool of progenitor cells (Nagashima et al., 2013). To investigate whether careg:EGFP was present in proliferative cells, we performed immunofluorescence analysis using G1/S phase markers, namely, PCNA and MCM5, at different time-points after injury. First, we found that the immunostaining pattern of both these antibodies was very similar, validating their specificity (Supplementary Figure 3). The earliest careg:EGFP expression started on day 1 after injury and became evident at 2 dpMNU (Figure 2 and Supplementary Figure 3). Thus, this reporter can function as an early transgenic sensor of tissue activation after damage. At 2 and 3 dpMNU, we observed a remarkable colocalization between careg:EGFP⁺ cells and PCNA⁺ or MCM5⁺ nuclei within the inner nuclear layer (Figure 2 and Supplementary Figure 3), which comprise MG nuclei among other neurons (Figure 1C). On day 7, careg:EGFP⁺ cells contained PCNA⁺ or MCM5⁺ nuclei, however, clusters of small proliferative cells were mostly GFP-negative. This suggests that careg:EGFP is predominantly expressed in activated MG, but not in the derived progenitors.

At 15 and 30 dpMNU, careg:EGFP remained in the inner nuclear layer, whereas a few proliferative cells were detected mostly in the outer nuclear layer; these cells were not demarcated by careg:EGFP. This indicates that the activation of this reporter persists throughout the entire regenerative process in activated MG, including a phase of declining cell proliferation. At 90 dpMNU, the fully regenerated retina contained no careg:EGFP⁺ cells, demonstrating a reversible regulation of the reporter in the repaired organ (Figure 2 and Supplementary Figure 3). Based on these findings, we concluded that careg:EGFP provides a unique transgenic tool for labeling activated MG from the primary to the terminal phase of regeneration.

Methylnitrosourea (MNU) treatment predominantly damages rods, as shown by the Tunel assay and quantification of cells in the outer nuclear layer (Tappeiner et al., 2013). However, the phenotypic characterization of the MNU-injured retina remains incomplete. To better understand which cellular changes may activate careg:EGFP upregulation in MG, we aimed to assess how MNU-injury disrupts photoreceptor organization (Figures 3A, B). We applied the 4C12 antibody that detects rod cell bodies, their inner segments, including myoid, and their outer segments (Sotolongo-Lopez et al., 2016). We combined this marker with Phalloidin, which labels filamentous actin (F-actin) of photoreceptor inner segments and the outer plexiform layer (Nadolski et al., 2020). At 5 dpMNU, 4C12-immunoreactive cell bodies of rods were present, but showed a contracted and disorganized appearance, whereas the myoid structure was no longer detectable (Figure 3D). Furthermore, no overlap between 4C12 and F-actin was apparent in the myoid of the inner segment, suggesting the loss of distinctive rod morphology (Figure 3D). In addition, Phalloidin staining of the outer plexiform layer was abolished after MNU-treatment, suggesting defects at the synaptic ends of photoreceptors and second level neurons. At 30 dpMNU, 4C12 and F-actin displayed a pattern comparable to the uninjured retina, suggesting regeneration (Figure 4A). These data demonstrate that MNU-treatment induces transient degenerative processes in rods, perturbing interconnections between photoreceptors and neurons.

In our attempts to perform lineage tracing analysis with the previously validated careg:Cre-ERT2 driver (Pfefferli and Jaźwińska, 2017), we identified that the ubi:loxP-EGFP-Stop-loxP-mCherry transgene, here abbreviated as ubi:EGFP, is specifically expressed in UV and blue cones in the retina (Figure 3C). The ubi:EGFP was not activated in rods or ZPR-1-immunoreactive double red/green cones (Supplementary Figures 4A, B). Although CreERT2-loxP-mediated recombination was inefficient in the retina, as shown by negligible switching from EGFP to mCherry upon hydroxytamoxifen treatment, we identified that ubi:EGFP alone can serve as a transgenic tool to demarcate UV and blue cones.

We aimed to use this line for testing whether MNU treatment affected any of the ubi:EGFP-positive cone types. To strengthen this approach, we tested two monoclonal antibodies, Xap-2 and Xap-1, which were generated against the Xenopus retina (Harris and Messersmith, 1992). The Xap-2 antigen recognizes outer segments of rods in frogs and killifish (Choi et al., 2011; Berrosteguieta et al., 2022). The Xap-1 antigen has been mapped to Grp78 (Heat shock protein a5), which is detected in outer segments of rods and cones in frogs, but only of cones in mice, monkeys, and pigs (Nookala et al., 2010). The ability of the antibody to bind to photoreceptors has been correlated with proper outer segment formation in cones (Wohabrebbi et al., 2002). Interestingly, we found that both antibodies immunoreacted with the zebrafish retina, whereby the outer segment of UV cones was labeled by both markers (Figures 3E, F). In addition, Xap-2 immunodetected the outer segments of other photoreceptors with a similar intensity as that of UV cones, whereas Xap-1 displayed a dotty pattern that was much more diffuse in other photoreceptors. Double antibody staining with Xap-2 and Xap-1 suggests their colocalization in the outer segment of UV cones (Figure 4B).

Equipped with these new markers, we assessed morphology of ubi:EGFP-positive cones after MNU-treatment. At 5 dpMNU, ubi:EGFP⁺ short UV-cones were missing, and no Xap-2 and Xap-1 immunostaining was observed at the position normally corresponding to the outer segment of these cones (Figures 3E, F). Although retina morphology was disturbed, ubi:EGFP⁺/Xap-2⁺ blue cones were detected in MNU-treated retinas (Figures 3E, F). Similarly, ZPR-1-immunoreactive double red/green cones were also present, although their shape was disturbed, suggesting morphological abnormalities to double cones (Supplementary Figure 4C). We concluded that MNU injury not only damages rods, but also leads to degeneration of UV cones. This finding suggests that these two components of the outer nuclear layer are susceptible to this chemical treatment, triggering the regenerative program that can be monitored by careg:EGFP.

Our immunofluorescence analysis demonstrated that careg:EGFP is activated mainly in a subset of MG. To identify the molecular profile of these cells, we performed a scRNA-seq experiment with retinas dissected from adult careg:EGFP zebrafish. We selected three time-points at 3, 7, and 10 dpMNU, corresponding to the onset, the peak and the progressing-exit of the proliferative phase, respectively. Given that exposure to a chemical compound could on its own induce transcriptional changes, our control retinas were dissected from fish at 3 days after treatment with heat-inactivated MNU (Figure 5A). We confirmed that the administration of heat-inactivated MNU did not cause upregulation of careg:EGFP or injury, and thus, validating this control (Figure 5B).

Single-cell RNA sequencing analysis in Seurat pipeline included elimination of damaged or dying cells and doublets/triplets, which accounted for 12 ± 2% of all cells per condition (Supplementary Figure 5A). Following integration analysis of two replicates per condition, 17 distinct cell populations were identified, whose identity was annotated using known retina markers (Supplementary Table 2; Hoang et al., 2020; Ogawa and Corbo, 2021). Among them, we identified two subpopulations of GABAergic amacrine cells and three groups of bipolar cells, which we merged into one cluster per cell type, as they were beyond the focus of this study. The analysis captured only very few retinal ganglion cells, as none of the clusters was characterized by a unique rbpms2b expression. Interestingly, we identified two groups of rods and cones, each of which we maintained separately, as clusters A and B, respectively (Figures 5C, D). As the topic of our study on photoreceptor regeneration, we analyzed these subpopulations of each photoreceptor type.

Given that the composition of recovered cell types can be influenced by the status of cell adhesion and tissue integrity, we expected to obtain a higher yield of detached rods from MNU-treated retinas, which comprises a damaged nuclear layer. Indeed, the scRNA-seq data showed a much higher proportion of rods at 3, 7, and 10 dpMNU compared to uninjured retina. This result indicates a bias in cell survival and capture efficiency, as reported for other injury models of vertebrate retinas (Macosko et al., 2015; Clark et al., 2019; Hoang et al., 2020).

First, we aimed to identify the distinctive features between the two rod clusters. In uninjured retinas, rod cluster A contained only 91 cells, corresponding to 3.5% of all cells, whereas regenerating retinas contained 6-times more cells in this cluster, namely, 22, 28, and 23% at 3, 7, and 10 dpMNU, respectively (Figure 5E, Supplementary Figures 5B, C, and Supplementary Table 3). This suggests that this population of rods expanded during regeneration.

Comparison between rod cluster A versus B reveled approximately 400 differentially expressed genes (DEGs) (Figure 6A and Supplementary Table 4). Interestingly, both subpopulations displayed an inversed enrichment of paralogous genes for three phototransduction factors, namely, rhodopsin/rhodopsin-like (rho/rhol), phosphodiesterase 6 gamma paralog a/b (pde6ga/b), and guanylate cyclase activator 1 paralog a/b (guca1a/b) (Figure 6B). To assess the functional difference between both cell populations, we performed Gene Ontology (GO) analysis of their DEGs. The cluster A comprised factors involved in translational processes and oxidative phosphorylation, suggesting increased cellular growth and metabolism (Figure 6C and Supplementary Table 4). Unlike cluster A, cluster B was characterized by genes linked to visual perception, photoreceptor cell outer segment organization, non-motile cilium assembly, photoreceptor cell development, and circadian gene expression, suggesting a mature status of rods in this group (Figure 6D). In cluster B, we identified several factors associated with retina diseases, such as retinitis pigmentosa (rpgrip1 and rpgrb) and cone-rod dystrophy (prom1 and crx) (Figure 6E and Supplementary Table 4). Importantly, cluster B displayed upregulated expression of the phototransduction pathway components, namely, retinol-binding protein 4-like (rbp4l), recoverin a (rcvrna), G protein-coupled receptor kinase 1a (grk1a), rod arrestin (sagb), guanine nucleotide binding protein, beta polypeptide 1a/b (gnb1a/b), regulator of G protein signaling 9 (rgs9b), phosphodiesterase 6a/b (pde6a/b), guanylate cyclase 2f (gucy2f), guanylate cyclase activator 1b (guca1b), and solute carrier family 24 member 1 (slc24a1) and cyclic nucleotide gated channel subunit beta 1a/b (cngb1a/b). Taken together, cluster A was different than cluster B, due to a lower expression of the visual perception genes, upregulation of translation genes, and a metabolic switch to oxidative phosphorylation. These characteristics suggest that cluster A contains immature rods.

Next, we aimed to better characterized the molecular signature of immature rods. Interestingly, two subunits of the transducin complex (gnat1 and gngt1), were expressed at higher levels in immature than in mature rods (Figure 6E and Supplementary Table 4). In addition, a factor involved in the maintenance of the retina outer nuclear layer, retinal outer segment membrane protein 1b (rom1b), was also upregulated in immature rods. Among new factors, which have not been linked to retina functions, we identified ppdpfa (pancreatic progenitor cell differentiation and proliferation factor a) and meig1 (meiosis/spermiogenesis associated gene 1), which are unique to this cluster (Supplementary Table 1). The latter gene is particularly interesting, because its mammalian ortholog is required for sperm axoneme assembly (Zhang et al., 2009). Given that various ciliated cells express meig1,¹ and ciliogenesis is highly conserved among species (Avidor-Reiss and Leroux, 2015), we predict that meig1 might be involved in the formation of the basal body/axoneme backbone during differentiation of the outer segment in zebrafish rod photoreceptors. This finding suggests that this axoneme-morphogenesis gene demarcates immature/maturating rods during assembly of the connecting cilium for the stabilization of the outer segment.

To uncover the molecular changes in photoreceptors of MNU-damaged retinas, we analyzed DEGs in immature and mature rod clusters between control and at different time point post-MNU treatment. In the immature rod cluster (rod cluster A), we found 74, 122, and 72 DEGs at 3, 7, and 10 dpMNU, respectively, compared to this cluster of uninjured retinas (Supplementary Table 5). In the cluster of mature rods (rod cluster B), we found 178, 148, and 182 DEGs at 3, 7 and 10 dpMNU, respectively (Supplementary Table 6). Within each cluster, many of DEGs were common for subsequent time-points after injury, suggesting persisting changes of the transcriptional profiles during regeneration. The most significant DEGs were graphically depicted in volcano plot for a comparison between 3 dpMNU versus control (Figure 6F). We noticed several common DEGs in both clusters after injury, such as regulators of circadian rhythm, nfil3-5 and aanat2, which were downregulated in both clusters, whereas retinal binding protein 4l (rbp4l) and retinal degeneration 3-like (rd3l, zgc:162144) genes were upregulated. Furthermore, the genes highly expressed in cluster A, such as pde6ga and guca1a, as well as a several oxidative phosphorylation and metabolic genes were also upregulated in both rod clusters (Supplementary Tables 5, 6). Beside these similarities, several interesting differences were also observed between DEGs in both populations of rods. Specifically, mature rods (cluster B) downregulated gnat1 and gngt1, the genes that are enriched in immature rods. On the other hand, immature rods (cluster A) upregulated their marker genes, such as rho and rom1b, in response to injury. Taken together, retina regeneration after MNU-injury is characterized by extensive transcriptomic changes of both immature and mature rods.

Next, we extracted cone clusters A and B, and reanalyzed them to distinguish between UV and non-UV photoreceptors (Figure 7A). The cluster A demonstrated high expression of UV-sensitive opn1sw1, whereas cluster B displayed more expression of other visual opsin-1 genes that are sensitive for non-UV spectra (Figures 7B, C and Supplementary Table 7). Both cone clusters were characterized by a nearly inverse abundance of arrestin 3 (arr3) paralogous transcripts, whereby cluster A highly expressed arr3b, but was devoid of arr3a (Figures 7C, D and Supplementary Table 7). Among upregulated genes in cluster A as compared to B, we identified spock3, efna1b, tgfa, tbx2a, cngb3.2 guca1e, kcnv2b, all of which are associated with UV-light sensitivity (Ogawa and Corbo, 2021). We concluded that cluster A is enriched in UV-cones, whereas cluster B mostly contains non-UV cones.

Next, we analyzed dynamics of gene expression upon injury. In the UV-cone cluster A, we found 95, 18, and 68 DEGs at 3, 7, and 10 dpMNU, respectively, compared to uninjured retinas (Supplementary Table 8). In the cluster of non-UV cones (cluster B), we found 142, 217, and 348 DEGs at 3, 7, and 10 dpMNU, respectively (Supplementary Table 9). These numbers suggest that both clusters of cones were markedly affected in response to injury. Among the most significant DEGs, several genes were similarly deregulated in both clusters. At 3 dpMNU, common downregulated genes were neurod1, opn6b (a non-visual opsin), rorcb, cry3a, nfil3-5, ipmkb, grk1b, and kera, whereas common upregulated genes were rbp4l, guca1c, rcvrn3, ckbb, and cluster-specific arr3 paralogs (Figure 7E and Supplementary Tables 8, 9). We have also identified many deregulated genes that are associated with visual, metabolic and cell survival processes (Figure 7F and Supplementary Tables 8, 9). These massive changes of transcriptome suggest that MNU treatment functionally impaired cone photoreceptors.

In order to determine the molecular identity of careg:EGFP-expressing cells, we extracted all cells expressing EGFP transcripts, and we displayed them in UMAP plots (Figure 8A). While control retinas comprised only 14 EGFP-positive cells (0.54% of all cells), this number was 5- to 10-times higher in regenerating retinas (Figure 8B; 5.40% at 3 dpMNU, 2.64% at 7 dpMNU, and 2.66% at 10 dpMNU). This upregulation is consistent with our immunofluorescence analysis. Among all clusters, EGFP-positive cells predominantly mapped to MG, as EGFP transcripts were detected in nearly 10% of these cells, which is an outstanding proportion among other cell types (Figures 8C, D). Thus, we focused on this cluster for further analysis.

In the MG cluster, careg:EGFP-positive cells showed an enrichment for genes specific to this cell type, including six3b, glulb, gfap, pleca, cahz, icn, txn, and apoeb (Figure 8E and Supplementary Table 10). Consistent with regenerative activation, careg:EGFP-positive cells expressed higher levels of the proliferation-related gene mdka, and differentiation genes, such as crabp1a, id1, ddr1, and lgals2a. Furthermore, we also identified genes encoding extracellular matrix proteins, such as stm, mmp9, hbegfa, col15a1b, and col18a1a, and ion binding proteins, such as fxyd6l, icn, sncga, mt2, selnop, and cabp5a. Altogether, our analysis highlights a molecular distinction between careg:EGFP-negative and careg:EGFP-positive MG during retina regeneration in zebrafish.

Recent studies have identified a relationship between the mammalian target of rapamycin complex 1 (mTORC1) pathway and photoreceptor survival in mice (Wang et al., 2022). The phosphorylated 40S ribosomal protein S6 (rpS6) is commonly used as a readout of mTORC1 activity in various contexts, including neuroscience (Biever et al., 2015). Inflammation-induced mTOR signaling is essential for retina regeneration in adult zebrafish after a stab injury (Zhang et al., 2020). To test whether this pathway is involved in the regulation of regenerative plasticity in the zebrafish retina after a non-invasive chemical injury, we compared the expression of selected relevant genes in our scRNA-seq analysis. Among the relevant DEGs, we identified mtor, ribosomal protein rpS6, and its kinases, rpS6kb1a/b in careg:EGFP-positive MG, suggesting the involvement of TOR signaling (Figure 9A). To address the hypothetical link between this pathway and the retinal response after MNU treatment, we assessed immunoreactivity against TOR regulation by phosphorylation of rpS6 (p-rpS6) in careg:EGFP regenerating retina. Consistent with a recent report on the needle-poke injury model (Zhang et al., 2020), we found that anti-rpS6 antibody was strongly induced at 1, 2, and 3 days post-MNU treatment. At this initial phase, in particularly at 2 dpMNU, we observed a colocalization between careg:EGFP and p-rpS6 (Figure 9B). However, at 7 and 11 dpMNU, this colocalization was no longer present, and p-rpS6 and careg:EGFP+ areas were not overlapping (Figure 9B). These results suggest that TOR signaling is only transiently associated with careg:EGFP-expressing MG shortly after injury, whereas both markers become uncoupled in proliferative progenitors.

To determine if careg:EGFP is regulated by TOR signaling, we inhibited this pathway with 1 μM Rapamycin (Figure 9C). We verified that this treatment completely abrogates p-rpS6 immunoreactivity, as shown at 2 and 7 dpMNU (Figures 9D, E). Using PCNA immunostaining, we found that this treatment also decreased cell proliferation (Figures 9D, F). Despite these effects, careg:EGFP expression remained unaffected (Figures 10A–C). Interestingly, at 22 dpMNU, immunostaining with rod marker 4C12 and Phalloidin revealed that the restoration of rods and photoreceptor synaptic processes was similar between DMSO-treated control and Rapamycin-treated retinas (Figures 10D, E). We concluded that TOR signaling is activated in MG to increase cell proliferation, but is not essential for careg:EGFP expression and the subsequent restoration of retina morphology in the MNU-induced injury model.