Midkine-a Is Required for Cell Cycle Progression of Müller Glia during Neuronal Regeneration in the Vertebrate Retina

In the retina of zebrafish, Müller glia have the ability to reprogram into stem cells capable of regenerating all classes of retinal neurons and restoring visual function. Understanding the cellular and molecular mechanisms controlling the stem cell properties of Müller glia in zebrafish may provide cues to unlock the regenerative potential in the mammalian nervous system. Midkine is a cytokine/growth factor with multiple roles in neural development, tissue repair, and disease.


Introduction
Cell division is an essential biological process during development, homeostasis, and repair. In the CNS of adult mammals, stem cells reside in specialized niches, and these cells maintain the ability to divide and generate new neurons (Kriegstein and Alvarez-Buylla, 2009;Ming and Song, 2011). In the vertebrate retina, Müller glia harbor molecular features of stem and progenitor cells (Dyer and Cepko, 2000). In mammals, Müller glia respond to injury by partial dedifferentiation and entering the G 1 phase of the cell cycle (Bringmann et al., 2006). However, in general, this reprogramming does not lead to cell division, and structural remodeling and the loss of retinal homeostasis are the typical sequalae (Bringmann et al., 2009;Karl and Reh, 2010;Hamon et al., 2016). Importantly, in the limited instances where regeneration does occur, new neurons functionally integrate into existing synaptic circuits (Jorstad et al., 2017;Yao et al., 2018), indicating that in the mammalian retina the limitations of neuronal regeneration hinge on a more complete neurogenic response in Müller glia.
In zebrafish, Müller glia can adopt the features of stem cells (Karl and Reh, 2010;Goldman, 2014;Gorsuch and Hyde, 2014;Lenkowski and Raymond, 2014;Hamon et al., 2016). In uninjured retinas, Müller glia reside in a quiescent state and function to maintain retinal homeostasis. Neuronal death triggers Müller glia to reprogram into a stem cell-like state, enter the cell cycle, and undergo a single asymmetric division to produce rapidly dividing, multipotent retinal progenitors with the ability to regenerate retinal neurons (Nagashima et al., 2013;Lenkowski and Raymond, 2014). Several signaling pathways have been identified that regulate the initial response of Müller glia (Karl and Reh, 2010;Goldman, 2014;Gorsuch and Hyde, 2014;Lenkowski and Raymond, 2014;Hamon et al., 2016). Ascl1, Lin28, and Stat3 have been identified as "core" transcriptional regulators that govern signaling cascades required for Müller glia to divide (Fausett and Goldman, 2006;Ramachandran et al., 2010;Nelson et al., 2012).
Midkine is a growth factor/cytokine that has multiple roles in neural development, repair, and disease (Sakamoto and Kadomatsu, 2012;Winkler and Yao, 2014;Sorrelle et al., 2017). In malignant tumors, Midkine promotes proliferation and metastasis (Muramatsu, 2011) and is also involved in CNS inflammation (Muramatsu, 2011;Weckbach et al., 2011;Herradon et al., 2019). The diverse functions of Midkine are transduced through receptors, which may function individually or as members of a multiprotein complex (Muramatsu, 2011;Weckbach et al., 2011;Xu et al., 2014). During retinal development in zebrafish, midkine-a is expressed by retinal progenitors and functions to govern elements of the cell cycle (Calinescu et al., 2009b;Uribe and Gross, 2010;Luo et al., 2012). Postmitotic neurons downregulate midkine-a. Retinal injury rapidly induces midkine-a in Müller glia (Calinescu et al., 2009b;Gramage et al., 2014Gramage et al., , 2015. Induction of midkine-a following injury has been reported for a variety of tissues with the capacity to regenerate (Ochiai et al., 2004;Lien et al., 2006), suggesting that Midkine may universally regulate aspects of tissue regeneration. The molecular mechanisms whereby Midkine governs regeneration are not well understood.
Using a Midkine-a loss-of-function mutant, we demonstrate that, following a retinal injury, Midkine-a is required for reprogrammed Müller glia to progress from G 1 to S phases of the cell cycle. Following photoreceptor death, Müller glia in Midkine-a mutants reprogram into a stem cell state and enter G 1 phase of the cell cycle. However, for the vast majority of Müller glia, subsequent entry into the S phase and mitotic division are blocked, resulting in failure to regenerate cone photoreceptors. Further, Midkine-a is required for the upregulation of id2a, which inhibits the retinoblastoma (Rb) family of cell cycle inhibitors. In addition, the G 1 -arrested Müller glia undergo reactive gliotic remodeling, hallmark of pathology in the mammalian retina. Finally, we provide evidence that activation of the Midkine receptor, anaplastic lymphoma kinase (Alk), is required for proliferation in Müller glia.

Materials and Methods
Zebrafish. Fish were maintained at 28°C on a 14/10 h light/dark cycle with standard husbandry procedures. Adult WT, AB-strain zebrafish (Danio rerio; ZIRC, University of Oregon, Eugene, OR) and the transgenic reporter line Tg(gfap:eGFP) mi2002 (Bernardos and Raymond, 2006) were of either sex and used between 6 and 12 months of age. All animal proce-dures were approved by the Institutional Animal Care and Use Committee at the University of Michigan.
Western blots. Western blot analyses were performed as previously described (Calinescu et al., 2009a). Briefly, proteins were extracted from the heads of 30 -50 WT and mdka mi5001 embryos or adult retinas (6 retinas from 3 animals per sample) in cold RIPA lysis buffer containing protease and phosphatase inhibitor mixture (Cell Signaling Technology). Proteins were separated in 12% Mini-PROTEIN TGX Precast gel (Bio-Rad) and were transferred to PVDF membranes (GenHunter). After blocking in 5% nonfat dry milk in Tris-buffered saline containing 0.3% Tween 20, membranes were incubated with rabbit anti-Midkine-a antisera or rabbit anti-STAT3 (Nelson et al., 2012) followed by HRPconjugated secondary antibody (1:1000) (Calinescu et al., 2009a). Immunolabeled proteins were detected using the enhanced ECL detection system for chemiluminescence assay (GE Healthcare). Actin was used as a loading control.
RNAseq. Embryos at 30 hpf were manually dechlorinated. Deyolking was performed by triturating with glass pipette in cold Ringer's solution containing 1 mM EDTA and 0.3 mM PMSF in isopropanol. Total RNA from 30 embryos was extracted using TRIzol (Invitrogen). Purity of RNA was analyzed with Bioanalyzer (Agilent Technologies). Samples with an RNA integrity number of acceptable quality (Ͼ7) were used for Illumina RNA-seq library preparation. Deep sequencing was performed on an Illumina GAIIx Sequencer (Illumina).
Read quality trimming and quality assessments. Trim Galore! (version 0.2.7; Babraham Institute) was used to trim adapter sequences and poorquality bases (below Phred of 20) from the reads while removing any reads that were Ͻ20 nt long, using the default parameters. Trim Galore! makes use of cutadapt (version 1.4.2) (-f fastq-e 0.1-q 20-O 1-a AG-ATCGGAAGAGC file.fq.gz). The quality of the reads was assessed before and after trimming with FastQC (version 0.10.1).
Differential expression analysis and annotation. The gene-level counts output from RSEM were filtered to remove noise before normalization with trimmed means of M, such that only genes with a FPKM (fragments per kilobase of exon per million reads mapped) value Ͼ 1 in all replicates of any genotype were retained. Counts per million were determined using edgeR (version 3.10.2), genes with a counts per million Ͻ 1 in all samples were removed, and remaining counts were trimmed means of M normalized. Limma (version 3.24.15) was used to voom transform the filtered count data by empirically deriving and applying quality weights to the samples. These weighted values were used to calculate differential expression using limma. Annotations for each gene were added using biomaRt (version 2.24.0), including both the D. rerio Entrez gene identifiers and the corresponding Mus musculus Entrez orthologous gene identifiers.
Gene ontology and pathway analysis overview of workflow. Gene ontology term enrichment analysis was performed using a log2-fold change (log2FC) ranked list from limma (log2FC Ͼ 1 and false discovery rate Ͻ 0.05) as input into clusterProfiler (version 2.2.4). This analysis determines which Molecular Function, Biological Process, or Cellular Component gene ontology terms are positively or negatively enriched in the mutant embryos compared with WT, at a false discovery rate Յ 0.05, while taking into account the magnitude and direction of change. Pathway database, Reactome pathway analyses, were used. A log2FC ranked list of all differential gene expression data from limma (log2FC Ͼ 1 and false discovery rate Ͻ 0.05) was input into ReactomePA (1.12.3).
All the zebrafish genes in the dataset were manually annotated with their murine orthologs using biomaRt, and a Reactome pathway analyses was performed using zebrafish gene annotations (from zebrafish differential expression data) and zebrafish pathway annotations. The analysis was performed with zebrafish Entrez gene identifiers to determine the Reactome pathways that were positively or negatively enriched at a false discovery rate Յ 0.05.
EdU and BrdU labeling. Proliferating cells were labeled with the S-phase markers, EdU (Thermo Fisher Scientific) or BrdU (Millipore Sigma). Embryos were incubated on ice in 1.5 mM EdU dissolved in embryo rearing solution containing 15% diethylsulfoxide for 20 min. Embryos were then returned to room temperature for 10 min before fixation. EdU-labeled cells were visualized using Click-iT Assay kit (Thermo Fisher Scientific). To label dividing cells in adults, animals were housed in the fish system water containing 5 mM BrdU.
Light lesion. To selectively damage photoreceptors, an ultra-high intensity light lesion was used as previously described (Bernardos et al., 2007). In brief, zebrafish were exposed to 120,000 lux light from an EXFO X-Cite 120 W metal halide lamp for 30 min and then returned to the aquarium system.
qPCR. Total RNA from whole retinas was extracted using TRIzol (6 retinas from 3 fish per sample) (Invitrogen). RNA was quantified using Nanodrop spectrophotometer (Thermo Fisher Scientific). Reverse transcription and qPCR were performed according to the manufacturer's instructions using QIAGEN QuantiTec Reverse Transcription kit and Bio-Rad IQ SYBR Green Supermix, respectively. Reactions were performed using a CFX384 Touch Real-Time PCR Detection Systems (Bio-Rad). Primer sequences are listed in the supporting Table 1.
ALK inhibitor treatment. Zebrafish were housed from 24 to 72 h post lesion (hpl) in system water containing 10 M TAE684 (Abcam, ab142082) constructed from a 20 mM stock solution in 0.1% DMSO. Control groups were housed in system water containing the 0.1% DMSO. Solutions were changed daily.
Experimental design and statistical analysis. In radial sections, cells counted in three nonadjacent sections in each retina were averaged. Three to seven retinas were analyzed. In flat-mount preparation, ZO1 profiles with perimeter Ͼ3.5 m were identified as cone photoreceptors. For each retina, cones in 5625 m 2 area were counted using National Institutes of Health ImageJ (https://imagej.nih.gov/ij/). A total of six retinas were analyzed.
For qPCR experiment, three biological replicates were prepared for each time point, and three technical replicates were evaluated for each sample. For quantification of fold changes, ⌬⌬ C(t) method was used, and the housekeeping gene, gpia, was used to normalize the data (Livak and Schmittgen, 2001).
Statistical analysis was performed in JMP pro software using the nonparametric Mann-Whitney-Wilcoxon and ANOVA with post hoc Tukey HSD test (SAS Institute). A p value Յ 0.03 was considered significant.

Loss-of function mutant, mdka mi5001
We generated a CRISPR-Cas9-mediated Midkine-a loss-offunction mutant, mdka mi5001 , which carries a 19 bp deletion in the exon three of midkine-a. This deletion results in a predicted premature stop codon (Fig. 1A) and absence of protein in Western blot analysis (Fig. 1B). Immunostaining of both larval and adult retinas showed absence of protein in these tissues (Fig. 1C). Mdka mi5001 larvae progress normally through early developmental stages and at 48 hpf show only slight reduction in body pigmentation, shortened body length, and smaller eyes (Fig. 1D). The pigmentation defect recovers by 72 hpf (Fig. 1D). Notably,  the mdka mi5001 mutants replicate the delayed retinal development described previously following morpholino-mediated knockdown of Midkine-a ( Fig. 1E) (Luo et al., 2012).

Transcriptome analysis
Larvae from adult mdka mi5001 mutants were initially evaluated using transcriptome analysis of whole embryos at 30 hpf. This identified 638 differentially expressed genes (log2 fold change Ն 1 and a false discovery rate Յ 0.05) (  (Luo et al., 2012) and directed us to evaluate the role of Midkine-a in regulating proliferation in Müller glia.

Regeneration of cone photoreceptors is compromised in the mdka mi5001 mutant
Persistent, growth-associated neurogenesis is a hallmark of teleost fish (Hitchcock et al., 2004). In the growing eye and retina, stem and progenitor cells at the ciliary marginal zone generate new retinal neurons, with the exception of rod photoreceptors (Cerveny et al., 2012). Fate-restricted, proliferating rod precursors, derived from sporadic division of Mü ller glia and sequestered in the outer nuclear layer, selectively give rise new rod photoreceptors (Raymond and Rivlin, 1987;Bernardos et al., 2007;Stenkamp, 2011). This growth-associated neurogenesis occurs normally in the retinas of mdka mi5001 mutants, and there is no apparent alteration in the maturation or variety of cell types in the mdka mi5001 retina (Fig. 1E).
In response to neuronal cell death, Müller glia in zebrafish dedifferentiate and undergo a single asymmetric division to produce retinal progenitors, which rapidly divide, migrate to areas of cell loss, and differentiate to replace the ablated neurons (Nagashima et al., 2013). To assess photoreceptor regeneration in the mdka mi5001 , we used a photolytic lesion that selectively kills photoreceptors; photoreceptors undergo apoptotic cell death by 1 day post lesion (dpl) (Vihtelic and Hyde, 2000;Bernardos et al., 2007). In WT retinas, by 1 dpl, Müller glia can be labeled with antibodies against the late G 1 or early S phase marker, PCNA and, by 3 dpl, the Müller glia-derived progenitors form radial neuro-genic clusters that span the inner nuclear layer ( Fig. 2A). In contrast, in the mdka mi5001 retinas, PCNA labeling was completely absent at 1 dpl, and only a few cells were PCNA ϩ at 3 dpl ( Fig.  2 A, B). Based on the size and location of their nuclei, we infer that these cells are Müller glia. However, at 5 dpl, there were no differences in the number of PCNA ϩ cells in the outer nuclear layer of WT and mutant retinas (Fig. 2 A, C). In WT retinas, photoreceptor progenitors in the outer nuclear layer begin withdrawing from the cell cycle at 4 dpl (Bernardos et al., 2007). This, coupled with the delayed proliferation in the mutant retinas, can explain the similarity in the number of PCNA ϩ cells in the outer nuclear layer at 5 dpl.
We next asked whether PCNA ϩ cells in mdka mi5001 are capable of progressing further through the cell cycle. We exposed WT and mutants to BrdU between 48 and 72 hpl and killed the animals immediately for histology. In WT retinas at 3 dpl, BrdU ϩ cells were present in the inner and outer nuclear layers, and we In the mdka mi5001 mutant, Müller glia fail to proliferate in response to photoreceptor cell death. A, Immunocytochemistry for PCNA (magenta) in WT and mdka mi5001 at 1 and 3 dpl. In WT, Müller glia become positive for PCNA at 1 dpl. At 3 dpl, PCNA ϩ progenitors form neurogenic clusters. Mutant retinas lack PCNA-labeled cells at 1 dpl. Very few, isolated cells are positive for PCNA in the mdka mi5001 at 3 dpl. B, C, The number of PCNA ϩ cells in WT and mdka mi5001 in the inner (B; 1 dpl: p ϭ 0.0017; 2 dpl: p Ͻ 0.0001; 3 dpl: p Ͻ 0.0004, ANOVA with post hoc Tukey; F ratio ϭ 116.1834) and outer (C; 3 dpl; p ϭ 0.0001, ANOVA with post hoc Tukey; F ratio ϭ 17.5589) nuclear layers. The mdka mi5001 retinas have significantly less PCNA ϩ cells at 1, 2, and 3 dpl compared with WT retinas. A total of 3 sections were counted and averaged in each retina. A total of 3 retinas were analyzed. Scale bar, 30 m. *p Ͻ 0.01.
infer that these BrdU ϩ cells are Müller glia-derived progenitors (see Nagashima et al., 2013). In the mdka mi5001 mutants at 3 dpl, BrdU ϩ cells were also observed in the inner and outer nuclear layers, though, relative to WT retinas, there were significantly fewer cells in the mutant retinas (Fig. 3 A, B; compare Fig. 2A). To establish the identity of the BrdU ϩ cells in the inner nuclear layer in the mutant retinas, the transgenic reporter line, Tg(gfap: eGFP) mi2002 (Bernardos and Raymond, 2006), was crossed into the mdka mi5001 background. This showed that, in 3 dpl, the BrdU ϩ cells in the inner nuclear layer are Müller glia. The presence of a few closely paired nuclei indicates that some of these Müller glia can progress through the cell cycle (Fig. 3C). These results show, in the mdka mi5001 mutants, a few Müller glia can progress through the cell cycle, but relative to WT animals, many fewer of these cells divide and their division is significantly delayed.
We next assayed photoreceptor regeneration using specific cone and rod photoreceptor markers, Zpr1 and Zpr3, respectively, and flat-mount retinal preparations immunostained with the cell junction marker, ZO1. In WT retinas, regenerated cones appear as early as 5 dpl, and by 14 dpl the regeneration of cone photoreceptors is largely complete (Fig. 4 A, C,D). In contrast, at 5 dpl in the mdka mi5001 retinas, regenerated cones are nearly completely absent, and at 7 dpl only a few immature cone photoreceptors are present (Fig. 4A). The absence of regenerated cones persists through 14 and 28 dpl, demonstrating that, in the mutant retinas, regeneration of cone photoreceptors is not simply delayed (Fig. 4D). These results indicate that cone photoreceptor regeneration is permanently compromised in the absence of Midkine-a. In contrast, rod photoreceptors are regenerated slowly but to apparently normal numbers in the mdka mi5001 mutants (Fig. 4B). In WT retinas, mature rod photoreceptors appear between 7 and 14 dpl (Fig. 4B), whereas in the mdka mi5001 mutants, rod photoreceptors are immature at 7 dpl but completely replenished by 28 dpl (Fig. 4B). Together, these results demonstrate that Midkine-a is required for Müller glial to proliferate in response to cell death, and the absence of Midkine-a leads to a failure of cone photoreceptor regeneration. In contrast, rod photoreceptors regenerate normally in the mutants.

Müller glia in the mdka mi5001 mutants undergo gliotic remodeling
In all vertebrate retinas, neuronal death induces a gliotic response in Müller glia (Bringmann et al., 2006(Bringmann et al., , 2009Jadhav et al., 2009). Although the initial reactive gliosis is neuroprotective, persistent gliosis results in dysregulation of retinal homeostasis, glial remodeling and scar formation, and the subsequent death of neurons (Bringmann et al., 2006). In zebrafish, the gliotic response of Müller glia is transient and interrupted by cell cycle entry (Thomas et al., 2016). To determine whether the failure of Müller glia proliferation in Midkine-a mutants leads to a mammalian-like gliotic response, the expression of GFAP was compared in the WT and mdka mi5001 retinas. In unlesioned retinas, immunostaining for GFAP labels the basal processes of Müller glia (Bernardos and Raymond, 2006). In WT retinas at 28 dpl, GFAP immunolabeling resembles that in unlesioned retinas (Fig. 5A). In contrast, in the mdka mi5001 retinas at 28 dpl, GFAP immunolabeling is present throughout the cytoplasm, extending apically into the inner nuclear layer (Fig. 5A). Enhanced expression of GFAP is a marker of gliosis in mammalian Müller glia (Bringmann et al., 2006). To further characterize the gliotic response in the mutant retinas, we again used the transgenic reporter Tg(gfap:eGFP) mi2002 . Computing the planimetric density of Müller glia in flat-mount preparations at 28 dpl showed no significant difference in the number of Mü ller glia in wildtype and mutant retinas, suggesting that Mü ller glia do not die. However, in the mdka mi5001 ; Tg(gfap: eGFP) mi2002 line, Müller glia remained hypertrophic, as evidenced by elevated eGFP levels (Fig. 5B, arrows; Movies 1, 2), and these cells adopt abnormal morphologies, including expanded lateral extensions in the inner plexiform layer and migration of the somata into the outer plexiform layer (Fig. 5C,D; Movies 3,  4). This hypertrophic morphology is also revealed in flat mounts stained with the cell junction marker, ZO1, in which the apical profiles of Müller glia have expanded to fill the planar surface of the outer limiting membrane previously occupied by cones (Fig.  4C, dashed line). The abnormal gliotic remodeling observed in the mutant retinas is a hallmark of persistent reactive gliosis in mammals.

Müller glia in the mdka mi5001 dedifferentiate in response to photoreceptor death
In response to neuronal cell death, Müller glia spontaneously reprogram, upregulating stem cell-associated genes before entering the cell cycle (Goldman, 2014;Gorsuch and Hyde, 2014;Lenkowski and Raymond, 2014;Hamon et al., 2016). Immunostaining retinas at 1 and 2 dpl for the stem-cell associated proteins, Rx1 and Sox2, labeled elongated, polygonal nuclei,  The mdka mi5001 mutant retinas fail to regenerate cone photoreceptors. A, Immunocytochemistry for red/green cone photoreceptor marker, Zpr1. In WT retina, immature cone photoreceptors start to appear at 5 dpl, and regeneration largely completes by 14 dpl. In the mdka mi5001 mutant, regenerating photoreceptors are absent at 5 dpl. At 7 dpl, very few cone photoreceptors appear. The number of cone photoreceptors is less at 14 dpl compared with WT. B, Immunocytochemistry for rod photoreceptor marker Zpr3 following lesion. In WT retina, regenerating rod photoreceptors appear by 7 dpl. In the mdka mi5001 retinas, rod photoreceptors slowly regenerate by 28 dpl. C, Flat-mounted retinal preparation immunostained with ZO1 in unlesioned and 14 dpl. In unlesioned retina of both WT and mdka mi5001 , cone photoreceptors form a crystalline mosaic array in the planar apical surface of the retina (Livak and Schmittgen, 2001;Nagashima et al., 2017). Higher magnification of boxed region indicates the alignment of cones in the mosaic array (asterisks) with flattened cell boundaries (arrowheads). At 14 dpl in WT, cone photoreceptors regenerate (asterisks), although the crystalline mosaic array is not restored. In the mdka mi5001 retina, cone profiles are instead replaced by irregularly shaped, expanded Müller glial apical processes (dotted line). D, Counts of ZO1-labeled cone photoreceptors at 14 and 28 dpl. Significantly fewer cones are regenerated in the mdka mi5005 mutant (white) compared with WT (gray). n ϭ 6. 14 dpl: p ϭ 0.0051; 28 dpl: p ϭ 0.0051, nonparametric Mann-Whitney-Wilcoxon. onl, Outer nuclear layer; inl, inner nuclear layer; gcl, ganglion cell layer. Scale bars: A, B, 30 m; C, 10 m. *p Ͻ 0.01. characteristic of Müller glia (Nagashima et al., 2013;Gorsuch et al., 2017), in both WT and mutant retinas (Fig. 6A). Further, the Rx1 ϩ and Sox2 ϩ nuclei were displaced apically in both (Fig. 6A), revealing the interkinetic nuclear migration that is associated with cell cycle progression in Müller glia (Fig. 6B) (Nagashima et al., 2013). We also evaluated the reprogramming in Müller glia by qPCR for the core transcriptional factors, ascl1a, stat3, and lin28 (Fausett and Goldman, 2006;Ramachandran et al., 2010;Nelson et al., 2012). At 30 and 36 hpl, which is before when Müller glia divide, ascl1a, stat3, and lin28 are significantly upregulated in both WT and mutant retinas, although the expression level of ascl1a is slightly reduced in the mutants (Fig. 7 B, C). These data indicate that, in the absence of Midkine-a, Müller glia respond to photoreceptor death by reprogramming into a stem cell-like state.

Midkine-a is partially responsible for ascl1a expression via phosphorylation of Stat3
Following a retinal lesion, phosphorylation of Stat3 is required for the upregulation of ascl1a in Müller glia (Nelson et al., 2012;Zhao et al., 2014). In WT and mdka mi5001 retinas at 1 and 2 dpl, STAT3 protein was induced (Fig. 7 D, E). Consistent with previ- Figure 5. Following photoreceptor death, Müller glia in the mdka mi5001 mutant undergo gliotic remodeling. A, Immunocytochemistry for Gfap in WT and mdka mi5001 retinas at 28 dpl. In WT, the Gfap immunosignal is restricted to the inner third of radial processes. No obvious signal is detected at the inner nuclear layer. The mdka mi5001 upregulates Gfap, and signals are seen at the cell body of Müller glia in the inner nuclear layer. B, Single optical planes from z-stack series of the Tg(gfap: EGFP) reporter flat-mount retinal preparation in the mdka mi5001 background. In the ganglion cell and inner plexiform layers, some Müller glia show signs of hypertrophy, including increased levels of the EGFP transgene signal (arrows). C, Cross section view of 3D reconstructed image in the Tg(gfap:GFP); mdka mi5001 (green) retina at 28 dpl, immunolabeled with Zpr1 (red) in a flat-mount preparation. Yellow arrows indicate displaced Müller glia somata in the outer plexiform layer. D, Cross section and flat-mounted views of the 3D reconstructed image. Displaced Müller glia (magenta asterisk) retain basal radial process (magenta arrows). opl, Outer plexiform layer; inl, inner nuclear layer; ipl, inner plexiform layer; gcl, ganglion cell layer. Scale bars: A, 30 m; B, 20 m.

Absence of cell cycle progression in mutant Müller glia
Following reprogramming, Müller glia begin entering the cell cycle around 24 hpl and complete the asymmetric cell divisions by 42 hpl (Nagashima et al., 2013). We next asked whether Müller glia in the mdka mi5001 possess the ability to enter the cell cycle by quantifying the expression of G 1 phase cyclins, cyclin d1(ccnd1) and cyclin e1 (ccne1). These cyclins are expressed during G 1 and function to drive G 1 -to-S phase transition (Dyer and Cepko, 2001). We isolated mRNA at 30 and 36 hpl, knowing that cell cycle progression is not completely synchronous among the population of Müller glia, but that these time points will allow us to capture gene expression changes in Müller glia and exclude Müller glia-derived progenitors. This analysis showed that mdka mi5001 upregulates ccnd1 and ccne1 significantly at both 30 and 36 hpl (Fig. 8 A, B), indicating that, following photoreceptor death in the mdka mi5001 mutants, Müller glia enter the G 1 phase of the cell cycle. In WT retinas, cell cycle entry is followed by upregulation of S phase cyclin, ccna2 (Fig. 8C). In contrast, there is no upregulation of ccna2 in the mdka mi5001 retinas, indicating that Müller glia in mutants fail to progress from G 1 to S (Fig. 8C). This was confirmed using the S-phase label, BrdU, between 24 and 30 hpl. In WT retinas, Müller glia are uniformly labeled with BrdU; whereas in the in the mdka mi5001 retinas, there are no BrdUlabeled cells (Fig. 8D, n ϭ 6 retinas). Consistent with these results, the expression of the cell cycle regulators, cyclin-dependent kinase 4 and 6, is dysregulated in mutant retinas (Fig. 8 E, F ). Together, these results indicate that, following photoreceptor cell death in the mdka mi5001 retinas, cell cycle progression of Müller glia is compromised, demonstrating that, in reprogrammed Müller glia, Midkine-a regulates the G 1 -S phase transition.
During retinal development, Midkine-a governs cell cycle kinetics through Id2a (Luo et al., 2012). Id proteins play important roles in cell cycle regulation during development and in cancer (Lasorella et al., 1996;Sikder et al., 2003). In WT retinas, id2a expression is markedly upregulated at 30 hpl, as Müller glia progress through the cell cycle, and rapidly returns to baseline levels by 48 hpl, when the single asymmetric mitotic division is complete (Fig. 8G). This transient induction of id2a is completely absent in the mdka mi5001 retinas (Fig. 8G). In cancer cells, Id2 proteins antagonize the Rb family of cell cycle inhibitors, thereby allowing progression from G 1 to S phase of the cell cycle (Lasorella et al., 2001;Sikder et al., 2003). Previous analyses of the Müller glia specific transcriptome show that p130, one of the Rb gene family, exhibits highest expression among Rb genes in quiescent Müller glia (Sifuentes et al., 2016;Nieto-Arellano and Sánchez-Iranzo, 2019). Consistent with these data, we validated that, in WT retinas, the expression of p130 decreases as Müller glia progress through the cell cycle (Fig. 8H ). In contrast, in mdka mi5001 retinas at 30 and 36 hpl, p130 levels are elevated above the those found in quiescent Müller glia (Fig. 8H ). These results suggest that Id2a is downstream of Midkine-a, and in Müller glia Id2a functions to inhibit Rb genes.

Signaling through the ALK receptor is responsible for Müller glial proliferation
ALK is a member of the superfamily of receptor tyrosine kinases. ALK is involved in the initiation and progression of many cancers, including neuroblastoma (Morris et al., 1995;Webb et al., 2009;Hallberg and Palmer, 2013). Midkine and its related protein pleiotrophin are the only ligands known to activate ALK (Stoica et al., 2001(Stoica et al., , 2002. To determine whether Alk functions as a Midkine-a receptor on Müller glia during photoreceptor regeneration, double immunocytochemistry was performed for pAlk and PCNA following a photolytic lesion. In WT retinas, pAlk colocalizes with PCNA, indicating activation of Alk in dividing Müller glia and Müller glia-derived progenitors (Fig. 9A). In contrast, both pAlk and PCNA immunolabeling were absent in mdka mi5001 retinas, indicating that, in the retina, Midkine-a is required for ALK phosphorylation. To test whether activation of ALK is required for proliferation among Müller glia, WT animals were housed from 24 to 72 hpl in the ALK inhibitor, TAE684. Inhibiting the activation of ALK phenocopied the proliferation defect observed in the mdka mi5001 mutants (Fig. 9 B, C). These data indicate that phosphorylation of Alk is required for Müller glia to proliferate and identify ALK as a putative receptor for Midkine-a during the initial asymmetric division in Müller glia and the subsequent regeneration of cone photoreceptors.

Discussion
In mammals, neuronal damage in the retina stimulates transient entry of Müller glia into the cell cycle; however, any subsequent proliferative response is very limited (Dyer and Cepko, 2000;Hamon et al., 2019;Rueda et al., 2019). In zebrafish, Müller glia respond to neuronal death by spontaneously entering and transiting the cell cycle, giving rise to Mü ller glia-derived progenitors that amplify in number and functionally replace ablated neurons. Numerous studies have identified transcriptional regulators and signaling cascades that promote Mü ller glia reprogramming in both mammals and fish (Karl and Reh, 2010; Goldman, 2014; Gorsuch and Hyde, 2014; Lenkowski   Stat3 (pStat3) in the WT and mdka mi5001 mutant. In unlesioned retina, immunosignal for pStat3 is not detected in the inner nuclear layer. Following photoreceptor lesion, Müller glia in the inner nuclear layer upregulate pStat3 in WT, whereas mdka mi5001 mutants have reduced phosphorylation of Stat3. Scale bar, 30 m. inl, Inner nuclear layer; ipl, inner plexiform layer. *p Ͻ 0.01. and Raymond, 2014;Hamon et al., 2016). The molecular mechanisms that govern cell cycle kinetics in Mü ller glia, while essential, have received relatively little attention. Here we provide the first evidence that Midkine-a, ostensibly acting as an extrinsic regulator of proliferation, governs the transition from G 1 to S phases of the cell cycle in injury-induced, reprogrammed Mü ller glia.
Our data support the mechanistic model shown in Figure 10. In unlesioned retinas, Müller glia remain quiescent in the G 0 phase (Fig. 10A). In response to photoreceptor cell death, nearby Müller glia upregulate reprogramming-associated genes, and enter the cell cycle (Fig. 10B). Midkine-a, signaling through Alk receptors, promotes the expression of ascl1a via phosphorylation of Stat3 (Fig. 10B). Midkine-a also induces the brief, transient upregulation of Id2a, which suppresses cell cycle inhibitor p130, thereby allowing Müller glia to enter both S and the subsequent phases of the cell cycle (Fig. 10B). In Midkine-a loss-of-function mutants, Müller glia initiate reprogramming into a stem cell state and enter G 1 phase of the cell cycle, but fail to activate Id2a and fail to transition from G 1 to S and mitosis (Fig. 10 B, C). The consequence of this cell cycle block is the selective failure in the regeneration of cone photoreceptors.
Following photoreceptor death, reprogrammed Müller glia enter the cell cycle within 24 hpl and undergo a single division by 42 hpl (Nagashima et al., 2013). This temporal sequence is closely coordinated, but not completely synchronized. We timed experiments using RNA isolated from whole retinas such that cell cycle-related gene expression could be evaluated in Müller glial stem cells while excluding Müller glia-derived progenitors. Following photoreceptor death in zebrafish, id2a is upregulated transiently at 30 hpl and returns to baseline levels by 36 hpl. This indicates that Midkine-a-dependent Id2a expression is required for the proliferation of Müller glia, but not Müller glia-derived progenitors. The family of Id proteins is involved in intrinsic control of proliferation during development and in cancer (Ruzinova and Benezra, 2003;Sikder et al., 2003). Id2 regulates the Rb proteins, and high levels of Id2 can suppress the Rb tumor suppressor pathway, which blocks progression from G 1 to S phase of the cell cycle (Lasorella et al., 2000(Lasorella et al., , 2001. In adult mice, the Rb-family member p130 maintains quiescence in muscle satellite cells, which retain the capacity to self-renew and regenerate myoblasts (Carnac et al., 2000). During sensory hair cell regeneration in zebrafish inner ear, p130 is downregulated immediately following injury (Jiang et al., 2014). Consistent with these data, our in silico screening identified p130 as a highly expressed gene in quiescent Müller glia, suggesting that p130 expressed in Müller glia of uninjured retinas functions to restrict their proliferation (Sifuentes et al., 2016;Nieto-Arellano and Sánchez-Iranzo, 2019). Our data suggest that, in response to cell death, Midkine-a signaling blocks p130 through upregulating Id2a, allowing Müller glia to progress through the cell cycle. We suggest that the brief upregulation and downregulation of id2a are a mechanism that allows Müller glia to divide, but restricts these cells to a single mitotic cycle. Further, our data suggest that Midkine-a is required for the rising phase of this transient id2a expression. Notably, increased levels of Id2 are also present in anaplastic large cell lymphomas that result from constitutive activation of the Midkine receptor, ALK (Mathas et al., 2009). It is not known whether ALK in Müller glia functions alone or as a member of multiprotein complex to relay the Midkine-a signal. Receptor protein tyrosine phosphatase-(RPTP-z) is also a known receptor for Midkine and can activate the intracellular kinase domain of ALK, and may function as a coreceptor to transduce Midkine-a signaling in Müller glia (Mathas et al., 2009;Hallberg and Palmer, 2013).
Following photoreceptor death in the mdka mi5001 mutants, Müller glia initially fail to progress through the cell cycle, although a small number of Müller glia eventually do so. As a consequence, the regeneration of cone photoreceptors is permanently compromised, whereas the regeneration of rod photore-ceptors is not. This suggests the initial proliferative response of the Müller glia gives rise to cone progenitors. Previous reports suggest that separate lineages give rise to regenerated cone and rod photoreceptors, respectively (Morris et al., 2008;Thummel et al., 2010;Gorsuch et al., 2017), and our results are consistent with these reports. We favor the interpretation that fate-restricted rod precursors persist in the outer nuclear layer and contribute to the regeneration of rod photoreceptors. However, we cannot exclude the possibility that regenerated rods in the mdka mi5001 mutants originate from the few latent Müller glia that progress through the cell cycle.
In vitro experiments demonstrated that inhibition of Midkine successfully suppresses proliferation of cancer stem cells (Mirkin et al., 2005;Erdogan et al., 2017). Therefore, Midkine silencing is proposed as a potential therapy for limiting cell cycle progression in cancer stem cells (Muramatsu and Kadomatsu, 2014). Our data also provide molecular insights into the potential role of Midkine in tumorigenesis, especially in regulating the cell cycle among cancer stem cells.
Constitutive activation of glial cells and formation of a glial scar are detrimental to the function of the CNS. An intriguing phenotype in the mdka mi5001 mutants is the cell death-induced gliotic remodeling of Müller glia. A previous report showed that pharmacological suppression of cell cycle progression following photoreceptor death results in hypertrophy and increased GFAP in Müller glia (Thomas et al., 2016). Together, these results suggest that, in zebrafish Müller glia, the molecular mechanisms that promote cell cycle progression are required to limit the initial gliotic response. Although it is not clear whether entry into the cell cycle initiates reactive gliosis in mammalian retinas, levels of cell cycle proteins appear to be a critical variable in the gliotic response (Dyer and Cepko, 2000;Levine et al., 2000;Vázquez-Chona et al., 2011;Ueki et al., 2012).
Our data significantly expand the understanding of retinal regeneration in zebrafish and more fully define the function of Midkine-a in governing the eukaryotic cell cycle. We provide convincing evidence that Midkine-a regulates proliferation of reprogrammed Müller glial during the regeneration of cone photoreceptors. In the absence of Midkine-a, zebrafish Müller glia respond similarly to Müller glia in mammals, with only a limited ability to regenerate neurons. In developing mammalian retinas, Midkine has been identified as component in the core transcriptional repertoire of retinal progenitors (Livesey et al., 2004). It remains to be determined whether Midkine-dependent cell cycle machinery is present in the Müller glia of adult mammals or whether manipulation of Midkine signaling in adult mammals could promote neuronal regeneration.