Ndr kinases regulate retinal interneuron proliferation and homeostasis

Ndr2/Stk38l encodes a protein kinase associated with the Hippo tumor suppressor pathway and is mutated in a naturally-occurring canine early retinal degeneration (erd). To elucidate the retinal functions of Ndr2 and its paralog Ndr1/Stk38, we generated Ndr1 and Ndr2 single knockout mice. Although retinal lamination appeared normal in these mice, Ndr deletion caused a subset of Pax6-positive amacrine cells to proliferate in differentiated retinas, while concurrently decreasing the number of GABAergic, HuD and Pax6-positive amacrine cells. Retinal transcriptome analyses revealed that Ndr2 deletion increased expression of neuronal stress genes and decreased expression of synaptic organization genes. Consistent with the latter, Ndr deletion dramatically reduced levels of Aak1, an Ndr substrate that regulates vesicle trafficking. Our findings indicate that Ndr kinases are important regulators of amacrine and photoreceptor cells and suggest that Ndr kinases inhibit the proliferation of a subset of terminally differentiated cells and modulate interneuron synapse function via Aak1.


Results
Ndr KO validation. We generated congenic homozygous Ndr1/Stk38 and Ndr2/Stk38l single KO mice to investigate the roles of Ndr kinases in retinal development and maintenance (Fig. 1, see methods). Ndr2 was deleted in all tissues by crossing Ndr2/Stk38l flox/flox mice (obtained from the Knockout Mouse Project, UC Davis), in which Ndr2 exon 7 is flanked by loxP sites to congenic mice expressing Cre recombinase (ACTB-Cre) (Fig. 1A). The LacZ ORF within the CSD Knockout First allele is not in frame with Ndr2 exon 6, so no Ndr2-LacZ fusion protein is expected to be produced. We validated Ndr2 KO mice by PCR, DNA sequencing, immunoblot and immunohistological strategies (Figs 1B-D and S1). Although RT-PCR experiments indicated that an Ndr2 transcript containing exons 4-5 was detectable in Ndr2 KO mouse retinas, immunoblots probed with an antibody to the conserved N terminal region of Ndr1/2 revealed no evidence of truncated Ndr2 or Ndr2-LacZ fusion protein (Supp. Fig. S1C). Immunoblots probed with an Ndr2-specific antibody (generated from unique peptide sequence within the Ndr2 C-terminal region) revealed a single 55 kD immunoreactive band in wild-type (WT) mouse eye extracts that was absent from Ndr2 KO protein extracts (Figs 1D and S1D). Likewise, comparative immunofluorescence microscopy revealed no specific Ndr2 immunoreactivity in adult Ndr2 KO mouse retinas, whereas Ndr2 localized broadly throughout differentiated retinas of WT mice and was prominent in photoreceptor inner segments (IS), the outer plexiform layer (OPL), inner plexiform layer (IPL) and ganglion cell layer (GCL), suggesting that Ndr2 is important for the function of multiple retinal cell types (Fig. 1C).
Because erd phenotypes (caused by an Ndr2 mutation) appear in young dogs after retinal neuron differentiation and lamination 7,8 , we limited our analysis of Ndr KO phenotypes to young adult mice (usually ~1 month old (P28). Ndr1 KO and Ndr2 KO mice exhibited no obvious adverse health or behavioral traits, although we did not analyze the structure and functions of non-retinal tissues. Notably, Ndr1 or Ndr2 deletion did not drastically impair vision in young mice (P28), as both Ndr1 KO and Ndr2 KO mice displayed similar ERG and visual placement responses as WT mice (Supp. Fig. S1).
Ndr deletion disrupts photoreceptor homeostasis. Since a major phenotype of canine erd retinopathy is progressive loss of retinal lamination, we analyzed retinal structure in age-matched adult Ndr KO and WT mice. The overall appearance and histological organization of Ndr1 KO and Ndr2 KO retinas were similar to that of WT retinas ( Fig. 2A), observed up to 6 months of age. We compared the relative thicknesses of the outer nuclear layer (ONL) and inner nuclear layer (INL) of Ndr2 KO and WT mice by counting the number of rows of nuclei and found no significant differences (Fig. 2B). In contrast, the ONL and INLs in the central retina of Ndr1 KO mice were thicker than WT by ~1-3 nuclei (p < 0.05), suggesting a role for Ndr1 in photoreceptor development or homeostasis (Fig. 2B). These data indicate that Ndr1 and Ndr2 deletion does not significantly impair SCiEnTiFiC REPoRTS | (2018) 8  and excised by the cre recombinase under control of the actinB promoter to produce Ndr2 KO mice. LacZ is indicated by the blue box, Neo cassette is indicated by the orange box. RT-qPCR primers for Exons 13-14 are indicated by red arrows. (B) RT-qPCR data confirms Ndr2 deletion. cDNA was isolated from brain and eye tissue from P28 wild type (WT) and Ndr2 KO mice. Data are from 4 sets of RT-qPCRs, targeting exons 13 to 14, with each sample run in duplicate (p < 0.05, calculated by one-sample test). (C) Ndr2 immunofluorescence was performed on P28 WT and Ndr2 KO retinas. Nuclei were labeled with Hoechst 33342. IPL, inner plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; IS, inner segments; OS, outer segments. Scale bar = 20 μm. (D) Immunoblot of eye protein extracts probed with anti-Ndr2 and antiactin antibodies. Uncropped images of this immunoblot are presented in Suppl. Fig. S1. (E) Two independent Ndr1/Stk38 alleles (Ndr1 ∆4 , Ndr1 ∆6 ) were generated by distinct CRISPR-cas9 procedures and confirmed by DNA sequencing. The chromatograms of WT and Ndr1 KO alleles are presented. The Ndr1 ∆4 allele contains a mutation in Ndr1 exon 4, in which CTC (underlined) is replaced by AGCG (red) to yield a frame shift mutation. The Ndr1 ∆6 allele contains a single base deletion in exon 6 (blue T in WT chromatogram) to yield a frame shift mutation. Red asterisks represent stop codons introduced by indels. (F) RT-PCR reveal the presence of Ndr1 transcript containing exons 4-5 and exons 13-14 in retina cDNA from Ndr1 ∆4 and WT mice. There is no detectable Ndr1 transcript in Ndr1 ∆6 mice. GAPDH RT-PCR data serve as positive controls.
retinal strata development and organization in young adult mice. We therefore conducted immunohistochemical analyses to look for other phenotypic markers of retinal disease in Ndr KO mice.
To determine if Ndr1 or Ndr2 deletion affects photoreceptors in young adult mice, we examined rod opsin localization by immunofluorescence microscopy. In fully differentiated retinas from P28 or older WT mice, rod opsin was uniformly restricted to rod outer segments (OS). In Ndr1 and Ndr2 KO retinas, most rod opsin localized to the rod OS, however some rod opsin conspicuously mislocalized to the inner segments (IS), the perinuclear cytoplasm in the ONL (Fig. 3A, see inset) and the rod synaptic terminals in the OPL (Fig. 3A, arrowhead).
In principle, opsin mislocalization in Ndr KO retinas may be caused by a variety of mechanisms, including aberrant opsin expression, post-translational processing or trafficking 41 . We conducted quantitative immunoblots on retinal protein extracts to determine if Ndr deletion disrupts rod opsin protein expression or electrophoretic mobility, as might be the case if Ndr influences rhodopsin glycosylation or Golgi trafficking. The relative levels and electrophoretic mobility of rhodopsin monomer and dimers were indistinguishable in Ndr1 KO, Ndr2 KO and WT retinal extracts (Fig. 3B) As a complementary method to detect retinal degeneration in Ndr KO mice, we analyzed the localization and expression of the apoptosis effector caspase-3 by immunofluorescence microscopy. Caspase-3 is not prominently expressed in differentiated retinas of healthy young adult mice, but is upregulated and proteolytically cleaved to an active form in apoptotic cells, in response to inherited or chemically-induced retina degenerations [42][43][44] . As expected, there were no active caspase 3 immunoreactive cells in mouse retinas from adult WT mice (Fig. 4B). In contrast, there were many caspase 3-positive cells in the INL and GCL of Ndr1 and Ndr2 KO retinas, indicative of apoptosis (Fig. 4B). Moreover, ~20-40% caspase 3-positive cells in the INL also express Pax6 or Syntaxin 1, suggesting that many of those cells are amacrine cells (Fig. 4B,C   Ndr deletion promotes cell proliferation in differentiated retinas. The aberrant photoreceptor proliferation observed in canine erd, which is caused by an Ndr2 mutation, suggests that Ndr kinases regulate photoreceptor proliferation [7][8][9][10]45 . To determine if Ndr1 or Ndr2 deletion promotes cell proliferation in differentiated mouse retinas, we probed P28 retina sections with an antibody to phospho-histone H3 S10 (pHH3), a specific marker for mitotic cells [46][47][48] . Fully developed mouse retinas (≥P14) do not normally contain proliferating cells, as retinal neurons and photoreceptors are terminally differentiated 8,49 . In support, there were no pHH3-positive cells in P28 WT retinas (Fig. 5A). In contrast, there were many (~17-19 cells/ 500 um length) pHH3-positive cells in the INL and GCL of Ndr1 KO and Ndr2 KO retinas (Fig. 5B). We also detected a few pHH3-positive cells in the ONL of Ndr1 KO mice (Fig. 5B). Parallel immunofluorescence experiments with antibodies to other cell proliferation markers, PCNA (S phase), Cyclin A (S, G2, M) and Ki67 (G1, S, G2, M phase) 50-52 also revealed multiple proliferating cells in the INL of Ndr1 KO and Ndr2 KO retinas, and none in WT retinas (Supp. Fig. S3, arrows). Since most ONL nuclei belong to photoreceptors, it is likely that the mitotic (pHH3-positive) cells in the ONL of Ndr1 KO mice are aberrantly proliferating photoreceptors, whereas the proliferating INL cells could be derived from one or more interneuron types, Müller glial cells, or retinal progenitor cells.   Fig. S6). Pax6 can be expressed in some Müller cells and retinal progenitor cells (RPC) 56,57 . Thus, we probed differentiated retinas with other amacrine markers, including HuD, Gad65 and calretinin, to determine if Ndr deletion influences amacrine cell development/maintenance. HuD is a regulator of RPC differentiation and is expressed in some mature amacrine cells [58][59][60] . Glutamate decarboxylase 65 (GAD65) normally localizes to neurites and nerve endings of GABAergic amacrine cells in the IPL of healthy retinas 54,55,61 and calretinin is a calcium binding protein that is expressed in a subset of glycinergic and GABAergic amacrine cells 62 . There was a significant decrease (~60%) in HuD-positive cells in the INL of Ndr1 and Ndr2 KO retinas in comparison to WT retinas (Supp. Fig. S7). Moreover, ~10% HuD-positive cells in Ndr1 and Ndr2 KO retinas were immunoreactive to pHH3, indicating that they are proliferating (Figs 6E and S7E). GAD65 immunoreactivity decreased by ~20% in Ndr1 KO retinas and by 40% Ndr2 KO retinas, suggesting Ndr deletion reduces the abundance of GABAergic amacrine cells ( Fig. 6D and F). In addition, there was a modest decrease in the number of calretinin-positive cells in the INL of Ndr2 but not Ndr1 KO retinas (Supp. Fig. S7). These data suggest that Ndr deletion reduces the abundance of mature amacrine cells, while simultaneously promoting the proliferation of some Pax6 and HuD-positive INL cells. Alternatively, Ndr deletion might reduce the expression of some amacrine cell markers (such as Pax6, GAD65, HuD, calretinin) without significantly diminishing amacrine cell number. Intriguingly, we also observed a modest (5-10%) increase in calbindin-positive horizontal cells in Ndr KO retinas, suggesting that Ndr also influences horizontal cell development (Supp. Fig. S4).
Since Pax6 and HuD are expressed in some RPCs 56-60 , it is possible that some of the proliferating Pax6 and HuD-positive cells in P28 Ndr KO retinas ( Fig. 6) are progenitor cells. To test this possibility, we probed retinas for the presence of Nestin, an intermediate filament protein that is expressed prominently in RPCs and is normally excluded from mature amacrine cells. We observed a few Nestin-positive cells in P28 Ndr1 and Ndr2 KO retinas, but none in corresponding WT retinas (Supp. Fig. S8A). Notably, in Ndr KO retinas, Nestin appeared to localize to elongated cytoplasmic processes near the plexiform layers, as opposed to the perinuclear localization typical for Nestin in RPCs or developing retinal neuroblasts Supp. Fig. S8A. Since Nestin can be expressed in mature Müller glia in response to neuronal stress 63 , the aberrant Nestin expression in Ndr KO retinas may reflect neuronal stress rather than the presence of progenitor cells. Moreover, because there are far fewer Nestin-positive cells than proliferating cells in P28 Ndr KO retinas, it seems unlikely that most of the proliferating cells are progenitor cells.
Ndr deletion disrupts amacrine cell interneuron organization. Ndr deletion appears to disrupt the cellular organization of a subset of INL interneurons. Specifically, in WT mouse retinas, most Pax6-positive amacrine cells localize prominently to the lower half of the INL with a few scattered throughout the GCL (Fig. 5), as described [64][65][66][67] . In contrast, in Ndr KO retinas, Pax6-positive cells in the INL was not limited to the lower half of the INL and appeared to localize throughout the INL. Moreover, in Ndr2 KO retinas, the mitotic calretinin-positive amacrine cells localized predominantly to the GCL, whereas they localized to both the INL and GCL in Ndr1 KO retinas (Fig. 6C), perhaps suggesting that Ndr2 influences a more limited subset of amacrine cell subtypes than Ndr1. We observed no obvious defect in horizontal or bipolar cell distribution in Ndr KO retinas (Supp. Figs S4 and S5).    the INL of Ndr1 and Ndr KO retinas co-expressed pHH3, suggesting that only a small percentage of Müller cells undergo proliferation in Ndr KO retinas (Fig. 7A,B,D). Moreover, the relative number and appearance of GS-positive Müller cells in Ndr KO retinas were similar to that of WT (Fig. 7C). Thus, the limited occurrence of Müller cell proliferation associated with Ndr deletion does not significantly alter Müller cell number or morphology in adult mice.

Ndr2 deletion alters neural retinal gene expression.
To determine how Ndr deletion influences gene expression in fully developed retinas, we screened for genes potentially regulated by Ndr2 using RNA-seq transcriptomic profiling. We set a data threshold of ≥2 fold (log2FC ≥I1I) and identified 340 differentially expressed genes (DEGs) composed of 190 up-regulated and 150 down-regulated genes when comparing Ndr2 retinas to that of age-matched WT retinas ( Fig. 8A and Supp. Table S4). We carried out enrichment analyses on these DEGs using Database for Annotation, Visualization and Integrated Discovery (DAVID) and Gene Ontology Consortium. Functional groups (gene ontologies) for up-regulated genes included "structural constituents of eye lens", "ubiquinol-cytochrome-c reductase activity" and "NADH dehydrogenase (quinone) activity" (Fig. 8B and Supp. Table S5). The presumed functions of many of these up-regulated genes are consistent with enhanced neuronal stress. In support, 39 of the 190 up-regulated genes are associated with the biological process gene ontologies that are commonly associated with stress, such as "oxidative phosphorylation" (cytochrome c), "respiratory  Table S6). Crystallin proteins are known to have chaperonin and anti-apoptotic functions in stressed neuronal cells [72][73][74] . Among the 150 down-regulated DEGs in Ndr KO retinas, 18 are involved in "regulation of synaptic plasticity" and "synapse organization", 31 are involved in "nervous system development" and 15 are involved in "histone modification and organization" (Fig. 8B and Supp. Table S5). In addition, some of the most down-regulated DEGs in Ndr KO retinas belong to the "muscle contraction" gene ontology (Supp. Table S5). The misregulated genes, which include actin and myosin subunits, are associated with the actin cytoskeleton and are expressed in many cell types. Cytoskeletal gene misregulation may also reflect cellular stress [75][76][77][78] .
To validate the RNA-seq findings, we conducted RT-qPCR with independently prepared mouse retinal cDNA from Ndr2 KO and WT mice and assessed the expression of selected genes involved in the gene ontologies identified above (Fig. 8C). We also randomly selected 4 of the top differentially expressed genes (Perp, S100a6, Ckm, Atp2a1) for RT-qPCR (Table S6). RT-qPCR analysis confirmed that several genes belonging to the "structural constituent of eye lens" and the "oxidative phosphorylation" gene ontologies were up-regulated by ≥2 fold (log2FC ≥I1I) in Ndr2 KO mouse retinas compared to WT retinas (Fig. 8C). Likewise, RT-qPCR experiments confirmed that expression of several genes from the "muscle contraction" (actin cytoskeleton) and "regulation of synaptic plasticity and organization" gene ontologies are decreased by ≥2 fold in Ndr KO retinas compared to WT retinas (Fig. 8C).
To determine whether Ndr1 regulates the same genes identified by the Ndr2 KO RNA-seq screen, we conducted RT-qPCR analyses of Ndr1 KO retinas (Supp. Fig. S9). Many genes displayed similar expression patterns in Ndr1 KO as in Ndr2 KO retinas. However, the expression of several genes belonging to the "synaptic plasticity and organization" gene ontology group were up-regulated in Ndr1 KO retinas but down-regulated in Ndr2 KO retinas. Collectively, these data suggest that Ndr deletion directly or indirectly induces retina neuronal stress and compromises retinal synapse and actin cytoskeletal function.
Ndr deletion disrupts expression of the vesicle trafficking regulator Aak1 kinase. Previous studies indicate that Ndr kinases regulate vesicle trafficking and brain neuronal and dendritic spine morphogenesis via phosphorylation of the trafficking regulator Aak1 protein kinase 29 . We analyzed Aak1 localization in WT and Ndr KO mice to determine if Ndr influences retinal neuron function via Aak1. Aak1 localizes predominantly to the inner and outer plexiform layers in WT mouse retinas, consistent with a role for Aak1 in neuronal vesicle trafficking, neuronal signaling and synapse functions (Fig. 9A). In addition, Aak1 is faintly detectable in the photoreceptor inner segments (IS), suggesting a role in photoreceptor vesicle trafficking. Strikingly, there is a  substantial decrease (>50%) in Aak1 immunofluorescence levels in the INL, IPL and IS of Ndr1 and Ndr2 KO retinas relative to WT retinas, with no concurrent increase in fluorescence in other retinal regions (Fig. 9). The same Aak1 antibody was used in control immunoblots of brain protein extracts and reveals a single immunoreactive 90 kD protein, the predicted molecular weight for Aak1. Parallel immunoblots of retinal protein extracts reveal a 50 kD immunoreactive band that decreases in abundance in Ndr KO retinas by ~30-50% relative to WT (Supp. Fig. S10). We observed the same immunolocalization and immunoblot results using two different Aak1 antibodies, thus the 50 kD immunoreactive protein likely represents a truncated form of Aak1 that may be a product of proteolytic processing or alternative slicing, as observed in other tissues 79 . RT-qPCR experiments indicate that Aak1 transcript levels are similar in Ndr KO and WT retinas (Supp. Fig. S10). Taken together with the immunofluorescence microscopy (IFM) results, these data suggest that Ndr deletion compromises Aak1 protein expression or stability in retinal interneurons and photoreceptors and suggest a possible mechanism for Ndr kinases in maintaining retinal interneuron, synapse and photoreceptor function. Our immunohistocytological data indicate that Ndr1 and Ndr2 are especially important for amacrine cell maintenance. Amacrine cells are a large class of synaptically active inhibitory interneurons that mostly reside in the INL (although some displaced amacrine cells localize to the GCL) and have dendritic arbors that project into the IPL, where they link retinal bipolar cells to retinal ganglion cells 61,84,85 . There are multiple loosely defined amacrine cell subtypes (>25, depending on reference) that are categorized by various markers, neurotransmitters and cell morphology 86 . Like other retinal interneurons, amacrine cells are considered to be terminally differentiated. Strikingly, our data suggest that Ndr1 or Ndr2 deletion causes a subset of amacrine cells in the INL and GCL to proliferate in differentiated retinas. At least 50% of the mitotic cells prominently express the pan-amacrine cell marker Pax6 and other amacrine cell proteins (calretinin, HuD and syntaxin 1), strongly suggesting that many of the mitotic cells are amacrine cells [53][54][55][56] (Fig. 6). Notably, deletion of either Ndr gene also dramatically reduces the overall number of Pax6-positive, HuD-positive and GABAergic amacrine cells in the INL of P28 mouse retinas (Figs 5, 6 and S7). The apparent decrease in Pax6-positive, HuD-positive and GABAergic amacrine cells in Ndr KO mice may be a consequence of increased cell death, as suggested by elevated active caspase 3 expression, or may reflect a role for Ndr kinases in promoting cell differentiation or the expression of some amacrine cell markers (Pax6, HuD, GAD65, calretinin). Taken together, our data suggest that Ndr1 and Ndr2 help maintain the differentiated state of a subset of amacrine cells by preventing them from proliferating and by ensuring proper gene expression.

Discussion
While our data strongly suggest that most of the mitotic cells in the INL of Ndr KO mice are amacrine cells, some mitotic cells do not express Pax6, suggesting that they could belong to other cell types. The Pax6-negative mitotic cells in the INL of P28 NDR KO mice are unlikely to be horizontal or rod bipolar cells because they fail to express markers for those cells (Supp. Figs S4 and S5) 86 . Moreover, most of the mitotic cells appear below the outer-most region of the INL, which is normally populated by horizontal and bipolar cells. A few mitotic cells express the Müller cell marker GS, suggesting they may be Müller glial cells or Müller progenitors. Intriguingly, in some non-mammalian vertebrates, such as zebrafish and Xeonopus, some Müller cells retain the capacity to proliferate and differentiate into other retinal cell types after retinal differentiation or in response to retinal injury [87][88][89][90][91][92] . While Müller cells are considered terminally differentiated in adult mice and other mammals 49,56,93 , recent experimental studies suggest that a small fraction of mouse Müller glial cells may reenter the cell cycle and express Pax6 in response to chemical-induced retinal injury 56,58,91,94 . Nevertheless, the Pax6-positive Müller cells from those conditions do not appear to exit S phase or label with antibodies to mitotic markers 56 . Thus, we favor the interpretation that the Pax6-positive mitotic cells in the INL of Ndr KO mice are amacrine cells and that most of the Pax6-negative mitotic cells are aberrant amacrine cells that fail to express Pax6. Alternatively, since Pax6 is also expressed in retinal progenitor cells (RPCs) in developing retinas, it is possible that some of the Pax6-positive mitotic cells in differentiated Ndr KO retinas are RPCs or amacrine cell progenitors.
The origin and fate of proliferating cells in the INL of adult Ndr KO retinas are not known. In principle, the proliferating cells may be derived from RPCs that persist and proliferate after the normal period of RPC proliferation in developing retinas. Alternatively, some developing amacrine cells in Ndr KO retinas may be cell cycle delayed or arrested in the retinal INL of young adult Ndr KO mice, thereby leading to the appearance of actively proliferating cells in differentiated retinas. Although we have not ruled out those possibilities, we do not favor either of those explanations. Notably, the proliferating cells appear to be devoid of the RPC marker Nestin and there are fewer HuD-positive cells in Ndr KO than in WT retinas, suggesting that the proliferating cells are not derived from canonical RPCs. Moreover, it seems unlikely that Ndr deletion leads to cell cycle arrest in the developing retina, as the failure to complete cell division would likely result in decreased INL thickness and induce more robust apoptosis and retinal degeneration than observed. A third possibility is that some of the proliferating cells may be derived from differentiated or nearly differentiated amacrine cells. Intriguingly, a subset of amacrine cells in differentiated mouse retinas express the stem cell marker Lgr5, a receptor for Wnt family ligands 95,96 . As Lgr5-positive amacrine cells may be capable of proliferating and differentiating into other interneuron cell types, including photoreceptors 95,96 , it is tempting to speculate that the mitotic Pax6-positive mitotic cells that we observe in the INLs of adult Nrd KO mice are derived from a similar pool of proliferation-competent amacrine cells. If so, Ndr signaling may function to help maintain the balance between amacrine cell proliferation and differentiation. Since there is no significant change in retinal thickness in young adult Ndr KO mice, our data suggest that there is a balance between INL cell proliferation and cell death in Ndr KO retinas. Indeed, misregulation of some cell cycle regulators can lead to increased apoptosis 81,[97][98][99] . Thus, some of the proliferating INL cells in NDR KO retinas may ultimately become apoptotic, while others may have amacrine cell fates. Further studies using methods to disrupt Ndr function in mature retinas are needed to address the role for Ndr in maintaining retinal cell fate and differentiation.

Role of Ndr in cell proliferation.
Our data suggest that mouse Ndr1 and Ndr2 inhibit amacrine cell proliferation in differentiated retina, although the responsible molecular mechanisms remain unknown. Numerous studies link the Ndr subfamily of protein kinases, especially Lats1/2 kinases, to regulation of cell proliferation 83 . The role of Ndr-related Lats1/2 kinases, which are terminal kinases in the canonical Hippo tumor suppressor pathway, in regulation of cell proliferation is well established. Lats1/2 kinases inhibit proliferation by phosphorylating the transcription activator Yap 100 , thereby preventing Yap nuclear translocation and Yap-dependent expression of cell cycle and anti-apoptosis genes 83,101 . Consequently, loss-of-function alleles of Lats and other Hippo pathway components cause tissue overgrowth 83 . It is possible that Ndr1 and Ndr2, which are related to Lats1/2 kinases, negatively regulate retinal cell proliferation via similar Yap-dependent mechanisms. In support, Ndr1/ Ndr2 kinases inhibit intestinal epithelial cell proliferation via a Yap phosphorylation 35 .
Yap is an important regulator of early retinal development, as it controls RPC proliferation and differentiation and influences RPE development 102,103 . Recently, Yap was shown to be expressed in a subset of retinal Müller cells in adult mouse retinas 104 . However, Ndr1 and Ndr2 deletion does not appear to significantly influence Müller cell proliferation or abundance in adult mice, suggesting that Ndr does not inhibit Müller cell proliferation via Yap phosphorylation. Moreover, we did not observe any significant change in Yap protein localization in the INL of differentiated retinas from P28 Ndr KO mice (unpublished data). These data argue against the model that Ndr kinases inhibit Müller cell and amacrine cell proliferation via Yap phosphorylation. Nevertheless, the remaining Ndr kinase in single Ndr KO mouse may impede detection of robust Yap-related phenotypes, thus it may be necessary to delete both Ndr1 and Ndr2 genes to definitively test this model. Alternatively, since Ndr1 and Ndr2 may negatively regulate amacrine cell proliferation via indirect mechanisms or via another, as yet unidentified, substrate, further analyses of putative Ndr substrates will be necessary to elucidate the precise molecular mechanisms of Ndr in modulating amacrine cell proliferation.
Ndr and amacrine cell homeostasis. Despite an apparent increase in amacrine cell proliferation in differentiated retinas of Ndr KO mice, there were fewer Pax6-positive and HuD-positive cells and significantly less Gad65 expression in Ndr KO retinas than in WT retinas. Pax6 is a pan-amacrine cell marker and is essential for RPC multipotency and amacrine cell differentiation 57 and HuD is a regulator of RPC differentiation and is expressed in some mature amacrine cells [58][59][60] . Thus, these data suggest that Ndr promotes amacrine cell development and/or maintenance. While a decrease in amacrine cell differentiation in Ndr KO retinas could potentially cause a correlative increase in development of other INL cell types, we found no evidence for significant changes in relative number or distribution of Müller, horizontal or bipolar cells (Figs 7, S4 and S5, and data not shown) in the INL of Ndr KO mice. Moreover, despite the decrease in Pax6, HuD and Gad65-positive cells in Ndr2 KO retinas, the relative INL thickness (measured by nuclei) of the central retina is similar to WT controls, arguing against a precipitous loss of amacrine cells. Thus, we favor the interpretation that Ndr deletion does not decrease the overall number of amacrine cells, but instead alters the cellular physiology of a subset of amacrine cells, leading to decreased expression of some amacrine cell proteins, such as Pax6, HuD and Gad65. Moreover, since Gad65 is marker for GABAergic amacrine cells 105 , the reduction in Gad65 immunoreactivity in Ndr KO retinas suggest that Ndr1 and Ndr2 are important for maintaining the proper balance of amacrine cell subtypes within the INL, perhaps via a combination of transcriptional and posttranscriptional mechanisms.
Ndr and retinal neuron stress. Our gene expression data suggest that Ndr2 deletion induces neuronal stress. In support, Ndr2 deletion is accompanied by increased expression of genes associated with oxidative stress (such as the ROS scavenging enzyme NAD(P)H quinone oxidoreductase), mitochondrial dysfunction (cytochrome c proteins), protein misfolding (crystallins) and cytoskeleton misregulation (actin and myosin) (Supp . Table S5). Intriguingly, many of these classes of genes are affected in retina after toxic injury, such as methanol intoxication 106 , further supporting a role for Ndr2 in maintaining retinal homeostasis 107 . Oxidative and mitochondrial stress often arise from an imbalance between mechanisms that generate reactive oxygen species (ROS) and cellular detoxification mechanisms, and can lead to neuronal dysfunction and cell death 108,109 . In retina, ROS is commonly generated by light-induced signal transduction pathways, oxidization of polyunsaturated fatty acids, and RPE-mediated phagocytosis of photoreceptor outer segments 108,109 . Healthy retinal neurons maintain homeostasis under conditions of moderate OS, however an aberrant increase in ROS caused by constant exposure to light or mislocalized opsin can lead to activation of caspase-3 and apoptotic pathways causing cell death and visual impairment 110,111 .
Although the mechanism for Ndr kinases in preventing neuronal stress is not known, Ndr loss-of-function may indirectly promote neuronal stress via a variety of methods, including opsin misclocalization, impaired gene expression, neuronal dysfunction, increased apoptosis, or aberrant vesicle trafficking/synapse functions. Intriguingly, previous studies demonstrate that in presence of light, rhodopsin mislocalization to the ONL can cause oxidative stress that is not compensated by the RPE cells. Thus, it is possible that in Ndr KO mice, the mislocalized rhodopsin in the ONL and OPL might contribute to elevated ROS production and a correlative increased expression of ROS scavenging enzymes, such as NAD(P)H:quinone oxidoreductase. Alternatively, elevated neuronal stress in Ndr KO retinas may be an indirect consequence of misregulated interneuron homeostasis and increased apoptosis in the INL. Ndr and synapse regulation. Our gene expression analyses reveal that Ndr2 deletion decreases expression of genes involved in synapse function and modulation. These data suggest that Ndr directly or indirectly regulates interneuron neurite and synapse function. In agreement, Ndr2 prominently localizes to the synapse-rich inner and outer plexiform layers. Moreover, previous studies in several diverse organisms implicate Ndr kinases in regulating neuronal morphogenesis 17,112 . Notably, mutations in Drosophila and C. elegans Ndr kinase homologs (Trc and Sax-1) cause defects in neuronal tiling and dendritic spine morphology and overexpression of Ndr kinases promote neurite formation and branching in cultured cells 113,114 . Yeast Ndr also regulates vesicle trafficking, polarized secretion and morphogenesis [12][13][14]20,24,115 , which are essential for neuronal cell development and function [116][117][118] .
It is likely that retinal Ndr kinases regulate interneuron function by modulating vesicle trafficking. In support, mammalian Ndr2 influences integrin trafficking and integrin-dependent neurite growth in hippocampal neurons 28 . Yeast Ndr regulates trafficking of secretory vesicles via phosphorylation of Sec2/Rabin8, a GEF for a Rab GTPase 20 . In mouse, Ndr1 and Ndr2 phosphorylate Rabin8 and Aak1 protein kinase, both of which are involved in neuronal vesicle trafficking 29,119 . Mutations that disrupt Aak1 phosphorylation cause neuronal branching and dendritic spine defects in hippocampal cell cultures, supporting a role for Ndr and Aak1 in and neuronal morphogenesis and synapse function 29 .
Aak1 kinase function is not fully understood and has not been investigated in retina. Previous studies indicate that Aak1 regulates clathrin coated vesicle trafficking during endocytosis via phosphorylation of the AP2 adapter complex and regulates trafficking of components involved in the Notch-signaling pathway, which is important for amacrine and Müller cell fate specification and retinal development [32][33][34]120,121 . We demonstrated that Aak1 localizes to the synapse-rich plexiform layers and is significantly diminished in Ndr1 and Ndr2 KO mutants. Thus, we hypothesize that retinal Ndr kinases modulate interneuron vesicle trafficking and synapse function via Aak1. Intriguingly, both Ndr1 and Ndr2 kinases are required to maintain WT levels of retinal Aak1. Thus, loss of Ndr signaling might impair retinal interneuron vesicle trafficking and neuronal signaling in retinas via diminished Aak1 levels/activity.
Ndr KO mice as a model for erd. Many retinal phenotypes in Ndr KO mice are shared with canine erd, including rod opsin mislocalization, increased cell proliferation and apoptosis, and impaired gene expression 7,8,10 . However, the progressive loss and disorganization of ONL in erd dogs suggests that erd more severely compromises photoreceptor integrity than does mouse Ndr1 or Ndr2 deletion. In addition, the rod opsin mislocalization phenotype and presence of TUNEL-positive cells and mitotic cells within the ONL suggest that photoreceptors are stressed by Ndr deletion. Ndr deletion also appears to disrupt the organization of Pax6-positive amacrine cells (Fig. 5A) and increase the fragility of fixed retinas (data not shown), further implicating Ndr kinases in maintenance of retinal structure and organization.
The differences between canine erd and mouse Ndr KO phenotypes may be attributed to a number of factors, including species-dependent differences in retinal homeostasis or Ndr function, the presence or absence of as yet unidentified genetic modifiers, intrinsic differences between mutant Ndr alleles, and differences in the relative importance of each Ndr gene with respect to specific cell types or functions. Notably, it has not been determined if the mutant canine Ndr2 erd allele is a complete loss-of-function allele or if it encodes a catalytically inactive truncated Ndr2 protein. The canine Ndr2-erd allele contains a 4 nt deletion in intron 3 and an exonic SINE insertion, leading to the deletion of the exon 4 from the canine STK38L transcripts 7 . This mutation could, in principle, could result in the expression of mutant Ndr2 protein that lacks amino acids encoded by exon 4. Thus, the differences in the mouse Ndr2 KO and canine erd alleles may account for phenotypic differences. It is also possible that there are species-specific differences in Ndr function or retinal development and maintenance mechanisms that account for the apparent phenotypic differences in mice and dogs. Regardless of differences between canine erd and mouse Ndr KO phenotypes, our experiments reveal that Ndr1 and Ndr2 are important regulators of retinal interneurons and broaden an understanding of retinal Ndr kinases. Additional studies are needed to further elucidate the shared and cell-type specific functions of Ndr1 and Ndr2 kinases in retinal development and disease.

Materials and Methods
Animal care IACUC compliancy. All
HRMT assay and Sanger sequencing (Fig. 1E) were used to identify insertion/deletion (Indels) caused by non-homologous end joining repair at the cut site in Ndr1 exon 4 (Stk38 ∆4 ) and exon 6 (Stk38 ∆6 ). Since mutant mice generated by zygote injection are frequently mosaic, mutant founder Ndr1 animals were outcrossed to C57BL/6J mice to establish stable and uniform transgenic Ndr1 KO lines. Ndr1 ∆4 mice were genotyped by Kompetitive Allele Specific PCR (KASP) (LGC Genomics) according to manufacturer's protocols 128  Visual function assessment. Visual placement response assays were performed on ≥3 animals per genotype (1-2 months old), as in 129 . Briefly, mice are held 25 cm above a clean surface and assayed for limb extension as animals are lowered toward the surface. Numbers were attributed for each of the following behavioral responses: 0 = no response, 1 = response upon nose contact, 2 = response upon vibrasse contact, 3 = response before vibrasse contact ≥18 mm. Normal mice reach their forelimbs toward the impending clean surface (scoring = 3), while blind animals will not (scoring 1-2). Ndr1 KO and Ndr2 KO mice scored indistinguishably from wild type mice (scoring = 3). Electroretinography (ERG) recordings were performed on ≥3 animals per genotype (1-4 months old) by the Noninvasive Assessment of Visual Function Facility (Penn Vision Research Center supported by the NIH core grant P30 EY001583), as in 130 . Protein extract preparation and Immunoblot analysis. Brain, eyes or retinas from ≥three animals were homogenized in cold RIPA buffer containing proteases inhibitors (Roche Complete Mini-EDTA free protease inhibitors, 100 μM leupeptin, 100 mM NaVO4, 20 mM NaF) using a BeadBug ™ Microtube Homogenizer (Benchmark Scientific Model D1030E) as in 10 . 50 μg extract samples were electrophoresed on 4-12% SDS-polyacrylamide gels and transferred to nitrocellulose membranes (Li-COR #926-31090). Membranes were incubated in Odyssey ® Blocking Buffer (Li-COR, #P/N 927-50000) for 1 h at room temperature, treated with primary antibody (Supp . Table S2) in Blocking Buffer +0.1% Tween 20 for 12-6 h at 4 °C and incubated for 1 h at room temperature in secondary antibody (goat anti-rabbit IRDye 800CW or goat anti-mouse IRDye680RD; LiCOR). Membranes were scanned on Odyssey Fc Dual-Mode Imaging System (Li-COR) and data quantified using Image Studio Software (Li-COR), according to manufacturer's protocols. Actin was used as a reference protein for quantitative immunoblots. All commercially available antibodies used for immunoblots and immunofluorescence experiments are listed in Supp. Table S2. Affinity purified rabbit polyclonal anti-Ndr2 antibody was generated using an Ndr2-specific peptide antigen (QPVPNTTEPDYKSK, corresponding to amino acids 421-434) (YenZym Antibodies LLC, San Francisco, CA), as previously described 29 .
Histological analysis and quantification. For histological analysis, right eyes from 1 month to 3 months old mice (n ≥ 4) were harvested, fixed in Excalibur alcoholic solution, a modified Davidson solution (Paula Pierce, Excalibur Pathology Inc.), paraffin embedded and sectioned (5 um). Retinal tissues were deparaffinized and counterstained with Hematoxylin and Eosin (H&E) 10 . To measure retinal thickness, H&E retinal sections were scanned and digitized using the Aperio scanscope CS-OT (Leica) and visualized using Aperio Image Scope software. Three retinal sections per mouse were used for quantitative evaluation of ONL and INL thickness, measured as the number of rows of nuclei at specific locations (central = optic nerve head (ONH) (±100 µm) and peripheral = ONH ±2000 µm (±100 µm)). The number of nuclei were counted and averaged, as in 131 . Immunohistochemistry and analysis. For immunohistochemistry, left eyes (n ≥ 3 per assay) were fixed in 4% paraformaldehyde in PBS for 15 minutes on ice, incubated overnight in PBS containing 15% and 30% sucrose and embedded in optimal cutting temperature (OCT) compound. 10 μm frozen sections were air-dried at RT and immunohistochemistry was realized as in 10 . Nuclei were stained with 2 ug/ml Hoechst 33342 (Thermo-scientific #62249) and slides were mounted in Gelvatol pH = 8.5 (Sigma-Aldrich). For TUNEL assays frozen retinal sections (10 um) were incubated 1 h in blocking solution, and TUNEL assays was done following standard procedures from the manufacturer (In Situ Cell Death Detection Kit, Fluorescein, Roche #11684795910). Microscopy and imaging/quantification. Immunofluorescence microscopy (IFM) was conducted using an Axioplan microscope (Carl Zeiss Meditec, Thornwood, NY) equipped with a Spot RT-KE slider 7.4.1 camera (Diagnostic Instruments Inc., Sterling Heights, MI) and controlled by Spot 5.1 software and a Leica DM6000 widefield Fluorescence microscope equipped with a Hamamatsu Orca 03 G CCD camera (Hamamatsu Photonics K.K., Japan) and controlled by LAS X software. For confocal fluorescence microscopy, a Leica TCS SP5 II scanning laser confocal microscope (Leica Microsystems, Wetzlar, Germany) controlled by Leica Application Suite Advanced Fluorescence (LAS AF) software. Confocal images were captured using a 40x oil immersion objective (HCX PL APO CS, 1.25-0.75 NA).
For each IFM experiment presented, a minimum of 3 mice per genotype were captured, processed, analyzed using the same settings and quantified at identical threshold settings to reduce background fluorescence. An average of 4 regions of interest (500 μm length, central retina) per mouse were selected to quantify pHH3 and Pax6-positive nuclei. To quantify calretinin, calbindin, caspase-3, HuD, syntaxin, glutamine sythetase (GS) nuclei, an average of 2 regions of interest (500 μm length, central retina,) per animal were analyzed using ImageJ or Metamorph (Molecular Devices). To quantify GAD65, and AAK1 immunofluorescence, an average of 5 regions of interest were selected within the IPL (100 × 100 μm), the OPL (100 × 20 μm), and the IS (100 × 20 μm) in central retina and analyzed using Metamorph (Molecular Devices).
Data were normalized (Limma VOOM function) and differentially expressed genes (DEGs) were identified using linear modeling and Bayesian statistics (Limma) 135 . An in-house data analysis program provided by the PennVet Bioinformatics Core was used to average the data and identified the potentially transcriptionally induced and repressed genes in P28 mouse Ndr2 KO retinas relative to WT controls. 341 potentially regulated genes (FC ≥ I2I) were classified into functional groups based on known gene ontology (GO) functions and pathways using the following online software and databases: Database for Annotation, Visualization and Integrated Discovery (DAVID), and Gene Ontology Consortium (GO).
To validate selected DEGs, RT-qPCR experiments were done in compliance with standard MIQE (Minimum Information for Publication of Quantitative Real-Time PCR Experiments) guidelines 136 . For RT-qPCR, RNA samples were reverse-transcribed using the High-Capacity cDNA Reverse Transcription Kit following standard procedures from the manufacturer (Applied Biosystems, Foster City, CA, #4368814). The RT-qPCR reactions contained 40 ng cDNA, 1x SYBR Green PCR Master Mix (Applied Biosystems, #4309155), and 250 nM of each unlabeled forward and reverse primer. Reactions were performed in 384-well reaction plate using the QuantStudio ™ 6 Flex Real-Time PCR System (Applied Biosystems). GAPDH was found to be the most stable housekeeping gene in all tested samples, and used for normalization and calculation of the ratio of Ndr KO vs. WT using the ΔΔCT method 131,136 . Statistical significance of DEGs (p < 0.05; fold change (FC ≥ I2I) was assessed by one-sample T-test. Primers for RT-qPCR are listed in Supp. Table S3.
Data availability. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.