The m6A reader YTHDF2 is a negative regulator for dendrite development and maintenance of retinal ganglion cells

The precise control of growth and maintenance of the retinal ganglion cell (RGC) dendrite arborization is critical for normal visual functions in mammals. However, the underlying mechanisms remain elusive. Here, we find that the N6-methyladenosine (m6A) reader YTHDF2 is highly expressed in the mouse RGCs. Conditional knockout (cKO) of Ythdf2 in the retina leads to increased RGC dendrite branching, resulting in more synapses in the inner plexiform layer. Interestingly, the Ythdf2 cKO mice show improved visual acuity compared with control mice. We further demonstrate that Ythdf2 cKO in the retina protects RGCs from dendrite degeneration caused by the experimental acute glaucoma model. We identify the m6A-modified YTHDF2 target transcripts which mediate these effects. This study reveals mechanisms by which YTHDF2 restricts RGC dendrite development and maintenance. YTHDF2 and its target mRNAs might be valuable in developing new treatment approaches for glaucomatous eyes.


Editor's evaluation
In this study, you propose a role for the m 6 A reader YTHDF2 in regulating the dendritic arbor size of retinal ganglion cells (RGCs). You show that retina-specific loss of Ythdf2 leads to the expansion of RGC dendritic arbors in the horizontal plane and a widening of the inner plexiform layer (IPL), with Ythdf2 conditional knockouts showing a modest increase in visual acuity in an optomotor assay. You point to a number of factors downstream of YTHDF2 not previously known to be involved in retinal dendritic development, and propose a role for YTHDF2 in a glaucoma model, in which loss of YTHDF2 is shown to prevent RGC loss. This study presents a careful phenotypic analysis of manipulation of YTHDF2 and provides a foundation for studies on how YTHDF2-mediated mechanisms are integrated into programs of dendritic development and RGC survival.

Introduction
The mammalian retina is an ideal model system to study neuronal development and neural circuit formation. The retinal ganglion cells (RGCs) are the final and only output neurons in the vertebrate retina and their dendrites collect the electrical information concerning the visual signal from all other cells preceding them. One of the major focuses of research in the retina is to understand how RGC dendrite arborization arises during development (Prigge and Kay, 2018). Existing evidences supported that homotypic repulsion controls retinal dendrite patterning (Lefebvre et al., 2015). However, in mice which had most RGCs genetically eliminated, the dendrite size and shape of remaining RGCs appeared relatively normal (Lin et al., 2004). Thus, the fact that the dendrites of remaining RGCs did not expand to neighboring areas by the remaining RGCs supports the existence of the intrinsic limit for RGC dendrite patterning, which cooperates with the homotypic repulsion to determine the dendrite size of RGCs (Lefebvre et al., 2015). However, such intrinsic limiting mechanisms remain elusive.
Glaucoma is one of the leading causes for blindness. The major risk factors for glaucoma include increased intraocular tension. Studies have shown that glaucoma causes pathological changes in RGC dendrites before axon degeneration and soma loss were detected in different model animals (Weber et al., 1998;Shou et al., 2003;Morgan et al., 2006). Thus, elucidation of mechanisms governing RGC dendrite arbor maintenance bears clinical significance.
N 6 -methyladenosine (m 6 A) is the most widely distributed and extensively studied internal modification in mRNA (Dominissini et al., 2012;Meyer et al., 2012;Nachtergaele and He, 2018). m 6 A modification has been shown to regulate brain development and functions in the nervous system (Livneh et al., 2020;Yu et al., 2021a). By effectors, most of these studies have focused on its demethylases ('m 6 A erasers') and methyltransferases ('m 6 A writers'). Since the fate of m 6 A-modified transcripts is decoded by the m 6 A-binding proteins ('m 6 A readers'), how the readers mediate these functions and what are their neural target mRNAs remain to be elucidated. In addition, more precisely controlled spatial-temporal ablation of the m 6 A readers instead of null knockout is required to elucidate their functions and mechanisms in nervous system.
In this study, we identified an m 6 A-dependent intrinsic limiting mechanism for RGC dendrite arborization and maintenance. Conditional knockout (cKO) of the m 6 A reader YTHDF2 in the developing mouse retina increases RGC dendrite branching and improves visual acuity. YTHDF2 also mediates acute ocular hypertension (AOH)-induced RGC degeneration, the experiment model for glaucoma, and Ythdf2 cKO in the retina alleviates AOH-induced RGC dendrite shrinking and neuronal loss. The regulation of RGC dendrite development and maintenance by YTHDF2 is mediated by two distinct groups of m 6 A-modified target mRNAs which encode proteins that promote dendrite arborization during development and maintain dendrite tree during injury, respectively. Therefore, our study reveals mechanisms by which YTHDF2 restricts RGC dendrite development and maintenance, which sheds light on developing new treatment approaches for glaucomatous eyes.

Knockdown of YTHDF2 leads to a robust increase of RGC dendrite branching
To examine whether m 6 A modification and its reader proteins play a role in the dendrite development, we utilized the retina as the model system. We first checked their expression patterns in the developing mouse retina. Immunostaining with a widely used m 6 A antibody demonstrated that RGCs had high m 6 A modification levels ( Figure 1-figure supplement 1A). Consistent with the m 6 A distribution, the m 6 A reader YTHDF2 is highly expressed in RGCs ( Figure 1A; Figure 1-figure supplement 1B). Conversely, the expression of YTHDF2 in other layers and cells of the retina is much lower or absent ( Figure 1A; Figure 1-figure supplement 1B-D). Another two m 6 A readers YTHDF1 and YTHDF3 show similar expression patterns (Figure 1-figure supplement 1E, F). The strong expression of YTHDFs and high level of m 6 A modification in RGCs suggest that the m 6 A reader YTHDFs might play roles in RGC development. We dissected and dissociated the retinal cells and cultured in vitro. We generated lentiviral shRNAs against YTHDFs, which showed similarly efficient knockdown (KD) of YTHDFs in RGC cultures in vitro ( Figure 1B; Figure 1-figure supplement 1G,H). In these YTHDFdeficient RGC cultures, the first and most obvious phenotype that we observed is the robust increase Note that all RGCs marked by the pan-RGC marker RBPMS express YTHDF2 while all YTHDF2-expressing cells are RBPMS + RGCs. GCL, ganglion cell layer. Scale bars: 20 μm. (B) Western blotting (WB) confirming efficient KD of YTHDF2 in cultured RGCs using shYthdf2. Data of WB quantification are mean ± SEM and are represented as dot plots: ***p = 0.00012 (n = 3 replicates); by unpaired Student's t test. (C) Examination of RGC dendrite development after YTHDF2 KD. As shown, significantly increased branching of dendrites marked by MAP2 immunofluorescence was observed in cultured RGCs marked by Brn3a. Dendrite traces were drawn for the corresponding RGCs. Scale bar: 10 μm. (D) Quantification of dendrite branching (C) using Sholl analysis. As shown, numbers of interactions are significantly greater in shYthdf2 groups (n = 68 RGCs) than shCtrl groups (n = 72 RGCs) in Sholl radii between 10 and 30 μm. Data are mean ± SEM. ****p = 4.32E-05 (10 μm), ***p = 0.00038 (15 μm), ****p = 2.85E-05 (20 μm), ***p = 0.00084 (25 μm), *p = 0.020 (30 μm), by unpaired Student's t test.
The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Source data for Figure 1B.
Source data 2. Source data for Figure 1B.
Source data 3. Source data for Figure 1B.      of dendrite branching of cultured RGCs treated by shYthdf2 (Figure 1C and D;J). In contrast, the dendrite branching of RGCs with YTHDF1 KD using shYthdf1 was not significantly different from control shRNA (Figure 1-figure supplement 1K), while YTHDF3 KD using shYthdf3 caused a slight (statistically significant in several Sholl radii) decrease of RGC dendrite branching compared with control shRNA (Figure 1-figure supplement 1L). These results suggest that the m 6 A reader YTHDF2 might play an important role in controlling dendrite branching of RGCs.
cKO of Ythdf2 in the retina increases RGC dendrite branching in vivo without disturbing sublaminar targeting To further explore whether YTHDF2 physiologically regulates RGC dendrite branching in vivo, we generated Ythdf2 cKO mouse (Figure 2A). We used the Six3-cre mouse line (Furuta et al., 2000), which has been widely used in the field to generate retina-specific knockouts (Lefebvre et al., 2012;Riccomagno et al., 2014;Sapkota et al., 2014;Krishnaswamy et al., 2015). YTHDF2 expression is efficiently eliminated in the Ythdf2 cKO retina compared with their littermate controls at E12.5 (  2A-D), suggesting that YTHDF2 is not involved in the generation or development of these cells. This is in line with the low or no YTHDF2 expression in these cells. The RGC number or density was not affected in the Ythdf2 cKO retina ( Figure 2C and D), demonstrating that Ythdf2 knockout does not disturb RGC neurogenesis. We then cultured RGCs from the Ythdf2 cKO retina. The dendrite branching of Ythdf2 cKO RGCs was significantly increased compared with littermate controls ( Figure 2E and F). RGCs include over 40 subtypes (Sanes and Masland, 2015;Baden et al., 2016). We thus examined the RGC dendrite branching within different subtypes. One of the RGC subgroups responds preferentially to movement in particular directions and is named the ON-OFF directionally selective RGCs (ooDSGCs). Expression of CART (cocaine-and amphetamine-regulated transcript), a neuropeptide, distinguishes ooDSGCs from other RGCs (Kay et al., 2011a). The dendrite branching of ooDSGCs marked by CART/Brn3a co-staining in Ythdf2 cKO retinal cultures also increased compared with control ( Figure 2G and H). These data further confirm that the m 6 A reader YTHDF2 regulates dendrite branching of RGCs.
Next, we wanted to confirm this phenotype in vivo by checking specific RGC subtypes. Intravitreal injection of an AAV reporter expressing ZsGreen visualized the dendrite morphology of ooDSGCs marked by CART immunostaining ( Figure 3A). ooDSGCs showed dramatically increased dendrite branching in Ythdf2 cKO retina compared with control retina by Sholl analysis ( Figure 3A and B). The intrinsically photosensitive RGCs (ipRGCs) are unique and melanopsin-expressing cells, which exhibit an intrinsic sensitivity to light (Hattar et al., 2002). We analyzed the morphology of ipRGCs visualized by wholemount immunostaining of melanopsin and found that the dendrite branching of ipRGCs was significantly increased in the Ythdf2 cKO retina (Figure 3C and D;. A similar trend was observed in the SMI-32 + αRGCs ( Figure 3E and F). These results strongly indicate that the m 6 A reader YTHDF2 negatively regulates RGC dendrite branching in vivo and Ythdf2 cKO promotes RGC dendrite arborization.
In the retina, RGCs target their dendrites in different sublaminae of the inner plexiform layer (IPL). Since the IPL sublaminar targeting of RGC dendrites is critical for normal visual functions, we wondered whether the increased dendrite branching caused by Ythdf2 cKO was also accompanied by altered sublaminar patterning of RGC dendrites. We used a Thy1-GFP reporter (line O) which labels a few RGCs (Feng et al., 2000). As shown in Figure 3-figure supplement 1F,G, GFP intensity is generally higher in IPL of the Ythdf2 cKO retina compared with their littermate controls, which further proves the increased RGC dendrite branching and density. However, the sublaminar pattern of GFP signals looks similar between cKO and littermate control (Figure 3-figure supplement 1F,G). Sublaminar dendrite patterning of the ipRGC subtype visualized by immunostaining of melanopsin also demonstrated the similar phenotype ( Figure 3-figure supplement 1H,I). These data suggest that YTHDF2 has a general control of RGC dendrite branching but has no striking effect on the sublaminar targeting of RGC dendrite. These results are consistent with the previous findings that the RGC dendrite targeting is determined genetically and several transcription factors controlling laminar choice have been identified in RGCs and amacrine cells (Cherry et al., 2011;Kay et al., 2011b;Lefebvre et al., 2015;Liu et al., 2018).

IPL of Ythdf2 cKO retina is thicker and has more synapses
The increased dendrite branching of RGCs further prompted us to check whether Ythdf2 cKO changes IPL development. Immunostaining of P6 retina vertical sections using a MAP2 antibody demonstrated that IPL thickness significantly increased in Ythdf2 cKO retina ( Figure 4A and B). As a control, the thicknesses of other retinal layers showed no difference between the Ythdf2 cKO and control mice ( Figure 4-figure supplement 1A-D). Quantification of MAP2 IF intensity in IPL suggested that the IPL of Ythdf2 cKO retina became denser with dendrites ( Figure 4A and C). These results suggest that the increased dendrite branching results in a thicker and denser IPL in the Ythdf2 cKO retina.
The IPL of retina is concentrated with synaptic connections, which contain synapses among and between bipolar-amacrine-ganglion cells. The increased RGC dendrite branching and denser IPL in the Ythdf2 cKO retina prompted us to wonder whether there are changes in synaptic connections in IPL. We used co-staining of the presynaptic marker Bassoon and the postsynaptic marker PSD-95 to count the colocalization puncta of Bassoon + /PSD-95 + . We found that the numbers of Bassoon + / PSD-95 + excitatory synapses in IPL of Ythdf2 cKO retina are significantly larger than that of control retina ( Figure 4D and E). As a control, the numbers of the excitatory ribbon synapses marked by the colocalization of Bassoon + /PSD-95 + in OPL (outer plexiform layer) show no difference between Ythdf2 cKO and control retinas (Figure 4-figure supplement 1E,F).
All these data verify that the IPL of Ythdf2 cKO retina is thicker and has more synapses.
Visual acuity is improved for the Ythdf2 cKO mice The features of RGC dendrites, including their size, shape, arborization pattern, and localization, influence the amount and type of synaptic inputs that RGCs receive, which in turn determine how RGCs respond to specific visual stimuli such as the direction of motion (Liu and Sanes, 2017). The increased dendrite branching, the thicker and denser IPL, and the more synapses in the IPL inspired us to further explore whether the visual responses of the Ythdf2 cKO mice were changed or not. Ythdf2 cKO mice looked normal and had similar body weight and size compared with control mice for either sex (male in Figure 5A and B; female in Figure 5C Furuta et al., 2000). We used an optomotor response (OMR)-based assay (Prusky et al., 2004;Umino et al., 2008;Shi et al., 2018) to monitor visual functions of Ythdf2 cKO mice ( Figure 5E). Surprisingly, the Ythdf2 cKO mice showed modestly improved visual acuity compared with the control mice, measuring spatial frequency threshold as 0.45 ± 0.0043 c/deg (cycle per degree) and 0.43 ± 0.0085 c/deg, respectively ( Figure 5F, male mice). Similar phenotype was observed in female mice ( Figure 5G). These results suggest that the visual acuity is modestly improved in the Ythdf2 cKO mice.
Wholemount immunostaining using a Brn3a antibody was carried out in P20 retina (C). Numbers of Brn3a + RGC per 10,000 μm 2 of retina were quantified and showed no difference between the Ythdf2 cKO and their littermate controls (D    YTHDF2 target mRNA were identified with transcriptomic and proteomic analysis Next, we continued to explore the underlying molecular mechanisms of the effects on dendrite branching caused by Ythdf2 cKO in the retina. First, we wanted to know what transcripts YTHDF2 recognizes and binds. We carried out anti-YTHDF2 RNA immunoprecipitation (RIP) in the retina followed by RNA sequencing of the elute (RIP-Seq). Two biological replicates of anti-YTHDF2 RIP-Seq identified 1638 transcripts (Supplementary file 1). Functional annotation of YTHDF2 RIP targets revealed significant enrichment in cellular component terms such as neuron part and neuron projection, and biological process terms such as cellular component organization and neuron projection development. We further zoomed in to check neural terms in cellular component ( Figure 6A) and biological process ( Figure 6B). We found that substantial numbers of YTHDF2 target transcripts are involved in cytoskeleton, dendrite, and their organization and development ( Figure 6A and B), which is consistent with the dendrite branching phenotype observed in the Ythdf2 cKO retina.
The working model for YTHDF2 is that it binds and destabilizes its m 6 A-modified target transcripts . Since the destabilization of mRNAs will eventually decrease their protein levels, we carried out proteome analysis using mass spectrometry (MS) in acute shYthdf2-mediated KD of cultured RGCs, in order to identify directly affected targets. Three biological replicates of YTHDF2 KD followed by MS (YTHDF2 KD/MS) identified 114 proteins which were upregulated by YTHDF2 KD (Supplementary file 2). Functional annotation of these proteins revealed significant enrichment in neuron development-and cytoskeleton-related terms ( Figure 6C), which is similar to anti-YTHDF2 RIP-Seq results.
The online version of this article includes the following figure supplement(s) for figure 3:    branching (Xie et al., 2010;Yan et al., 2015), and are required for normal brain functions (Penzes et al., 2001;Xie et al., 2007;Cahill et al., 2009;Russell et al., 2014;Lu et al., 2015;Herring and Nicoll, 2016). Strn (Striatin) was first identified in striatum, and functions as a B subunit of the serine/ threonine phosphatase PP2A and is also a core component of a multiprotein complex called STRIPAK (striatin-interacting phosphatase and kinase complex) (Benoist et al., 2006;Li et al., 2018). Strn was reported to regulate dendritic arborization only in striatal neurons but not in cortical neurons . However, whether and how Kalrn and Strn work in the retina was still unknown. Ubr4 (ubiquitin protein ligase E3 component N-recognin 4) is also known as p600 and has been shown to play roles in neurogenesis, neuronal migration, neuronal signaling, and survival (Parsons et al., 2015). However, whether Ubr4 regulates dendrite development remains elusive.
YTHDF2 controls the stability of its target mRNAs which encode proteins regulating RGC dendrite branching MS analysis after YTHDF2 KD has shown that the protein levels of these target mRNAs were upregulated (Supplementary file 2). IF using antibodies against Strn and Ubr4 detected specific signals in the IPL which were increased in Ythdf2 cKO retina compared with control retina (Figure 7-figure supplement 1A). Enrichment of these proteins in IPL implies that these proteins might function locally in RGC dendrites to regulate dendrite development.
We next wanted to know whether YTHDF2 controlled the protein levels of these m 6 A-modified target mRNAs through regulation of translation or transcript stability. As shown in Figure 7-figure supplement 1B-E, the mRNA levels of Kalrn7, Kalrn9, Kalrn12, Strn, and Ubr4 were dramatically increased after KD of YTHDF2, KO of Ythdf2, or KD of METTL14, supporting that YTHDF2 might regulate stability of these target mRNAs. We further evaluated potential changes in the stability of these target mRNAs in an m 6 A-dependent manner. We further verified this by directly measuring the stability of these target mRNAs. As shown in Figure 7A, all the target mRNAs showed significantly increased stability in the Ythdf2 cKO retina compared with controls. These results suggest that YTHDF2 controlled the protein levels of its m 6 A-modifed target mRNAs by decreasing their stability.
Next we explored the functions of these YTHDF2 target mRNAs in RGC dendrite development. We first generated siRNAs against these transcripts (Figure 7-figure supplement 1F). We then checked the effects on RGC dendrite branching after KD of these target mRNAs by siRNAs in cultured RGCs. As shown in Figure 7B, KD of Kalrn7, Kalrn9, Kalrn12, Strn, or Ubr4 led to significant decreases of RGC dendrite branching. Interestingly, the siCocktail against all these target mRNAs further significantly reduced the RGC dendrite branching compared with each individual siRNA (Figure 7-figure supplement 1G), suggesting that these targets may work in different pathways to regulate the RGC dendrite morphology. We further examined whether these target mRNAs mediate YTHDF2-regulated RGC dendrite branching. As shown in Figures 2E-H , and 3, cKO of Ythdf2 led to increased dendrite branching of RGCs both in vitro and in vivo. Transfection of siRNAs against these target mRNAs rescued dendrite branching increases in cultured Ythdf2 cKO RGCs ( Figure 7C). We continued to generate and performed intravitreal injection of AAV viral shKalrn12 and shUbr4, which significantly rescued dendrite branching increases of CART + ooDSGCs and SMI-32 + αRGCs in Ythdf2 cKO retina in vivo ( Figure 7D).
Taken together, we identified a group of YTHDF2 target mRNAs that encode proteins regulating RGC dendrite branching, which mediate YTHDF2-controlled RGC dendrite branching.
Ythdf2 cKO retina is more resistant to AOH The glaucomatous eyes are symptomatized with progressive neurodegeneration and vision loss (Agostinone and Di Polo, 2015). High intraocular pressure is a major risk factor in glaucoma and has been shown to cause pathological changes in RGC dendrites before axon degeneration or soma loss is detected in different model animals (Weber et al., 1998;Shou et al., 2003;Morgan et al., The online version of this article includes the following figure supplement(s) for figure 4: Figure supplement 1. Thickness or synapse numbers in outer plexiform layer (OPL) shows no difference between the Ythdf2 conditional knockout (cKO) and control retinas.

Figure 4 continued
2006). Our findings that Ythdf2 cKO in retina promotes RGC dendrite branching during development inspired us to wonder whether YTHDF2 also regulates RGC dendrite maintenance in the acute glaucoma model caused by AOH. We utilized the AOH model made with control and Ythdf2 cKO mice to check whether Ythdf2 cKO in the retina could alter the pathology in the glaucomatous eyes. RGC dendrite branching is significantly decreased after AOH operation compared with non-AOH in    (Figure 8-figure supplement 1A,B). Interestingly, the Ythdf2 cKO retina with AOH operation maintains significantly higher dendrite complexity compared with the glaucomatous eyes of Ythdf2 fl/fl control mice ( Figure 8A and B). In addition, there are significant RGC neuron losses in both genotypes after AOH ( Figure 8C and D). However, the reduction of RGC number in the Ythdf2 cKO retina is less than control retina ( Figure 8C and D). These results support that Ythdf2 cKO protects retina from RGC dendrite degeneration and soma loss caused by AOH.
Next we wanted to know whether and how YTHDF2 target mRNAs mediate these effects in the AOH models. We first checked the expression of YTHDF2 target mRNAs identified in the developing retina (Supplementary file 3) in the adult Ythdf2 cKO and control retina. We found that two target mRNAs Hspa12a and Islr2 show upregulation in the adult Ythdf2 cKO retina compared with control ( Figure 8-figure supplement 1C). m 6 A modification of Hspa12a and Islr2 mRNAs was further verified by anti-m 6 A pulldown (Figure 8-figure supplement 1D). Hspa12a encodes heat shock protein A12A which is an atypical member of the heat shock protein 70 family and has been shown to be downregulated in diseases such as ischemic stroke, schizophrenia, and renal cell carcinoma (Pongrac et al., 2004;Mao et al., 2018;Min et al., 2020). Islr2 encodes immunoglobulin superfamily containing leucine-rich repeat protein two and is poorly studied. Here, we found that Hspa12a and Islr2 are downregulated in the retina after AOH operation (Figure 8-figure supplement 1E), which is likely caused by upregulation of YTHDF2 in the AOH-treated retina (Figure 8-figure supplement 1F-H). We therefore hypothesized that AOH upregulates YTHDF2 which in turn downregulates its targets Hspa12a and Islr2, thus causing RGC dendrite degeneration and soma loss. If this is the case, overexpression of Hspa12a and Islr2 might protect RGC dendrite from AOH-triggered degeneration. We thus generated AAV harboring overexpression constructs of Hspa12a and Islr2 which were intravitreally injected to wild type retinas. After the AOH induction, the retinas overexpressing Hspa12a and Islr2 maintain significantly more complex RGC dendrite arbor and show better RGC survival compared with control AAV (Figure 8E-G).
These data verify that loss-of-function of YTHDF2 and gain-of-function of its targets Hspa12a and Islr2 have neuroprotective roles in the glaucomatous retina.

Discussion
Functions and mechanisms of mRNA m 6 A modification in the dendrite development were not known. Here, we revealed a critical role of the m 6 A reader YTHDF2 in RGC dendrite development and maintenance. YTHDF2 have two phases of function to control RGC dendrite development first and then maintenance through regulating two sets of target mRNAs. In early postnatal stages, the target mRNAs Kalrn7, Kalrn9, Kalrn12, Strn, and Ubr4 mediate YTHDF2 functions to regulate RGC dendrite development. In the adult mice, another set of target mRNAs Hspa12a and Islr2 mediate YTHDF2 function to regulate RGC dendrite maintenance.

Positive and negative regulators for dendrite development
The general principle for dendrite arborization is that the dendrite arbor cannot be either too big or too small in order to precisely sample a presynaptic target area during neural circuit formation (Lefebvre et al., 2015). Numerous extrinsic and intrinsic mechanisms have been found to regulate dendritic arbor patterning, which involves both positive and negative factors to achieve balanced control of dendritic growth (Jan and Jan, 2010;Dong et al., 2015;Ledda and Paratcha, 2017). For the secreted and diffusible cues, BDNF promotes dendrite branching and complexity (Cheung et al., 2007); the non-canonical Wnt7b/PCP pathway is a positive regulator of dendrite growth and branching (Rosso et al., 2005); the non-canonical Wnt receptor Ryk works as a negative regulator by limiting the extent of dendritic branching (Lanoue et al., 2017). For the contact-mediated signals, the cadherins Celsr2 and Celsr3 regulate dendrite growth in an opposite manner in cortical pyramidal and Purkinje neurons, and hippocampal neurons, respectively (Shima et al., 2004;Shima et al., 2007). For m 6 A pulldown followed by RT-qPCR. ND, not detected. Data are mean ± SEM and are represented as dot plots (n = 3 replicates): **p = 0.0016 for Kalrn7; ****p = 1.40E-08 for Kalrn9; ****p = 1.46E-06 for Kalrn12; ****p = 5.46E-06 for Strn; ****p = 4.90E-06 for Ubr4; by unpaired Student's t test.  (Cubelos et al., 2015); interestingly, the functions of Sp4 in dendrite development are dependent on the cellular context of its expression, for example, Sp4 promotes dendrite growth and branching in hippocampal dentate granule cells but limits dendrite branching in cerebellar granule cells (Ramos et al., 2007;Zhou et al., 2007). Here, we identified another negative regulator YTHDF2 which works posttranscriptionally, and loss-of-function of YTHDF2 increased dendrite complexity during development and protected RGC degeneration from AOH. We have validated this effect on several RGC subtypes. Since there are dozens of RGC subtypes, it is technically challenging but still interesting to test whether this effect is universal to all subtypes or there is specificity for RGC subtypes.

Neuroprotective genes in retinal injuries and degeneration
Transcriptome analyses have revealed differentially expressed genes after retinal injuries such as AOHinduced glaucoma and optic nerve crush (ONC), and the upregulated genes are of importance for discovering new treatment approaches (Jakobs, 2014;Tran et al., 2019). One of the previous studies has identified Mettl3, encoding the m 6 A writer, as an upregulated gene after ONC (Agudo et al., 2008). Here, we found Ythdf2, encoding an m 6 A reader, was also upregulated in the retina after AOH. We further found that Hspa12a and Islr2, two targets of YTHDF2 in adult retina, were downregulated in glaucomatous retinas. Overexpression of Hspa12a and Islr2 protected retina from AOH-caused RGC dendrite degeneration. Our findings in this study suggest that YTHDF2 and its neuroprotective target mRNAs might be valuable in developing novel therapeutic approaches to treat neurodegeneration caused by glaucoma and other retinal injuries.

Continued
Continued on next page
GeneSilencer Transfection Reagent (Genlantis) was used in siRNA transfection following the manufacturer's protocols. Culture medium was changed after 1 day of lentiviral shRNA infection or siRNA transfection. For lentiviral shRNA assay, puromycin (Thermo or Sigma, 1 μg/ml) was added after 2 days of infection. Immunofluorescence, RNA, or protein preparation was performed after shRNA or siRNA worked for 3 days. For AAV intravitreal injection, P0-P1 mouse pups were anesthetized in ice and then eyes were pierced at the edge of corneal by 30G × 1/2 needle (BD, 305106) under stereomicroscope. Then 1 μl AAV was intravitreally injected with 10 µl Syringe (Hamilton, 80330) following the pinhole. P15 or adult mice were anesthetized with 2.5% Avertin and then eyes were pierced at the side of corneal and the outer segment of sclera by 30G × 1/2 needle successively. Two μl AAV was intravitreally injected with 10 µl Syringe following the pinhole on the sclera. All subsequent experiments such as AOH operation and immunostaining were carried out after at least 3 weeks (10 days for ZsGreen/ CART labeling of ooDSGCs in Ythdf2 cKO and control mice in Figure 3A).
For cultured neurons, after twice of PBS washing, cells were fixed for 15 min with 4% PFA in 0.1 M PB at RT, then washed with PBS three times and blocked in PBST for 20 min at RT. Antibody incubation conditions are the same as tissue sections.
All images were captured on Nikon A1R confocal microscope or Zeiss LSM 800 confocal microscope with identical settings for each group in the same experiment. A region of interest, length or thickness in immunofluorescence experiments were obtained with ImageJ. The number of neurons in specific area was counted blindly and manually. To quantify RGC dendrite lamination in IPL with Thy1-GFP, z-stack and maximum projection were performed during the analysis. GFP intensity values across IPL depth were measured by ImageJ/Analyze/Plot Profile function . To quantify the numbers of Bassoon + /PSD-95 + excitatory synapses in IPL, the colocalization puncta was measured by ImageJ/Analyze/Puncta Analyzer as described previously (Ippolito and Eroglu, 2010).

Sholl analysis
For confocal images of cultured RGCs, MAP2 signals in original format were analyzed with simple neurite tracer and then quantified with Sholl analysis (5 μm per distance from soma center) which was a widely used method in neurobiology to quantify the complexity of dendritic arbors using ImageJ (Schindelin et al., 2012;Binley et al., 2014). Retina wholemount data were captured in z-stack mode (0.5-1 μm per slide) with confocal microscopes. ZsGreen, eGFP, and SMI-32 signals were directly analyzed with simple neurite tracer and then z projection of all tracers was quantified with Sholl analysis (10 μm per distance from soma center), while melanopsin signals were maximum-projected before tracing.

OMR assay
Ythdf2 cKO and control mice aged about 6 weeks were dark-adapted overnight before experiment and used in the OMR assay following the previously reported protocols (Douglas et al., 2005;Sergeeva et al., 2018). Using the Matlab program, 0.2 c/deg (15 s per direction of rotation) was first used for mice to adapt this experiment, and 0.3, 0.35, 0.4, 0.43, 0.45, 0.47, 0.5 and 0.55 c/deg (30 s per direction of rotation) were used in the following recordings. Mouse behaviors were analyzed in real time during the experiment and re-checked with video recordings. Finally, data for each mouse were determined by the minimal spatial frequency between left and right OMR.

CTB labeling of optic nerve
To label RGC axon terminals in mouse brain, RGC axons were anterogradely labeled by CTB conjugated with Alexa Fluor 555 (Invitrogen, C34776) through intravitreal injection 48 hr before sacrifice. After PFA perfusion, the brains were fixed with 4% PFA in 0.1 M PB overnight, dehydrated with 15% sucrose and 30% sucrose in 0.1 M PB overnight at 4°C sequentially, embedded with OCT for coronal section, and cryosectioned at 12 μm with Leica CM1950 Cryostat. After PBS washing, the sections were mounted with VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratory). The images were captured on Tissue Genostics with identical settings for each group in the same experiment with the TissueFAXS 7.0 software.

RIP and sequencing
For RIP experiment, we used the EZ-Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore) following the manual with minor modifications. Briefly, 1 × 10 7 retinal neurons were subjected to each 100 μl lysis buffer. The amount of YTHDF2 antibody (Proteintech, 24744-1-AP) and control IgG used for immunoprecipitation is 5 μg, respectively. RIP experimental steps, RNA sample preparation and sequencing, and sequence data analysis followed the procedures reported previously (Yu et al., 2021a).

MS analysis
E15.5 retinal neurons were cultured and infected with lentiviral shYthdf2 or shCtrl. Sample collection and lysis, protein and peptide preparation were performed following procedures reported previously (Yu et al., 2021b). Proteins with fold changes greater than 1.3 and p values less than 0.05 were considered to be regulated by YTHDF2 KD with statistical significance.

Anti-m 6 A immunoprecipitation
Total retinal RNA was extracted from P0 WT mouse pups. Immunoprecipitation of m 6 A-modified transcripts was carried out with Magna MeRIP m 6 A Kit (Merck-Millipore, 17-10499) following the manual. m 6 A antibody (Synaptic Systems, 202003) and corresponding control IgG were used in this experiment. The RNA samples pulled down from the experiment were used for RT-qPCR.