Activation of Wnt signaling reduces ipsilaterally projecting retinal ganglion cells in pigmented retina

Retinal ganglion cells (RGCs) are the output neurons from the eye to the brain. During retinal development, RGCs are produced at the interface of the retinal pigment epithelium (RPE) and neural retina, and are specified into two subtypes based on their laterality of projection – ipsilaterally or contralaterally. The proper proportion of ipsi- and contralateral RGC projections is crucial for binocular vision (Erskine and Herrera, 2014; Herrera et al., 2017; Petros et al., 2008). In mammalian albinism, a genetic disorder of melanin biogenesis in both skin and the RPE, the number of ipsilateral RGCs is reduced (Herrera et al., 2017). When pigmentation is restored in the RPE during RGC neurogenesis, the reduced ipsilateral projection is rescued, suggesting that the pigmented RPE plays a role in controlling the number of ipsilateral RGCs (Cronin et al., 2003). However, how pigment affects RGC specification and thus ipsi- versus contralateral RGC projection laterality even in pigmented retina is not understood.

In the developing albino mouse retina, ipsilateral RGC neurogenesis and subtype specification are disrupted (Bhansali et al., 2014; Herrera et al., 2003; Rebsam et al., 2012). During this period, albino RPE cells have irregular cell shape, fewer melanosomes and aberrant expression of junctional proteins (Iwai-Takekoshi et al., 2016). Because retinal neurons, including RGCs, are produced at the interface of the RPE and the neural retina, perturbed RPE cell integrity may result in aberrant RPE-neural retina communication and, in turn, altered ipsilateral RGC neurogenesis and subtype specification in albino mouse retina.

Here, to test the hypothesis that the RPE expresses extrinsic regulators influencing RGC neurogenesis, we investigated whether pigmented and albino mouse RPE cells express different sets of genes and whether the genes that are altered could affect ipsilateral RGC production. We found that the embryonic albino RPE expresses Wnt2b to a greater extent than pigmented RPE. Moreover, lithium treatment in vivo, known as an activator of Wnt signaling, led to a reduced number of RGCs that express Zic2, a transcription factor regulating the specification and guidance of ipsilateral RGCs (Herrera et al., 2003). These results suggest that Wnt signaling is a potential molecular link between RPE and RGC generation and specification, especially for ipsilateral RGC production.

To unravel a role for the RPE in mechanisms of RGC specification and to better understand the albino phenotype, we compared albino and pigmented RPE by performing a microarray analysis on the RPE isolated from embryonic albino and pigmented retina (Fig. 1). We focused on the RPE at embryonic day (E) 13.5, when ipsilateral RGCs are produced and differentiate in the ventrotemporal (VT) retina (Dräger, 1985; Herrera et al., 2003). Pigmented and albino RPE have distinct molecular features: 220 differentially expressed probes were identified in the albino RPE, corresponding to 191 different genes, of which 176 were upregulated and 15 were downregulated (Table S1). For validation of the microarray, we analyzed expression of select genes by RT-qPCR and in situ hybridization (Fig. S1A,B). Gene profiling with Genespring (Table S1) did not identify secreted signaling genes, which we hypothesized would be released from the RPE to act in the retina. To identify signaling pathways that could be altered in albino RPE compared with pigmented RPE, we applied Gene Set Enrichment Analysis (GSEA), which determines whether an a priori defined set of genes shows statistical significance between two biological phenotypes, in our case pigmented versus albino RPE (Fig. 1A, Tables S2-S4). In pigmented RPE, three gene sets involved in receptor and ligand binding were significant. The albino RPE transcriptome was enriched in genes involved in cancer, signaling, promoter activity, cell differentiation, and cell cycle and proliferation. Gene sets associated with cytoskeleton and cell junction were enriched in albino compared with pigmented RPE, supporting our findings on disorganized RPE cell integrity in albino retina (Iwai-Takekoshi et al., 2016). Gene sets associated with signaling pathways and enriched in albino RPE included Hh, Notch, Bmp, Sfrp and Wnt (Fig. 1B. Table S3). Because GSEA focuses on sets of genes rather than individual genes, we searched the literature to find candidate genes in the pathways that were recognized by the GSEA analysis (Bao and Cepko, 1997; Liu et al., 2003; Wang et al., 2016). Bmp4 (Fig. 2), Shh and Sfrp2 (Fig. S1) expression is similar in both genotypes. Notch2 expression is very faint in the pigmented RPE but evident in albino RPE (Fig. S1). Compared with Notch2, which is expressed in the entire albino RPE, Wnt2b expression is of interest because it is expressed in the peripheral RPE, which is adjacent to where ipsilateral RGCs settle in the neural retina. We thus focused on Wnt2b mRNA expression in the pigmented and albino RPE (Fig. 1C). In pigmented retina, as melanin masks in situ hybridization signals, we bleached melanin after the in situ hybridization color reaction. At E13.5, Wnt2b is expressed in the periphery of the ciliary margin zone (CMZ), as previously reported (Cho and Cepko, 2006; Kubo et al., 2003; Liu et al., 2003), both in albino and pigmented retina. However, at E15.5, when expression of Zic2 is expressed in the majority of ipsilateral RGCs (Bhansali et al., 2014; Herrera et al., 2003), Wnt2b expression is expanded toward central retina in albino RPE compared with pigmented RPE (1.5-fold increase in albino RPE compared with pigmented RPE: Fig. 1D). RT-qPCR also indicated a trend of increased expression of Wnt2b in albino RPE (1.4-fold increase, Fig. S1D).

To ascertain whether this alteration of Wnt expression in albino RPE at E15.5 is specific to Wnt2b, we examined other Wnt signaling components expressed in the peripheral retina. Wnt inhibitory factor 1 (Wif1) (Ha et al., 2012) is enriched in pigmented RPE (Table S4), but a difference in intensity and pattern between genotypes was not apparent (Fig. 1E). Furthermore, expression of Fzd7, a Wnt receptor (Liu et al., 2003), was specifically enriched in the ventral CMZ, which is proposed as a neurogenic site for RGCs in embryonic retina (Marcucci et al., 2016). Although the spatial pattern and intensity of Fzd7 were similar in both pigmented and albino retina (Fig. 1F), the asymmetric Fzd7 expression implicates ventral retinal specific Wnt signaling. These results indicate that the pigmented and albino RPE differentially express Wnt2b, which is temporally and spatially correlated with ipsilateral RGC production.

Previous studies have shown that a constitutively active Wnt model in the mouse retina displayed increased ectopic expression of CMZ markers such as Otx1 and Msx1 (Ha et al., 2012; Liu et al., 2007). We examined whether the expanded expression of Wnt2b in the albino mouse RPE is associated with changes in patterning of peripheral retinal structure. Otx1 and Bmp4 expression (Martinez-Morales et al., 2001; Zhao et al., 2002) is similar in pigmented and albino CMZ at E13.5 (Fig. 2A). At E15.5, when Wnt2b is expanded in albino compared with pigmented retina (Fig. 1C,D), Otx1, Bmp4 and Msx1 expression in the CMZ was again the same in both genotypes (Fig. 2B). Of note, Msx1 is expressed more intensely in ventral than dorsal retina in both albino and pigmented retina, as previously described (Marcucci et al., 2016) (Fig. 2B).

Next, we considered the possibility that Wnt2b expression in the RPE overlying the CMZ may influence neurogenic potential of the CMZ. Connexin 43 (Cx43) interacts with regulators of cell proliferation, such as cyclin D1 (Swayne and Bennett, 2016). Intense Cx43 labeling is observed around progenitor cells in the CMZ of adult newt (Umino and Saito, 2002) and is associated with cell proliferation in albino rat retina (Tibber et al., 2007). We found that Cx43 is more highly expressed in albino peripheral retina compared with pigmented retina at E13.5 (Fig. 2C). To determine whether the Cx43 upregulation in albino retina reflects structural changes at the border of CMZ and neural retina, we examined expression of genes found in the neural progenitor layer and differentiated neural retina (Sox2 and Fzd5) (Wang et al., 2016). The spatial pattern of these genes was similar in albino and pigmented retina (Fig. 2D). The intense expression of Cx43 in the albino CMZ may reflect alterations in the pace of RGC neurogenesis (Bhansali et al., 2014).

Because retinal area is unchanged in albino and pigmented mice at E15.5 (Bhansali et al., 2014), it is reasonable to assume that expansion of Wnt2b area in the albino RPE reflects an upregulation of Wnt activity. We assume that Wnt signaling has dual (or multiple) functions in retinal development, depending on the activation level: excess levels of Wnt activity lead to macroscopic deficiencies in the structure of peripheral retina and slight upregulation of Wnt signaling may affect only development of RGC subtypes, such as ipsilateral RGCs versus contralateral RGCs.

To test whether Wnt signaling is involved in ipsilateral RGC neurogenesis, we activated Wnt signaling in vivo by treating pregnant mice with lithium chloride (Lancaster et al., 2011; Liu et al., 2007) and quantifying the number of RGCs in the VT retina (Fig. 3A). We characterized ipsilateral RGCs by co-expression of Zic2 (Fig. 3B-E) and islet 1/2 (Isl1/2) (expressed in all RGCs). As expected from previous studies (e.g. Bhansali et al., 2014; Herrera et al., 2003; Marcucci et al., 2016), the number of ipsilateral RGCs is reduced by about half in control (NaCl-injected) albino retina compared with control (NaCl-treated) pigmented retina (Fig. 3F). In contrast, following lithium treatment, the number of ipsilateral RGCs in pigmented retina was reduced to numbers similar to those in control albino retina (Fig. 3F). The number of total RGCs (expressing Isl1/2) (Fig. 3G) and the expression zone of Otx1 (Fig. 3H) were also similar in lithium-treated and control retina, suggesting that peripheral retinal structure was not affected by the lithium treatment. Wnt activation in the lithium-treated pigmented and albino retina was confirmed by increased expression of Wnt target genes Axin2 and Lef1 beyond the levels observed in control retina, as spreading of an existing domain (Fig. 3I,J). Lithium activates the Wnt pathway by inhibiting GSK3β activity, which results in stabilization of β-catenin and transcriptional activation of target genes. Therefore, the lithium response is more likely to be induced in cells that are in a responsive state rather than in a broader region. These results suggest that Wnt activation is involved in the adjustment of the number of ipsilateral RGCs.

Marccuci et al. (2016) proposed that, in the albino mouse retina, diminished neurogenesis in the ventral CMZ correlates with reduced ipsilateral RGCs. To test the possibility that Wnt activation by lithium treatment affects mitosis in the ventral retina and results in fewer ipsilateral RGCs, we compared the number of PH3-positive M-phase nuclei in the CMZ and neural retina (Fig. 4A,B). A similar number of PH3-positive cells across treatments and genotypes was observed in all regions of analysis, including dorsal/ventral CMZ and dorsal/ventral neural retina (Fig. 4C). These results suggest that, even with a reduction in the number of Zic2-positive cells in lithium-treated pigmented retina, at least at the end point of lithium treatment, proliferation of retinal cells was not affected.

How does Wnt signaling affect RGC output? Wnt/β-catenin signaling functions as an extrinsic regulator of cell cycle control and cell fate specification in the central nervous system (Megason and McMahon, 2002). Previous studies demonstrated that increasing Wnt signaling induces progenitor cell markers in chick retina (Kubo et al., 2003) and delays neuronal differentiation in mouse cortex (Chenn and Walsh, 2002). Decreased Wnt activity is accompanied by reduced proliferation, but treatment with lithium rescues this phenotype in mouse cerebellum (Lancaster et al., 2011). Moreover, gene profiling comparing E14 mouse retinal explants treated with lithium to activate Wnt signaling to control retinal explants revealed cell cycle inhibitors and genes involved in neuronal differentiation (Ha et al., 2012). These studies indicate that activation of Wnt signaling is likely associated with a slowing of the cell cycle and inhibition of differentiation.

The wave of ventrotemporal RGC production is delayed in the albino retina compared with pigmented retina (Bhansali et al., 2014). This delay is linked to the altered expression of genes that regulate ipsilateral and contralateral fate. Specifically, fewer Zic2-positive RGCs are born before E15.5 and more Zic2-negative RGCs after E15.5 (Bhansali et al., 2014). The decrease in RGCs at E15.5 in albino mice is specific to the Zic2-expressing region in ventrotemporal retina, which is probably why retinal area is unchanged in albino and pigmented mice at E15.5. These results suggest that there is a brief time window of competence for Zic2 expression to be fated as ipsilateral RGCs (∼E13) and Isl2 to be fated as contralateral RGCs (∼E15) (Bhansali et al., 2014). Wnt activation by lithium treatment from E12.5 to E14.5 in the present study did not affect the number of differentiated RGCs (Isl1/2 positive) but led to fewer ipsilateral (Zic2-positive) RGCs at E15.5 (Fig. 4). Together with findings by Bhansali et al., the present data raise the possibility that activated Wnt signaling underlies delayed neurogenesis/differentiation in the albino retina and in lithium-treated pigmented retina, affecting the competency of retinal progenitor cells to express Zic2, and consequently leading to a reduced number of ipsilateral RGCs.

Wnt activation of lithium treatment reduced Zic2-positive ipsilateral RGCs in pigmented but not in albino retina. Our result may indicate that ipsilateral RGCs affected by lithium treatment in pigmented mice are subsets of RGCs that might be absent in normal albino retina. Subsets of ipsilateral RGCs have been discussed but not characterized. Cyclin D2 knockout mice have 20% fewer ipsilateral RGCs than wild type (Marcucci et al., 2016), suggesting that subsets of ipsilateral RGCs are produced by regulation of cyclin D2.

We hypothesize that Wnt2b protein is released from peripheral RPE to the neural retina and signals to retinal precursor cells via Fzd receptors, especially Fzd7, which is enriched in the ventral CMZ, or Fzd5, which is enriched in ipsilateral RGCs (Wang et al., 2016). RT-qPCR analysis in isolated RPE at E15.5 did not show obvious upregulation of Wnt target genes in albino RPE (Fig. S1). This could be explained by the co-expression of Wnt2b and Wif1. Whether Wnt and Wif1 interact with each other to play a role in RGC subtype specification is a pressing issue. The combination of Wnt2b and Wif1 expression in the peripheral RPE may represent a tunable system by which the production of different cell types (ipsilateral and contralateral RGCs) can be controlled in a temporal and spatial manner. The VT retina produces ipsilateral RGCs in a limited time window then switches to produce contralateral RGCs.

Wnt activation by lithium treatment, as performed in this study, is not Wnt2b specific. Conditional inactivation of Wnt expression, for example, inhibition of ligand secretion by deletion of Wntless in a RPE-specific manner, could directly test whether secretion of Wnt2b from RPE impacts the number of ipsilateral RGCs. Conditional inactivation of Wnt signaling in the presumptive RPE induces abnormal development of the RPE (Fujimura et al., 2009). To maintain normal development of the RPE and analyze effects on RGC neurogenesis, manipulation of Wnt signaling could be specific to the period of RGC neurogenesis (E13-E15) and perturbed in the RPE itself. In summary, our past and present results support the hypothesis that cellular and molecular defects of the albino RPE during RGC neurogenesis have an impact on communication with the neural retina and the production of factors (potentially Wnts) that could affect ipsilateral RGC specification.

Mice were obtained from The Jackson Laboratory (Bar Harbor, ME, USA; MGI: 1855985) and maintained in a timed-pregnancy breeding colony at Columbia University. Conditions and procedures were approved by the Columbia University Institutional Animal Care and Use Committee in protocols AC-AAAG9259 and AC-AAAG8702. Heterozygous mice (Tyr^+/c-2J) were crossed with homozygous tyrosinase mutants (Tyr^c-2J/c-2J) to generate litters containing homozygous Tyr^c-2J/c-2J embryos (‘albino’) and heterozygous Tyr^+/c-2J embryos (‘pigmented’). Pigmented littermates were used as controls for albino embryos. Females were checked for vaginal plugs at approximately noon each day. Embryonic day (E) 0.5 corresponds to the day when the vaginal plug was detected, with the assumption that conception took place at approximately midnight.

Gene expression profiling was performed on E13.5 isolated RPE. To isolate RPE, the cornea, the lens and the neural retina were detached from eyecup and extraocular tissue was removed. The isolated RPE eyecup was treated in ice-cold RNAprotect cell reagent (Qiagen) and processed for RNA isolation using RNeasy Plus Mini Kit (Qiagen). For each biological replicate (pigmented and albino), RNA was isolated from pooled RPE tissues of three or four littermate embryos, generating more than 100 ng/µl for each sample. RNA quantity and quality were assessed using Agilent 2100 Bioanalyzer (Agilent). RNA samples (100 ng for each as input RNA) were labeled with biotin by 3′ IVT Expression Kit (Affymetrix). The labeled RNA was hybridized on Mouse Genome 430 2.0 Array chips (Affymetrix) and analyzed using GeneSpringGx11 software (Agilent). Differentially expressed genes were identified from three biological replicates by average expression level greater than 20 in at least one population, at least 1.5-fold differential expression and corrected P-value less than 0.05 by Benjamini-Hochberg multiple testing correction. Gene set size filters (min=15, max=500) of Gene Set Enrichment Analysis (GSEA) resulted in filtering out 1898/10295 gene sets. The remaining 8397 gene sets were used in the analysis. GSEA sets with FDR (q value)<0.05 were hand-curated into thematic categories to highlight transcriptional differences between populations. Analysis was carried out with GSEA software from the Broad Institute (Cambridge, MA, USA).

cDNA from E13.5 RPE sheets (collected from 10-12 embryos from three or four litters) was retrotranscribed from purified RNA using Superscript III Reverse Transcriptase (Invitrogen). Quantitative PCR (qPCR) was performed using a Stratagene MX3000 with SYBR Green PCR Kit (Applied Biosystems) and a Takara Thermal Cycler Dice with SYBR Fast qPCR kit (Kapa Biosystems). Changes in gene expression were quantified using the 2^−ΔΔCT method with normalization to hypoxanthine phosphoribosyltransferase (HPRT). qPCR-specific primers are listed in Table S5. A total of three biological independent experiments were performed in triplicate (n=3).

For in situ hybridization, embryos were fixed by immersion with 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS, pH 7.4) at 4°C overnight. After fixation, tissue was washed with PBS, cryoprotected with 30% sucrose in PBS for 24 to 72 h at 4°C and frozen in dry ice. For immunohistochemistry, embryos were fixed by immersion in 4% PFA in PBS for 1 to 1.5 h and cryoprotected in 10% sucrose in PBS. Consecutive coronal sections (20 µm) were collected through the retina with a cryostat (Leica Biosystems) on Fisher frosted microscope slides.

Lithium treatment was performed as previously described by Lancaster et al. (2011). Briefly, we injected intraperitoneally equimolar lithium chloride or sodium chloride (10 µl of a 600 mM stock solution in normal saline per gram body weight) in timed pregnant dams every 24 h from E12.5 until E14.5.

In situ hybridization was performed as described previously (Kuwajima et al., 2012) with specific antisense DIG-labeled riboprobes for Wnt2b (a gift from Dr T. Jessell, Columbia University, NY, USA), Fzd7, Sfrp2 and Shh (gifts from Dr P. Bovolenta, Universidad Autónoma de Madrid, Spain), Otx1 and Msx1 (gifts from Dr X. Zhang, Columbia University, NY, USA and Dr V. Wallace, University of Toronto, Canada), Bmp4 (a gift from Dr J. Dodd, Columbia University, NY, USA), Fzd5 (a gift from Dr G. Papaioannou, Columbia University, NY, USA), Notch2 (a gift from Dr G. Fishell, Harvard University, Boston, MA, USA), and Axin2 (a gift from Dr G. Oliver, Northwestern University, Evanston, Il, USA and Dr F. Constantini, Columbia University, NY, USA). For Cx43 and Sox2 (Wang et al., 2016) and Wif1, the unique sequence was amplified by PCR from E13.5 mouse RPE cDNA using primers generated from the mouse sequence. In some cases, to reduce the concentration of melanin in pigmented RPE, which masks the in situ hybridization signals, sections were bleached after color reaction of in situ hybridization as described previously (Foss et al., 1995; Iwai-Takekoshi et al., 2016; Orchard and Calonje, 1998). In brief, sections were incubated with 0.25% KMnO₄ (Sigma 23851) in PBS at room temperature for 30 min, washed in PBS and then incubated in 1% oxalic acid (Sigma, O-0376) at room temperature for 1 min and washed again in PBS. Experiments were repeated at least three times in different animals from different litters and produced consistent results.

For immunohistochemistry, slides were rinsed in PBS and antigen retrieval was performed prior to blocking by incubating slides in 10 mM sodium citrate and 0.05% Tween-20 (pH 6.0) at 95°C for 20 min. Slides were blocked in 10% normal goat serum and 0.2% triton in PBS for 30 min, then incubated in primary antibodies diluted in 1% normal goat serum and 0.2% triton in PBS overnight at 4°C. Incubation with specific secondary antibodies was performed for 2 h at room temperature. The following primary antibodies were used: rabbit anti-Zic2 (1:10,000; a gift from Dr S. Brown, University of Vermont, Burlington, VT, USA), mouse anti-Isl1/2 (1:100; a gift from Drs S. Morton, Columbia University, NY, USA and T. Jessell), rabbit anti-PH3 (1:200, Millipore, #07-081), anti-Lef1 (1:200, Cell Signaling, #2230S). Secondary antibodies used included donkey anti-rabbit Alexa488 and Alexa594, and donkey anti-mouse IgG Alexa488 and Alexa594 (1:500, Life Technologies). Hoechst 33258 (Life Technologies, #H3569) 1 µg/ml in PBS was used to counterstain cell nuclei.

The Wnt2b-positive region in the eye was analyzed from sections of retina after in situ hybridization at E15.5, imaged using a Zeiss AxioImager M2 microscope. Four retinal sections were analyzed from each animal: two sections anterior to the section with optic nerve exit, the section with optic nerve exit and one posterior to the section with optic nerve exit. Contours of RPE perimeter and the Wnt2b-positive region in the RPE were traced at 20× magnification using Neurolucida software (v11, MBF Biosciences; RRID: SCR_001775). The ratio of the Wnt2b-positive perimeter over RPE perimeter was calculated for each retinal section and averaged from five animals.

The expression domains of Axin2 and Lef1 in tissue from lithium-treated animals were analyzed in sections of E15.5 retina after in situ hybridization and immunohistochemistry. Two sections per embryo and three embryos per condition were analyzed. The contours of the peripheral retina (a sector of 300 µm long, as measured on the superficial aspect of the retina) and the Axin2-positive region in this region were traced at 10× magnification using ImageJ (NIH; RRID: SCR_003070). The ratio of the Axin2-positive area over the peripheral retinal area was calculated for dorsal and ventral retina of each retinal section. Average mean intensity±s.e.m. of Axin2 expression: control (NaCl) pigmented, 25.06%±2.13; control (NaCl) albino, 38.50%±5.70; lithium pigmented, 32.61%±4.15; lithium albino, 52.89%±5.67. For Lef1 expression, the contours of the Lef1-positive region in the CMZ were traced at 20× magnification using ImageJ. The mean intensity of the traced region was measured from dorsal and ventral CMZ in each section. Average mean intensity±s.e.m. of Lef1 expression: control (NaCl) pigmented, 89.32±4.99; control (NaCl) albino, 75.41±7.83; lithium pigmented, 101.90±9.04; lithium albino, 122.38±7.75.

Immunostained sections were imaged using a Zeiss AxioImager M2 microscope equipped with ApoTome, AxioCam MRm camera and Neurolucida software as previously described (Bhansali et al., 2014). A max projection merged stack of eight images (2 µm steps) was acquired with the ApoTome and 20× objective and used for analysis. After all images were renamed for blinding to genotype and treatment, background intensity was measured by ImageJ. We counted cells that met a threshold fluorescence intensity of at least three times the intensity of background. After subtraction of background, Zic2- and Isl1/2-positive cell numbers were quantified within a sector of 240 µm long as measured in the superficial aspect of the retina by using Meta Imaging Series Metamorph, as previously described (Bhansali et al., 2014). To delineate the region of interest, four consecutive sectors of 60 µm were traced starting at the most peripheral Isl1/2-positive RGCs in VT retina. Total cell counts from two consecutive sections caudal to the optic nerve were obtained from one animal and averaged from seven animals.

PH3 cell number was counted within a 200 µm long sector of the ciliary marginal zone (CMZ) and a 400 µm long sector from the ciliary margin toward central retina (see Fig. 4B). To delineate the region corresponding to the CMZ, two consecutive 100 µm long sectors were traced from the most peripheral Isl1/2 cells to the peripheral tip of the retina. To delineate the regions corresponding to the neural retina, VT retina was divided into four consecutive sectors of 60 µm each with the first sector traced starting at the most peripheral Isl1/2 RGCs. Total cell counts from four consecutive sections (two rostral to the optic nerve, one at the optic nerve and one caudal to the optic nerve) were obtained from one animal and averaged from counts from eight animals.

To determine statistical significance, all data were analyzed with GraphPad InStat version 3.1. After we confirmed the data met the assumptions of the tests, statistical significance was determined using an unpaired t-test and one-way ANOVA. A two-tailed P-value was used. Error bars represent s.e.m. *P<0.05; **P<0.01; ***P<0.001. For RT-qPCR analysis, all measurements were shown as dot plots instead of determining statistical significance because the total number of biological experiments is three.

We thank Mika Melikyan for mouse breeding; John Peregrin, Sania Khalid and Corrine Quirk for technical support; and Jane Dodd and members of the Mason lab for discussions and comments on the manuscript. We also thank Dr Lorraine Clark and members at Genomics Core at the Taub Institute, Columbia University Medical Center for assistance with the gene expression analysis based on Affymetrix GeneChips.