ASCL1 reprograms mouse Müller glia into neurogenic retinal progenitors

Many non-mammalian vertebrates have the ability to regenerate new neurons after retinal injury. In fish, initially quiescent MG respond to chemical- or light-induced damage by re-entering the cell cycle and de-differentiating to form multipotent progenitors; these progenitors give rise to all retinal neuron subtypes and repair the lost neurons (Raymond et al., 2006; Bernardos et al., 2007; Fausett et al., 2008; Thummel et al., 2008; Ramachandran et al., 2010). However, mammals have a restricted ability to regenerate the retina. In rodents, MG can form new neurons in response to damage and growth factor stimulation, but only in very limited numbers (Ooto et al., 2004; Close et al., 2006; Osakada et al., 2007; Wan et al., 2008; Takeda et al., 2008; Karl et al., 2008; Del Debbio et al., 2010). Instead, mammalian MG normally respond to damage by becoming reactive and undergoing gliosis (Bringmann et al., 2009; Dyer and Cepko, 2000). We are interested in finding mechanisms to enhance the regenerative response of mammals, and recent advances in somatic cell reprogramming towards neural fates have potential application in the mammalian retina.

Recent reports have demonstrated highly efficient conversion between distant cell lineages using a minimal number of transcription factors (Vierbuchen and Wernig, 2011). Glial cells are ideal candidates for reprogramming, as they share many properties with progenitor cells (Berninger, 2010; Nelson et al., 2011). In astrocytes, forced expression of a number of transcription factors promotes neuronal conversion (Heins et al., 2002; Berninger et al., 2007; Blum et al., 2011; Heinrich et al., 2010; Addis et al., 2011; Corti et al., 2012). Moreover, Ascl1 is emerging as a key factor involved in neuronal fate conversion. Vierbuchen et al. (Vierbuchen et al., 2010) demonstrated that viral expression of Ascl1 along with two other transcription factors, Brn2 (Pou3f2 – Mouse Genome Informatics) and Myt1l, could directly convert fibroblasts into neurons, although Ascl1 alone was sufficient to induce significant conversion.

Recently, Ascl1a was shown to be required for retinal regeneration in the fish. ascl1a is upregulated in proliferating MG within six hours of injury, and Ascl1a knockdown blocks MG proliferation and de-differentiation into progenitors (Fausett et al., 2008; Ramachandran et al., 2010; Ramachandran et al., 2012; Wan et al., 2012). Because Ascl1 is not upregulated in the mouse retina after N-methyl-D-aspartate (NMDA)-induced damage (Karl et al., 2008), we hypothesized that the limited regenerative capacity of mammalian MG might be due, in part, to their failure to activate Ascl1 after damage. Further support for this hypothesis is demonstrated by the neurogenic role of Ascl1 during normal retinal development. In the developing mouse retina, Ascl1 maintains progenitors by driving expression of components of the Notch pathway (Nelson et al., 2009; Jasoni and Reh, 1996). Knockout of Ascl1 increases the number of MG relative to other cell types (Tomita et al., 2000; Akagi et al., 2004), whereas Ascl1 overexpression in progenitors biases cell production towards photoreceptor, and possibly bipolar, cells in mice (Hatakeyama et al., 2001), amacrine cells in the chick (Mao et al., 2008), and bipolar and amacrine cells in human cells (Gamm et al., 2008).

In the present study, we demonstrate that viral expression of ASCL1 is sufficient to activate a neurogenic program in mammalian MG, both in dissociated cultures and in the intact retina. ASCL1 remodels the chromatin at retinal progenitor genes, activates their expression and downregulates glial genes. The reprogrammed MG differentiate into cells that resemble neurons in morphology, gene expression and their responses to neurotransmitters. Our results suggest that stimulating neurogenesis in MG with ASCL1 could provide an alternative strategy for repair of the retina after disease or injury.

Mice were housed at the University of Washington; protocols were approved by the University of Washington Institutional Animal Care and Use Committee. C57BL/6J mice (Jackson) were used except where indicated. Hes5-GFP mice (Basak and Taylor, 2007) have been previously characterized as an MG reporter line in vivo and for fluorescence-activated cell (FAC)-sorting (Nelson et al., 2011). Rlbp1-cre^ERT2 mice were derived from plasmid described by Vázquez-Chona et al. (Vázquez-Chona et al., 2009) and were crossed to R26-stop-flox-CAG-tdTomato mice (Jackson). αPax6-cre mice (R. Ashery-Padan, Tel-Aviv University, Tel-Aviv, Israel) (Marquardt et al., 2001) were crossed to R26-stop-flox-rtTA mice (Jackson) (Belteki et al., 2005). NMDA damage was performed as previously described (Ueki et al., 2012). Tamoxifen (Sigma) was administered intraperitoneally at 100 mg/kg in corn oil.

PLOC-hASCL1-IRES-turboGFP(nuc) (Open Biosystems) was used to generate PLOC-IRES-turboGFP(nuc). hASCL1-IRES sequence was inserted into the pTRE3G-mCherryn vector (Tet-On 3G, Clontech), and inserted into the pLVX-Tight-Puro vector (Clontech) using In-Fusion Cloning (Clontech). pLVX-Tet-On Advanced (Clontech) was used to express rtTA protein in vitro.

MG from postnatal day (P)12 mice were cultured [Neurobasal + N2, epidermal growth factor (EGF), 10% fetal bovine serum (FBS)] as previously described (Ueki et al., 2012) with 1 μM 5-ethynyl-2′-deoxyuridine (EdU). Lentiviral particles were added in Optimem (Gibco) or neural medium (Neurobasal + N2, B27, 1% tetracycline (tet)-free FBS), and 3-6 hours later medium was replaced. hBDNF (R&D Systems, 10 ng/ml), bFGF (R&D Systems, 100 ng/ml) and rGDNF (R&D Systems, 10 ng/ml) were added for longer cultures. 4-Hydroxytamoxifen (4-OHT; Sigma) was included where indicated at 10 μM.

Retinas of αPax6-cre;R26-stop-flox-rtTA mice were explanted at P12 as previously described (Ueki et al., 2012). Explants were infected with pLVX-tetO-hASCL1-IRES-mCherryn, and 3 μg/ml doxycycline was added at 2 days in vitro (div).

Cells were lysed in Trizol (Invitrogen), and RNA was extracted, followed by DNase-1 (Qiagen) digestion and RNA cleanup (Qiagen). Microarray data were generated using GeneChip Mouse Gene 1.0 ST Array (Affymetrix) at the Institute for Systems Biology (Seattle, WA, USA) (see Nelson et al., 2011). Microarray data were deposited in Gene Expression Omnibus under accession number GSE45835. cDNA was synthesized (iScript cDNA Synthesis Kit, Bio-Rad), and qPCR was performed (SsoFast EvaGreen Supermix, Bio-Rad) on a Bio-Rad thermocycler. Reactions were performed in triplicate and values normalized to Actb. ΔΔCt between ASCL1-infected and uninfected or GFP-infected MG was expressed as fold change (2^ΔΔCt). Standard error of the mean (s.e.m.) was determined from ΔΔCt values and log transformed. One-way Student's t-test was performed on ΔΔCt values compared with 0 (Yuan et al., 2006). qPCR primers are listed in supplementary material Table S1.

For Ascl1 ChIP, P0 retinas or cultured P12 MG were digested with papain into a single cell suspension and fixed with 1% formaldehyde for 10 minutes at room temperature (RT). Cells were sonicated at 4°C. Ascl1 IP was performed with 40 μl anti-mouse IgG magnetic beads (Invitrogen) and 4 μg mouse anti-Mash1 antibody (BD Pharmingen) or 4 μg mouse IgG (Millipore) against chromatin from 1×10⁶ (P0 retinas) or 2.5×10⁵ (cultured MG) cells per IP according to LowCell # ChIP Kit (Diagenode). IP and wash buffers were as described by Castro et al. (Castro et al., 2006). DNA sequences were quantified by qPCR and averaged from three to six independent experiments. For histone ChIP, cell suspensions were fixed with 0.5% formaldehyde (10 minutes at RT) and sonicated. IP was performed with 20 μl anti-rabbit IgG magnetic beads (Invitrogen) and 2 μg rabbit anti-H3K27me3 (Active Motif), rabbit anti-H3K27Ac (Abcam) or rabbit IgG (R&D Systems). Values were averaged from three to five independent experiments. See supplementary material Table S1 for ChIP primers.

Cells were collected in lysis buffer and processed as described by Ueki et al. (Ueki et al., 2012). Antibodies used were: anti-Cralbp (1:5000; Abcam), anti-glutamine synthetase (1:10,000; Millipore), anti-β-actin (1:10,000; Abcam) and horseradish peroxidase-conjugated goat anti-mouse (1:10,000; Bio-Rad).

Coverslips were fixed in 2% paraformaldehyde (PFA) and standard protocols were used (Ueki et al., 2012). EdU staining was carried out using the Click-iT EdU Kit (Invitrogen). Explants were fixed in 2% PFA and immunolabeled as whole mounts. For cryosections, retinas were fixed in 2% PFA and sectioned at a thickness of 12 μm. Primary antibodies used were: anti-Insm1 (1:100; Genway), anti-Mash1 (1:100; BD Biosciences), anti-Ascl1 (1:250; Gift of J. Johnson, UT Southwestern, TX, USA), anti-GsiB4 Lectin (1:500; Vector Laboratories), anti-calretinin (1:1000; Swant), anti-Cralbp (1:1000; Gift of J. Saari, University of Washington, WA, USA), anti-RFP (1:500; Clontech), anti-RFP (1:500; Abcam), anti-Tuj1 (1:500; Covance), anti-Map2 (1:200; Sigma M9942), anti-S100β (1:1000; Sigma), anti-GFP (1:500; Abcam), biotinylated anti-Otx2 (1:100; R&D Systems), anti-Islet1 (1:50; Developmental Studies Hybridoma Bank), anti-Sox2 (1:250; Abcam), anti-Sox2 (1:250; Santa Cruz), anti-Id1 (1:200; BioCheck), anti-Pax6 (1:600; Covance), anti-Hes5 (1:100; Santa Cruz), anti-Sox9 (1:500; Millipore), anti-phospho-histone H3 (PH3) (1:500; Millipore), anti-NG2 (1:100; Chemicon). Secondary antibodies from Invitrogen, Molecular Probes and Jackson ImmunoResearch were used at 1:400 or 1:500.

Imaging was performed using an Olympus Fluoview confocal microscope or Zeiss Observer D1 with AxioCam. Three to six random fields per coverslip (20×) were counted. Single slice images (1 μm thick) were counted from four to eight random fields per explant. Seven explants were analyzed for each neuronal marker. z-stacks (0.5 μm thick per slice) were also captured to analyze colocalization of signals in three dimensions. Time-lapse imaging was performed on an Axio Observer Z1 (Zeiss) using Axiovision 4.7 software (Zeiss); CO₂, O₂ and temperature were kept at standard cell culture conditions.

Dissociated cells on coverslips were incubated in a solution of Fura-2 (Invitrogen) and Pluronic F-127 (Invitrogen) in Hank's balanced salt solution (HBSS+) (Gibco) for 30 minutes at 30°C, then washed in HBSS+ for 30 minutes. Chemicals (NMDA, 100 μM; glycine, 10 μM; kainate, 10 μM; ATP, 100 μM; KCl, 50 mM; all obtained from Sigma) were diluted in HBSS and applied by bath perfusion at constant flow rate for 60 seconds at 30°C before a 60-second washout with HBSS. Images were taken every 4 or 6 seconds at 340 and 380 nm alternating excitation, and ratios of fluorescence intensity signals (F340/380) were determined (Metafluor; Molecular Devices). F340/380 values were normalized to baseline activity. Each experiment was conducted on multiple coverslips over multiple days, and at least three experiments were conducted for each agonist.

Whole-cell voltage clamp and current clamp recordings were made from ASCL1-GFP-expressing cells using an Axopatch 200B amplifier (Molecular Devices). Electrodes (2-4 MΩ) were filled with 148.5 mM potassium gluconate, 9 mM NaCl, 1 mM MgCl₂, 10 mM HEPES and 0.2 mM EGTA (330 mOsm, pH 7.2). The extracellular solution contained 119 mM NaCl, 5 mM KCl, 2.5 mM CaCl₂, 30 mM glucose, 20 mM HEPES and 1 μM glycine. Action potentials were elicited by injecting current into cells in whole-cell current clamp mode. Responses to 50 μM kainate (in extracellular solution) were recorded in response to 3-8 seconds of drug application, delivered with a local puffer pipette controlled by a Picospritzer (Parker Hannifin) while holding the membrane at −60 mV under voltage clamp. Spontaneous miniature postsynaptic currents were recorded during continuous 10-second sweeps while holding the membrane at −60 mV under voltage clamp.

To study efficient transcription factor-mediated conversion, we needed a reliable and well-characterized cell culture model of mouse MG. We recently demonstrated that MG from postnatal day 11-12 (hereafter referred to as P12) mouse retina have mature glial properties and can expand as dissociated cultures (Ueki et al., 2012). By this age, all of the progenitors have completed their final mitotic cell divisions and differentiated as retinal neurons or MG. Upon retinal dissociation, most neurons die within one day, and only a small number survive after initial passage. MG proliferate over the next week, and >95% of cells express common MG markers after 4-5 days in vitro (div). Other cell types were not found to be a major source of contamination by immunolabeling and microarray analysis for astrocyte-, neuron- and endothelial-specific markers (Ueki et al., 2012), as well as by immunolabeling for the pericyte marker NG2 (Cspg4 – Mouse Genome Informatics) (supplementary material Fig. S1A).

To test whether ASCL1 was sufficient to reprogram MG into neurogenic progenitors, we carried out the experimental design detailed in Fig. 1A. P12 MG were cultured for one week, and EdU was added throughout this period to distinguish MG and their progeny from surviving post-mitotic neurons. Cells were passaged to remove surviving neurons and then infected with lentiviral particles expressing GFP or ASCL1-GFP in neural medium (1% FBS + B27). More than 85% of P12 MG had incorporated EdU by 4 days post-infection (dpi) (~10 div), in both GFP and ASCL1-infected cultures (Fig. 1B,C). Ascl1 expression was robust at 4 dpi; 85.4±2.7% of EdU+ ASCL1-infected P12 MG expressed Ascl1 protein compared with 1.8±0.9% of GFP-infected MG (Fig. 1B,D). At 4 dpi, ASCL1 mRNA was highly expressed; Fig. 1E shows the qPCR for human ASCL1 in ASCL1-infected cultures compared with paired GFP-infected cultures. ASCL1-infected MG had >2¹¹ times (2000-fold) more ASCL1 than the GFP-infected MG.

Because P12 MG do not normally express Ascl1, we initially assessed whether ASCL1 would bind to its predicted targets (Nelson et al., 2009; Castro et al., 2011) in these cells after overexpression. Chromatin immunoprecipitation (ChIP) with an Ascl1 antibody was performed in newborn mouse retina, when endogenous Ascl1 and its predicted targets are highly expressed (Fig. 2A, gray bars). There was significant pull-down of Ascl1 at the promoters of Dll3, Hes6, Mfng, Hes5 and Dll1 in P0 retina. Predictably, these targets were not enriched for Ascl1 in GFP-infected P12 MG (supplementary material Fig. S2). However, 4 days after ASCL1 infection, Ascl1 was significantly enriched at these promoters (Fig. 2A; supplementary material Fig. S2, blue bars), indicating that ASCL1 can bind its developmental targets in P12 MG.

Next, microarray analysis was used to understand the extent to which ASCL1 may activate developmental expression patterns in the infected MG. Although many progenitor genes are normally expressed in MG, we recently characterized a subset of genes, primarily proneural factors and Notch pathway components, that are highly expressed in Hes5− GFP+ FAC-sorted progenitors, but not in MG (Nelson et al., 2011). In Fig. 2B, we compared the expression levels of these genes in Hes5− GFP+ retinal progenitors (P0 RPCs) with those in dissociated P12 MG cultures infected with GFP or ASCL1 at 4 dpi. ASCL1 induced the expression of many of these progenitor-specific genes in P12 MG. Many of the genes that are highly upregulated in ASCL1-infected glia are known targets of Ascl1, such as Dll1, Dll3, Hes5, Hes6, Gadd45g, Fabp7, Id1, Id3 and Mfng, whereas others are not known Ascl1 targets, e.g. Neurog2, Mycn, Olig2 and Fgf15. The Notch pathway genes Heyl, Hey1, Hes1 and Dner were also increased. A gene ontology (GO) analysis of the genes that were upregulated in MG after ASCL1 infection (Eden et al., 2007; Eden et al., 2009) revealed that the biological function terms with the highest P-values were those associated with ‘nervous system development’ and ‘regulation of neurogenesis’; terms related to synaptic transmission, conduction of nerve impulses and cell cycle/DNA replication were also enriched in the genes upregulated in ASCL1-infected MG cells (supplementary material Table S2). Although these changes are consistent with ASCL1 reprogramming MG to the retinal progenitor state, not all progenitor-specific genes were upregulated by ASCL1 infection (supplementary material Fig. S3). Genes expressed specifically in early embryonic retinal progenitors, such as Hmgb3 and Sfrp2, are examples of genes that were not increased.

These changes in gene expression were then validated by qPCR by comparing expression levels of progenitor genes in infected MG relative to those present during development. Figure 2C shows levels of expression in newborn (P0-P1) mouse retina or ASCL1-infected P12 MG relative to GFP-infected P12 MG. At 4 dpi, there was very good correspondence between increases in gene expression observed in the microarray data and the qPCR experiments. Almost all genes were significantly increased in ASCL1-infected P12 glia at 4 dpi, though in most cases the level of expression did not reach that observed in the P0 progenitors. At 6 and 8 dpi, progenitor genes were still expressed in ASCL1-infected glia; however, by 13 dpi, the level of the most highly upregulated genes, Hes5, Mfng and Heyl, were decreased (supplementary material Fig. S4). These data suggest that the induction of progenitor genes in MG by ASCL1 infection is, in part, transient.

To test whether similar changes could be induced in MG from adult retinas, we used a recently developed protocol to isolate MG from adult (P30) mouse retina after NMDA-induced neurotoxic damage to the neurons. Adult glia from NMDA-damaged retinas proliferated in vitro, and the majority of cells were labeled with EdU and the glial marker S100β (supplementary material Fig. S1B). Typically <1% of the cells in the cultures were isolectin B4+ endothelial cells (supplementary material Fig. S1B). When adult MG were infected with ASCL1, we found similar levels of activation of progenitor genes as in P12 MG (Fig. 2C, red bars).

Microarray and qPCR results were further validated using immunofluorescence. Hes5, Insm1 and Dll1 are highly expressed in developing retinal progenitors (Nelson et al., 2007; Nelson et al., 2009) and are not highly expressed in MG in vivo or in vitro on our microarray analysis. By contrast, after 4 days of ASCL1 overexpression ~15% of EdU+ P12 MG were immunoreactive for Hes5 (Fig. 3A,B) or Insm1 (Fig. 3C,D) with a smaller percentage (~10%) immunoreactive for Dll1 (not shown). The percentage of Hes5+ cells remained relatively constant from 3 to 7 dpi, consistent with the qPCR time course data.

The GO analysis of the microarray data further suggested that ASCL1-infected MG have a higher level of some cell cycle genes. Time-lapse recordings of the MG after ASCL1 infection showed that the infected cells continued to undergo mitotic divisions (Fig. 4A), and ASCL1-infected MG had a greater number of cells expressing the mitotic marker PH3 (Fig. 4B,C). Approximately 50% of these PH3+ cells also expressed the progenitor marker Hes5 (Fig. 4D).

The fact that ASCL1 is able to directly bind the promoters and/or proximal enhancers of progenitor genes Hes5, Dll1, Hes6 and Dll3 and activate their expression suggests that chromatin at these sites can be reprogrammed to an active state by ASCL1. To determine directly the chromatin status at progenitor genes, we probed the 5′ proximal regions of these genes for active and repressive histone modifications in both P0 retinal progenitors and P12 MG. Histone 3 lysine 27 tri-methylation (H3K27me3) represses expression of genes with this histone modification (Cao et al., 2002; Boyer et al., 2006), whereas H3K27 acetylation (H3K27Ac) is correlated with the promoters and enhancers of actively expressed genes (Karlić et al., 2010). ChIP-qPCR was performed for both modifications at the 5′ proximal promoters (or 5′ enhancer for Dll1). We observed a significant increase in the repressive H3K27me3 modification (Fig. 5A) and a significant decrease in the active H3K27Ac modification (Fig. 5B) between progenitors and MG at all four gene loci.

Because repressive chromatin modifications are established at genes that can be subsequently activated by ASCL1, we reasoned that ASCL1 might reprogram the chromatin at these gene loci from a state of repression to activation. We therefore performed ChIP for H3K27me3 and H3K27Ac on dissociated P12 MG infected with ASCL1 or GFP at 4 dpi. We observed a significant decrease in H3K27me3, except at Hes6 (Fig. 5C), and an increase in H3K27Ac (Fig. 5D) at progenitor gene loci in ASCL1-infected MG. These data demonstrate that ASCL1 has the ability to bind genes with repressed chromatin and reprogram the chromatin to an active state.

We also assessed the chromatin state of the Ascl1 promoter in MG. We found that, like its targets, the promoter of Ascl1 acquires the H3K27me3 repressive histone modification in MG, unlike in P0 retinal progenitors (supplementary material Fig. S5A), while at the same time losing the H3K27Ac active modification (supplementary material Fig. S5B). Overexpression of ASCL1 in MG was able to partly remodel the Ascl1 promoter, causing an increase in the H3K27Ac modification (supplementary material Fig. S5D); however, there was a persistence of the repressive modification H3K27me3 at these loci (supplementary material Fig. S5C), which might prevent significant expression of endogenous Ascl1 in the reprogrammed MG.

Because ASCL1-infected MG robustly upregulated progenitor genes, we next assessed whether these changes were accompanied by a loss of glial identity. Using microarray analysis, we analyzed a set of MG-specific genes, previously characterized in MG by Ueki et al. (Ueki et al., 2012) (Fig. 6A). These genes are highly enriched in dissociated GFP-infected P12 MG at levels comparable to those in FAC-sorted Hes5− GFP+ MG. Many of these genes, including Slc1a3, Sox9, Rlbp1, Aqp4 and Glul, were downregulated following ASCL1 infection of P12 MG. We confirmed that two of these glial genes, Rlbp1 (also known as Cralbp) and Slc1a3 (also known as Glast), were downregulated by qPCR following ASCL1 infection at 6, 8 and 13 dpi (Fig. 6B). Additionally, immunofluorescence of the glial marker S100β was highly reduced by 4 days after Ascl1 induction (Fig. 6C). We also carried out western blots for two MG-specific proteins, Rlbp1 and GS (Glul), and found that both of these proteins were reduced after ASCL1 infection (Fig. 6D,E).

We next assessed the ASCL1-reprogrammed MG for neuronal properties. By 6 days after ASCL1 infection, these cells had lost their glial morphology and adopted a neuronal appearance. Whereas cultured MG have large, flat cell bodies, neuron-like cells with small, round nuclei and long, thin processes were observed after ASCL1 infection (Fig. 7A).

These morphological changes were associated with robust expression of general neuronal markers. By 6 dpi, ASCL1-infected MG highly expressed the pan-neuronal marker βIII-tubulin (Tubb3; also known as Tuj1) (Fig. 7B). qPCR confirmed that Tubb3 was significantly enriched in ASCL1-infected MG compared with controls, and that P12 and adult MG upregulated Tubb3 to similar levels (Fig. 7C). To confirm that these neurons were newly generated from MG, we cultured MG from Rlbp1-cre^ERT2;R26-flox-stop-tdTomato P12 retinas (supplementary material Fig. S6A). These mice express cre-recombinase under the control of the MG-specific Rlbp1 (also known as Cralbp) promoter, and, following tamoxifen administration, Cralbp+ MG expressed tdTomato both in vivo and in dissociated P12 cultures (supplementary material Fig. S6B,C). Following ASCL1 infection, Rlbp1-cre^ERT2;tdTomato-derived MG adopted neuronal morphology and expressed Map2 (Fig. 7D; Fig. 8A) and Tuj1 (not shown).

In addition to the lineage-tracing experiments, we also followed the progeny of the MG with EdU. The majority of the MG incorporate EdU in vitro (Fig. 1), whereas any neurons that survived from the initial retinal dissociation do not. We found that the majority of Tuj1+ and Map2+ cells in the ASCL1-infected MG cultures were EdU+ and were, thus, newly generated from proliferating MG. We quantified the number of EdU+ cells that labeled with these markers and found that >25% expressed Tuj1 or Map2 in ASCL1-infected P12 MG at 9-12 dpi (Fig. 8C). We also found many Tuj1+EdU+ and Map2+EdU+ double-labeled cells in adult MG at 10 dpi (Fig. 8B,D). Thus, progeny of ASCL1-reprogrammed MG can differentiate into cells with neuronal morphology and expression of pan-neuronal markers.

As ASCL1 can induce pan-neuronal markers and morphology, we investigated whether more specific retinal neuronal genes could be induced in cells derived from ASCL1-infected MG. We analyzed retinal specification genes that are normally expressed after progenitors have exited the cell cycle: Neurod4, Otx2, Neurod1, Prox1, Crx, Isl1 and Atoh7. All of these early post-mitotic neuronal markers were upregulated by 6 dpi in both P12 and adult MG (Fig. 8E).

These immature retinal neuronal markers were validated further by immunolabeling. Otx2 is a marker of early photoreceptor and bipolar cells (Brzezinski et al., 2010; Omori et al., 2011). Many of the progeny (~30%) of ASCL1-infected MG developed into cells expressing Otx2 (Fig. 8F,I). Another early bipolar marker, Islet 1 (Isl1), was expressed in ASCL1-infected EdU+ MG (Fig. 8G,J), although in a much smaller percentage of the cells. Calretinin (calbindin 2 – Mouse Genome Informatics), a marker of amacrine cells and some bipolar cells, was expressed by ~20% of EdU+ ASCL1-infected MG (Fig. 8H,K). Although the cells appeared to be progressing in their differentiation program, the majority of the differentiating neuronal cells continued to express progenitor markers. Otx2 and Id1 double-labeled cells are shown in Fig. 8L,M, but we also found Hes5 and Sox9 expression in cells labeled with neuronal markers Tuj1 and Map2 (not shown).

We also used qPCR and immunolabeling to determine whether mature retinal neuronal markers were expressed in the progeny of the ASCL1-infected MG. We observed a low, but consistent, level of expression of genes that are normally expressed in mature retinal neurons (supplementary material Fig. S7). The qPCR showed upregulation of Gad1 (also known as Gad67) and Bhlhe22 (also known as Bhlhb5), genes that are normally expressed in mature amacrine and bipolar neurons, respectively. In addition, ASCL1-infected MG expressed the bipolar markers Cabp5 and Vsx1, and the rod photoreceptor marker Nrl. However, the expression levels of most of these genes were very low, and other later-expressed photoreceptor genes, such as Opn2 (rhodopsin) and Opn1sw (also known as S opsin), were not detectable in the reprogrammed MG. These results indicate that ASCL1-reprogrammed MG can develop into cells expressing many early markers of retinal neuronal differentiation; however, a declining percentage expresses later markers and only a very few develop more mature marker expression. Interestingly, the predominant cell type generated in these cultures expresses markers consistent with the bipolar cell fate.

To assess whether ASCL1-induced neurons could have functional activity, we looked for gene expression changes in neurotransmitter receptors (Fig. 9A). By 4 dpi, ASCL1-infected MG upregulated many neurotransmitter receptors, the most prevalent of which were nicotinic cholinergic and kainate receptors. At the same time, ASCL1-infected MG downregulated P2Y and P2X purinergic receptors, which are normally highly expressed in MG (Wurm et al., 2009; Wurm et al., 2011).

We next tested whether cells derived from the reprogrammed MG could respond to pharmacological agonists using Ca²⁺ imaging. At 10-12 dpi, cells were loaded with the ratiometric Ca²⁺ indicator dye Fura-2. NMDA (plus glycine to potentiate responses), kainate (KA), ATP and KCl were bath-applied to cells, and ΔF340/380 responses were measured. ASCL1-infected MG were derived from Rlbp1-cre^ERT2;tdTomato retinas; tdTomato+ cells with neuronal morphologies (Fig. 9B, arrowheads) had detectable ΔF340/380 responses to NMDA and failed to respond to ATP, unlike neighboring cells with glial morphology (Fig. 9C). In some cases, we were able to locate the field after fixation, and, in the example shown, the cell was later identified by Tuj1 and EdU immunolabeling (Fig. 9B).

We next assessed overall changes in responsiveness to these agonists in ASCL1- or GFP-infected wild-type MG or retinal neurons from newborn mice (P0, 5-12 div). Figure 9D shows signals, plotted as area under the curve, for individual cells responding to NMDA, KA and ATP. ASCL1 infection caused a significant increase in responsiveness to NMDA and KA and a significant decrease in responsiveness to ATP. These differences were more pronounced in ASCL1-infected MG that had a neuronal or transitional morphology compared with those with a glial morphology or GFP-infected MG (supplementary material Fig. S8B). More modest differences were observed in response to KCl (supplementary material Fig. S8A). Thus, neuronal cells derived from reprogrammed MG responded to ionotropic glutamate agonists but had reduced responsiveness to purinergic receptor agonists. We also confirmed the Ca²⁺ imaging results by whole-cell patch clamp recordings (supplementary material Fig. S9). Injection of 40 pA could elicit action potentials from ASCL1-infected MG cells, application of kainate induced robust inward currents, and when the ASCL1-infected MG were co-cultured with immature retinal neurons, mini-excitatory postsynaptic currents (EPSCs) could be observed.

To test whether ASCL1 could induce neuronal conversion of MG in the intact retina, Hes5-GFP explants, which label MG, were infected with ASCL1. We first validated that MG in P12 Hes5-GFP retinas cultured as explants re-entered the mitotic cell cycle in response to EGF and incorporated EdU (Fig. 10A) (Ueki et al., 2012), as previous studies in other species suggest that MG have to re-enter the cell cycle prior to their de-differentiation to progenitor cells. MG typically migrate to the outer nuclear layer during regeneration in those species where this process occurs naturally, and we find that the same occurs in the explant cultures of mouse retina when stimulated to proliferate with EGF. To induce ASCL1 expression reliably, we then explanted retinas from αPax6-cre;R26-stop-flox-rtTA P12 mice, resulting in reverse tetracycline transactivator (rtTA) expression in the peripheral retina (Fig. 10B). EGF was added to promote MG proliferation, and EdU was included to track MG and their progeny. Explants were infected with tetO-ASCL1-mCherryn lentivirus in neural medium, and doxycycline was added after two days to activate ASCL1 and mCherry expression. ASCL1 and mCherry nearly perfectly colocalized in HEK293T cells as well as in explants (supplementary material Fig. S10); therefore, mCherry was used as a reliable marker for ASCL1-expressing cells.

Thirteen days after the induction of ASCL1-mCherry, retinal explants were immunolabeled with neuronal markers. Similar to what was observed in dissociated ASCL1-infected MG cultures, there were significant numbers of EdU+mCherry+ cells (ASCL1-infected MG) that expressed bipolar cell markers (Fig. 10C-E). Over 80% of ASCL1-infected MG (EdU+mCherry+) expressed Otx2 compared with <5% of uninfected cells (EdU+mCherry−) (Fig. 10C). Approximately half of the ASCL1-expressing proliferating MG progeny also expressed Islet 1, compared with <1% of the uninfected, EdU+mCherry− population (Fig. 10D). ASCL1-infected MG also expressed protein kinase C (PKC), a marker of more mature bipolar cells. Approximately 5% of ASCL1-infected MG (EdU+mCherry+) expressed PKC compared with <1% of uninfected MG (EdU+mCherry−) (Fig. 10E). Many of the MG-derived cells that expressed bipolar markers were in the inner nuclear layer, with the existing bipolar cells, though some were also located ectopically in the outer nuclear layer. No significant cell death was observed by DAPI staining (data not shown), suggesting that newborn cells survive in the native environment.

Here, we report that mammalian MG can be reprogramed to a neurogenic state by forced expression of the proneural transcription factor ASCL1. ASCL1 remodeled repressive chromatin at its targets to an active state and induced expression of progenitor genes, while downregulating MG genes. Reprogrammed MG produced cells that express pan-neuronal markers, exhibit neuronal morphology and upregulate many retinal-specific neuronal markers. Furthermore, the MG-derived ASCL1-induced neurons exhibited functional activity in response to appropriate neurotransmitter receptor agonists. Finally, ASCL1-reprogramming of MG in the intact retina revealed that newly generated neurons primarily differentiated as bipolar cells that integrated within the pre-existing neuronal population.

ASCL1 was sufficient to induce the re-expression of many progenitor genes, while inhibiting the glial differentiation program. Direct targets of Ascl1, Dll1, Dll3, Hes5, Hes6 and Mfng (Ueno et al., 2012; Castro et al., 2011), were bound and re-activated in the MG, and progenitor genes that are not direct Ascl1 targets, such as Neurog2 and Mycn, were also increased; moreover, the Ascl1 promoter was partly remodeled to a more active state. Taken together, these results suggest that ASCL1 initiates extensive reprogramming of the MG.

The response of mouse MG to ASCL1 expression is reminiscent of observations in zebrafish after retinal damage, in which MG de-differentiate into progenitors after upregulating ascl1a (Fausett et al., 2008; Ramachandran et al., 2010; Thummel et al., 2008; Raymond et al., 2006). We observed that ASCL1 overexpression upregulates Notch pathway components and Insm1, which are downstream of Ascl1a and required for de-differentiation in the regenerating fish retina (Ramachandran et al., 2012; Wan et al., 2012).

ASCL1 appears to reprogram MG through a transitional progenitor state, and stimulates proliferation; however, we cannot say whether this is an obligatory step in the process of reprogramming MG, as we did not determine whether the neuronal cells derived from the MG have undergone a mitotic division after infection; moreover, direct neuronal reprogramming of astrocytes (Heinrich et al., 2010) to neurons with Neurog2 does not require mitotic division.

The reprogramming of MG by ASCL1 includes remodeling of the chromatin at the promoters of progenitor genes, including Ascl1 itself, for at least two important histone modifications. During glial differentiation, progenitors acquire the repressive chromatin mark H3K27me3 and lose the activation mark H3K27Ac at Ascl1 and its targets. However, at targets where Ascl1 was bound by ChIP, ASCL1-expressing MG lost this repressive mark and gained the activation mark (although it is difficult to compare quantitatively the epigenetic state of these chromatin modifications between the progenitors and the reprogrammed MG owing to heterogeneity in the populations). Several models have been put forth to explain reprogramming in contexts in which transcription factors are forcibly expressed (Vierbuchen and Wernig, 2012). One model suggests a ‘permissive enhancer’, i.e. a transcription factor is able to bind at sites of open chromatin at enhancers of activatable genes (Taberlay et al., 2011). An alternative model suggests that these factors can act as ‘pioneers’ and bind repressed elements to pave the way for other factors (Zaret and Carroll, 2011; Cao et al., 2010).

MG-derived progenitor-like cells differentiated to form cells that shared morphological, immunohistochemical and functional characteristics with retinal neurons. ASCL1-reprogrammed cells adopted a distinctly neuronal appearance by 6 dpi, which coincided with their expression of pan-neuronal markers. These cells were responsive to ionotropic glutamate agonists, NMDA and kainate, and showed decreased responsiveness to the purinergic receptor agonist ATP. These effects are comparable to those from studies of neurogenic reprogramming in postnatal cortical astrocytes (Addis et al., 2011; Berninger et al., 2007; Blum et al., 2011; Corti et al., 2012; Heinrich et al., 2010; Heins et al., 2002) and pericytes (Karow et al., 2012). Berninger et al. (Berninger et al., 2007) and Heinrich et al. (Heinrich et al., 2010) found that Ascl1 or Neurog2 expression could convert fate-mapped postnatal astrocytes into morphologically distinct neurons by 4 dpi. However, astrocyte-derived neurons did not form synapses and mature membrane properties until 2-3 weeks in vitro, suggesting that maturation may take significantly longer than initial reprogramming and cell-type specification.

Newly generated neuronal cells from ASCL1-reprogrammed MG primarily differentiated as bipolar cells. In dissociated MG, ASCL1 induced early markers of many retinal neuron subtypes; however, later neuronal markers were more restricted to bipolar cells (Otx2, Islet 1, calretinin, Vsx1, Cabp5, Bhlhb5 and PKC) and, possibly, amacrine cells (Gad67, Bhlhb5, calretinin). The bias towards bipolar cells was also apparent in retinal explants, in which ASCL1-infected MG generated a large number of Otx2+ and Islet 1+ cells and a smaller number of PKC+ rod bipolar cells. This is consistent with the role of Ascl1 during normal development, as Ascl1 is expressed in late progenitors, which give rise to amacrine cells, bipolar cells and photoreceptors, but not ganglion cells (Brzezinski et al., 2011), and deletion of Ascl1 in mice leads to a reduction in bipolar cells and photoreceptors (Akagi et al., 2004; Tomita et al., 2000; Brzezinski et al., 2011).

Previous reports have suggested that dissociated MG, from both mouse and human, can act as stem cells by self-renewing and differentiating towards multiple neural lineages (Das et al., 2006; Nickerson et al., 2008; Lawrence et al., 2007; Giannelli et al., 2010). However, many of the stem and progenitor markers used in these studies, such as nestin and Sox2, are expressed in cultured MG (Karl et al., 2008; Bhatia et al., 2011; Lin et al., 2009), and they are not sufficient to specify the neural stem/progenitor fate. Recently, reports have suggested that MG spontaneously produce neurons in vitro (Giannelli et al., 2010) or when co-cultured with hippocampal explants (Das et al., 2006). However, neither experiment distinguished putative new neurons from those that survived the dissociation (e.g. using a thymidine analog or lineage-tracing method). Significant numbers of neurons can survive in dissociated retinal cell cultures, even after passage, and thus a method to distinguish surviving neurons from newly generated ones is crucial.

Additionally, reports have claimed that MG in mouse and rat retina can regenerate new neurons after damage if treated with exogenous factors. Wnt3a, EGF, fibroblast growth factor (FGF), insulin-like growth factor (IGF), retinoic acid (RA), Notch, N-methyl-N-nitrosourea (MNU) and α-aminoadipic acid (α-AA) have all been shown to stimulate a small number of MG to re-enter the mitotic cell cycle (Ooto et al., 2004; Close et al., 2006; Osakada et al., 2007; Wan et al., 2008; Takeda et al., 2008; Karl et al., 2008; Del Debbio et al., 2010). Some of the progeny of the bromodeoxyuridine (BrdU)+ MG were reported to differentiate characteristics of various types of retinal neurons, depending on the study and the treatment. However, only one of these studies used confocal imaging and 3D reconstruction to characterize definitively the BrdU+ cells (Karl et al., 2008), and that study was only able to find a small number of new amacrine cells generated from the BrdU+ MG (3.6%).

It is unclear why ASCL1-reprogrammed neurons do not more efficiently differentiate and mature. ASCL1 alone was sufficient to activate pan-neuronal and early markers of retinal neurons robustly, but we found lower levels of genes expressed by more mature retinal neurons, such as rhodopsin and S opsin. ASCL1 can activate its direct targets, including Otx2, which commits progenitors to photoreceptor or bipolar fates. However, direct targets of Otx2, such as Nrl and Pde6b, are not robustly expressed, and only 5% of ASCL1-infected MG differentiated to PKC+ bipolar cells, although much higher percentages expressed Otx2 and Isl1. Several potential mechanisms could account for the limited ability of MG-derived neurons to mature fully. Continued expression of progenitor markers, such as Sox9, Id1 and Hes5, may limit full differentiation, though this might also be due to continued expression of ASCL1 in the cells. It may also be the case that factors in the retinal environment, not present in the dissociated cultures, are needed for differentiation and/or survival of the new neurons derived from the ASCL1-reprogrammed MG. Evidence for this comes from the fact that a much higher percentage of the ASCL1-reprogrammed MG express Otx2 and Isl1 in the explant cultures than in the dissociated cell cultures. Epigenetic restrictions in the MG might also limit full reprogramming. Although repressive histone modifications are remodeled at the Ascl1 targets, other genes might persist in a repressed state. Genes required in photoreceptors, e.g. rhodopsin, might be repressed by different mechanisms than are progenitor and early neuronal specification genes. A recent report by Powell et al. (Powell et al., 2012) found that retinal regeneration in fish required expression of the cytidine deaminases apobec2a and apobec2b, which are involved in DNA demethylation.

Viral overexpression of ASCL1 in MG suggests a strategy for stimulating regeneration of the mammalian retina. This approach is complementary to transplantation of stem cell-derived neurons for retinal repair, as it allows for MG to be targeted for reprogramming within their native environment. Viral reprogramming of MG cells with ASCL1 predominantly generates bipolar neurons, so other reprogramming factors will probably be needed to direct the MG to photoreceptors and ganglion cells. Nevertheless, our study shows that viral reprogramming may allow MG to serve as a source of new retinal neurons to treat retinal degenerative diseases and one day provide a basis for gene therapeutic approaches.

We thank Drs Olivia Bermingham-McDonogh, Rachel Wong and Anna La Torre for helpful manuscript comments; Kristin Sternhagen for cloning expertise; and Ajay Dhaka for help with Ca²⁺ imaging experiments.