The transcriptome of retinal Müller glial cells

Müller glial cells are the major type of glia in the mammalian retina. To identify the molecular machinery that defines Müller glial cell identity and function, single cell gene expression profiling was performed on Affymetrix microarrays. Identification of a cluster of genes expressed at high levels suggests a Müller glia core transcriptome, which likely underlies many of the functions of Müller glia. Expression of components of the cell cycle machinery and the Notch pathway, as well as of growth factors, chemokines, and lipoproteins might allow communication between Müller glial cells and the neurons that they support, including modulation of neuronal activity. This approach revealed a set of transcripts that were not previously characterized in (Müller) glia; validation of the expression of some of these genes was performed by in situ hybridization. Genes expressed exclusively by Müller glia were identified as novel markers. In addition, a novel BAC transgenic mouse that expresses Cre in Müller glia cells was generated. The molecular fingerprint of Müller glia provides a foundation for further studies of Müller glia development and function in normal and diseased states.

Indexing terms: murine retina; Mü ller glia; microarray; molecular markers; BAC transgenic mouse; transcriptome Glial cells are found throughout the central nervous system. In the mammalian retina, three types of glia have been identified: Mü ller glia, astrocytes, and microglia. Astrocytes originate in the more medial aspects of the brain, and enter into the retina through the developing optic nerve (Stone and Dreher, 1987;Watanabe and Raff, 1988). Microglial cells are thought to be of mesodermal origin (Chan-ling, 1994). The only glial cells derived from the retinal neuroepithelium are the Mü ller glia. They arise late in development from multipotent progenitor cells (Sidman, 1961;Young, 1985;Turner and Cepko, 1987). In the mature retina, they are normally not mitotic but can re-enter the cell cycle under certain circumstances (Sahel et al., 1990;Dyer and Cepko, 2000).
Mü ller glial cells are radial and form the outer and inner limiting membranes. They surround neuronal cell bodies and processes within the retina and perform a wide range of functions, including structural stabilization, regulation of ion homeostasis, neurotransmitter recycling, and neuronal survival (Newman and Reichenbach, 1996;Bringmann et al., 2000;Mata et al., 2002). Recently it was shown that their specific cylindrical shape, their regular and parallel orientation within the retina, and the high refractive index of their stalk allow them to capture and transfer light efficiently through the retina (Franze et al., 2007). Mü ller glia can undergo reactive gliosis, when they divide and upregulate glial fibrillary acidic protein (GFAP) in response to neuronal damage (Ridet et al., 1997;Pekny and Pekna, 2004). Gliotic Mü ller glial cells also show altered expression of various enzymes, ion channels, and receptors. The significance of this response is poorly understood. The role of Mü ller glial cells in disease has become even more intriguing, because an intimate molecular and functional relationship between Mü ller glia and multipotent progenitors has become appreciated in various systems, leading to the suggestion that Mü ller glia might serve as a source of retinal neurons following injury (Fisher andReh, 2001, 2003;Ooto et al., 2004;Blackshaw et al., 2004, Osakada et al., 2007. Although extensive studies of the physiological and morphological properties of Mü ller glia have been performed (Reichenbach and Robinson, 1995;Newman and Reichenbach, 1996), much less is known of their molecular biology. In the retina, gene expression in the abundant rod photoreceptors is most well characterized (Blackshaw et al., 2001), and advances have been made in understanding the molecular nature of specific neuronal cell types such as amacrine and ganglion cells (Ivanov et al., 2006;Trimarchi et al., 2007). Considerably less is understood about the molecular complexity of Mü ller glia. In this study, we have employed microarray analysis of single cells to obtain a detailed profile of the Mü ller glial transcriptome. Expression of Mü ller glia enriched or specific transcripts was tested by in situ hybridization on retinal sections and on dissociated retinal cells. Many novel Mü ller glial genes have been found. Clusters of genes with characteristics related to Mü ller glial cell function, such as Kϩ and water homeostasis, neuronal signaling, and recycling of photopigments, were readily identified.

MATERIALS AND METHODS Isolation of single Mü ller glial cells and cDNA amplification
Single-cell cDNA generation was achieved as described (Brady and Iscove, 1993;Dulac and Axel, 1995;Tietjen et al., 2003;Trimarchi et al., 2007). In short, retinae from C57/BL6 adult mice were dissected in phosphate-buffered saline (PBS; pH 7.4) and then dissociated with papain (Worthington Biochemical, Freehold, NJ) for 15-20 minutes at 37°C and gentle manual trituration. Cells were pelleted and resuspended in PBS containing 0.1% bovine serum albumin (BSA). Dissociated cells were visualized under an inverted microscope, and Mü ller glial cells were identified based on their distinct morphology. They were transferred with a drawn glass needle into wash buffer (PBS/0.1% BSA) and then seeded into polymerase chain reaction (PCR) tubes containing cold lysis buffer. cDNA was generated from single cells by reverse transcription (Superscript II [Invitrogen, La Jolla, CA]). After poly(A) tailing of the synthesized first-strand cDNA product, PCR amplification of the samples was performed by using LA Taq polymerase (Takara Mirus Bio, Shiga, Japan; Trimarchi et al., 2007). The cDNA smears were analyzed on agarose gels.

Affymetrix array hybridization
Ten micrograms of each single-cell cDNA were digested for 13 minutes at 37°C with 1 U of RNase free DNase I (Roche, Indianapolis, IN) in One-Phor All buffer (100 mM Tris acetate, 100 mM Mg acetate, 500 mM K acetate).
Mouse genome 430 2.0 Genechip® oligonucleotide arrays were prehybridized in a Genechip® Hybridization Oven 320 (Affymetrix, Santa Clara, CA) in 1X MES buffer (100 mM MES, 1 M NaCl, 20 mM EDTA, 0.01% Triton X-100) with 0.5 mg/ml acetylated BSA and 0.1 mg/ml herring sperm DNA. The single cell cDNAs were heated to 99°C for 5 minutes, cooled to 45°C, added to the microarrays, and incubated in the hybridization oven at 45°C for 16 hours. The slides were then scanned on a Genechip Affymetrix Scanner 7G. Signals from microarrays were globally scaled with a target intensity set to 500 by using Affymetrix Microarray Suite Software (MAS 5.0). Signal levels for all Affymetrix probe sets on the mouse 430 2.0 microarray are listed in Supplemental Table T4 for the five Mü ller glial cells profiled in this study as well as the 23 single cells (21 immature cells, 2 adult rod photoreceptors) used for comparison.

Data analysis: hierarchical clustering and Fisher's exact test
Gene Cluster software (Eisen et al., 1998) was used to determine gene association by hierarchical clustering. The results were visualized with Treeview software (Eisen et al., 1998). The data were filtered such that any probe set that failed to reach a signal of 1,000 in any single cell was eliminated; 1,000 was found to be a distinct value, where the threshold of a present and absent call on the array was very clear. The data for the remaining probe sets were log transformed and normalized according to the software instructions.
Additionally, for each probe set that was detected with a signal of Ͼ1,000 in at least one microarray, the expression values were scaled by dividing each value by the maximum value for that probe set. The resulting scaled values were then binned into five equally sized bins as described (Trimarchi et al., 2007). All probe set pairs were analyzed for association by using the following procedure. First, a contingency table with five rows and five columns was obtained that recorded the joint distribution over bins for a given probe set pair. A P value for significant association was then calculated from this table by using Fisher's exact test. For multiple Affymetrix identifiers that correspond to a single gene, the P values were generated as follows. First, for each gene, all Affymetrix IDs that map to this gene were identified. Then, for each microarray, the maximal expression value for these Affymetrix IDs was identified. This maximal value was then used as the expression value for that gene in that array. This collapsing method is referred to as "each-max-value". A new ID was then generated by a comma-separated list of all Affymetrix IDs that were collapsed for a given gene. For each array and gene, the maximal value among the Affymetrix IDs was always used. The resulting expression profile of a gene was thus a combination of values measured by any of the collapsed Affymetrix probe sets and therefore does not always correspond to a single Affymetrix probe set profile.

In situ hybridization, Xgal histochemistry, and immunohistochemistry
RNA tissue hybridization was performed following the detailed protocol described (Trimarchi et al., 2007), which is based on previous techniques developed by Murtaugh et al. (1999). Retinae were dissected from young adult (ϾP21) C57/BL6 or CD1 mice. Tissues were fixed in 4% paraformaldehyde (PFA) in PBS (pH 7.4), dehydrated, and embedded in paraffin, or cryoprotected in 30% sucrose in PBS and embedded in Tissue-Tek OCT compound. In situ hybridization (ISH) was performed on 20-m retinal sections. The protocol used for dissociated cell in situ hybridization (DISH) was performed as in Blackshaw et al. (2004) and Trimarchi et al. (2007), in combination with immunohistochemistry. For antibody incubation and detection, slides were incubated in blocking buffer (PBS, 0.1% Triton-X100, and 5% goat serum) for 30 minutes at room temperature in a humidified chamber. Primary antibody (anti-Glul, mouse monoclonal antibody against glutamine synthetase; Chemicon/Millipore, Temecula, CA, MAB302; as immunogen glutamine synthetase purified from sheep brain) was added 1:500 diluted in PBS and incubated at room temperature for 1.5 hours. This antibody recognizes a single 45-kDa protein in adult retinal tissue by Western blot analysis (Chang et al., 2007), and staining of adult murine retinal sections revealed a characteristic glial expression pattern (Rhee et al., 2007).
Slides were then washed three times with PBT (PBS and 0.1% Tween-20), and then cy2-labeled secondary antibody (donkey anti-mouse IgG, cyanine cy2, Jackson Im-munoResearch, West Grove, PA, 715-225-150) in PBT (1: 250) was added. The slides were incubated at room temperature for 1 hour and afterwards washed again three times with PBT. At the end, slides were DAPIstained and mounted in gel/mount (Biomeda, Foster City, CA). For immunohistochemistry on 20-m OCT retinal sections, slides were blocked in blocking buffer (PBS, 0.1% Triton-X100, and 10% goat serum) for 45 minutes at room temperature in a humidified chamber. Primary antibody, polyclonal anti-Pax6 (purified rabbit antibody; Covance, PRB-278P; generated against the peptide QVPGSEP-DMSQYWPRLQ of the mouse Pax6 protein and purified on a peptide affinity column; Marquardt et al., 2001), and monoclonal anti-Glul (Chemicon/Millipore, MAB302) were added at 1:150 and 1:500, respectively, diluted in blocking buffer, and incubated overnight at 4°C. The anti-Pax6 antibody reacts strongly with a single band of approximately 50 kDa on Western blots of retinal tissue extracts (Davis and Reed, 1996).

Photography
Photography of immunohistochemical staining patterns on sections was done on an upright confocal microscope (Leica TCS SP2). Pictures of ISH and DISH experiments were acquired on either a Zeiss Axiophot microscope or a Nikon eclipse E1000 microscope, both equipped with a Nikon DXM 1200F digital camera. Brightness and contrast of the section ISH and DISH images were adjusted by using Adobe Photoshop 7.0 (San Jose, CA). Digital magnification was also done by using Adobe Photoshop 7.0.

Generation of Pdgfra BAC transgenic mice
The Cre recombinase gene, followed by the SV40 late polyA site and an FRT-flanked kanamycin cassette, was cloned into a pBSKS vector (Cre-pA-kan). Cre-pA-kan was amplified by PCR primers containing the following 50-bp homology arms and NotI sites (Lee et al., 2001;Liu et al., 2003): CGCCCCGCGGCCGCGGCTAATGCT-GTTTATGTTTGGCTTTTGTCATTTGCAGGTCTCAGG-AGCTATGCCCAAGAAGAAGAGGAAGGTGTCCAA and ATAATTGCGGCCGCATACCTGTGAGGAGACAGCTGA-GGACCAGAAAGACCTGGTGGGAGGTCCCTCCCGGC-GGATTTGTCCTACTCAGGAGAGCG. The PCR product was digested by NotI and cloned into the NotI sites of pBSKS (pdg-Cre-pA-kan-pdg). The pdg-Cre-pA-kan-pdg was digested by NotI and recombined into the RP23-116F7 BAC clone; the loxP sites of the recombined BAC were removed, and the BAC was linearized with AscI before pronuclear injection (Lee et al., 2001;Liu et al., 2003). The BAC was injected into C57/BL6 blastocysts, and transgenic animals were afterwards crossed to C57/ BL6 animals. carried out by using modified oligo dT primers. cDNA preparations from five individual Mü ller glial cells were hybridized to Affymetrix Genechip® Mouse genome 430 2.0 arrays. These arrays provide almost complete coverage of the mouse genome with 45,000 probe sets. The identity of cells as Mü ller glial cells was confirmed by expression of known marker genes for Mü ller glia. Glutamine synthetase (Glul), clusterin (Clu), dickkopf homolog 3 (Dkk3) (Blackshaw et al., 2004), and S100 calcium binding protein A16 (Seigel et al., 1996) were successfully detected ( Fig. 2 and Supplemental Fig. S1). All of the cells share expression of these key marker transcripts.
The values for the Affymetrix signals from the five Mü ller glial cells were averaged (Supplemental Table T2). There were 7,377 Affymetrix IDs with averages Ͼ 1,000, of which 871 had values Ͼ 10,000. Many known Mü ller glial specific genes, such as Glul, aquaporin 4 (Aqp4), and Clu/ApoJ were some of the most highly expressed genes ( Fig. 2), perhaps facilitating their previous discovery as Mü ller glial markers (Jenne and Tschopp, 1992). Additionally, this comparison of highly expressed genes with the data in the literature (Nagelhus et al., 1998; Black-  Biological processes and molecular functions were systematically assigned to genes with average expression values Ͼ 10,000 by using IPA (Ingenuity Systems, Redwood City, CA). Genes with the highest significance can be categorized into protein synthesis, cell-to-cell signaling, cellular assembly and organization, cellular function and maintenance, carbohydrate metabolism, cell morphology, and lipid metabolism (Supplemental Table T6).

Comparison of expression of Mü ller glia enriched transcripts to expression in other retinal cells
The Affymetrix signal values from the Mü ller glial cells were compared with those of several different types of single retinal single cells in order to gain an appreciation of how Mü ller glial cells differ from neurons and also to identify markers for Mü ller glial cells. This comparison included 2 immature rod photoreceptors, 2 adult rod photoreceptors, and 19 immature amacrine and ganglion cells (Trimarchi et al., 2007). The data were sorted according to decreasing average values from the five Mü ller glia samples (Supplemental Table 1) and are represented in Figure  2 as a heat map. The 100 transcripts expressed with the highest averaged values show an almost uniform general expression pattern across all Mü ller glial samples. Many genes that had high expression values in Mü ller glia were also highly expressed in rod photoreceptors and immature amacrine and ganglion cells. Some of these genes are involved in generic cellular functions, e.g., ribosomal subunits, eukaryotic translation elongation factor 1 ␣ 1 (Eef1a1), ubiquitins (Ubb and Ubc), and glyceraldehyde-3-phosphate dehydrogenase (Gapd).
This initial comparison revealed a dozen genes that appear to be preferentially expressed by Mü ller glia, including the tRNA ligase BING4 and the transcription factor Sox2. They also include the lipid transporter ApoE, the major apolipoprotein in the central nervous system (Pitas et al., 1987). Interestingly, ApoE has also been implicated in cell proliferation (Mahley, 1988) and in modulation of the innate and acquired immune response (Laskowitz et al., 2000). Another gene involved in lipid metabolism that was found enriched in Mü ller glia is the diazepam binding inhibitor (Dbi), which is also known to be expressed in glial cells of both the central and peripheral nervous systems (Yanase et al., 2002). Other Mü ller glia enriched genes are the water channel, Aqp4, and Abca8a, a transporter with ATPase activity, the isoprenoid binding protein Rlbp1, and the receptors Gnai2 and Gpr37. Gnai2 (also called Gi2) mediates signaling from the endothelin-B receptor to maintain mitogenic activity of neural progenitors in the brain (Sinohara et al., 2004).
Expression of the Gpr37 and Dkk3 transcripts in Mü ller glia was shown in a previous study using serial analysis of gene expression (SAGE; Blackshaw et al., 2004). Dkk3 is a gene involved in cell differentiation and negative regulation of Wnt signaling. Further, caveolin1 (Cav), which is important in nitric oxide metabolism and vesicle-mediated transport, Spbc25, a component of the Ndc80 kinetochore complex (McCleland et al., 2004), and genes with unknown biological function such as Itm2b, Jagn1, and 2310032F03Rik can be classified as enriched in Mü ller glia. The data suggest that these genes are specific for Mü ller glia in the adult murine retina. However, some genes are also expressed in retinal progenitor cells, as has been shown for Sox2 and Dkk3 (Blackshaw et al., 2004), and for Spbc25 (data not shown; Trimarchi and Cepko, in preparation). Mü ller glia enriched transcripts for synaptoporin (Synpr) and type IX procollagen alpha1 (Col9a1) might also be Mü ller glia specific, although both of these are also detectable in one immature ganglion cell (P0 A6). Some of the potentially Mü ller glia specific transcripts, such as BING4, Prss2, Ube1c, and Gnb1l could only be detected in three of five Mü ller glial cells. This might be an indication of heterogeneity within Mü ller glial cells, or it might be due to a technical issue.
The Mü ller glia marker clu, expected to be specific to Mü ller glia in this comparison, is also detected in one of the rod photoreceptors. Because the retinal pigmented epithelium (RPE) expresses clu as well, it is possible that this expression occurred through RPE contamination of the rod single-cell preparation (Trimarchi et al., 2007). A summary of these data can be found in Table 1. Verification of the expression pattern of these genes by another method is required, such as in situ hybridization on retinal sections, as was done for a subset in this study (below).

Identification of genes characteristic for Mü ller glial cells
To identify Mü ller glial enriched or specific genes, the analytical tool Cluster 3.0 (Eisen et al., 1998) and Treeview software (Eisen et al., 1998) were used for clustering the single cell profiles, as well as to create a visual map of the data. The five Mü ller glial cells were once again compared with immature amacrine and ganglion cells and with rod photoreceptors. However, instead of considering only the most highly expressed transcripts, as was done in the previous analysis (Fig. 2), the complete data set obtained from the microarrays was used for comparison. The data were filtered such that any probe set that failed to reach a signal of 1,000 in any single cell was eliminated. The data for the remaining probe sets was log transformed and normalized according to Eisen et al. (1998).
A representative part of the resulting expression level matrix is shown in Supplemental Figure S1 and a portion of that same cluster around Aqp4 is shown enlarged in Figure 3A. Interestingly, tight clustering of genes in a given cell type can be observed, with Figure 3A showing a portion of a Mü ller glial cluster, and Figure 3B showing a rod photoreceptor cluster. All Mü ller glial cells shared a similar gene expression profile over a wide range of genes (Supplemental Fig. S1). Other cell types showed expression values that were considerably distinct from those of the Mü ller glia. The data in Figure 3A of genes enriched in Mü ller glia suggest a Mü ller glia core transcriptome, which likely comprises genes encoding some of the key functions of Mü ller glia. One of the most prominent examples is the inwardly rectifying potassium channel (Kcnj10). Mü ller glia have been previously shown to play an active role in regulating retinal functions by allowing transport of small molecules through specific ion channels. Other potassium channels expressed in Mü ller glial cells include Kir4.1 and Kir2.1 (Newman and Reichenbach, 1996;Bringmann et al., 2000). Genes included in this cluster are also ApoE and Abca8a, which have been identified to be among the most highly expressed genes in the previous analysis (Fig. 2). Genes not previously iden-tified in Mü ller glia, such as two members of the lysosomal cystein protease family, cathepsin O (Ctso) and cathepsin H (Ctsh), and rhophilin, which is involved in actin cytoskeleton reorganization, have been found to be present in this cluster as well.
The expression of genes known to encode functions that might be relevant for Mü ller glia physiology was examined. For example, as noted above, there are retinal progenitor genes expressed in Mü ller glia, such as Dkk3 and Chx10 (Blackshaw et al, 2004;Rowan and Cepko, 2004), perhaps in keeping with their ability to produce neurons in some species under some conditions, or respond to injury by re-entering the cell cycle.
Expression of the paired-type homeobox, Pax6, was noted in four out of five of the single Mü ller glial cells (Fig.  5). Pax6 is a marker for progenitor cells and also for ganglion cells, amacrine cells, and horizontal cells in the mature retina. It has not been reported to be expressed in mature Mü ller glia. However, expression is consistent with the expression of other progenitor genes in Mü ller glia.
Examination of the expression of cell cycle genes by Mü ller glia showed that cyclinD3, Cdc14A, Cdk10, and Spbc25 were enriched in Mü ller glia. Several genes of the Notch pathway were expressed in significant amounts in Mü ller glia, such as Notch1 and Notch2, and might regulate the re-entry of Mü ller glia into the cell cycle under specific (e.g., pathological) conditions (Fig. 4). Growth factors such as Fgfs have been suggested to be important for short-range interactions among retinal cells. The facts that aFGF can be released from rod outer segments by a phosphorylation-dependent mechanism, and that apical processes of Mü ller glial cells surround the photorecep-tors, suggest that these factors may mediate communication between glial cells and neurons (Mascarelli et al., 1991). Fgfs, neurotrophic factors, and growth factor receptors are also suggested to be important in Mü ller glial proliferation (Milenkovic et al., 2003). Activation of several growth factor-and G-protein-coupled receptors has been shown to induce proliferation in cultured Mü ller glial cells (Ikeda and Puro, 1994). FGF r1, Ngf R, and Ogfr were highly expressed among Mü ller glia, whereas Eph and Igfb4 were expressed among a few Mü ller glial cells.
Under pathological conditions, Mü ller glia might additionally act as modulators of the immune and/or inflammatory response. Retinal glial cells have been shown to be a major source of cytokines after retinal detachment (Nakazawa et al., 2006). Recently, MCP-1 (monocyte chemoattractant protein 1) expression was shown in Mü ller glia after retinal detachment (Nakazawa et al., 2007). Components of the chemokine system that could be detected in Mü ller glia were Xcr1 and Cxcl16 (Fig. 4). Mü ller glia might also be involved in the prompt clearing of photoreceptor debris under pathological conditions. During Drosophila development, it has been shown that glial cells engulf degenerating axons through recognition mediated by drpr/ced-1 and ced-6/CED-6 (Awasaki et al., 2006). It has been proposed that apoptotic cells and degenerating axons of mature neurons are removed by a similar mechanism (Awasaki et al., 2006;MacDonald et al., 2006). The ability of Mü ller glia to phagocytose retinal cell debris and foreign substances has been previously shown in vitro (latex beads) and in vivo (melanin) after retinal detachment (Mano and Puro, 1990;Stolzenburg et al., 1992;Francke et al., 2001). However, two of the genes, Megf10 (multiple EGF-like-domains 10)/ced-1 and Abca1 (ATP-

Heterogeneity of Mü ller glia
Although the focus of this study was to define the transcriptome of Mü ller glia, it was noted that certain genes were expressed heterogeneously among the Mü ller glial cells analyzed (Fig. 4, Supplemental Figs. S1, S2), including common housekeeping genes such as cytoplasmic ␤-actin, ␤-2-microglobulin, heat shock protein 1 ␤, and TATA box binding protein. This might be due to technical or real differences. Indeed, other studies employing Affymetrix microarrays to examine housekeeping gene expression among different tissues, as well as our SAGE study of the retina during development, similarly found that housekeeping genes vary considerably in their expression levels (Warrington et al., 2000;Blackshaw et al., 2004;Lee et al., 2007).
Little is known about the significance and extent of Mü ller glial cell heterogeneity, although it is clear that molecular differences exist because the homeodomaincontaining transcription factor Chx10 is expressed in only a subset of all Mü ller glial cells (Rowan and Cepko, 2004). Heterogeneous expression of Chx10 could also be observed in the current study (Fig. 4). Heterogeneity of gene expression among three of the Mü ller glial cells was observed across a large cluster (Supplemental Fig. S1). This was further analyzed for a few of these genes by using in situ hybridization (see below). A recent expression analysis of astrocytes has revealed extensive molecular heterogeneity among that class of glial cells as well (Bachoo et al., 2004). Furthermore, morphological studies have identified extensive anatomical differences between Mü ller glial cells in the central versus peripheral chick retina (Anezary et al., 2001). Further characterization of the glial genes identified in this study will help clarify the diversity of Mü ller glial cells and perhaps shed light upon a potential functional diversity.

Mü ller glia enriched genes: Fisher's exact test
To examine the relationships of gene expression patterns by using a method other than hierarchical clustering, the data were subjected to an analysis using the Fisher's exact test (Trimarchi et al., 2007). This test calculates the probability that a particular expression pattern for a pair of genes across a number of arrays would occur by chance. Supplementary Table T3A shows the genes that have similar a pattern to that of ApoE, with P values more significant than 10 Ϫ4 . This method identified many of the same genes identified by using hierarchical clustering around ApoE. The Fisher's exact test was also applied to look for genes with a similar distribution to that of Aqp4 (Supplementary Table T3B). As expected, very similar genes were obtained as for ApoE, confirming the observation that ApoE and Aqp4 are enriched/specific in Mü ller glia. Some highly expressed and potentially Mü ller glia specific genes shown and discussed in Figure 2 could be identified through this method as well.

Validation of expression in Mü ller glia
To validate the expression of genes expressed in Mü ller glial cells, in situ hybridization on retinal sections was performed for a subset of genes (Fig. 5). The corresponding Affymetrix signal values are represented in the top panel of Figure 5 as a heatmap. Because the cells used for comparison were mainly immature cells, some of the genes that are indeed specific to Mü ller glia in the adult retina are also expressed in those immature cells that still contain a certain expression profile of (late) progenitor cells. The similarity of Mü ller glia gene expression to progenitor cell gene expression has been discussed previously ( Fig. 4 and Blackshaw et al., 2004). Mü ller glia transcripts were localized most prominently in the middle portion of the inner nuclear layer (INL), consistent with the location of the transcript in the cell body. Transcripts with this predicted restriction in expression pattern are Dkk3, ApoE, Spbc25, Apg4B, Pak3, Car2, Rlbp1, deltex 2 homolog (Dtx2), and pleiotrophin (Ptn) (Fig. 5A-F,H,L,M). Certain genes, such as Dbi, GPR37, and Glul ( Fig.  5G,N,K, respectively), gave punctuate staining in the middle portion of the INL, as well as diffuse expression in the outer nuclear layer (ONL), consistent with expression in ascending glial processes, and in the ganglion cell layer (GCL), consistent with expression in Mü ller glial descending processes and endfeet. Expression in the INL was either throughout the INL, as is the case for clu (Fig. 5J), or in a subset of cells in the middle of the INL, as is the case for Dbi (Fig. 5G). The pan INL pattern is suggestive of a gene being expressed in several cell types in the INL, whereas the more specific middle of the INL pattern is more consistent with expression restricted to Mü ller glia. Even transcripts expressed in only a subset of Mü ller glial cells profiled, such as Apg4b, Dtx2, and Ptn could be verified and detected by in situ hybridization (Fig.  5D,M,L, respectively).
It was of interest to determine whether some of the genes with moderate to high levels of Affymetrix signals in profiled Mü ller glia were specific to Mü ller glia. Specificity might give some insight into function and would indicate that such genes are suitable for use as specific markers of Mü ller glia. Although analysis of candidate Mü ller glial genes by section ISH is a fairly rapid and informative assay, it was unclear for some genes whether their expression was restricted to Mü ller glial cells. To determine specificity of expression more precisely, ISH on dissociated cells was carried out in combination with immunocytochemistry for the known glial marker glutamine synthetase (Glul). Dissociated cell in situ hybridization (DISH) for Itm2b revealed that it was exclusively expressed in Glul-positive cells. A modulator of the Notch pathway, Apg4b, as well as Dkk3, Aqp4, and ApoE were shown to be fairly specific to Mü ller glia as well (Fig. 6). DISH also allowed an assessment of heterogeneity of expression within Muller glial cells. The Rlbp1 transcript, expressed in only three of five cells in the microarray analysis, was found in only a subset of Mü ller glia, in 70% of the cells marked with the anti-Glul (Supplemental Fig.  S2). In contrast, there was a complete overlap of Itm2b and clusterin transcript expression with anti-Glul staining (Fig. 5, top panel, Supplemental Fig. S2).
Additional validation of GPR37 as a Mü ller glia specific gene in the murine retina was possible through the use of a transgenic mouse line, GPR37 tm1Dgen , available from the Jackson Laboratory. The GPR37 tm1Dgen mice express a transgene consisting of lacZ driven by the endogenous GPR37 promotor. Xgal staining on retinal sections showed lacZ expression in Mü ller glial cells (Fig. 5O).
Pax6 expression was further investigated by immunocytochemistry by using anti-Pax6 and anti-Glul antibodies (Fig. 5Q,R; see Supplemental Fig. S3 for magenta-green versions of images). Analysis of the sections with confocal microscopy showed that Pax6 is indeed expressed in adult Mü ller glial cells, as well as in some types of retinal neurons, as previously reported for neurons (de Melo et al., 2003), but not previously reported for Mü ller glial cells.

Creation of a mouse line expressing Cre for genetic access to Mü ller glia
To create a line of mice for future experiments in which gene expression specifically in Mü ller glial cells could be manipulated, a BAC was engineered to express Cre from a promoter that was predicted to be enriched or specific to glial cells (Kessaris et al., 2006). This BAC was used to create a transgenic mouse. Mice that successfully transmitted the BAC were assessed for Cre activity by crossing to mice carrying a nuclear lacZ floxed indicator allele, RC::PFwe (Farago et al., 2006). Strong nuclear lacZ expression was detected in all or nearly all Mü ller glial cells in the progeny of this cross (Fig. 5P). This may be a fortuitous finding, as inspection of the microarray data for Mü ller glia did not show expression of Pdgfra. Some Xgal staining was observed in the ONL and GCL, but it was not distinctly nuclear. Control Xgal staining of littermates did not show this lacZ activity, and thus the origin of the ONL and GCL staining is unclear.

DISCUSSION
The goal of this study was to define the transcriptome of mature Mü ller glial cells. A single cell profiling approach was used because total tissue preparations of the retina would only contain approximately 5% of cells as Mü ller glia (based on the percentage of cells showing Glul immunoreactivity after dissociation into single cells). A PCR-based amplification method was used to generate enough cDNA from a single cell to use as probe on Affymetrix microarrays. The microarray results were then validated by section ISH and DISH. Validation was particularly important because PCRbased protocols may lead to a higher false-positive rate, compared with the higher false-negative rate that is generally associated with linear amplification methods (Van Gelder et al., 1990;Iscove et al., 2002;Che and Ginsberg, 2004;Subkhankulova and Livesey, 2006). The microarray data from the single cells confirmed the high expression of many known Mü ller glial genes and also revealed the expression of many genes not previously known to be transcribed in Mü ller glia. By characterizing single, freshly isolated Mü ller glial cells, this study also circumvented the need to use FACS or in vitro culture, or to manipulate these cells further, and thereby minimized induced gene expression changes. We have found this method to be fairly sensitive and reproducible in the current study, as well as a previous study of developing amacrine and ganglion cells (Trimarchi et al., 2007).
We also created and characterized a Cre line for recombination and expression in Mü ller glial cells. This novel The Journal of Comparative Neurology line will provide genetic access for manipulations of Mü ller glia in mice, in addition to the recently published lentiviral vector-based approach applied in rats by Greenberg et al. (2007). The utility of such methods might be especially interesting in retinal disease backgrounds.

Relationship of Mü ller glia to progenitor cells and stem cells
A SAGE study, coupled with large-scale in situ hybridizations, showed that genes enriched or specific to Mü ller glia in the mature retina are also expressed in multipotent Fig. 6. Dissociated cell ISH confirms expression in Mü ller glia. A-J: Retinae from mature mice were dissociated, plated on slides, and processed for detection by ISH for the indicated probes (ApoE, Dkk3, Apg4b, Itm2b, Rlbp1). Merged images show DAPI stain and immuno-staining for Glul and ISH. Insets are digital magnifications. Scale bar ϭ 25 m in A (applies to A,B), C (applies to C,D), E (applies to E,F), G (applies to G,H), and I (applies to I,J). retinal progenitor cells (Blackshaw et al., 2004). Similarly, a microarray analysis of genes enriched in 4N retinal progenitor cells showed enrichment for genes expressed in Mü ller glia, relative to retinal neurons (Livesey et al., 2004). In addition, some genes expressed at a relatively high level in mature Mü ller glia, such as Glul and Car2, are also expressed in late retinal progenitor cells. To this list of progenitor genes expressed in Mü ller glia, we can now add Pax6, a gene required for formation of the retina and other ocular structures (Hill et al., 1991;Glaser et al., 1992). This extensive overlap in gene expression between Mü ller glial cells and progenitors, coupled with the observation that Mü ller glia can divide in response to some stimuli, might indicate that Mü ller glia are a form of retinal progenitor cell.
Late retinal progenitor cells might become Mü ller glia without an irreversible cell fate determination event, such as occurs when a neuron is generated. Progenitor cells becoming Mü ller glia might retain much of the progenitor gene expression program, including those genes that allow a cell to divide (Fig. 3C), whereas upregulating and/or initiating expression of genes required for specialized Mü ller glial functions, such as ApoE, Glul, and GFAP. Candidate genes for regulation of this transition from retinal progenitor cell to Mü ller glia are the genes Notch1, Hes1, Hes5, and Rax, as overexpression of each of these genes in retinal progenitor cells leads to production of cells with at least some properties of Mü ller glia (Bao and Cepko, 1997;Furukawa et al., 2000;Hojo et al., 2000;Jadhav et al., 2006a,b). In contrast, overexpression of positive bHLH genes, such as Math 3, NeuroD, and many others, leads to a reduction in Mü ller glial production (Morrow et al., 1999;Cai et al., 2000;Inoue et al., 2002). The transition of a late progenitor cell to a Mü ller glia cell might thus depend on the balance of positive and negative bHLH genes, which might normally be regulated by Notch signaling. It appears that Notch activation generally favors formation of cells with glial characteristics not only in the retina, but also elsewhere in the nervous system (Morrison et al., 2000;Gaiano and Fishell, 2002). The potential of Mü ller glial cells to produce neurons in the adult retina might suggest that a discrete set of genes limits this ability.
In light of this notion, it is interesting to consider the Notch pathway genes further, not only as regulators of the formation of Mü ller glia as discussed above, but also as regulators of the re-entry into the cell cycle and/or as negative regulators of the production of neurons by Mü ller glia. The single-cell profiles for mature Mü ller glia include Notch1, Notch2, Hes1, Nrarp, and Cntn1, as well as regulators of Notch signaling, such as (Fig. 4), which bear examination in this context. Hes1, a downstream target of NICD, is also highly expressed by neural stem cells in the forebrain (Ohtsuka et al., 2001;Chojnacki et al., 2003), perhaps reflecting a similar role.

Similarity to other glial cell types
A comparison of the Mü ller glia transcriptome with those of other glial cell types of the central nervous system (Bachoo et al., 2004) failed to reveal common clusters of gene expression. This might be due to technical differences or might suggest a different genetic constitution among them. However, there were some transcripts with shared expression among Mü ller glia and other glial cell types. Mü ller glia enriched transcripts that are also enriched in astrocytes are peroxiredoxin 6, Basigin (also called CD147), caveolin, CD9 antigen, carboxypeptidase E, cystatin C, phospho protein enriched in astrocytes (Pea15), Ptn, Ret4, Slc2a1, and Slc3a2 (Bachoo et al., 2004;Vincent et al., 2005). Preliminary comparisons also imply that Mü ller glial cells are less related to Schwann cells (peripheral glia), because many transcripts highly expressed in Mü ller glia are depleted in Schwann cells, i.e., ApoE, clu, cystatin C (Cst3), enolase1 ␣ (Eno1), Gapd, Glul, malate dehydrogenase (Mdhl), and myotrophin (Mtpn) (Vincent et al., 2005).
Lastly, on the basis of the investigation presented here, we hope to continue our efforts to understand better the role(s) of Mü ller glia in retinal degeneration. Several diseases of the retina, such as retinitis pigmentosa and macular degeneration, lead to a severe loss of photoreceptor neurons and involve "activation" of Mü ller glia. The identification of the cellular events and changes that take place within Mü ller glia during the progressive neuronal cell death might be applicable to a therapeutic strategy.