Retina Transcriptome and DNA Methylome Signatures Associated With Retinal M¨uller Glia Development, Injury Response, and Aging

Siyuan Lin, Jingyi Guo, and Shuyi Chen State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, China

Correspondence: Shuyi Chen, State PURPOSE. The purpose of this study was to systematically characterize and correlate the Key Laboratory of Ophthalmology, transcriptome and DNA methylome signatures of mouse Muller¨ cells that may underlie the Zhongshan Ophthalmic Center, Sun development, physiological functions, and regeneration capacity of these cells. Yat-sen University, Guangzhou 510060, China; METHODS. Mouse Muller¨ cells under normal, injury, and aging conditions were sorted from [email protected]. Muller¨ cell–specific green fluorescent (GFP)-expressing mice. RNA sequencing was Submitted: April 17, 2019 used to sequence transcriptomes, and reduced representation bisulfite sequencing was used Accepted: September 22, 2019 to sequence DNA methylomes. Various bioinformatics tools were used to compare and correlate the transcriptomes and DNA methylomes. Citation: Lin S, Guo J, Chen S. Tran- scriptome and DNA methylome sig- RESULTS. Muller¨ cells express a distinct transcriptome that is in line with their retinal natures associated with retinal Muller¨ supporting roles and dormant retinogenic status. Injury changes the Muller¨ cell transcriptome glia development, injury response, dramatically but fails to stimulate the cell cycle machinery and retinogenic factors to the states and aging. Invest Ophthalmol Vis Sci. observed in early retinal progenitor cells (RPCs). Muller¨ cells exhibit a less methylated 2019;60:4436–4450. https://doi.org/ genome than that of early RPCs, but most regulatory elements for Muller¨ cell– and RPC- 10.1167/iovs.19-27361 specific are similarly hypomethylated in both Muller¨ cells and RPCs, except for a subset of Muller¨ cell–specific functional genes. Aging only subtly affects the transcriptome and DNA methylome of Muller¨ cells.

CONCLUSIONS. Failure to reactivate the cell cycle machinery and retinogenic factors to necessary levels might be key barriers blocking Muller¨ cells from entering an RPC-like regeneration state. DNA methylation might regulate the expression of a subset of Muller¨ cell– specific functional genes during development but is likely not involved in restricting the regeneration activity of Muller¨ cells. Keywords: Muller¨ cells, transcriptome, DNA methylome, regeneration, cell cycle

uller¨ cells are the primary glia of the retina and play to proliferate by downregulating the cell cycle inhibitor M essential supporting roles in maintaining the structural p27Kip1.5 This proliferation tendency of Muller¨ cells is exploited and physiological homeostasis of the retina. Muller¨ cell nuclei most effectively in lower vertebrates, such as zebrafish; Muller¨ are located in the midstratum of the inner nuclear layer of the cells in these species react to injuries with dedifferentiation, retina, but their cell bodies extend across the entire thickness extensive proliferation, and differentiation to all types of retinal of the retina and project numerous short lateral branches to neurons to repair the retina.6,7 However, the regeneration ensheath all nearby retinal neurons. This special radial structure activity of Muller¨ cells decreases to a negligible level in of Muller¨ cells allows them to intimately interact with retinal mammals.8 Nonetheless, the dramatic regeneration ability of neurons in both healthy and diseased conditions. The well- Muller¨ cells in lower vertebrates has inspired researchers to established functions of Muller¨ cells include maintaining the explore ways to use endogenous Muller¨ cells to regenerate ion and water homeostasis of the retinal microenvironment, retinal neurons in situ in mammals, with the ultimate goal of providing nutrition for retinal neurons, recycling neurotrans- developing regeneration methods for treating human retinal mitters and photopigments, protecting retinal neurons from degeneration diseases. Encouragingly, by overexpressing tran- oxidative stress, regulating retinal blood flow, and contributing to the blood–retinal barrier.1 Consistent with the essential scription factors (TFs) coupled with epigenetic manipulations functions of Muller¨ cells, depletion or malfunction of Muller¨ or signaling pathway stimulation, some retinal interneurons and cells causes severe disruption of retinal structure and visual photoreceptors have been successfully regenerated from 9–11 function, eventually leading to retinal degeneration.2–4 reprogrammed Muller¨ cells in situ in mice. However, thus Upon assaults to the retina (physical, chemical, or patho- far, only limited types of retinal neurons can be regenerated, logical), Muller¨ cells respond with a series of gliotic reactions, and retinal ganglion cells, the major cells damaged in glaucoma, including reduced potassium conductance and membrane which is the most prevalent retinal degeneration disease, seem depolarization, cellular hypertrophy, and upregulation of resistant to regeneration from Muller¨ cells. Therefore, efforts intermediate filaments, such as GFAP, nestin, and vimentin1 are needed to investigate the molecular barriers blocking the More interestingly, upon injury, Muller¨ cells exhibit a tendency regeneration capacity of Muller¨ cells in mammals and to

Copyright 2019 The Authors iovs.arvojournals.org j ISSN: 1552-5783 4436

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4437

develop more efficient protocols to regenerate various types of (Shanghai, China) for RNA extraction, library preparation, retinal neurons in situ. sequencing, and data analyses. The cDNA libraries were From a developmental perspective, Muller¨ cells share the constructed using the TruSeq Stranded Total RNA with Ribo- same progenitor as retinal neurons. Muller¨ cells and retinal Zero Gold kit (Illumina, San Diego, CA, USA), and sequenced neurons are all generated by retinal progenitor cells (RPCs) in a by HiSeq XTen (Illumina) on a 150 base pair (bp) paired-end highly ordered sequential differentiation process. Muller¨ cells run. A total of 6 to 8 3 107 reads were generated for each are the last cell type generated by RPCs toward the end of sample, and on average, 96% reads were mapped to the mouse retinogenesis.12,13 A number of expression analyses have genome. noted interesting overlaps of gene expression patterns between Muller¨ cells and RPCs, which partially explains the RNA-Seq Data Processing regeneration potential of this mature retinal cell type.14,15 RPCs change their competence to generate retinal cells during Before read mapping, clean reads were obtained from the raw retinogenesis such that early RPCs generate early-born retinal reads by removing the adaptor sequences and low-quality neurons, including retinal ganglion cells, horizontal cells, reads. The clean reads were then aligned to mouse genome amacrine cells, and cone photoreceptors, and they gradually (GRCm38/mm10) using HISAT2.19 HTseq20 was used to get transform to late RPCs to generate later-stage retinal cells, gene counts, and the fragments per kilobase per million including bipolar cells, rod photoreceptors, and Muller¨ mapped fragments (FPKM) method as used to normalize the cells.16,17 As the last cell type differentiated from late RPCs, gene expression. We applied the limma algorithm21 on the Muller¨ cells are closer to late RPCs than to early RPCs value of Log2(FPKMþ0.5) to filter differentially expressed genes molecularly,14 which may explain why Muller¨ cells are more (DEGs) under the following criteria: (1) fold change > 2or< prone to be reprogrammed to late-stage retinal neurons such as 0.5; (2) P value < 0.05, false discovery rate (FDR) < 0.05. For rods and bipolar cells.9,10 However, early RPCs proliferate more long noncoding RNA (lncRNA) cis prediction, we identified actively and possess the full potential to generate all types of genomic localization of lncRNAs and paired mRNAs that are retinal neurons. Exactly how Muller¨ cells are different from less than 10 kb upstream or downstream away from the early-stage RPCs at the genome-wide level awaits further lncRNA. Gene Ontology (GO) term analysis was performed investigation. using DAVID bioinformatics resources.22 For gene functional In this study, we systematically measured, compared, and association network analysis, the gene coexpression network correlated the transcriptomes and DNA methylomes of Muller¨ modeling algorithm was used23 based on the normalized cells and early RPCs, as well as the transcriptomes and DNA expression values of genes.24 We focused on cell type–specific methylomes of Muller¨ cells under injury and aging conditions, TFs and genes in enriched GO terms in each cell type. For each to explore the possible molecular mechanisms governing the pair of genes, we calculated the Pearson correlation and chose development, physiological functions, and regeneration capac- the significant correlation pairs (FDR < 0.05) to construct the ity of Muller¨ cells. network. Zebrafish high-throughput sequencing data were downloaded from Gene Expression Omnibus (GEO): zebrafish Muller¨ cells (SRR4241537, SRR4241538, SRR4241539), 36hpf METHODS zebrafish eyes (SRR5398205-SRR5398212), zebrafish retinal neuroepithelial cells (SRR8417667, SRR8417669, Animals SRR8417671). All animal studies were performed in compliance with the ARVO Statement for the Use of Animals in Ophthalmic and Reduced Representation Bisulfite Sequencing Vision Research and were approved by the Institutional Animal (RRBS) Care and Use Committee of Zhongshan Ophthalmic Center. Rlbp1-GFP mice were kindly provided by Edward M. Levine Muller¨ cells and RPCs were collected as for RNA-Seq, and two from the University of Utah and maintained on a C57BL/6J to four eyes were pooled for one RRBS library preparation. background. This mouse strain specifically and persistently Genomic DNA was extracted with a Genomic DNA Purification expresses green fluorescent protein (GFP) in Muller¨ cells Kit (Promega, Madison, WI, USA) and stored at 808C. RRBS (Supplementary Fig. S1A).18 For retinal injury, 2-month-old was performed following a published protocol with minor Rlbp1-GFP mice were anesthetized with ketamine (130 mg/kg) modifications. Briefly, genomic DNA was digested using MspI and xylazine (9 mg/kg); then, left eyes received intravitreal (NEB, Ipswich, MA, USA), followed by end-repair, A-tailing, injection of 2 lL 0.1 M N-methyl-D-aspartic acid (NMDA) adapter ligation, bisulfite conversion, and PCR library prepa- (Sigma-Aldrich Corp., St. Louis, MO, USA) using injection ration using a NEXTflex Bisulfite-Seq kit (Bioo Scientific, Austin syringes with 32-gauge needles (Hamilton, Reno, NV, USA). TX, USA) and EZ DNA methylation-gold kit (Zymo Research, Irvine, CA, USA) following the manufacturer’s instructions. RNA Sequencing (RNA-Seq) Libraries were sent to Novel Bioinformatics Co., Ltd. for sequencing and data analyses. The libraries were quality Muller¨ cells were obtained through fluorescence-activated cell controlled with Agilent 2000 (Agilent, Santa Clara, CA, USA) sorting (FACS) of GFPþ cells from dissociated retinal cells of 2- and then sequenced by HiSeq XTen on a 150-bp paired-end month-old or 1.5-year-old Rlbp1-GFP mice (Supplementary Fig. run. A total of 2 to 5 3 107 reads were generated for each S1B). Sorted Muller¨ cells from four to six eyes were pooled for sample, and on average, 65% reads were mapped to the mouse one RNA-Seq experiment. RPCs were collected by papain genome. dissociation of neural retinas of embryonic day 12.5 embryos derived from mating between Rlbp1-GFP mice. RPCs from five Bisulfite-PCR Sequencing or six embryos of the same litter were pooled for one RNA-Seq experiment. At least two replicates were performed for each Genomic DNA was extracted as for RRBS; then, DNAs were type of cell. Samples were collected and stored in TRIzol bisulfite-converted using EZ DNA methylation-gold kit (Zymo (Thermo Fisher Scientific, Waltham, MA, USA) in a 808C Research) and amplified using EpiMark Hot start Taq DNA freezer until submitted to Novel Bioinformatics Co., Ltd. polymerase (NEB). Amplicons were excised from the agarose

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4438

gel and cloned into pGEM-T easy vector (Promega) for mitotic RPCs in early postnatal day Hes1-expressing retinal sequencing. Data were analyzed using QUMA.25 cells analyzed by Ueno et al.31 (as illustrated by the abundant expression of cell cycle genes and nervous system develop- RRBS Data Processing mental genes in these cells) (Supplementary Fig S2B; Supple- mentary Table S1). Raw sequence reads were quality-trimmed using TrimGalore Gene Ontology term enrichment analysis was performed to (Babraham Institute, Cambridge, UK) to remove the adaptor examine the biological processes that are enriched or depleted sequences and low-quality reads. Then, the clean data were in Muller¨ cells. The Muller¨ cell transcriptome was significantly aligned to mouse genome (GRCm38/mm10) using Bismark.26 enriched for GO terms that are in line with their essential Differences in methylation between groups were measured supporting roles in maintaining retinal structural and physio- using the CpG sites (regions of DNA where a cytosine logicalhomeostasis(Fig.1B).Forexample,Muller¨ cells nucleotide is followed by a guanine nucleotide) with read abundantly express a variety of binding , transporters, coverage more than 10 by a logistic regression model built in and recycling enzymes for photopigments and neurotransmit- methylKit,27 allowing us to identify differentially methylated ters to support the visual perception function of retinal regions (DMRs, 200 bp) with at least a 25% difference in neurons (e.g., Rlbp1, Rbp3, Glul, Slc17a7, and Cyp2d22) methylation levels and a q value less than 0.01. Finally, the (Supplementary Figs. S4A, S4C). Muller¨ cells also express large related genes and genome features were assigned to DMRs numbers of ion, organics, and water channel proteins to using ChIPseeker package.28 maintain the ever-changing microenvironment challenged by active retinal neuron activity (e.g., Atp1a1, Lrrc8b, Ttyh1, Promoter and TF-Binding Site Methylation Level Aqp4, and Kcnj10) (Supplementary Fig. S4B). Moreover, Muller¨ cells express large groups of adhesion molecules and Calculation extracellular matrix proteins that differ from those of RPCs, Promoters were defined as 2000-bp sequence flanking which reflects the intimate direct interaction and structural transcription start site (TSS). High-density CpG promoter supporting relationship of Muller¨ cells with other cells in the (HCP), intermediate-density promoter (ICP), and low-density retina (e.g., Spon1, Vtn, Itgb5, Vcam1, and Itga9) (Supple- CpG promoter (LC) were annotated as previously published.29 mentary Fig. S4D). Additionally, Muller¨ cells express high levels TF-binding sites were downloaded from the ChIP-Atlas of the angiogenic growth factor Vegfa, as well as two key database (http://chip-atlas.org/; in the public domain); then, receptors for VEGF ligands, Kdr and Flt1 (Supplementary Fig. the binding sites were converted to mm10 reference build with S4E), suggesting that Muller¨ cells are not only important the UCSC LiftOver tool (https://genome.ucsc.edu/cgi-bin/ sources of VEGF growth factors for the retinal vasculature but hgLiftOver; in the public domain). For each genomic region, are also regulated by the signaling pathway. Interestingly, the DNA methylation level was calculated as the average DNA Muller¨ cells also express several key regulators in the circadian methylation levels of all CpG sites with read coverage more rhythm regulatory pathway, including Cry2, Per3, Per1, and than 10 within the region. Per2, suggesting that Muller¨ cells are involved in circadian rhythm regulation (Supplementary Fig. S4F). In contrast to the wide distribution of GO terms in Muller¨ Data Deposit cell–upregulated genes, genes depleted in Muller¨ cells com- All sequencing data of this study have been deposited at GEO pared to those in early RPCs were mostly related to cell cycle (GSE124532). regulation, involving genes participating in cell division, mitotic nuclear division, and DNA replication (Fig. 1C). A large number of genes involved in various aspects of cell cycle RESULTS regulation, including cell cycle drivers-cyclins (Ccna2, Ccnb1, Ccnb2, Ccnd1, Ccnd2, Ccne1, and Ccne2) and cyclin-depen- The Transcriptome of M¨uller Cells Is Distinct From dent kinases (Cdk1, Cdk2, and Cdk4), DNA replication factors That of Early RPCs (Mcm2, Mcm3, Mcm4, Mcm5, Mcm6, and Mcm7), mitosis progression (Ndc80, Plk1, Cdca2, and Bub1b), and other key RNA sequencing was performed on FACS-sorted 2- to 3-month- regulators, such as Mki67, Aurka,andE2f1,wereall old mouse Muller¨ cells, as well as on early RPCs from significantly downregulated in mature Muller¨ cells compared embryonic day 12.5 embryos. Gene expression comparison to those in RPCs (Supplementary Fig. S5A). The drastically revealed that Muller¨ cells and early RPCs expressed dramati- different expression patterns of cell cycle regulators in Muller¨ cally different transcriptomes (Fig. 1A). Compared to early cells and RPCs are in line with the dramatically different RPCs, Muller¨ cells had 2450 significantly upregulated protein- proliferation capacities of the two cell types and may represent coding genes and 3717 significantly downregulated protein- the fundamental molecular barrier Muller¨ cells need to coding genes (Fig. 1A; Supplementary Table S1) (fold change > conquer when they reenter the proliferative progenitor/stem 2, FDR < 0.05). By comparing these 2450 Muller¨ cell–enriched cell-like state. In addition to cell cycle regulators, a large genes with the published Muller¨ cell–specific and –enriched number of nervous system development regulatory genes were genes revealed by single cell microarray analysis,30 20 out of 32 dramatically downregulated in Muller¨ cells compared to those ‘‘Muller¨ genes’’ revealed by Roesch et al.30 were also enriched in early RPCs, reflecting the much more potent neurogenic in the Muller¨ cell transcriptome of this study (Supplementary capacity of early RPCs compared with Muller¨ cells (Fig. 1C; Fig. S2A; Supplementary Table S1). Among these genes are Supplementary Fig. S5B). The dramatically downregulated many well-known Muller¨ cell–specific genes, such as Abca8a, neurogenic factors in Muller¨ cells include those TFs essential Rlbp1, Clu, Aqp4, and Kcnj10, which were also verified for the differentiation of early retinal neurons, such as Atoh7, individually by RT-quantitative (q)PCR (Supplementary Fig. S3). Sox11, Foxn4, Pou4f1, Neurog2, Ascl1, Sox4, and Isl132–34 We also compared our list of Muller¨ cell–enriched genes with (Supplementary Figs. S5B, S5D). On the other hand, several the list of genes enriched in postnatal day 1 (p1) and p4 Hes1- important RPC proliferation and multipotency regulatory TFs, GFPþ retinal cells31 but found that only a small portion of the including Rax, Pax6, Vsx2, , and Six3,35 were abundantly lists overlapped, which was likely caused by the mixed and almost equally expressed in Muller¨ cells and RPCs, while population of differentiating Muller¨ cells and uncommitted the late RPC marker and Muller¨ cell fate regulator Sox936 was

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4439

FIGURE 1. Muller¨ glia and early RPCs express distinct transcriptomes. (A) The volcano plot shows the distribution of the fold changes of each mRNA transcript in Muller¨ cells versus early RPCs (from embryonic day 12.5 mouse embryos). The transcripts significantly enriched in RPCs (FC < 0.5, FDR < 0.05) and Muller¨ cells (FC > 2, FDR < 0.05) are highlighted by blue and red color, respectively. (B) GO terms related to biological processes enriched in the genes upregulated in Muller¨ cells. (C) GO terms related to biological processes enriched in the genes upregulated in RPCs. (D) Functional association gene network in Muller¨ cells. (E) Functional association gene network in RPCs. TFs with the highest degree of gene–gene interactions are highlighted in (D, E)bydarker texts and lines. The color scale represents Log2FPKM.

more abundant in Muller¨ cells than in early RPCs (Supplemen- in the cells. The analyses predicted hundreds of functional tary Fig. S5D), which is consistent with published reports15,37 association relationships between TFs and key pathway and supports the idea that Muller¨ cells maintain a certain level components in Muller¨ cells or RPCs (Figs. 1D, 1E). For of RPC properties. Finally, GO term analysis showed that ‘‘DNA example, Mafk and were the top two TFs that showed methylation on the cytosine’’ process was more enriched in the highest degree of functional association with enriched RPCs than in Muller¨ cells (Fig. 1C). Indeed, two key DNA pathway components in Muller¨ cells (Fig. 1D). Mafk is a methyltransferases, Dnmt3b and Dnmt1, were highly ex- member of the small Maf TF family that has been shown to pressed in RPCs, while they were dramatically downregulated regulate the oxidative stress response.38 The gene association in Muller¨ cells (Supplementary Fig. S5C), suggesting that DNA network analysis predicted that Mafk is functionally associated methylation is more active in RPCs than in Muller¨ cells. with a number of ion, water, and neural transmitter transport- TFs are key regulators for gene expression. Muller¨ cells and ers as well as genes supporting visual perception in Muller¨ RPCs each express a unique group of TFs. We used a cells, indicating that Mafk might play important roles in coexpression functional association gene network modeling regulating the visual supporting functions of Muller¨ cells. strategy23 to predict potential regulatory relationships be- Similarly, Klf4, a well-known pluripotency promoter TF,39 tween cell type–specific TFs and genes in enriched pathways might also be an important TF for the expression of genes

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4440

essential for visual supporting functions of Muller¨ cells (Fig. only in aged Muller¨ cells (Fig. 2D; Supplementary Table S3). 1D). In RPCs, Barhl2, Tbx20, and Ebf3 were the three TFs that These genes included molecules involved in cell adhesion showed the highest degree of functional association with (such as Tenm4, Lama3, Itgb7), signaling pathway compo- enriched pathway genes. These three TFs have been shown to nents (such as Wnt6, Hgf, Plcb1, Npy), and cell cycle and stress play important roles in regulating cell fate determination, response regulators (such as Bach2, Trp63, Perp, Impdh2, including neurons.40–42 Our gene association network analysis Chaf1a). Though none of these genes have been implicated in indicates that these TFs not only regulate the expression of regulating the physiology of Muller¨ cells, studying the roles of retina developmental genes but also regulate a large number of these genes as well as those of genes most dramatically cell cycle genes, suggesting their additional roles in cell changed in aged Muller¨ cells might reveal mechanisms proliferation regulation (Fig. 1E). In addition, Mybl2 and regulating retinal aging and age-related retinal diseases. Hmga2, two DNA binding transcription regulators that have been shown to regulate cell proliferation in stem cells and 43–47 Expression Dynamics of Key Regulators for cancers, showed a high degree of functional association Retinal Regeneration with cell cycle genes, suggesting that these two genes are important regulators for controlling proliferation in RPCs (Fig. In lower vertebrates, Muller¨ cells respond to injury by actively 1E). The functional importance of these TF–gene associations reentering the cell cycle, dedifferentiating to an RPC-like state in Muller¨ cells and RPCs is worth further experimental testing. and regenerating all types of retinal neurons, while in mammals, this regeneration activity is reduced to a negligible 8 Gene Expression Changes in M¨uller Cells During level. Taking advantage of the quantitively characterized Injury and Aging transcriptomes of RPCs and Muller¨ cells in both normal and injury conditions in this study, we examined the expression To examine how Muller¨ cells respond to injury and aging at the dynamics of key regulators of retinal regeneration in Muller¨ transcriptome level, we performed RNA sequencing on Muller¨ cells during development, injury responses, and aging. Tran- cells sorted from retinas 2 days after NMDA treatment and from scriptome analysis revealed that a large number of cell cycle the retinas of 1.5-year-old mice. NMDA treatment kills retinal regulators were significantly downregulated in Muller¨ cells ganglion cells and amacrine cells, while Muller¨ cells respond to compared to those in RPCs (Supplementary Fig. S5A; Fig. 3A). NMDA injury with a gliosis reaction as in other pathological Interestingly, many of these cell cycle regulators, for example, conditions but only occasionally reenter the cell cycle in mice Ccnb1, Mki67, Cdk1, and Aurka, were upregulated in Muller¨ (Supplementary Fig. S6). Gene expression comparison showed cells upon injury (Fig. 3A; Supplementary Table S2), consistent that a large number of genes exhibited dramatic changes in with the known proliferation tendency of Muller¨ cells in mice expression in Muller¨ cells after NMDA injury, and more genes in response to injury (Supplementary Fig. S6).8 However, were upregulated than downregulated (fold change > 2, FDR although the upregulation of these cell cycle regulators was < 0.05) (Fig. 2A; Supplementary Table S2). GO term significant, their expression levels in Muller¨ cells after injury enrichment analysis showed that many genes participating in were still far lower than the levels in RPCs (Fig. 3A), indicating the translation process were dramatically upregulated in Muller¨ that Muller¨ cells failed to reactivate the cell cycle machinery to cells under the injury condition (Fig. 2B), suggesting elevated a necessary level, thus mostly remaining proliferation quies- translation activity of Muller¨ cells upon injury. Genes upregu- cent. lated in Muller¨ cells under the injury condition were also Cellular activity and cell fates are controlled by cell type– enriched for cytoskeleton organization, wound response, cell specific TFs. RPC proliferation and multipotency regulatory death, and immunity (Fig. 2B), consistent with the known TFs, such as Rax, Pax6, Vsx2, Sox2, and Six3, were abundantly reaction of Muller¨ cells to retinal injury.48 Downregulated expressed in Muller¨ cells, and their expression remained high genes in Muller¨ cells upon NMDA treatment were enriched for in Muller¨ cells in both injury and aging conditions (Fig. 3B). cell adhesion molecules (Fig. 2B), suggesting that Muller¨ cells Ascl1 is a TF expressed in a subpopulation of RPCs that give change their ways of interacting with surrounding retinal cells rise to all kinds of retinal neurons except retinal ganglion in response to retinal injury. In addition, the Wnt signaling cells.49 During regeneration, Ascl1 is essential for the pathway in Muller¨ cells seemed to be affected by injury (Fig. dedifferentiation and regeneration activities of zebrafish Muller¨ 2B), which will be examined further in the next section. cells50,51 and capable of reprogramming mouse Muller¨ cells Comparing gene expression levels in Muller¨ cells in young into retinal neurons in vitro and in vivo.9,52 Our RNA and old mice, there were 488 genes upregulated in aged Muller¨ sequencing data showed that Ascl1, as well as its downstream cells (fold change > 2, FDR < 0.05), and over 80% of them targets Lin28a/b, were highly expressed in RPCs, but were were upregulated more than 5-fold (Fig. 2C; Supplementary absent in Muller¨ cells and remained unexpressed under injury Table S3). For example, Msi, Trpm3, Gnas, Fgfr1, and Add1 and aging conditions, consistent with the published report.52 were the top five most upregulated genes that were expressed In addition, most early RPC-enriched neurogenic genes, 100 times more in aged Muller¨ cells than in young Muller¨ cells including key TFs for retinal neuron differentiation (Supple- (Supplementary Table S3). Genes upregulated in aged Muller¨ mentary Fig. S7, marked by the red arrow), remained cells are involved in a variety of biological processes, but no unexpressed in Muller¨ cells in the injured retina, indicating GO terms were enriched. On the other hand, there were only that Muller¨ cells were kept in a terminally differentiated glial 34 genes downregulated in aged Muller¨ cells, and cell state instead of a retinogenic RPC-like state. LOC102639800, Snrpn, Sap25, Ubl4a, and Rbm4 were the The Notch pathway has been widely studied for its top five most downregulated genes that showed over 45 times important roles in Muller¨ cell development and regenera- lower expression levels in aged Muller¨ cells than in young tion.11,53–60 To examine how the Notch pathway changes in Muller¨ cells (fold change > 2, FDR < 0.05) (Fig. 2C; Muller¨ cells during development, injury, and aging, we Supplementary Table S3). Interestingly, most genes that were extracted the expression data for the four Notch receptors upregulated or downregulated in aged Muller¨ cells were also and several key downstream target genes. Of the four Notch upregulated or downregulated in Muller¨ cells upon injury (Fig. receptors, Notch1 and Notch2 are the predominant receptors 2D), indicating that aging and injury share some common in the retina that were highly expressed in both RPCs and pathological properties. Thirty-eight genes showed significant Muller¨ cells. Notch3 was highly expressed in RPCs but upregulation and 10 genes showed significant downregulation significantly downregulated in Muller¨ cells, whereas Notch4

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4441

FIGURE 2. Transcriptome changes in Muller¨ cells during injury and aging. (A) The volcano plot shows the distribution of the fold changes of each mRNA transcript in Muller¨ cells 2 days after NMDA intravitreal injection. The transcripts significantly downregulated (FC < 0.5, FDR < 0.05) and upregulated (FC > 2, FDR < 0.05) are highlighted by blue and red color, respectively. (B) GO terms related to biological processes enriched in the genes upregulated and downregulated in Muller¨ cells after NMDA injury. (C) The volcano plot shows the distribution of the fold changes of each mRNA transcript in Muller¨ cells from aged versus young mice. (D) The Venn diagram shows the overlap of the genes upregulated or downregulated in Muller¨ cells from aged mice with the genes upregulated or downregulated in Muller¨ cells after NMDA injury.

was marginally expressed in both RPCs and Muller¨ cells (Fig. RPCs and Muller¨ cells. The data showed that only Wnt5b is 3C). Of the five major Notch pathway targets and pathway expressed in RPCs, while Muller¨ cells did not express any Wnt activity indicators, Hes1, Hes5, Hes6, Hey1, and Hey2, all ligands (Fig. 3D). On the other hand, various Fzd-family Wnt except Hey2 were expressed in both RPCs and Muller¨ cells, receptors were abundantly expressed in both RPCs and Muller¨ while Hey2 was expressed only in Muller¨ cells (Fig. 3C). Upon cells (Fig. 3D). In addition, negative regulators of the pathway, injury, Hes5 was significantly downregulated in Muller¨ cells, including axins and Dkk3, were also abundantly expressed in while all other Notch pathway components remained relatively both RPCs and Muller¨ cells (Fig. 3D). The pathway down- unchanged (Fig. 3C). These data demonstrate that the Notch stream TF and activity indicator Lef1 (the only Wnt down- signaling components are expressed in both RPCs and Muller¨ cells, and minor adjustments occur during Muller¨ cell stream TF detected in our sequencing data) was expressed in development and response to injury. RPCs but was nearly absent in Muller¨ cells (Fig. 3D). Upon The Wnt signaling pathway plays critical roles in zebrafish injury, Muller¨ cells slightly upregulated several Wnt ligands, Muller¨ cell regeneration activity61 and mouse Muller¨ cell including Wnt5a/b, Wnt7b, and Wnt9a, and significantly proliferation activity.62 We extracted the expression data for downregulated Fzd9, but Lef1 remained unexpressed (Fig. Wnt ligands, receptors, downstream mediators, and targets in 3D).

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4442

FIGURE 3. Expression dynamics of key regulators of retinal regeneration during Muller¨ cell development, injury response, and aging. (A) Expression dynamics of cell cycle regulatory genes (same list of genes as in Fig. 1E) in the different groups of cells. The fold change refers to the fold change of a gene in the respective group of cells versus that in young Muller¨ cells (Normal). (B) Expression levels of important RPC and retinal regeneration regulators in the different groups of cells. (C) Expression levels of the Notch pathway components in the different groups of cells. (D) Expression levels of the Wnt pathway components in the different groups of cells. RPCs: retinal progenitor cells of E12.5 mouse embryos; Normal: young Muller¨ cells; NMDA: Muller¨ cells from retinas treated with NMDA intravitreal injection; Aged: Muller¨ cells from 1.5-year-old mice.

Zebrafish Muller¨ cells react to retinal injury by dedifferen- tomes of Muller¨ cells of the two species react very differently tiation to an RPC-like state and regenerate the retina. Sifuentes to retinal injury, consistent with the dramatically different et al.63 used RNA-Seq to examine the transcriptome changes in reaction modes of Muller¨ cells to retinal injury in the two zebrafish Muller¨ cells after retinal injury. To get an idea of the species. status of basal fish Muller¨ cell transcriptome, we also extracted Finally, we also examined the expression of DNA methyl- the transcriptomes of 36hpf zebrafish eyes, when most retinal transferases and demethylating enzymes. As described above, cells are at the RPC stage, from a public database,64 and of the four DNA methyltransferases, Muller¨ cells expressed compared them with the transcriptomes of fish Muller¨ cells only Dnmt1 (Supplementary Figs. S5C, S6). Interestingly, without injury (0 hpls in Sifuentes et al.63). The analyses injury significantly stimulated the expression of Dnmt1 and showed that, at the basal situation, the fish Muller¨ cell Dnmt3a in Muller¨ cells, and a similar trend was also observed transcriptome lacked the expression for a large number of in aged Muller¨ cells (Supplementary Fig. S10). On the other important cell cycle regulators and neurogenic factors, while hand, three Tet family DNA demethylating enzymes were enriched for Muller¨ cell functional genes such as ion expressed at relatively constant levels in RPCs and Muller¨ cells, transporters (Supplementary Fig. S8), resembling that of mouse while Apobec2, a cytidine deaminase required for zebrafish Muller¨ cells and consistent with the differentiated glia status of Muller¨ cells to reenter the regeneration state,65 was not the cells. We compared our lists of DEGs in mouse Muller¨ cells expressed in mouse RPCs and Muller¨ cells (Supplementary Fig. after NMDA injury with DEGs in zebrafish Muller¨ cells 8 hours S10). The dynamic expression patterns of Dnmt family after lesion (8 hpl) and 16 hpl but found only small overlap methyltransferases suggest that Muller¨ cells may adjust their between the two species (Supplementary Fig. S9), and most DNA methylome in response to injury and during aging. zebrafish Muller¨ cell regeneration-associated genes and those genes most dramatically changed in zebrafish Muller¨ cells after Noncoding RNA Expression in M¨uller Cells and a light lesion remained unchanged in mouse Muller¨ cells RPCs (among the 12 genes listed in the main text Table of Sifuentes et al.,63 only Stat3 and Fab7 were slightly upregulated in Our RNA sequencing also detected noncoding RNAs (ncRNAs), mouse Muller¨ cells, and of the most dramatically changed and expression level comparisons showed that many ncRNAs genes in zebrafish Muller¨ cells from Table s2 and s3 of Sifuentes changed their expression levels during Muller¨ cell differentia- et al.,63 only Tspo, Adam8, Mvp, Mef2a, and Bik changed tion from RPCs. We detected 355 upregulated ncRNAs and 731 expression in mouse Muller¨ cells in the same trends) downregulated ncRNAs in Muller¨ cells (fold change > 2, FDR < (Supplementary Table S2), demonstrating that the transcrip- 0.05) compared with those in early RPCs (Fig. 4A). Noncoding

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4443

FIGURE 4. Noncoding RNA expression changes in Muller¨ cells during development, injury. and aging. (A) The volcano plot shows the distribution of the fold changes of each ncRNA in Muller¨ cells versus early RPCs. (B) The volcano plot shows the distribution of the fold changes of each ncRNA in Muller¨ cells 2 days after NMDA intravitreal injection versus in normal Muller¨ cells. (C) The volcano plot shows the distribution of the fold changes of each ncRNA in Muller¨ cells from 1.5-year-old mice versus in young normal Muller¨ cells. The transcripts significantly downregulated (FC < 0.5, FDR < 0.05) and upregulated (FC > 2, FDR < 0.05) are highlighted by blue and red color, respectively. (D) Heat maps of the lncRNAs (left) and mRNAs (right) that showed correlated expression level changes between Muller¨ cells and RPCs. (E) Heat maps of the lncRNAs (left) and mRNAs (right) that showed correlated expression level changes in Muller¨ cells after injury. The color scale represents Log2FPKM.

RNAs, especially lncRNAs, play important roles in controlling cleotides that is essential for genomic DNA replication. Rrm2 the transcription of neighboring genes.66,67 Cis-analysis corre- expression was 310 times lower in Muller¨ cells than in RPCs lating lncRNAs with their neighboring protein-coding genes (the last mRNA in Fig. 4D). LOC102639931 is a lncRNA located showed that changes in the expression levels of some lncRNAs 7637 upstream of Rrm2 gene. LOC102639931 expression was were accompanied by changes in the expression levels of their 15.9 times higher in Muller¨ cells than in RPCs (the last lncRNA neighboring mRNAs, either positively or inversely (Fig. 4D). in Fig. 4D). The correlated changes in lncRNAs and mRNAs Among 287 differentially expressed lncRNAs, 52 (18%) were indicate that some mRNA expression in Muller¨ cells and RPCs accompanied by coordinated expression level changes of might be regulated by their neighboring lncRNAs. Similarly, we neighboring mRNAs, with most located upstream of corre- also found that 667 ncRNAs were upregulated and 386 ncRNAs sponding mRNAs and positively coregulated (Supplementary were downregulated in Muller¨ cells in the injured retina (Fig. Fig. S11). For example, Rrm2 is a ribonucleotide reductase that 4B). Among them, 244 were lncRNAs, and the expression level catalyzes the formation of deoxyribonucleotides from ribonu- changes in 19 (8%) of them were accompanied by neighboring

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4444

mRNA gene expression changes (Fig. 4E; Supplementary Fig. S11). Aging did not markedly affect ncRNA expression level changes; only 178 ncRNAs and 41 ncRNAs were upregulated and downregulated, respectively, in aged Muller¨ cells (Fig. 4C). Studying the regulatory relationship of these correlated ncRNAs and mRNAs during Muller¨ cell development and physiological processes may reveal new molecular mechanisms governing gene expression in Muller¨ cells.

Methylomes of M¨uller Cells and RPCs DNA methylation is a crucial form of epigenetic modification that controls gene expression and genome stability.68,69 Our gene expression comparison showed that DNA methyltrans- ferases were more abundantly expressed in RPCs than in Muller¨ cells (Supplementary Figs. S5C, S10), indicating that DNA methylation was more active in RPCs than in Muller¨ cells. However, systematic characterization and comparison of the methylomes of Muller¨ cells with those of RPCs are lacking. Here, we used the RRBS technique to sequence the most informative CpG sites in the genomes of Muller¨ cells and early RPCs. Similar to other cell types,70,71 the methylomes of both Muller¨ cells and RPCs displayed a bimodal distribution in that most bases were either hypermethylated or hypomethylated (Supplementary Fig. S12A). Unsupervised hierarchical cluster- ing analysis showed that Muller¨ cells and RPCs are clearly separated from each other (Fig. 5A), demonstrating that Muller¨ cells and RPCs are also distinct from each other at the methylome level. Interestingly, the methylome of Muller¨ cells was more biased toward hypomethylation than that of RPCs (Fig. 5B), which coincided with the lower levels of expression of DNA methyltransferase in Muller¨ cells than in RPCs (Supplementary Figs. S5C, S10). Over the gene body, the region around the TSS was depleted of CpG methylation (Supplementary Fig. S12B), consistent with the findings for other cell types.71,72 Across other regions of the gene body, Muller¨ cells were generally less methylated than RPCs FIGURE 5. Methylomes of Muller¨ cells and RPCs. (A) Unsupervised (Supplementary Fig. S12B). Comparison of methylation levels hierarchical clustering shows the separation of the methylomes of on 200-bp tiles of the genomes showed that 15,955 genomic Muller¨ cells from those of RPCs. (B) Distribution of CpG methylation regions were differentially methylated between Muller¨ cells levels in the genomes of Muller¨ cells and RPCs. (C) The heat map shows the methylation levels of DMRs in Muller¨ cells and RPCs. (D) and RPCs (DMR, methylation level change > 25%) (Fig. 5C). Distribution of the genomic annotations of DMRs between Muller¨ cells Most DMRs showed demethylation in Muller¨ cells (13,847 and RPCs. DMRs), while only 2108 DMRs were more methylated in Muller¨ cells (Fig. 5C). Annotation of these DMRs showed that nearly half of these DMRs were distributed in intergenic regions; 32% were among introns, 13% were among exons, and 12% were hypomethylated in both Muller¨ cells and RPCs (Figs. 6Aa and located in promoter regions (Fig. 5D). We confirmed the Ab), and promoters in the ICP group were also generally methylation status of some DMRs by bisulfite conversion-PCR hypomethylated in both cell types (Figs. 6Ac, 6Ad). Promoters followed by Sanger sequencing (Supplementary Fig. S13). in the LCP group of RPC-specific genes exhibited variable methylation status from hypomethylated to hypermethylated in Correlation Between Promoter Methylation and both Muller¨ cells and RPCs (Fig. 6Ae). Interestingly, promoters Gene Expression During M¨uller Cell Development of Muller¨ cell–specific genes in the LCP group showed a trend of demethylation in Muller¨ cells compared to those in RPCs One important function of CpG methylation is to suppress (Fig. 6Af), suggesting that promoter demethylation might be 68,69 gene transcription by promoter methylation. We wanted to involved in the upregulation of this group of genes during the know how promoter CpG methylation is correlated with gene RPC to Muller¨ cell differentiation process. For example, Rlbp1 transcription in Muller¨ cells and RPCs. We focused on genes and Prss56 are two key Muller¨ cell functional genes that are expressed in Muller¨ cells or RPCs 5-fold more than those in the highly expressed in Muller¨ cells but absent from RPCs. The other cell type as Muller¨ cell–specific genes (1156 genes) or promoters of Rlbp1 and Prss56 contain 28% and 20% CpG RPC-specific genes (1621 genes). How the expression of a gene density, respectively. The promoters of these two genes were is regulated by promoter methylation has been suggested to depend on the density of CpG in the promoter.68,73,74 Thus, we methylated in RPCs but demethylated in Muller¨ cells (Fig. 6B). further divided Muller¨ cell/RPC–specific genes into the To examine whether promoter methylation was involved in the following groups: genes with high CpG density promoter dramatic downregulation of cell cycle regulators in Muller¨ (HCP), intermediate CpG density promoter (ICP), and low CpG cells, we calculated the CpG methylation levels of the density promoter (LCP). We then calculated the CpG promoters of cell cycle genes. The results showed that methylation levels of the promoter of each gene in Muller¨ promoters of cell cycle regulators are generally hypomethy- cells and RPCs. Promoters in the HCP group were mostly lated in both Muller¨ cells and RPCs (Supplementary Fig. S14),

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4445

FIGURE 6. Correlation between promoter methylation and gene expression during Muller¨ cell development. (A) The box plots show the distribution of the methylation levels of the promoters of RPC- or Muller¨ cell–enriched genes in Muller¨ cells and RPCs. Genes were divided into three groups based on the density of the CpG sites in the promoter: HCP: promoter with high CpG density; ICP: promoter with intermediate CpG density; LCP: promoter with low CpG density. (B) Integrative Genomics Viewer (IGV) (Broad Institute, Cambridge, MA, USA) views of the DNA methylation status of the genomic regions of two representative genes. The dotted red boxes highlight the differentially methylated promoter regions. (C) The box plots show the distribution of the methylation levels of the binding sites of the individual TFs in Muller¨ cells and RPCs.

suggesting that the expression of cell cycle regulators is not M¨uller Cell Methylome Changes During Injury and regulated at the DNA methylation level. Aging TF binding to their target sites and genomic DNA methylation have been proposed to reciprocally regulate each Gene expression examination revealed that Muller¨ cells upregu- other.68,75 Thus, we extracted the available genomic binding lated the expression of DNA methyltransferases Dnmt1 and sites of important RPC and Muller¨ cell regulatory TFs, including Dnmt3a after injury and in aging (Supplementary Fig. S10), Pax6, Sox2, Sox9,andHes1, and calculated the DNA suggesting that DNA methylation patterns were dynamically methylation levels of these sites. CpG methylation of the regulated in Muller¨ cells. Moreover, inhibition of DNA methyl- binding sites for these TFs was generally hypomethylated in ation perturbs the regeneration activity of Muller¨ cells in both Muller¨ cells and RPCs (Fig. 6C), suggesting that DNA zebrafish.76 To examine the CpG methylation patterns in Muller¨ methylation would not affect the genome targeting of these TFs cells during injury and aging, we performed RRBS sequencing of in Muller¨ cells. the genomes of Muller¨ cells isolated from retinas 2 days after

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4446

FIGURE 7. Muller¨ cell methylome changes during injury and aging. (A) The heat map shows the methylation levels of DMRs in Muller¨ cells before and after NMDA treatment. (B) The heat map shows the methylation levels of DMRs in young and aged Muller¨ cells. (C) Distribution of the genomic annotations of DMRs in Muller¨ cells before and after NMDA treatment. (D) Distribution of the genomic annotations of DMRs between young and aged Muller¨ cells. (E) IGV views of the DNA methylation status of the genomic regions of three key regulators of retinal regeneration. (F) The box plot shows the methylation levels of the Ascl1-targeting sites in Muller¨ cells under different conditions.

NMDA treatment and from 1.5-year-old mice. Searching for DMRs and 16% were distributed in the promoter regions in the injury of 200-bp tiles of Muller¨ cell genomes showed that a number of condition and aging condition, respectively. Ascl1 is an essential regions exhibited altered DNA methylation levels in response to TF for the regeneration of the zebrafish retina and is quickly injury (Fig. 7A, methylation level change > 25%), while aging had upregulated in Muller¨ cells after injury,50,51 while it fails to do so a mild effect on the DNA methylation patterns of Muller¨ cells in mammals (Fig. 3B).52 To test whether Ascl1 expression is (Fig.7B).DMRsinbothinjuryandagingconditionswerelargely sequestered by promoter methylation in Muller¨ cells in mice, we distributed in the intergenic region, introns and exons, and 8% examined the genomic region of Ascl1. The data showed that the

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4447

promoter region was hypomethylated in Muller¨ cells in all the differentiating into retinal neurons.6 In mice, cell cycle conditions tested, similar to the methylation pattern in RPCs (Fig. regulators were indeed upregulated in Muller¨ cells after injury, 7E). Similarly, Lin28a and Lin28b, two other important reflecting the tendency to reenter the cell cycle. However, regeneration regulators for Muller¨ cells,51 also had similar although upregulated, the expression levels of the cell cycle promoter methylation patterns in Muller¨ cells and RPCs (Fig. regulators were still far lower in Muller¨ cells than in RPCs, 7E), suggesting that promoter methylation is not the blockade for which explains the rare proliferation events of Muller¨ cells the expression of Ascl1, as well as its downstream effectors, after injury. The dramatic differences in the expression Lin28a and Lin28b,inMuller¨ cells, consistent with the patterns of a large group of genes necessary for driving cell previously published restriction PCR result.76 To examine cycle progression and the failed upregulation of these genes to whether Ascl1 binding to genomic targets might be influenced a necessary level after injury suggest that reactivating the cell by DNA methylation, we calculated the CpG methylation levels cycle machinery might be a key barrier for Muller¨ cells to enter of genomic Ascl1-targeting sites in Muller¨ cells under different an RPC-like state to regenerate retinal neurons. conditions, as well as in RPCs. Ascl1-targeting sites were slightly less methylated in Muller¨ cells than in RPCs, while this Notch and Wnt Signaling in M¨uller Cells During methylation level was slightly upregulated in injury conditions Development and Injury Response (Fig. 7F), suggesting that Ascl1-targeting sites in Muller¨ cells were not locked by DNA methylation. Notch signaling has demonstrated important roles during Muller¨ cell development and regeneration.11,53–60 Nelson et al.77 used RNA microarray to examine the transcriptomes of DISCUSSION Muller¨ cells from postnatal day 1 (P1) to P21 and demonstrated that the Notch pathway remains active in postmitotic Muller Muller¨ cells play essential roles in maintaining the structural and ¨ functional homeostasis of the neural retina and hold great cells and stabilizes the glial fate. The authors grouped Notch1, promise for regenerating retinal neurons in situ after damage. Notch2, Hes1, and Hes5 to gene cluster 1 or 9 (highly expressed at all ages from P1 to P21); , , and This study systematically quantified the transcriptomes of Muller¨ Notch3 Hes6 Hey1 cells in both physiological and pathological conditions, which to cluster 7 or 8 (trend down from P1 to P21); Notch4 to are distinct from early RPCs and reflect the physiological cluster 6 (no expression at all ages), and Hey2 to cluster 10 functions and regeneration capacity of Muller¨ cells. Genomic (increase from P1 to P21). Consistent with this microarray DNA methylome analyses showed that Muller¨ cells have a less analysis of postnatal maturing Muller¨ cells, our analysis on the methylated genome than early RPCs. Transcriptome and DNA components of the Notch signaling pathway showed exactly methylome correlation analyses suggested that CpG methylation the same trends of expression level changes in these genes might regulate the expression of a subset of Muller¨ cell–specific between early RPCs and mature Muller¨ cells (Fig. 3C). In functional genes during development, but likely is not involved zebrafish, the Notch pathway is suppressed in some Muller¨ in restricting the regeneration activity of Muller¨ cells. cells, which is required for the cells to enter the regeneration state.11,78,79 Our data showed that most Notch pathway components remained unchanged in Muller¨ cells upon retina The Transcriptomes of M¨uller Cells and Early RPCs injury, except that Hes5 was downregulated, suggesting that Are Distinct mouse Muller¨ cells might also downregulate Notch signaling in response to injury, yet the downregulation was far from Though studies have shown overlaps of gene expression sufficient to promote the regeneration activity of Muller¨ cells in patterns between Muller¨ cells and early RPCs,15 our tran- mice. Of note, Elsaeidi et al.11 used RT-qPCR on whole mouse scriptome comparison clearly demonstrates that the gene retina tissues to show that Hes5 expression remained expression profiles of Muller¨ cells and early RPCs are distinct unchanged upon injury, which is different from our result. on a genome-wide level. Consistent with the different retino- Because these authors used whole retina as the input, we genic statuses and physiological functions of each cell type, believe the discrepancy might be caused by technical genes involved in neurotransmitter and photopigment metabo- differences. lism, ion and water transportation, cell–cell interactions, and Wnt signaling plays important roles in Muller¨ cell regener- angiogenesis were highly and specifically expressed in Muller¨ ation.61,62 Our data showed that Muller cells did not express cells, while genes participating in cell cycle progression and ¨ any Wnt ligands, and the pathway downstream TF and activity nervous system development were highly expressed in RPCs. indicator was nearly absent in Muller¨ cells, in both normal Although Muller¨ cells express a number of RPC TFs, including Lef1 and injury conditions, suggesting the pathway is very weak or Pax6, Sox2, Rax, Six3,andVsx2, many other TFs required for absent in Muller¨ cells. On the other hand, Muller¨ cells specific retinal neuron differentiation, for example, Atoh7, abundantly expressed various Wnt receptors, suggesting the Foxn4, Ascl1, Neurog2,andIsl1,wereabsentinMuller¨ cells, which means that certain TFs need to be reactivated or capability of the cells to react to Wnt signals. supplemented exogenously for Muller¨ cells to regenerate retinal neurons. Gene functional association network analysis indicated CpG Methylation Might Regulate the Expression of that certain TFs, for example, Mafk, Klf4, Tef, Nr1d1,andDdit3 a Subset of M¨uller Cell–Specific Functional Genes in Muller¨ cells and Barhl2, Tbx20, Ebf3, Mybl2, and Hmga2 in During Development but Is Likely Not Involved in RPCs, might play more roles in determining the developmental Restricting the Regeneration Activity of M¨uller and physiological properties of Muller¨ cells or RPCs than other cell type–specific TFs, which is worth further investigation. Cells Genomic DNA methylation is dynamically regulated during Cell Cycle Re-Entry Might Be a Key Barrier for embryo development and shows different patterns in different 69–72 M¨uller Cells To Enter the Regeneration State After cell types. Whether and how DNA methylation is involved Injury in Muller¨ cell development and physiology are unknown. According to gene expression measurements, RPCs highly In zebrafish, Muller¨ cells regenerate the retina in response to express both maintenance DNA methyltransferase Dnmt1 and injury by reentering the cell cycle, proliferating, and then de novo DNA methyltransferase Dnmt3b, while Muller¨ cells

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4448

express only low levels of Dnmt1, suggesting that DNA 3. Bachleda AR, Pevny LH, Weiss ER. Sox2-deficient muller glia methylation activity is higher in RPCs. Consistent with this disrupt the structural and functional maturation of the finding, the overall genomic methylation level is lower in mammalian retina. Invest Ophthalmol Vis Sci. 2016;57: Muller¨ cells than in RPCs. When examining the relationship 1488–1499. between genomic DNA methylation and gene expression, we 4. Wohl SG, Jorstad NL, Levine EM, Reh TA. Muller glial found that the promoters of most cell type–specific genes were microRNAs are required for the maintenance of glial hypomethylated in both Muller¨ cells and RPCs, thus seemingly homeostasis and retinal architecture. Nat Commun.2017;8: not regulated by promoter methylation. However, promoters 1603. of a group of Muller¨ cell–specific genes with a low density of 5. Dyer MA, Cepko CL. Control of Muller glial cell proliferation CpG were less methylated in Muller¨ cells than in RPCs, and activation following retinal injury. Nat Neurosci.2000;3: suggesting that demethylation is involved in the upregulation 873–880. of this group of genes during differentiation from RPCs to 6. Goldman D. Muller glial cell reprogramming and retina Muller¨ cells. regeneration. Nat Rev Neurosci. 2014;15:431–442. Muller¨ cells in lower vertebrates exhibit efficient regener- 7. Pebay´ A, ed. Regenerative Biology of the Eye. Totowa, NJ: ation activity, while these cells are mostly dormant in Humana Press; 2014. mammals. One possible mechanism is that genomic regions 8. Wilken MS, Reh TA. Retinal regeneration in birds and mice. that must be activated during retinal regeneration are Curr Opin Genet Dev. 2016;40:57–64. sequestered in a hypermethylated state in Muller¨ cells in 9. Jorstad NL, Wilken MS, Grimes WN, et al. Stimulation of mammals. However, RRBS sequencing of the zebrafish Muller¨ functional neuronal regeneration from Muller glia in adult cell genome and methylation-sensitive restriction-PCR analysis mice. Nature. 2017;548:103–107. of a number of regeneration-associated genes in mouse Muller¨ 10. Yao K, Qiu S, Wang YV, et al. Restoration of vision after de cells indicated that regeneration-associated gene promoters are 76 novo genesis of rod photoreceptors in mammalian retinas. hypomethylated in quiescent Muller¨ cells. Consistently, our Nature. 2018;560:484–488. genome-wide methylation analysis of the mouse Muller¨ cell 11. Elsaeidi F, Macpherson P, Mills EA, et al. Notch suppression genome showed that promoters of most genes that are highly collaborates with Ascl1 and Lin28 to unleash a regenerative expressed in RPCs, including those that are involved in cell response in fish retina, but not in mice. J Neurosci. 2018;38: cycle progression and retinogenesis, are hypomethylated in 2246–2261. both Muller¨ cells and RPCs. The genome targeting sites of many 12. Agathocleous M, Harris WA. From progenitors to differenti- RPC multipotency TFs, such as Pax6, Sox2, and Ascl1, are also ated cells in the vertebrate retina. Annu Rev Cell Dev Biol. hypomethylated in both cell types. Thus, our methylome 2009;25:45–69. sequencing data suggest that genomic DNA methylation is 13. Cepko CL, Austin CP, Yang X, et al. Cell fate determination in probably not a barrier for Muller¨ cells to enter the RPC-like the vertebrate retina. Proc Natl Acad Sci U S A. 1996;93:589– regeneration state. 595. As the primary glia in the retina, Muller¨ cells intimately 14. Jadhav AP, Roesch K, Cepko CL. Development and neurogenic interact with retinal neurons in both healthy and diseased potential of Muller glial cells in the vertebrate retina. Prog conditions. Moreover, due to their special resistance to Retin Eye Res. 2009;28:249–262. stressors and unique potential to regenerate retinal neurons, 15. Blackshaw S, Harpavat S, Trimarchi J, et al. Genomic analysis Muller¨ cells are well suited as targets for therapeutic of mouse retinal development. PLoS Biol. 2004;2:E247. interventions to treat retinal diseases. Our study provides a comprehensive characterization of the molecular features of 16. Cepko C. Intrinsically different retinal progenitor cells Muller¨ cells at the transcriptome and DNA methylome levels, produce specific types of progeny. Nat Rev Neurosci. 2014; 15:615–627. which should be valuable for future mechanism investigations, new retinal regeneration method exploration, and translational 17. He J, Zhang G, Almeida AD, et al. How variable clones build an studies concerning Muller¨ cells. invariant retina. Neuron. 2012;75:786–798. 18. Vazquez-Chona FR, Clark AM, Levine EM. Rlbp1 promoter drives robust Muller glial GFP expression in transgenic mice. Acknowledgments Invest Ophthalmol Vis Sci. 2009;50:3996–4003. The authors thank Edward M. Levine, PhD, from the University of 19. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner Utah for providing the Rlbp1-GFP mice, Jingxuan Pan, PhD, and with low memory requirements. Nat Methods. 2015;12:357– Bei Jin, PhD, from the Zhongshan Ophthalmic Center for their 360. assistance with the fluorescence-activated cell sorting experi- 20. Anders S, Pyl PT, Huber W. HTSeq–a Python framework to ments, and Lai Wei, PhD, for assistance and suggestions on RNA- work with high-throughput sequencing data. Bioinformatics. Seq and RRBS experiments. 2015;31:166–169. Supported by Guangdong Provincial Department of Science and 21. Ritchie ME, Phipson B, Wu D, et al. limma powers differential Technology (2015B020225003), the National Natural Science expression analyses for RNA-sequencing and microarray Foundation of China (81870659), and the Ministry of Science studies. Nucleic Acids Res. 2015;43:e47. and Technology of China 973 program (2015CB964600). 22. Huang da W, Sherman BT, Lempicki RA. Systematic and Disclosure: S. Lin, None; J. Guo, None; S. Chen, None integrative analysis of large gene lists using DAVID bioinfor- matics resources. Nat Protoc. 2009;4:44–57. References 23. Pujana MA, Han JD, Starita LM, et al. Network modeling links breast cancer susceptibility and centrosome dysfunction. Nat 1. Bringmann A, Pannicke T, Grosche J, et al. Muller cells in the Genet. 2007;39:1338–1349. healthy and diseased retina. Prog Retin Eye Res. 2006;25:397– 24. Prieto C, Risueno A, Fontanillo C, De las Rivas J. Human gene 424. coexpression landscape: confident network derived from 2. Shen W, Fruttiger M, Zhu L, et al. Conditional Muller cell tissue transcriptomic profiles. PLoS One. 2008;3:e3911. ablation causes independent neuronal and vascular patholo- 25. Kumaki Y, Oda M, Okano M. QUMA: quantification tool for gies in a novel transgenic model. J Neurosci. 2012;32:15715– methylation analysis. Nucleic Acids Res. 2008;36:W170– 15727. W175.

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4449

26. Krueger F, Andrews SR. Bismark: a flexible aligner and 49. Brzezinski IV JA, Kim EJ, Johnson JE, Reh TA. Ascl1 expression methylation caller for Bisulfite-Seq applications. Bioinfor- defines a subpopulation of lineage-restricted progenitors in matics. 2011;27:1571–1572. the mammalian retina. Development. 2011;138:3519–3531. 27. Akalin A, Kormaksson M, Li S, et al. methylKit: a comprehen- 50. Fausett BV, Gumerson JD, Goldman D. The proneural basic sive R package for the analysis of genome-wide DNA helix-loop-helix gene ascl1a is required for retina regenera- methylation profiles. Genome Biol. 2012;13:R87. tion. J Neurosci. 2008;28:1109–1117. 28. Yu G, Wang LG, He QY. ChIPseeker: an R/Bioconductor 51. Ramachandran R, Fausett BV, Goldman D. Ascl1a regulates package for ChIP peak annotation, comparison and visualiza- Muller glia dedifferentiation and retinal regeneration through tion. Bioinformatics. 2015;31:2382–2383. a Lin-28-dependent, let-7 microRNA signalling pathway. Nat 29. Guo JU, Ma DK, Mo H, et al. Neuronal activity modifies the Cell Biol. 2010;12:1101–1107. DNA methylation landscape in the adult brain. Nat Neurosci. 52. Ueki Y, Wilken MS, Cox KE, et al. Transgenic expression of 2011;14:1345–1351. the proneural Ascl1 in Muller glia 30. Roesch K, Jadhav AP, Trimarchi JM, et al. The transcriptome of stimulates retinal regeneration in young mice. Proc Natl retinal Muller glial cells. J Comp Neurol. 2008;509:225–238. Acad Sci U S A. 2015;112:13717–13722. 31. Ueno K, Iwagawa T, Ochiai G, et al. Analysis of Muller glia 53. Hojo M, Ohtsuka T, Hashimoto N, et al. Glial cell fate specific genes and their histone modification using Hes1- specification modulated by the bHLH gene Hes5 in mouse promoter driven EGFP expressing mouse. Sci Rep. 2017;7: retina. Development. 2000;127:2515–2522. 3578. 54. Yaron O, Farhy C, Marquardt T, et al. Notch1 functions to 32. Xiang M. Intrinsic control of mammalian retinogenesis. Cell suppress cone-photoreceptor fate specification in the devel- Mol Life Sci. 2013;70:2519–2532. oping mouse retina. Development. 2006;133:1367–1378. 33. Hatakeyama J, Kageyama R. Retinal cell fate determination 55. Jadhav AP, Mason HA, Cepko CL. Notch 1 inhibits photore- and bHLH factors. Semin Cell Dev Biol. 2004;15:83–89. ceptor production in the developing mammalian retina. 34. Harada T, Harada C, Parada LF. Molecular regulation of visual Development. 2006;133:913–923. system development: more than meets the eye. Genes Dev. 56. Jadhav AP, Cho SH, Cepko CL. Notch activity permits retinal 2007;21:367–378. cells to progress through multiple progenitor states and 35. Chow RL, Lang RA. Early eye development in vertebrates. acquire a stem cell property. Proc Natl Acad Sci U S A. 2006; Annu Rev Cell Dev Biol. 2001;17:255–296. 103:18998–181003. 36. Poche RA, Furuta Y, Chaboissier MC, et al. Sox9 is expressed in 57. Mizeracka K, DeMaso CR, Cepko CL. Notch1 is required in mousemultipotentretinalprogenitorcellsandfunctionsinMuller newly postmitotic cells to inhibit the rod photoreceptor fate. glial cell development. JCompNeurol. 2008;510:237–250. Development. 2013;140:3188–3197. 37. Livesey FJ, Young TL, Cepko CL. An analysis of the gene 58. Scheer N, Groth A, Hans S, Campos-Ortega JA. An instructive expression program of mammalian neural progenitor cells. function for Notch in promoting gliogenesis in the zebrafish Proc Natl Acad Sci U S A. 2004;101:1374–1379. retina. Development. 2001;128:1099–1107. 38. Yamazaki H, Katsuoka F, Motohashi H, et al. Embryonic 59. Bernardos RL, Lentz SI, Wolfe MS, Raymond PA. Notch-Delta lethality and fetal liver apoptosis in mice lacking all three signaling is required for spatial patterning and Muller glia small Maf proteins. Mol Cell Biol. 2012;32:808–816. differentiation in the zebrafish retina. Dev Biol. 2005;278: 39. Takahashi K, Yamanaka S. A decade of transcription factor- 381–395. mediated reprogramming to pluripotency. Nat Rev Mol Cell 60. Lee HY, Wroblewski E, Philips GT, et al. Multiple require- Biol. 2016;17:183–193. ments for Hes 1 during early eye formation. Dev Biol. 2005; 40. Takeuchi JK, Mileikovskaia M, Koshiba-Takeuchi K, et al. 284:464–478. Tbx20 dose-dependently regulates transcription factor net- works required for mouse heart and motoneuron develop- 61. Ramachandran R, Zhao XF, Goldman D. Ascl1a/Dkk/beta- ment. Development. 2005;132:2463–2474. catenin signaling pathway is necessary and glycogen synthase kinase-3beta inhibition is sufficient for zebrafish retina 41. Liberg D, Sigvardsson M, Akerblad P. The EBF/Olf/Collier regeneration. Proc Natl Acad Sci U S A. 2011;108:15858– family of transcription factors: regulators of differentiation in 15863. cells originating from all three embryonal germ layers. Mol Cell Biol. 2002;22:8389–8397. 62. Yao K, Qiu S, Tian L, et al. Wnt regulates proliferation and neurogenic potential of Muller glial cells via a Lin28/let-7 42. Ding Q, Chen H, Xie X, et al. BARHL2 differentially regulates miRNA-dependent pathway in adult mammalian retinas. Cell the development of retinal amacrine and ganglion neurons. J . 2016;17:165–178. Neurosci. 2009;29:3992–4003. Rep 43. Heinrichs S, Conover LF, Bueso-Ramos CE, et al. MYBL2 is a 63. Sifuentes CJ, Kim JW, Swaroop A, Raymond PA. Rapid, sub-haploinsufficient tumor suppressor gene in myeloid dynamic activation of Muller glial stem cell responses in malignancy. Elife. 2013;2:e00825. zebrafish. Invest Ophthalmol Vis Sci. 2016;57:5148–5160. 44. Baker SJ, Ma’ayan A, Lieu YK, et al. B- is an essential 64. Seritrakul P, Gross JM. Tet-mediated DNA hydroxymethylation regulator of hematopoietic stem cell and myeloid progenitor regulates retinal neurogenesis by modulating cell-extrinsic cell development. Proc Natl Acad Sci U S A. 2014;111:3122– signaling pathways. PLoS Genet. 2017;3:e1006987. 3127. 65. Powell C, Elsaeidi F, Goldman D. Injury-dependent Muller glia 45. Oh IH, Reddy EP. The myb gene family in cell growth, and ganglion cell reprogramming during tissue regeneration differentiation and apoptosis. Oncogene. 1999;18:3017–3033. requires Apobec2a and Apobec2b. J Neurosci. 2012;32:96– 46. Fusco A, Fedele M. Roles of HMGA proteins in cancer. Nat 109. Rev Cancer. 2007;7:899–910. 66. Kopp F, Mendell JT. Functional classification and experimental 47. Copley MR, Babovic S, Benz C, et al. The Lin28b-let-7-Hmga2 dissection of long noncoding RNAs. Cell. 2018;172:393–407. axis determines the higher self-renewal potential of fetal 67. Engreitz JM, Haines JE, Perez EM, et al. Local regulation of haematopoietic stem cells. Nat Cell Biol. 2013;15:916–925. gene expression by lncRNA promoters, transcription and 48. Bringmann A, Iandiev I, Pannicke T, et al. Cellular signaling splicing. Nature. 2016;539:452–455. and factors involved in Muller cell gliosis: neuroprotective 68. Jones PA. Functions of DNA methylation: islands, start sites, and detrimental effects. Prog Retin Eye Res. 2009;28:23–51. gene bodies and beyond. Nat Rev Genet. 2012;13:484–492.

Downloaded from iovs.arvojournals.org on 10/01/2021 Transcriptome and DNA Methylome of Retinal Muller¨ Glia IOVS j October 2019 j Vol. 60 j No. 13 j 4450

69. Smith ZD, Meissner A. DNA methylation: roles in mammalian 75. Stadler MB, Murr R, Burger L, et al. DNA-binding factors shape development. Nat Rev Genet. 2013;14:204–220. the mouse methylome at distal regulatory regions. Nature. 70. Meissner A, Mikkelsen TS, Gu H, et al. Genome-scale DNA 2011;480:490–495. methylation maps of pluripotent and differentiated cells. 76. Powell C, Grant AR, Cornblath E, Goldman D. Analysis of DNA methylation reveals a partial reprogramming of the Muller glia Nature. 2008;454:766–770. genome during retina regeneration. Proc Natl Acad Sci U S A. 71. Guo H, Zhu P, Yan L, et al. The DNA methylation landscape of 2013;110:19814–19819. human early embryos. Nature. 2014;511:606–610. 77. Nelson BR, Ueki Y, Reardon S, et al. Genome-wide analysis of 72. Sanosaka T, Imamura T, Hamazaki N, et al. DNA Methylome Muller glial differentiation reveals a requirement for Notch analysis identifies transcription factor-based epigenomic signaling in postmitotic cells to maintain the glial fate. PLoS signatures of multilineage competence in neural stem/ One. 2011;6:e22817. progenitor cells. Cell Rep. 2017;20:2992–3003. 78. Conner C, Ackerman KM, Lahne M, et al. Repressing notch 73. Xie W, Schultz MD, Lister R, et al. Epigenomic analysis of signaling and expressing TNFalpha are sufficient to mimic retinal regeneration by inducing Muller glial proliferation to multilineage differentiation of human embryonic stem cells. generate committed progenitor cells. J Neurosci. 2014;34: Cell. 2013;153:1134–1148. 14403–14419. 74. Weber M, Hellmann I, Stadler MB, et al. Distribution, 79. Wan J, Goldman D. Opposing actions of Fgf8a on Notch silencing potential and evolutionary impact of promoter signaling distinguish two Muller glial cell populations that DNA methylation in the . Nat Genet. 2007;39: contribute to retina growth and regeneration. Cell Rep. 2017; 457–466. 19:849–862.

Downloaded from iovs.arvojournals.org on 10/01/2021