UHRF2 regulates cell cycle, epigenetics and gene expression to control the timing of retinal progenitor and ganglion cell differentiation

Development of the retina involves the highly coordinated generation of six neural and one glial cell type in correct numbers and at the appropriate time from a pool of multipotent retinal progenitor cells (RPCs) (Cepko, 2014; Bassett and Wallace, 2012; Centanin and Wittbrodt, 2014). Retinal ganglion cells (RGCs) are produced first during embryogenesis, followed by cone photoreceptors, horizontal cells and amacrine neurons (Waid and McLoon, 1995; Mu and Klein, 2004). Rod photoreceptors, Müller glia and bipolar cells are mostly generated after birth, and cell proliferation is complete by postnatal day (P) 10 in the mouse retina (Young, 1985). Many extrinsic and intrinsic factors balance the maintenance of RPCs with the production of new cell types by controlling cell cycle exit, interkinetic nuclear migration, and symmetric versus asymmetric cell division (Cepko et al., 1996; Agathocleous and Harris, 2009; Wallace, 2011; Seritrakul and Gross, 2019; Dyer and Cepko, 2001; Baye and Link, 2007; Del Bene et al., 2008). Extrinsic factors adjust and diminish RPC multipotential fate over the developmental window (Austin et al., 1995; Cepko et al., 1996; Xiang, 2013; Aldiri et al., 2017; Wallace, 2011; Bassett and Wallace, 2012). RPC fate is also regulated through highly coordinated changes to key lineage-driving transcription factors and the epigenetic status of their transcriptional target (Cepko, 2014; Brzezinski and Reh, 2015; Kim et al., 2016; Mo et al., 2016; Swaroop et al., 2010; Corso-Díaz et al., 2018; Seritrakul and Gross, 2019; VandenBosch and Reh, 2020).

DNA methylation of cytosine in CpG dinucleotides plays a key role in regulating gene expression during retinal development (Mo et al., 2016; Kim et al., 2016; Corso-Díaz et al., 2018; Dvoriantchikova et al., 2019). The methylation status of the genome is maintained during DNA replication by the coordinated activity of ubiquitin-like, containing PHD and RING finger domains, 1 (UHRF1) protein and DNA methyltransferase 1 (DNMT1). UHRF1 binds to newly replicated hemi-methylated DNA strands and recruits DNMT1 protein into proximity to accurately copy the methylation pattern to the daughter strand (Sharif et al., 2007; Bostick et al., 2007; Zhang et al., 2011; Rothbart et al., 2012). A loss of either DNMT1 or UHRF1 can lead to passive DNA demethylation during replication (Liu et al., 2013). Methyl groups can be directly removed through an iterative process not requiring cell division called active demethylation (Tahiliani et al., 2009; Koh and Rao, 2013; Rasmussen and Helin, 2016). This process begins when ten-eleven translocation (TET) enzymes oxidize 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC) (Kriaucionis and Heintz, 2009; Shen and Zhang, 2013; Tahiliani et al., 2009). Crucial for proper development of the retina and most tissues, 5hmC coordinates the expression of genes important for cellular differentiation (Ficz et al., 2011; Hahn et al., 2013; Perera et al., 2015; Seritrakul and Gross, 2017). It can be further oxidized to 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) by TET proteins (Ito et al., 2011) and these bases are enzymatically replaced to cytosine by the thymine DNA glycosylase (TDG) enzyme (He et al., 2011; Weber et al., 2016). TET activity is crucial for eye development (Xu et al., 2012), as tet2 and tet3 knockdowns reduced 5hmC locally to alter gene expression and impaired retinal tissue development (Seritrakul and Gross, 2017).

UHRF2, a paralog of UHRF1, was identified as an E3-ubiquitin ligase involved in cell cycle regulation (Mori et al., 2002, 2011). Although structurally homologous to UHRF1, ectopic UHRF2 expression cannot replace the methylation defect in Uhrf1^−/− embryonic stem cells (ESCs) or compensate for Uhrf1 deletion in mice, suggesting these two proteins possess distinct functions (Zhang et al., 2011; Pichler et al., 2011; Vaughan et al., 2018). UHRF2 was recovered from neural progenitor cells as a 5hmC binding protein (Spruijt et al., 2013). UHRF2 co-localizes with 5hmC in cancer cell lines, as seen using chromatin and DNA immunoprecipitation comparisons (Liu et al., 2016). Uhrf2 germline knockout mice are viable and exhibit no gross phenotypic defects, but do show altered neural gene expression and decreased 5mC at specific loci (Chen et al., 2017; Liu et al., 2017). The co-crystal structure of the UHRF2 SET and RING-associated (SRA) domain bound to 5hmC has been solved (Zhou et al., 2014). UHRF2 associates with the base-excision repair (BER) complex and can promote TDG-mediated active demethylation (Liu et al., 2021). UHRF2 protein is widely expressed in neural, intestinal and common lymphoid progenitor cells and generally remains expressed in post-mitotic tissues (Spruijt et al., 2013; Munoz et al., 2012; Lu et al., 2016; Li et al., 2020). However, UHRF2 function has not been investigated in RPCs.

In this paper, we show that UHRF2 levels are elevated in RPC-rich whole retinae at P0 compared with differentiated cells. Uhrf2 deletion from RPCs delayed cell cycle arrest and increased total cell numbers and KI67⁺ proliferating retinal cells. RGCs were overproduced in Uhrf2-deficient retinae, partially at the expense of RPCs, which were reduced. Other retinal cell types, including rod, cone, horizontal, bipolar and Müller glia, exhibited delayed differentiation that was mostly restored by P30. Uhrf2 deficiency reduced 5hmC both globally and locally within the gene bodies of the Tet3-Uhrf2-TDG active demethylation circuitry, contributing to their reduced expression. In contrast, the rod-specific gene rhodopsin (Rho), which undergoes demethylation during rod photoreceptor development, was less methyl- and hydroxymethylated across the promoter and its expression was further induced in Uhrf2-deficient retinae. These findings collectively indicate that UHRF2 regulates cell cycle of RPCs and controls 5hmC homeostasis to alter gene expression and augment progenitor differentiation.

We first examined the expression patterns of Uhrf2 in the developing and postnatal retina. RNA was isolated from control and Uhrf2-deficient retinae at P0 and P30 to determine Uhrf2 mRNA levels using quantitative PCR (qPCR) using primer sets that span exons 2-3, 7-8 and 15-16 (Fig. 1A). In each case, Uhrf2 mRNA levels were significantly reduced by ∼75% at P30 compared with P0. UHRF2 protein levels were assessed by immunoblotting at P0, P7 and P30 using three independent samples (Fig. 1B). UHRF2 protein levels were quantified and normalized against β-actin loading control and were more abundant at P0 than at P7 and P30 (Fig. 1C). We tested UHRF2 function by conditionally deleting the Uhrf2 gene from RPCs. Mice with Uhrf2 exon 3 sequence flanked by loxP sites (Fig. 1D) were crossed with mice carrying the visual system homeobox 2 (Vsx2) promoter driving expression of a Cre:Gfp transgene to generate Vsx2-Cre⁺; Uhrf2^fl/fl offspring (Rowan and Cepko, 2004). The Vsx2 promoter expresses Cre recombinase in RPCs so all neural and Müller glial retinal cell types are affected by the deletion. We confirmed the loss of Uhrf2 mRNA by qPCR analysis using primers spanning exons 2 and 3 (Fig. 1E) and UHRF2 protein by immunoblotting (Fig. 1F). Sections from control and knockout retina were cut and stained with Hematoxylin and Eosin (H&E) for histological examination (Fig. 1G). Uhrf2 deletion led to increased retinal cell layer thickness, which was particularly notable at P7. Uhrf2 deletion significantly increased total retinal cell numbers, compared with wild-type retinae, at P0, P3 and P7, but cell numbers were restored to levels observed in control cells by P30 (Fig. 1H).

We detected proliferative cells by immunostaining wild-type and Uhrf2-deficient retinae with anti-KI67 antibodies. Uhrf2 deficiency led to increased KI67⁺ cells early and throughout postnatal development (Fig. 2A). KI67⁺ cells were elevated at P0 and P3 throughout the neuroblastic layer (NBL) of Uhrf2-deficient retinae. Control retinal cells are almost all post-mitotic at P7; however, KI67⁺ retinal cells are still present in Uhrf2-deficient retinae at P7. No proliferating cells were detected at P30 in control or Uhrf2-deficient retinae. KI67⁺ cells were quantified in P0, P3, P7 and P30 control and Uhrf2-deficient retinae (Fig. 2B). The results demonstrate significant hypercellularity in Uhrf2-deficient retina through postnatal development that equilibrates in the mature retina. The retinoblastoma (pRB) cell cycle regulator is phosphorylated by cyclin/CDK phosphorylation during the G₁/S cell cycle transition (Dyson, 2016). We tested whether Uhrf2 deletion leads to increased serine 780 phosphorylation of pRB, observing a significant increase in pRB-P (S780) in extracts from knockout retinae (Fig. 2C). Retinal cells expressing CRE:GFP⁺ fusion from Vsx2 or negative (CRE:GFP⁻) were isolated from P0 control and Uhrf2-deficient retinae and assessed for cell cycle stages by flow cytometry (Fig. 2D). Uhrf2 loss significantly reduced the percentage of retinal cells in the G₁ phase and increased the numbers of S-phase and G₂/M-phase cells. We assessed apoptosis in the Uhrf2-deleted retinae as deregulation of the Rb/E2F pathway is often associated with increased cell death. The numbers of TUNEL⁺ apoptotic cells were detected in P0 control and Uhrf2-deficient retinae by immunofluorescence (IF) staining (Fig. 2E) and were comparable (Fig. 2F). TUNEL⁺ apoptotic cells were stained by immunohistochemistry (IHC) in P0 and P7 mice (Fig. 2G) and quantified and graphed (Fig. 2H). Levels of apoptosis were the same between genotypes at P0 but dropped in wild-type retinae at P7, with levels in Uhrf2-deleted retinae – significantly higher than in P7 control cells. Overall, retinal cells lacking Uhrf2 phosphorylate pRB and fail to undergo timely growth arrest during development, leading to excess KI67⁺ cells.

We determined how 5hmC levels are affected by location in control and Uhrf2-deficient developing retinal tissue using IHC. 5hmC staining was comparable at P0; however, individual cells with higher levels of 5hmC were identified in the outer NBL of control cells (Fig. 3A, top, red arrow). 5hmC visibly increased in much of the NBL of P3 and P7 control versus Uhrf2-deficient retinae. Each cell in the control retina stained 5hmC⁺ but, in contrast, many Uhrf2-deficient retinal cell types were 5hmC⁻, particularly evident in photoreceptor cells in the outer nuclear layer (ONL). Intriguingly, cells within the abnormally proliferating inner nuclear layer (INL) were a mixture of 5hmC⁺ and 5hmC⁻ cells. The 5hmC levels were comparable in cells of the ganglionic cell layer (GCL) of both retinae at P7. These differences noted at P7 were largely corrected by P30, when control and Uhrf2-deficient cells displayed a similar staining pattern (Fig. 3A, bottom).

We used isotope dilution high performance liquid chromatography (HPLC)-tandem mass spectrometry to accurately quantify 5mC, 5hmC and 5fC levels during retinal development in control and Uhrf2-deficient retinae at P7 (Fig. 3B, left). 5mC levels did not differ significantly but 5hmC levels were reduced by ∼30% after Uhrf2 deletion compared with control. Furthermore, 5fC, a reliable marker of active demethylation, was also significantly reduced in the Uhrf2-deleted retina. These findings indicate that UHRF2 is required for 5hmC and 5fC maintenance during retinal development. Long-term effects of Uhrf2 deletion on epigenetic marks of DNA at 8 months demonstrated that 5hmC levels were equal between control and Uhrf2-deficient retinae (Fig. 3B, right), but 5fC levels were significantly reduced in aged Uhrf2-deleted retinae, indicating a continued failure of active demethylation. DNA dot-blotting was used to qualitatively measure 5hmC levels in control and Uhrf2-deleted retinae at P7 (Fig. 3C). Global 5hmC levels were scanned and compared with Methylene Blue staining of total DNA and were reduced by ∼40% in Uhrf2-deficient retinae compared with control, consistent with mass spectrometry results.

We tested whether other ‘active demethylation’ components are also highly expressed in RPCs similar to Uhrf2. Expression of each of the Tet genes (Tet1, Tet2, Tet3) and Tdg was significantly reduced at P30 compared with expression levels at P0 (Fig. 3D), indicating that active demethylation pathway components are more highly expressed in proliferating progenitor than in differentiated retinal cells. The reduced levels of 5hmC observed in Uhrf2-deficient retinae at P7 could be caused by several factors. One possibility is that reduced expression of Tet genes at P7 contributes to the reduced 5hmC. Indeed, we observed significantly decreased expression of Tet1, Tet2, Tet3 and Tdg in Uhrf2-deleted retinal cells at P7 compared with control (Fig. 3E, left). Because 5hmC levels were restored in Uhrf2-deficient retinae by P30, we anticipated that expression levels of the Tet genes would also be restored. As expected, qPCR analysis indicated that Tet1, Tet2, Tet3 and Tdg were all returned to comparable levels in control and knockout retinal cells at P30 (Fig. 3E, right).

Further transcriptional differences in Uhrf2-deleted retinae at P7 were uncovered using RNA-sequencing (RNA-seq). Fragments per kilobase of transcript per million mapped reads (FPKM) values were generated based on the mapped files and used to assess specific gene expression differences. Transcripts displaying standard deviation >1 were identified, log transformed, mean centered and displayed using a heatmap (Fig. 3F). Relative transcript expression level is denoted by the color code blue (low) to yellow (high). In total, 54 genes were significantly induced, and expression levels of 275 genes were reduced in the Uhrf2-deleted retina. We used Gene Set Enrichment Analysis (GSEA) comparison of control and Uhrf2-deficient retinae to give a comprehensive view of the pattern of changes to gene expression (Fig. 3G) (Mootha et al., 2003; Subramanian et al., 2005). ‘Hallmark’ gene sets with a false discovery rate (FDR) lower than 0.25 were displayed. The normalized enrichment scores (NES) allow comparison of gene set analysis results. One set of reduced genes related to E2F targets relating to G₂M, mitosis DNA damage and apoptosis (Luo et al., 2013; Motnenko et al., 2018; Lu and Hallstrom, 2013). A second set lacked gene expression responsiveness to developmental extrinsic signaling pathways such as TGF-β, androgen response, Notch, Wnt/β-catenin and hedgehog. DNA repair and UV response, metabolism, oxidative phosphorylation and reactive oxygen species gene sets (shaded pink) were induced upon Uhrf2 deletion. Overall, the expression studies indicate that UHRF2 is transiently required to maintain normal 5hmC levels and gene expression during retinal development.

We next examined whether UHRF2 is required for proper retinal cell differentiation by performing IHC using cell-type specific marker antibodies to detect each retinal cell type in control and Uhrf2-deficient developing retinae. The number of Brn3b⁺ (encoded by Pou4f2) RGCs, the first retinal cells formed, were detected with anti-Brn3b antisera by IHC (Fig. 4A, red arrows) and were significantly overproduced in Uhrf2-deficient retinae at P0, P3 and P7 compared with control, but restored to comparable numbers at P30 (Fig. 4B). We used the Vsx2-Cre:GFP transgene to quantify GFP⁺ RPC numbers by flow cytometry across a developmental time course in control and Uhrf2-deficient retinae (Fig. 4C) (Rowan and Cepko, 2004). We noted significantly fewer GFP⁺ cells in Uhrf2-deficient retinae at P0, P3 and P7, but at P18 GFP⁺ cell numbers were equalized for both genotypes. At P3, Uhrf2-deleted retinae contained almost no GFP⁺ RPCs, whereas control retinae contained ∼35% GFP⁺ RPCs. These findings indicate that UHRF2 limits the production of RGCs from RPCs in normal retinae, but RGCs are significantly overproduced and RPCs reduced upon Uhrf2 deletion. A list of genes that promote particular retinal cell fate was generated based on representation in high quality lists (Clark et al., 2019; Bassett and Wallace, 2012; Dvoriantchikova et al., 2019; Xiang, 2013). RNA-seq derived gene expression levels were compared between control and Uhrf2-deficient retinae and represented as significantly increased (red), significantly decreased (pink) or not significantly altered (gray). For RPC-related genes, one gene (Pax6) was induced upon Uhrf2 deletion and five genes were significantly reduced in the knockout (Ikzf1, Sox9, Dll1, Notch1 and Ccnd1) (Fig. 4D). In contrast, four genes involved in RGC production were induced following Uhrf2 deletion (Eomes, Pou4f1, Barhl2 and Pou4f2) and two genes were reduced (Dlx1 and Dlx2) (Fig. 4E). Cone cells, detected with anti-cone arrestin antibody, were present but also delayed in production in Uhrf2-deficient retinae at P7 compared with control (Fig. 4F, red arrows). Cone cells remained significantly reduced by ∼40% in the Uhrf2-deficient retinae at P30 (Fig. 4G). Eight cone-specifying genes were reduced with none increased (Fig. 4H). These data indicate that Uhrf2 controls proper levels of the early produced RGC, RPC and cone cells; these changes correlate with altered expression of their related cell-fate driving transcription factors.

Rod photoreceptor cells were detected by IHC with anti-rhodopsin antibody. Rod production was strikingly delayed in the Uhrf2-deficient retina at P3 (Fig. 5A). By P7, many rhodopsin⁺ cells had abnormally traversed the outer plexiform layer (OPL) and penetrated the INL (Fig. 5A, middle, red arrows). Rhodopsin⁺ cells mislocalized to the INL were still detected in mature P30 Uhrf2-deficient retina (Fig. 5A, bottom, red arrows), and P414 aged retinae (Fig. S1). Rho⁺ cells were quantified in control and Uhrf2-deficient retinae and were significantly fewer in knockout retinae at P3 and P7, but levels were equal at P30 (Fig. 5B). Expression of genes that promote rod cell generation was also affected, with Rho significantly overproduced in Uhrf2-deficient retinae, but the expression levels of six other rod genes were reduced in the knockout (Fig. 5C). These data indicate that UHRF2 controls Rho and other rod-specific gene expression, rod cell positioning and production in the developing retina.

We detected bipolar cells in control cells using anti-PKCα (Fig. 5D) and anti-VSX2 antibodies (Fig. 5F), and defects were noted with each in P7 Uhrf2-deficient retinae. The number of PKCα⁺ cells was significantly reduced in Uhrf2-deleted retinae at P7 compared with control, but by P30 the levels were comparable between genotypes (Fig. 5E). VSX2⁺ cells were detected as expected in P7 control retinae. VSX2⁺ cell position was shifted abnormally to the center of the INL closer to the GCL in the knockout retina (Fig. 5F), potentially displaced by the rhodopsin⁺ cells abnormally located on the incorrect side of the OPL (Fig. 5A, middle). At P30, the number of VSX2⁺ cells appeared to be comparable between control and Uhrf2-deficient retinae. Bipolar-related gene expression indicates that one gene was increased in the knockout (Bhlhe23) and three (NeuroD1, NeuroD4 and Ascl1) were significantly reduced (Fig. 5G). These findings indicate that Uhrf2 deficiency contributes to a delay, but not failure, of bipolar cell development.

Horizontal cells were detected with anti-calbindin antibody in control retinae at P7. Their development was delayed in Uhrf2-deficient retinae (Fig. 6A, red arrows) but were comparable with control retinae at P30. The expression levels of five of seven horizontal cell genes queried was reduced at P7 in knockout cells compared with control (Fig. 6B). Uhrf2 deficiency also affected gliogenesis, as Müller glia and axon connections were not detected in the knockout retinae at P7 by IHC using anti-CRALBP antibody (Fig. 6C) but were present in both control and Uhrf2-deficient P30 retinae. Genes related to Müller glia production were mostly unaffected in the Uhrf2-deficient retina, with only Hey2 showing reduced expression (Fig. 6D). Syntaxin⁺ amacrine cell staining was comparable between control and Uhrf2-deficient retinae at P7 and P30 (Fig. 6E). One amacrine-promoting gene was increased (Neurod2) and three (Foxn4, Neurod1 and Neurod4) were reduced (Fig. 6F). In summary, Uhrf2 deficiency led to delays and, for some cells, long-term defects in production of both neural and glial retinal cell types. However, compensatory mechanisms allow most cells to develop normally after their initial delay in cellular production.

UHRF2 supports the completion of active demethylation in ESCs, where Uhrf2^−/− leads to 5mC accumulation at promoters bound and regulated by UHRF2 (Liu et al., 2021). We chose four target genes for further analysis to elucidate how Uhrf2 regulates levels of 5mC, 5hmC and gene expression in the developing retina. Tet3, Uhrf2 and Tdg make up the ‘active demethylation’ pathway, which facilitates the enzymatic replacement of 5mC marks back to cytosine. Rho is an important rod gene that undergoes active demethylation in rod photoreceptors during retinal development to support rod-specific expression (Kim et al., 2016; Dvoriantchikova et al., 2019). The Rho gene was significantly induced in Uhrf2-deficient retinae, seen using RNA-seq, whereas expression of Tet3, Uhrf2 and Tdg was significantly reduced (Fig. 7A). In order to determine how Uhrf2 loss could affect the expression of these genes, we employed two techniques to map 5hmC marks in the gene bodies, or 5mC and 5hmC in the CpG islands (CGIs).

Elevated 5hmC within gene bodies outperforms 5mC as a strong predictor of positive gene expression (He et al., 2021). We used 5hmC-DNA-immunoprecipitation followed by qPCR (hMeDIP-qPCR) to explore the Tet3, Tdg and Uhrf2 gene bodies for 5hmC accumulation during retinal development. Rho was not analyzed using this assay because it did not accumulate 5hmC in its gene body in this same study (Perera et al., 2015). 5hmC marks were significantly reduced in the gene bodies of Tet3, Tdg and Uhrf2 (Fig. 7B). Although it is not clear by what mechanism the presence of 5hmC in the gene body augments gene expression or how UHRF2 stabilizes it, Uhrf2-dependent reduction in gene bodies could contribute to decreased levels of gene expression (He et al., 2021). Furthermore, UHRF2 is prominently detected at introns by chromatin immunoprecipitation (Liu et al., 2021).

Next, we assessed the methylation and hydroxymethylation status of the CGIs located in these genes. We used simultaneous targeted methylation sequencing (sTM-seq) to generate base resolution detection of 5mC and 5hmC sequences in discrete regions of the mouse Rho, Tet3, Tdg and Uhrf2 genes. This approach allows for examination of multiple targeted genes at once. DNA is bisulfite (BS) converted to detect the combined levels of 5mC and 5hmC. A diagram showing the genes and primer locations is shown in Fig. S2. Rho, a rod-specific phototransduction gene, is highly methylated and poorly expressed in RPCs. As RPCs differentiate into rod cells, the Rho gene undergoes active demethylation and its expression levels rise (Kim et al., 2016; Dvoriantchikova et al., 2019). In non-rod cells however, Rho stays methylated and is repressed. Analysis of the Rho CGI by sTM-seq revealed that the Rho gene contains less 5mC and 5hmC at 7 of 8 CpGs tested in Uhrf2-deficient (yellow) compared with wild-type (blue) retina (Fig. 7C). Oxidative bisulfite (OxBS) ‘converts’ away 5hmC before BS treatment and this allows detection of the remaining 5mC only. 5mC levels were reduced at positions #6 and #8 in Uhrf2-deficient (purple) compared with wild-type (green) retina. 5hmC levels were calculated by subtracting the difference between the BS (5mC+5hmC) and OxBS (5mC). 5hmC levels were significantly reduced at positions #4, #5 and #8. Uhrf2 loss thus may have distinct effects at different CpGs in terms of maintaining cytosine methylation and hydroxymethylation. Analysis of the Tet3, Uhrf2 and Tdg genes did not reveal significant changes to 5mC or 5hmC in the CGI region analyzed by sTM-seq (Fig. S3). Overall, these data indicate that UHRF2 maintains 5mC and/or 5hmC at specific CpGs in the promoter of Rho and this affects Rho expression in the knockout. UHRF2 also augments gene expression of members of the active demethylation pathway by mediating 5hmC accumulation in their gene bodies.

In this study we demonstrate that UHRF2 regulates cell cycle and pRB phosphorylation during the G₁ phase, controls site-specific methylation and hydroxymethylation at individual CpGs, and changes expression of cell cycle, signaling pathways and DNA repair genes during retinal development (Fig. 8A). Deleting Uhrf2 from RPCs caused their premature reduction, overexpression of several key RGC-promoting transcription factors and increased Brn3b⁺ cell numbers in the RGC layer. In contrast, there was a decreased number of cone cells, the next differentiating cell type. Most other retinal cell types exhibited delayed differentiation that was later restored, with rod and bipolar cells occasionally positioned improperly (Fig. 8B). These data demonstrate that UHRF2 possesses a role to promote the transition from retinal progenitor to differentiated cells. UHRF2-dependent changes to retinogenesis could be caused by alterations to cell cycle arrest or progression, by changes to the underlying epigenetic status of a cell, or a combination of both.

Here, we show that deleting Uhrf2 from the retina reduced the percentage of cells in the G₁ phase (Fig. 2B) concomitant with phosphorylation of pRB (Fig. 2C). Retinoblastoma (pRB) is a crucial cell-cycle protein that regulates cell fate during the G₁ phase for cells to permanently exit the cell cycle, arrest growth and differentiate (Weinberg, 1995; Nevins, 1998; Dyson, 2016; Zhang et al., 2004). UHRF2 physically associates with pRB as demonstrated both by direct co-immunoprecipitation (Mori et al., 2011) and by unbiased immunoprecipitation of UHRF2 followed by mass spectrometry identifying pRB among the bound proteins (Lai et al., 2016). UHRF2 was first characterized as a cell cycle regulatory protein that induced G₁ arrest by ubiquitylating and degrading the pRB-targeting G₁ cyclins D1 and E1, thereby inhibiting cells from undergoing G₁/S passage (Mori et al., 2002; Li et al., 2004). UHRF2 also physically associates with E2F1, a pRB-binding transcription factor, and this can positively regulate expression of the key G₁/S promoting cyclin Ccne1 (Lu and Hallstrom, 2013). Deletion of Uhrf2 in vivo in the retina caused excess KI67⁺ cell numbers so it might be expected to see increased E2F activity by GSEA. But instead we observed reduced E2F activity, and these genes seem specifically related to G₂M and mitotic spindle targets (Fig. 3G). Uhrf2 deletion also caused reductions in Ccnd1 (Fig. 4D) and Rb1 (Fig. 5C) mRNA by RNA-seq. Relatively few total cells were proliferating in the knockout compared with the total number of cells assayed at P7, and proliferation could have ceased before KI67 expression was extinguished. Further, it is unclear whether the Ser780 pRB-P antibody is detecting mono-phosphorylation of pRB at that site or whether Ser780 is one of the many sites included among pRB hyperphosphorylation which is linked to S-phase entry and proliferation. In contrast, deleting Dnmt1 using the same Vsx2-Cre line increased the numbers of cells in G₁ and reduced cells in S/G₂/M (Rhee et al., 2012). DNMT1, UHRF1 and UHRF2 each can physically associate with pRB, and the E3-ubiquitin ligase activity of both UHRF1 and UHRF2 can situationally induce the degradation of DNMTs and cyclins. This suggests a potentially complex interplay between these cell cycle phase regulators during G₁ phase passage (Robertson et al., 2000; Jeanblanc et al., 2005; Jia et al., 2016; Bronner et al., 2007). The overall effects of Uhrf2 deletion on G₁ phase progression may pertain to the transition from early- to late-G₁ phase, particularly when considering the reduced gene expression patterns to several G₁ regulating signaling pathways such as Notch and Wnt.

DNA methylation and demethylation play a major role in controlling gene expression and cell fate during retina tissue development (Nasonkin et al., 2013; Corso-Díaz et al., 2018). Other epigenetic regulators, such as UHRF2, are expressed in RPCs and exert precise control over DNA methylation and cell fate during retinal development (Nasonkin et al., 2011; Rhee et al., 2012; Seritrakul and Gross, 2014, 2017, 2019; Singh et al., 2017; Perera et al., 2015; Corso-Díaz et al., 2018). Zebrafish retinae lacking dnmt1, for example, do not maintain adult retinal stem cells (Angileri and Gross, 2020). Although Uhrf1 has not yet been deleted from RPCs, Uhrf1 conditional deletion from neural stem cells causes global DNA hypomethylation, premature expression of cell cycle inhibitors and differentiation-promoting genes, and stem cell exhaustion (Ramesh et al., 2016; Blanchart et al., 2018). UHRF2 was isolated specifically from neural progenitor cell extracts but not ESCs or mature brain as a 5hmC-binding protein, unlike its counterpart UHRF1, which was recovered from ESCs (Spruijt et al., 2013). UHRF2 is highly expressed in different progenitor cell types and, unlike UHRF1 and although an E2F target gene, it remains expressed in differentiated neural and hematopoietic tissues (Pichler et al., 2011; Spruijt et al., 2013; Lu et al., 2016; Munoz et al., 2012; Li et al., 2020). Therefore, the roles of UHRF proteins are likely dependent on the developmental status of retinal cells.

In this study we show that Tet3 and Tdg are also more highly expressed with Uhrf2 in RPC-rich proliferating murine retinae (P0) than in mature post-mitotic retinal cells, similar to observations in other progenitor cells (Xu et al., 2012; Hon et al., 2014; Rasmussen and Helin, 2016; Seritrakul and Gross, 2017; Onodera et al., 2021). In the case of tet2^−/−/tet3^−/− mutant zebrafish, RGCs are specified but then fail to fully differentiate. Photoreceptor cell development was compromised in tet2 and tet3 knockouts in zebrafish, which suggested that TET activity mediates active demethylation in these cells during development (Seritrakul and Gross, 2017). This is particularly intriguing because the retinae of tet2^−/−/tet3^−/− mutant zebrafish also displayed unusually high proliferative rates and also were shown to overactivate Notch and Wnt signaling. Notch and Wnt affect the timing of cell cycle exit and differentiation of progenitors into RGCs (Waid and McLoon, 1995; Austin et al., 1995; Silva et al., 2003). RPCs change their sensitivity to specific extrinsic signals over the developmental time-course (Cepko et al., 1996; Livesey and Cepko, 2001; Dyer and Cepko, 2001; Ohnuma and Harris, 2003). In our study, Uhrf2 deficiency led to gene expression defects of several extrinsic signaling pathways crucial to retinal development, such as TGF-β, Notch, Wnt and sonic hedgehog (SHH), suggesting that UHRF2 may respond to and regulate these signals to control proliferation, gene expression or cell fate (Fig. 3G). Inhibition of Notch signaling by Uhrf2 deletion could contribute to the observed loss of RPC cells (Fig. 4E), which may be more susceptible to increased levels of RGC-promoting transcription factors (Fig. 4G). SHH production by RGCs auto-inhibit further production of RGCs, suggesting that reduced SHH in Uhrf2-deficient retinae could conceivably contribute to excess RGC production (Wang et al., 2005). These findings suggest that, although TET proteins and UHRF2 are both capable of augmenting 5hmC levels in vivo, they may have differential effects on retinal cell fate.

Photoreceptor specific genes such as Rho, Gnat1 and Cnga1 are methylated with low expression in RPCs and RGCs, but are demethylated and expressed in rod cells where required (Merbs et al., 2012; Kim et al., 2016; Dvoriantchikova et al., 2019). Methylation of the Rho CGI is compromised in Dnmt1 conditional knockout retina (Rhee et al., 2012), and compound disruption of Dnmt1, Dnmt3a and Dnmt3b activity in mouse retina causes severe hypomethylation of Rho, along with defects in photoreceptor differentiation and OPL structure (Singh et al., 2017). Genome-wide methylation analysis of DNA derived from purified rod and cone cells indicated that their cell specific gene expression was associated with reduced methylation at the promoter and gene bodies of those targets (Mo et al., 2016). In the retina, Rho expression is elevated in mice at postnatal week 3; however, 5hmC did not detectably accumulate within Rho during this developmental time-frame (Perera et al., 2015). Uhrf2 deletion in the retina had two primary effects on detection of Rho. Rho expression was significantly induced in Uhrf2-deficient retina (Fig. 7A). Consistent with this observation, combined levels of 5mC and 5hmC were reduced across seven of eight tested Rho CpGs. These changes most typically were due to reduced 5hmC without 5mC being affected. Second, the number of RHO⁺ cells detectable by IHC in Uhrf2 conditional knockout is reduced compared with wild-type at P7. Further analysis is required to understand the mechanisms behind the unusual Uhrf2-related changes to Rho regulation in the retina.

UHRF2 was recently identified as a key participant with TET and TDG in the active demethylation pathway. 5hmC-binding by UHRF2 coordinates an allosteric change within the protein causing its direct ubiquitylation of XRCC1, which leads to incorporation of TDG into the base excision repair complex. Uhrf2 ablation from ESCs was associated with reduced neural progenitor cell differentiation into mature neurons in induction studies (Liu et al., 2021). This defect was associated with site-specific accumulation of 5mC at UHRF2-bound promoters, owing to a failure to complete active demethylation (Liu et al., 2021). In another study, Uhrf2^−/− germline knockout mice in contrast decreased 5mC at specific loci, affecting neural gene expression (Chen et al., 2017, 2018; Liu et al., 2017). These and our findings indicate that UHRF2 function may have different effects on CpGs under different times or contexts. It is unclear why some 5hmC marks remain protected from further oxidation and others accumulate over time, particularly in neural tissue, whereas other sites undergo active demethylation to cytosine (Bachman et al., 2014; Hahn et al., 2014). Individual CpG sites can be differentially regulated and each can have distinct effects on gene expression and cell fate. This was observed for the T-regulatory cell-specifying Foxp3 gene, which is dynamically regulated at different CpGs during differentiation (Yue et al., 2016). However, elimination of active demethylation by conditionally deleting Tdg from hematopoietic stem cells did not compromise gross hematopoietic development (Onodera et al., 2021). Thus it remains to be deduced how much of UHRF2 activity and its role in retinal development depends on active demethylation. Genes besides Rho, such as Tet3, Uhrf2 and Tdg, displayed different regulation and did not undergo significant changes to methylation or hydroxymethylation in their promoters. Instead, during retinal development these genes accumulated 5hmC marks in gene bodies, which is positively correlated with gene expression, and 5hmC levels were reduced at these positions in Uhrf2-deficient retinal cells. Future work may involve determining whether UHRF2 links histone or DNA epigenetic information to E3-dependent changes in associated proteins that may play a role in 5hmC homeostasis, gene expression, cell cycle or fate.

Sample sizes were estimated based on power analysis to produce significant outcomes (P<0.05) with 30% differences. No animals or samples were excluded from analysis. IHC and IF staining was performed blinded to animal genotype. P-values were determined by paired two-tailed Student's t-test with P<0.05 considered significant. Error bars represent standard deviation from the mean.

Vsx2-Cre:GFP Mus musculus (mouse) (Rowan and Cepko, 2004) and related PCR genotyping protocols have been previously described (Filtz et al., 2015; Xie et al., 2015). Mice with conditional deletion of Uhrf2 exon 3 in a C57BL/6 background were obtained from The Canadian Mouse Mutant Repository at The Hospital for Sick Children, Toronto, Canada. Vsx2-Cre and Uhrf2^fl/fl mice were bred together to generate Vsx2-Cre⁺; Uhrf2^fl/fl offspring. Male and female mice were used for experiments. Strain and ages of animals are listed in the figure legends. All mouse experiments were performed in accordance with University of Minnesota Institutional Animal Care and Use Committee procedures and guidelines.

RNA was isolated from retinal tissue using the RNeasy Mini Kit (Qiagen), after tissue disruption and homogenization with needles. RNA concentration was calculated using NanoDrop™ 2000/2000c Spectrophotometers (Thermo Fisher Scientific). Libraries were prepared from ∼500 ng total RNA with the TruSeq Stranded Total RNA Library Prep Kit according to the manufacturer's directions (Illumina). Paired-end 60 cycle sequencing was performed on HiSeq 2500 sequencers according to the manufacturer's directions (Illumina). Mapping was performed as previously described (Sarver et al., 2021). RT-qPCR was performed using StepOne Real-Time PCR system (Applied Biosystems). QuantiTect SYBR Green RT-PCR Kit was used following the manufacturer's instructions (Qiagen, 204245). We included primer sequences for all qPCR in Table S1. qPCR primer design and testing was performed according to MIQE guidelines (Bustin et al., 2009; Bustin and Wittwer, 2017).

For IF staining, mouse eyes were embedded in Tissue–Tek OCT compound (Fisher Scientific International) and frozen in liquid nitrogen-cooled isopentane. Then 8 μm thick frozen tissue sections were sectioned on a cryostat and mounted on aminopropyltriethoxysilane-coated slides. The slides were fixed in cold acetone and incubated in 10% normal serum for 30 min before adding in the primary antibody. Visualization was achieved using conjugated secondary antibodies and nuclear staining with 4,6-diamidino-2-phenylindol (DAPI). H&E staining, KI67, TUNEL and retinal marker IHC staining were performed by the BIONET shared resource as previously described (Xie et al., 2017). Briefly, tissue samples were fixed in 10% formalin and embedded in paraffin. The list of antibodies used is shown in Table S2. Biotinylated secondary antibody, diluted in TBST [20 mM Tris-HCl (pH 7.6), 150 mM NaCl, 0.01% Tween-20] per manufacturer's recommendation, was added to each section and incubated for 30 min at room temperature. The slide sections were developed using Vectastain ABC Kits (HRP) or Vectastain ABC-AP Kits (AP) following the manufacturer's instructions. The ApopTag Peroxidase kit (EMD Millipore, S7100) was used for TUNEL IHC analysis. Cell numbers were quantified by counting five fields from three independently isolated fixed and stained retinae and are graphed as percent cells compared with total retinal cells unless specified otherwise.

Retinal tissue was dissected from mouse eyes and dissociated with trypsin. After digestion, trypsin inhibitor was added and the cell suspension was filtered and pelleted. The cell pellets were fixed in ice-cold 70% ethanol for 1 h at 4°C, propidium iodide (PI) and RNAse A were added and cells were filtered for analysis. Cell cycle data were obtained using FACSCanto (BD Biosciences) and analyzed with FlowJo_V10 software. The retinal cells from Vsx2-Cre-negative mice (EGFP⁻), retinal cells from Vsx2-Cre-positive mice (EGFP⁺) and PI-positive retinal cells were used for gating at the University of Minnesota Flow Cytometry Resource.

Immunoblotting was performed as previously described (Lu et al., 2011; Lu and Hallstrom, 2012). Briefly, mouse retinal tissue was lysed in protein lysis buffer [50 mM Tris (pH 7.4), 300 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40, 50 mM NaF, 1 mM Na₃VO₄, 1 mM DTT, 1× Sigma protease inhibitor cocktail tablet] for 1 h. Protein concentration was measured using BCA Protein Assay Kit (Pierce). Equivalent amounts of protein were separated by SDS-PAGE, transferred to PVDF membrane (Millipore) and blocked in TBST [20 mM Tris-HCl (pH 7.6), 150 mM NaCl, 0.01% Tween-20] containing 5% nonfat dry milk. The list of antibodies used is shown in Table S2. Blots were then incubated with primary antibody (1:2000) overnight at 4°C, washed three times with TBST buffer and then incubated with the appropriate secondary antibody (1:2000) for 1 h at room temperature. Blots were processed using the ECL system (Amersham Biosciences). The membranes were stripped and reprobed with β-actin antibody to verify equal loading.

Genomic DNA was isolated from cells with a QIAamp DNA kit (Qiagen) using RNaseA to eliminate RNA. DNA was diluted at different specified concentrations and dot blotted onto a nitrocellulose membrane and cross-linked using an ultraviolet Stratalinker 1800 (Stratagene). The DNA-immobilized membrane was immunoblotted with rabbit anti-5hmC polyclonal antibody (Active Motif, 39791). Afterwards, total DNA was stained with 0.04% Methylene Blue in 0.5 M sodium acetate (pH 5.2) overnight, then scanned and quantified against 5hmC.

Genomic DNA was extracted from tissues and (1.5 µg) hydrolyzed with phosphodiesterases, DNAse I and Alkaline Phosphatase. Samples were spiked with internal standard for mass spectrometry, filtered and 5fC converted to biotinyl-fC and enriched using offline HPLC on an Atlantis T3 column and eluted with a gradient of 5 mM ammonium formate and methanol. dC was quantified by HPLC-UV during the HPLC cleanup using calibration curves to normalize the amounts of epigenetic modifications. HPLC fractions corresponding to 5hmC, 5mC and Biotinyl-fC were combined and analyzed by capillary HPLC-ESI-MS/MS on a Thermo TSQ Vantage mass spectrometer interfaced with a Thermo Dionex Ultimate3000 HPLC. Quantification was conducted in the SRM mode by monitoring the transitions m/z 242→126 for 5mC, m/z 254→133 for 5mC standard, m/z 258→142 for 5hmC, m/z 261→145 for 5hmC standard, m/z 569→453 for Biotinyl-fC and 581→460 for Biotinyl-fC standard. Chromatographic separation was achieved on a Zorbax SB-C18 column eluted at a flow rate of 15 µl/min with a gradient of 2 mM ammonium formate and methanol. The mass spectrometer was operated in negative mode and the parameters for ionization and fragmentation were optimized using authentic standards. Note that the percentage of dC values for both hmC and fC must be multiplied by (10-1) and (10-3), respectively, owing to their lower abundance. Values are presented as the percent base compared with total cytosine.

Retinal genomic DNA was sonicated and fragmented to between 100-400 base pair average size. hMeDIP-qPCR was performed following the manufacturer's instructions (Active Motif, 55010) with minor modifications. The eluted DNA was treated with proteinase K to remove the immunoprecipitating antibody and then was purified with MinElute PCR purification kit (Qiagen, 28004). The purified DNA was used to perform quantitative PCR. The qPCR reactions were carried out using the QuantiTect SYBR Green PCR Kit (Qiagen, 204145) in triplicate. Ct value was analyzed to calculate enrichment using the 2^−ΔΔCt method (Livak and Schmittgen, 2001). Primer locations for hMeDIP-qPCR were selected by locating 5hmC peaks using Integrative Genomics Viewer on an hMeDIP-seq dataset generated from mature retinae and are shown in Fig. S2 (Perera et al., 2015).

Genomic DNA from three control (Vsx2-Cre:GFP) and three experimental (Vsx2-Cre⁺; Uhrf2^fl/fl) samples were isolated following the published protocol (Asmus et al., 2019). Then 5 µg genomic DNA from each sample was spiked with 150 ng standard 5hmC control DNA (Zymo, D5405) which was used to assess BS and OxBS conversion rates (Fig. S3). The DNA was sonicated (Covaris S220) using the following parameters: sample volume, 130 µl; target BP (peak), 500; peak incident power (w), 105; duty factor, 5%; cycles per burst, 200; treatment time (s), 80. Followed by DNA sonication, BS and OxBS treatment were performed using the TrueMethyl oxBS-Seq module from Tecan per their instructions, but we increased the starting material to 1 µg. The sTM-seq library construction was modified from a published protocol. Briefly, three PCR steps were used to construct a sTM-seq sequencing library. The first PCR reactions were run as separate samples to amplify the target regions from the genomic DNA. The PCR products were purified using Agencourt AMPure XP beads. For the second PCR, consensus adaptor sequences were added to the 5′ end of the target sequence primer using the same PCR parameters, except the annealing temperature was 68°C and we repeated 22 cycles. For the third PCR, the resulting PCR products from different genes were pooled and then amplified with NEBNext Ultra II DNA library prep kit for Illumina (E7645S, New England Biolabs) using index primers from NEBNext Multiplex Oligos for Illumina (E7335S, New England Biolabs) following the manufacturer's instructions. The indexed PCR products were cleaned using Agencourt AMPure XP beads and sent for quality control analysis at the University of Minnesota Genomics Center. sTM-seq primers of the target genes were designed using Methyl Primer software v1.0 and the sequences are shown in Table S1. The libraries were sequenced on Mi-Seq in 150 bp paired end mode at the University of Minnesota Genomics Center.

To analyze sequencing, 150 base reads from Mi-Seq sequencing of PCR products generated to examine potentially differentially methylated regions were reduced to 90 base reads using trimmomatic v-33 to remove the first nine base pairs and base pairs found after position 95 in order to improve mapping efficiency and systematically remove sequencing primer contamination. Samples were mapped to the mouse genome, or a control sequence containing hydroxy-methylated DNA (for BS and OxBS treatment) using Bismark v 22.3 using HISAT2. Bismark compares mapping to a normal genome as well as to a genome where all Cs are reduced. By comparing these mappings where sequence is present, the methylation state of a given sample can be calculated at each C. Methylation percentage at each location was calculated using the bismark_methylation_extractor function to generate files which represent the methylation percentage at C positions within the genome. Files representing methylation frequency at specific locations (.cov) were imported into and analyzed in R.

We thank Elisabeth Parr and Kindra Knutsen for mouse genotyping and colony maintenance.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.195644.