Functional analysis of the Vsx2 super-enhancer uncovers distinct cis-regulatory circuits controlling Vsx2 expression during retinogenesis

Retinal development is achieved through sequential coordinated steps that ultimately lead to the generation of seven major retinal cell types, born in a conserved order (Cepko et al., 1996; Livesey and Cepko, 2001). The initial events in retinal development involve the expansion of the retinal progenitor cells (RPCs), through multiple rounds of cell divisions, before they exit the cell cycle and give rise to six neuronal and one glial cell type (Bassett and Wallace, 2012). The transcription factors that drive proliferation and terminal differentiation often function in networks that are cell-type specific and autoregulatory in nature (Mao et al., 2008; Wang et al., 2014). Of those transcription factors, Vsx2 is a homeodomain-containing transcription factor expressed in RPCs, and becomes restricted to retinal bipolar cells and, to a lesser extent, Müller glia (Burmeister et al., 1996; Kim et al., 2008; Liu et al., 1994; Philips et al., 2005; Rowan and Cepko, 2004, 2005). Mutations in VSX2 are associated with microphthalmia in humans and mice, impacting retinal proliferation and cell-type differentiation (Bar-Yosef et al., 2004; Burmeister et al., 1996; Ferda Percin et al., 2000; Green et al., 2003; Williamson and FitzPatrick, 2014).

Mouse studies revealed diverse roles of Vsx2 during retinogenesis (Burmeister et al., 1996; Clark et al., 2008; Dorval et al., 2006; Green et al., 2003; Horsford et al., 2005; Liu et al., 1994; Livne-Bar et al., 2006; Rowan and Cepko, 2005; Rowan et al., 2004; West and Cepko, 2022). Loss of Vsx2 function leads to a severe reduction in retinal proliferation and a complete absence of bipolar cells (Burmeister et al., 1996; Ferda Percin et al., 2000; Green et al., 2003; Kim et al., 2008; Livne-Bar et al., 2006). VSX2 also acts to maintain neural retinal identity by repressing retinal pigment epithelium (RPE) identity via constraint of Mitf expression, and is implicated in suppressing rod photoreceptor differentiation in the neonatal retina (Dorval et al., 2006; Horsford et al., 2005; Livne-Bar et al., 2006; Rowan et al., 2004).

Investigation of the Vsx2 function in postnatal mouse retina points toward a regulatory network that involves OTX2, VSX2 and PRDM1(BLIMP1) in establishing a binary cell fate decision of bipolar cells versus rod photoreceptors (Brzezinski et al., 2010; Katoh et al., 2010; Mills et al., 2017; Wang et al., 2014). Otx2 is a homeobox transcription factor that is required for the proper genesis of photoreceptors and bipolar cells, whereas Prdm1 is involved in the rod photoreceptor production (Brzezinski et al., 2010; Fossat et al., 2007; Goodson et al., 2020a,.b; Housset et al., 2013; Katoh et al., 2010; Koike et al., 2007; Nishida et al., 2003). Data indicate that retinal precursors expressing both Otx2 and Vsx2 give rise to bipolar cells, while precursors that co-express Otx2 and Prdm1 form rod photoreceptors (Goodson et al., 2020a,b; Katoh et al., 2010; Wang et al., 2014). In this model, Prdm1 negatively regulates Vsx2 expression, thus promoting rod cell fate by suppressing bipolar cell specification in rod precursors (Goodson et al., 2020a; Katoh et al., 2010; West and Cepko, 2022). Whether VSX2 inhibits Prdm1 expression in bipolar cell precursors has not been established (Goodson et al., 2020a; West and Cepko, 2022). VSX2 likely exerts its function primarily by engaging with cis-regulatory elements (Clark et al., 2008; Reichman et al., 2010) and direct and indirect protein-protein interactions (Wilson et al., 1993). How VSX2 impinges on chromatin structure to orchestrate retinogenesis remains unclear.

Temporal and spatial control of gene expression during development is usually achieved by noncoding DNA sequences that act distally to regulate transcription in a tissue-specific manner (Buecker and Wysocka, 2012; de Laat and Duboule, 2013). Approximately 98% of the human genome is composed of non-coding sequences, much of which are likely to have regulatory functions (Pennacchio et al., 2013). Previous studies, including ours, have identified the cis-regulatory landscape of retinal cells in mice and humans (Aldiri et al., 2017; Hartl et al., 2017; White et al., 2016). The challenge now is to define the intended target genes of these enhancers and to map promoter-enhancer physical interactions in the complex 3D genome during retinal development, which is essential to understanding the molecular mechanisms underlying retinal development and diseases.

We have previously identified a large cis-regulatory region (a super-enhancer) upstream of Vsx2 that is required for the bipolar cell genesis (Norrie et al., 2019). However, several key important questions remained unanswered. For example, why did this enhancer deletion result in microphthalmia? What is the fate of retinal precursor cells destined to become bipolar cells? And what are the molecular mechanisms underlying the activation of Vsx2 regulatory elements during normal development? To address these questions, we mapped VSX2 genomic occupancy during retinogenesis, revealing VSX2-associated developmental gene networks. VSX2 binds its enhancer and cooperates with PAX6 to induce its activity. We further show that deleting a small region within the Vsx2 regulatory structure in mice impacts retinal proliferation and proper bipolar cell differentiation. We also show that retinae with a large deletion in the Vsx2 regulatory landscape suffer an increase in photoreceptor production and elevation in the expression of the Prdm1. VSX2 interacts with OTX2, occupies a conserved Prdm1 enhancer and can suppress the ability of OTX2 to activate the Prdm1 enhancer in vitro. Our data provide mechanistic insights into the role of VSX2 as a predominant transcriptional effector during retinogenesis, and how it drives cell fate specification and differentiation in RPCs.

To understand the roles of VSX2 in the murine developing retina, we demarcated genome-wide VSX2-associated chromatin regions by performing ChIP-Seq at E14.5 retina, defining over 9000 binding sites (Fig. 1A; Table S1). Gene Ontology (GO) term analysis of VSX2-bound peaks using Genomic Regions Enrichment of Annotations Tool (GREAT) (McLean et al., 2010) revealed top enrichment in biological functions related to retinal neurogenesis and morphogenesis, and spinal cord development, consistent with known functions of Vsx2 (Fig. 1C) (Clovis et al., 2016; Ferda Percin et al., 2000; Kim et al., 2008). To further delineate how VSX2 is associated with neural differentiation, we performed motif analysis of VSX2-binding sites using hypergeometric optimization of motif enrichment (HOMER) (Heinz et al., 2010). This analysis underscores an over-representation of sequences related to the binding of homeodomain-containing transcription factors, especially LIM-homeodomain (Lhx) proteins, which play predominant roles in retinal cell fate acquisition (Fig. 1B) (de Melo et al., 2016; Roy et al., 2013; Zibetti et al., 2019). Interestingly, the analysis indicates a particular enrichment in the Q50 homeoprotein consensus binding motif (TAATT^A/_G; Fig. 1B) (Murphy et al., 2019). Inspection of VSX2 occupancy revealed a robust presence near distinct groups of genes that regulate neural progenitor programs, including components of signaling pathways (Sfrp2), epigenetic regulators (Smarca4) and transcription factors (Lhx2) (Fig. 1D). Given that Vsx2 is expressed in RPCs and the adult retina, we sought to compare VSX2-associated genome-wide profiles in these two stages. We performed ChIP-Seq on adult mouse retina, defining over 19,000 peaks (Fig. 1E-H; Table S1). Similar to the E14.5 retina, motif analysis of VSX2-binding sites indicated a prevalent sequence enrichment of the Q50 motifs, which are suggested to be associated with transcription repression in bipolar cells (Fig. 1F) (Hughes et al., 2018; Murphy et al., 2019). However, GO analysis suggests more diverse roles of VSX2 that are related to cell homeostasis (mitochondria localization) and neuronal functions (phototransduction and action potential) (Fig. 1G). Because VSX2 is primarily expressed in retinal bipolar cells in the adult retina, we examined its genomic distribution within known bipolar cell genes. The analysis revealed enrichment of VSX2 within bipolar cell signature genes (126/149; 84.6%), as illustrated for Neurod4 and Cabp5 (Fig. 1H, Table S2). Notably, the over-representation of biological terms related to phototransduction, a function primarily mediated by photoreceptors, suggests that photoreceptor genes are principal targets for VSX2 occupancy, in agreement with previous data (Dorval et al., 2006). Indeed, we found that 117/190 (61.6%) and 65/108 (60.2%) of rod and cone genes, respectively, contained VSX2 peaks, including Pde6c (cones), and Crx, Tulp1 and Rbp3 (rods) (Fig. 1H,J; Table S2). This finding is consistent with the hypothesis that VSX2 maintains retinal bipolar cell identity by directly repressing non-bipolar gene expression programs, including those related to photoreceptors, in bipolar cells (Dorval et al., 2006; Livne-Bar et al., 2006; Murphy et al., 2019).

We then compared VSX2 genome binding profiles in E14.5 and adult retina. Data indicate that VSX2-shared peaks represent 33% (3077/9083) and 16% (3077/19,025) of E14.5 and adults peaks, respectively, indicating that VSX2 genomic binding profiles in RPCs and postmitotic neurons are largely non-overlapping (Fig. 1I, Table S1). Among VSX2-binding sites that are common between E14.5 and adult retina are those related to nearby genes known to be expressed in the developing and adult retina, such as Notch1, Dkk3 and Isl1 (Fig. 1J). The VSX2 preferential binding profile in adult retina includes photoreceptor and bipolar gene signatures, as described above (Fig. 1H,J), whereas E14.5-specific sites are associated with genes that are implicated in early retinal differentiation programs, such as Neurog2 and Olig2 (Fig. 1J). We also noticed evidence of VSX2 occupancy of nearby genes not expressed in the retina and known to be upregulated at E14.5 upon loss of Vsx2 function, such as Tfec (Fig. 1J, discussed later), likely reflecting direct transcriptional repression mediated by VSX2 (Fig. 1J). Taken together, our analysis reveals distinct networks of genes regulated by VSX2 in the developing and adult retina, underscoring the diverse roles that Vsx2 plays in establishing proper transcriptional programs of retinal lineage fate commitment.

We noticed that VSX2 robustly binds conserved non-coding regions upstream of Vsx2 itself (termed EN1 and EN2 hereafter) in E14.5 and adult retina (Fig. 2A). When explored with our previously published E14.5 retinal epigenomic data (Aldiri et al., 2017), these areas have sharp ATAC-Seq peaks, rich in enhancer marks such as BRD4 and H3K4me1, and are occupied by RNA polymerase II (RNAPII), which is indicative of enhancer activities (Fig. 2A). We reasoned that VSX2 modulates its enhancer through a positive- feedback loop. To test this, we performed luciferase assays in 293T cells, transfecting plasmids with or without Vsx2 cDNA along with luciferase vectors containing EN1 or EN2 fragments. Indeed, VSX2 can significantly transactivate mouse and human EN1 and EN2 elements in 293T cells (Fig. 2B). Earlier work identified a point mutation (R200Q) in the VSX2 homeodomain that is associated with microphthalmia in humans (Ferda Percin et al., 2000), impairing the ability of VSX2 to bind DNA (Zou and Levine, 2012), and human retinal organoids carrying this mutation suffer from reduced retinal growth and loss of bipolar cells (Phillips et al., 2014). Transfection of Vsx2 harboring the R200Q point mutation led to a sharp decrease in the VSX2-dependent enhancer activities (Fig. 2B). We reasoned that VSX2 binds EN1 and EN2 through discrete motif sequences in its regulatory elements. Our earlier work defined a VSX2 consensus motif in EN2 (Norrie et al., 2019), but a recent study indicated that this motif is dispensable for Vsx2 expression (Honnell et al., 2022). Thus, we focused on EN1 and defined three sites with the homeodomain recognition motif (TAATTA) that overlap with the VSX2 ChIP-Seq peak (Fig. S1A). A minimal EN1 enhancer constituent (165 bp) containing motif 2 (CTCTAATTAG) lies at the peak center and is sufficient to drive luciferase activity in the presence of VSX2, whereas a deletion in this sequence diminished EN1 enhancer activity (Fig. S1B).

Given earlier studies suggesting that VSX2 can bind to homeodomain-containing transcription factors, we hypothesized that VSX2 may cooperate with other transcription factors expressed in RPCs to promote its expression (Mikkola et al., 2001; Wilson et al., 1993). One such factor is PAX6, which has been implicated in the regulation of Vsx2, as loss of Pax6 leads to a significant downregulation of Vsx2 in the mouse developing retina (Farhy et al., 2013). We first tested whether PAX6 alone can activate Vsx2 enhancers in luciferase assays and found no evidence for such a function (Fig. 2B). However, the introduction of both Vsx2 and Pax6 led to a multi-fold increase in EN1 and EN2 enhancer activities when compared with transfection of Vsx2 alone, underscoring a synergistic relationship between VSX2 and PAX6 on the Vsx2 enhancer (Fig. 2B). Moreover, transfection of Vsx2 harboring the R200Q point mutation, in combination with Pax6, led to a sharp decrease in VSX2-dependent EN1 and EN2 enhancer activities (Fig. 2B).

Reasoning that VSX2 and PAX6 interact, we performed co-immunoprecipitation (co-IP) assays in 293T cells and E14.5 retina, and found a robust physical association (Fig. 2C,D). Together, these data suggest that PAX6 functions as a modulator to promote VSX2-dependent transcriptional activation and support the presence of a transcriptional autoregulatory loop, whereby VSX2 governs its expression in RPCs, and likely bipolar cells, in association with transcription factors important for retinal development.

Next, we assessed the functional relevance of EN1 and EN2 in the mouse retina. We cloned EN1 and EN2 sequences into a vector (Stagia3) that contains a bicistronic reporter gene cassette (GFP-IRES-PLAP) driven by a minimal promoter (Billings et al., 2010). We then ex vivo electroporated wild-type E15 retina with EN1, EN2 or empty Stagia3 plasmids along with a CAG-mCherry vector as a control for electroporation efficiency (Emerson and Cepko, 2011). We cultured the retinae for 2 days and performed AP staining afterwards. Both EN1 and EN2 fragments produced robust AP activity, suggesting that they can function as regulatory enhancers in the developing mouse retina (EN1=8/8, EN2=8/8, Stagia3 empty vector=0/6; Fig. 2E, Fig. S1C). To experimentally test whether the Vsx2 regulatory elements EN1 and EN2 are functionally conserved in vivo, we cloned EN1 and EN2 with GFP as a reporter gene using the Tol2kit protocol (Kwan et al., 2007), and injected the constructs into zebrafish embryos at the one-cell stage. We found that the human and mouse EN1 exhibited enhancer activities in developing zebrafish eyes (37%, n=245 for mouse EN1; 59%, n=192 for human EN1; Fig. 2F and Fig. S1D-Q), while EN2 sequences lacked activity. In retinal sections from 72 h post-fertilization (hpf) zebrafish embryos, GFP expression was observed across retinal neurons, particularly in the inner nuclear layer (INL), while the presence of GFP-positive cells in the retinal outer nuclear layer (ONL) and retinal ganglion cells (RGCs) likely resulted from EN1 enhancer activities in early RPCs (Fig. 2F and Fig. S1P,Q). Importantly, EN1 activities were also detected in the proliferative domain (ciliary marginal zone; CMZ), consistent with the known expression of Vsx2 during zebrafish retinogenesis (Fig. 2F and Fig. S1P,Q) (Vitorino et al., 2009).

Based on these data and on our evolutionary analysis (see below), we decided to focus on testing the requirement of EN1 for Vsx2 expression and retinal development in vivo. We generated a mouse enhancer knockout that harbors EN1 deletion (termed VSX2-EN1 hereunder; Fig. 2A; Fig. S2A) using CRISPR/Cas9 technology (see Materials and Methods). Mice lacking the EN1 element exhibited a severe reduction in eye size at E14.5, recapitulating ocular defects associated with Vsx2 gene mutations (Fig. 3A) (Ferda Percin et al., 2000). To test whether EN1 is required for Vsx2 expression, we immunostained the developing retina at various stages (E14.5, P0, P4, P7 and P10) with VSX2 antibody and found a pronounced reduction in staining signal (Fig. S2B). To address whether loss of EN1 segment influenced retinal proliferation, we performed EdU labeling coupled with mitotic Ki67 immunostaining at E14.5 and found a strong reduction in both markers, particularly in the retina periphery (Fig. 3B). Moreover, the retinal proliferation defect is associated with the presence of excessive pigmented cells at the retinal periphery (Fig. 3C), consistent with deficiencies observed upon loss of Vsx2 function in Orj mice, and suggesting an essential role for EN1 enhancer in promoting retinal proliferation and suppressing RPE acquisition (Ferda Percin et al., 2000; Rowan et al., 2004). To better define the molecular defects resulting from loss of the EN1 enhancer constituent, we performed bulk RNA-Seq analysis on retinas at E14.5 (four replicas) and defined 1434 and 1577 transcripts that are downregulated and upregulated, respectively (>1.5-fold change; P-value <0.05) (Fig. 3D; Table S3). GO analysis of deregulated genes using the database for annotation, visualization and integrated discovery (DAVID) (Dennis et al., 2003) underscored a significant enrichment of biological functions related to neurogenesis and morphogenesis (Fig. S3A,B). As expected, Vsx2 was significantly downregulated along with genes related to retinal patterning (i.e. Vax2), proliferation (i.e. Ccnd1), differentiation (i.e. Ascl1 and Ptf1a) and signaling (i.e. Hhip, Gli1 and Dll3) (Fig. 3D). Genes that were significantly upregulated include Mitf (hence, pigmented cell expansion), Tfec and Cdkn1c, all of which have been reported previously to be robustly induced upon loss of Vsx2 function (Fig. 3D; Table S3) (Rowan et al., 2004).

We then investigated genetic programs that are directly regulated by VSX2 at E14.5. We incorporated RNA-Seq data from the VSX2-EN1 enhancer knockout with the VSX2 genomic profile at E14.5, revealing that 33.8% (1019/3016) of deregulated genes are associated with VSX2 occupancy, with a slight preference towards downregulated transcripts [(37% (535/1434) versus 30.6% (483/1577) of upregulated genes] (Fig. 3E-H; Table S3). Overall, these results demonstrate that the EN1 enhancer element is essential for Vsx2 expression, retinal proliferation and the proper expression of RPC neural differentiation programs.

We previously generated a large deletion in the Vsx2 regulatory landscape (termed VSX2-SE hereafter; Fig. 2A; Fig. S2A) that exhibits a complete loss of bipolar cells (Fig. 4A) (Norrie et al., 2019). This deletion includes elements important for bipolar cell differentiation (Goodson et al., 2020a) as well as EN2, but excludes the EN1 enhancer region (Fig. 2A). Unlike VSX2-EN1 enhancer knockout, loss of VSX2-SE did not affect eye size during embryogenesis or adulthood, suggesting that RPC proliferation was not impacted (Fig. S3C) (Norrie et al., 2019). To test this directly, we performed EdU labeling along with Ki67 immunostaining at E14.5 VSX2-SE retinae and found no evidence for aberrations in EdU incorporation or Ki67 staining (Fig. 3I). Still, VSX2 immunolabeling of the developing retina at stages E14.5, P0, P4, P7 and P10 suggests that VSX2 expression was impacted (Fig. S2B). Western blotting indicated a ∼35% reduction in VSX2 protein levels in the developing VSX2-SE^−/− retinae at E14.5 when compared with a ∼90% drop in the adult retina, suggesting that Vsx2 expression is moderately affected in RPCs but not as strongly as in the VSX2-EN1 knockout retinae (Fig. 3J). To corroborate these results and assess transcriptional programs impacted in E14.5 VSX2-SE retinae, we performed bulk RNA-Seq (three replicas). Limited changes in gene expression in comparison with VSX2-EN1 knockout were detected (59 and 151 transcripts were downregulated and upregulated, respectively), mostly impacting genes associated with generic metabolic responses and protein biogenesis functions (Table S4). Nonetheless, some interesting genes were affected, including Mitf, a known target for Vsx2 repression, and Pou2f1, a transcription factor implicated in cone photoreceptor genesis, both of which are occupied by VSX2 (Fig. 3H,K) (Javed et al., 2020). Vsx2 was downregulated by around 40%, yet this was insufficient to impact the expression of proliferation and cell cycle genes, consistent with EdU incorporation data (Fig. 3N; Fig. S3D and Table S4). We also compared the retinal transcriptomes of VSX2-SE and VSX2-EN1 knockouts at E14.5 and found little overlap (38 genes in total; Fig. 3L,M; Table S4). Notably, RNA-Seq data indicated that whereas the increase in Mitf expression was 15-fold in VSX2-EN1 deficient retinae, VSX2-SE retinae exhibit only a twofold change; hence, no obvious pigmentation phenotype was observed in VSX2-SE knockout retina (Fig. 3N,O). Together, these data underscore the differential requirements for Vsx2 enhancer constituents in its expression and suggest highly specialized non-redundant functions of the Vsx2 enhancer structure during retinogenesis.

We then examined the effect of deleting Vsx2 enhancer elements on retinal cell fate acquisitions. We have previously shown that VSX2-SE knockout completely lost retinal bipolar cells, as indicated by the absence of OTX2 in the inner nuclear layer of the adult retina (Fig. 4A) (Norrie et al., 2019). However, immunostaining against VSX2 revealed the presence of a residual expression in VSX2-SE^−/− retinae that completely overlapped with the Müller glia marker SOX9, demonstrating that this enhancer element is dispensable for Vsx2 expression in Müller glia (Fig. 4A).

Previous work has suggested that Vsx2 inhibits photoreceptor differentiation; our data indicate widespread occupancy of VSX2 on rod and cone genes (Fig. 1H,J) (Livne-Bar et al., 2006). Thus, we postulated that loss of Vsx2 in the VSX2-SE^−/− developing retina leads to alterations in cell fate, particularly toward photoreceptor production. To test this prediction, we stained VSX2-SE adult retinae with different cell-type markers and quantified the number of each cell type in comparison with wild-type retinae. Indeed, results indicated a significant increase in the proportions of photoreceptors (rods and cones) as well as Müller glia in the VSX2-SE-deficient retina compared with the wild-type retina (Fig. 4B). At the molecular level, we performed western blot analyses for CRX, a transcription factor expressed in photoreceptors, in the adult VSX2-SE^−/− retinae and detected a significant increase in its protein levels when compared with heterozygous and wild-type littermates (Fig. S4B). Thus, although the VSX2-SE element is dispensable for retinal proliferation, it is required for proper cell fate specification and differentiation.

We next analyzed cell fate in adult VSX2-EN1-deficient retinae. Whereas the retina is microphthalmic and the optic nerve is underdeveloped (Fig. S4C), immunostaining with cell-type markers indicated that all major retinal cell types were born, including bipolar cells (Fig. 4A). We confirmed the presence of bipolar cells by immunostaining the retina with several bipolar cell markers (PKCa, OTX2 and VSX2), albeit their numbers are significantly reduced (Fig. 4C,D). Overall, our functional analysis of the Vsx2 enhancer landscape revealed distinctive roles for Vsx2 enhancer components in promoting proliferation and cell fate specification during retinal development.

The absence of retinal bipolar cells from retinae lacking Vsx2 enhancer elements motivated us to quantitatively assess defects in retinal thickness. We performed optical coherence tomography (OCT) imaging on live adult VSX2-SE^−/− mice and their age-matched heterozygous and wild-type littermates. We found that, overall, the retina was about 20% thinner in VSX2-SE^−/− animals, which is attributed to the significant decrease in the INL and inner plexiform layer (IPL) thickness (Fig. 5A-D). The ONL was slightly but significantly thicker, in agreement with the increase in photoreceptor production in retinae lacking the VSX2-SE element (Fig. 5E).

Visual acuity tests of VSX2-SE^−/− mice indicated that they are blind (Norrie et al., 2019); however, how the physiological functions of the retina were impaired by the loss of the Vsx2 super-enhancer remains unclear. To assess this, we performed electroretinography (ERG) examinations on adult VSX2-SE^−/− animals and their age-matched heterozygous and wild-type littermates, and found that the b-wave (produced mainly by retinal bipolar cells) was completely absent in the VSX2-SE^−/− mice, whereas the a-wave signal (generated by photoreceptors in response to light) remained intact (Fig. 5F-I). In fact, at an elevated light intensity, the response of photoreceptors became slightly stronger in VSX2-SE^−/− animals (Fig. 5G). These data indicate that photoreceptor cells are functional in the VSX2-SE^−/− retinae, whereas the absence of bipolar cells prevents the propagation of neuronal signals generated by photoreceptors to other retinal neurons.

To further delineate potential molecular mechanisms underlying the overproduction of rod photoreceptors observed in the VSX2-SE^−/− retina, we focused our attention on early events of rod versus bipolar cell fate choice. Previous data suggested that retinal bipolar cells and rod photoreceptors share common precursors: the OTX2-expressing cells (Brzezinski and Reh, 2015; Fossat et al., 2007). In a subset of OTX2-positive precursor cells in the postnatal retina, OTX2 positively regulates the expression of either Vsx2 or Prdm1, promoting bipolar or rod cell fate, respectively (Katoh et al., 2010; Wang et al., 2014; West and Cepko, 2022). Within this context, overexpression of Prdm1 promotes rod formation by inhibiting Vsx2 expression (Brzezinski et al., 2010; Goodson et al., 2020a). Whether VSX2 represses Prdm1 expression remains unclear (Goodson et al., 2020a; West and Cepko, 2022). In WT developing retina, Prdm1 expression peaks at P4 and is downregulated thereafter (Fig. 6A; Brzezinski et al., 2010). Consistent with the overproduction of photoreceptors in the VSX2-SE^−/− adult retinae, we detected an increase in PRDM1 levels as early as P10 (Fig. 6A-C).

To explore this observation on a deeper level, we reasoned that VSX2 might directly bind and repress Prdm1 regulatory elements. We first profiled the Prdm1 enhancer landscape during retinogenesis using our previously published ATAC-Seq data (Aldiri et al., 2017) and found several peaks correlating with its expression (Fig. 6D). ChIP-Seq analysis of VSX2 in the adult murine retinae revealed that VSX2 occupies regulatory elements near to Prdm1, including an enhancer region that is necessary and sufficient for the retinal expression of Prdm1 (Fig. 6D) (Mills et al., 2017; Wang et al., 2014). VSX2 occupancy on this Prdm1 regulatory element is observed in the adult retina, when Prdm1 expression has subsided, but not in the E14.5 retina, when the expression is active (Fig. 6D; (Brzezinski et al., 2010)).

We then tested whether VSX2 can directly suppress Prdm1 enhancer activities using luciferase assays (Fig. 6E). Although data hinted at a repressive activity by VSX2, it was not statistically significant (Fig. 6E). Earlier studies have shown that OTX2 is required for Prdm1 expression, an effect mediated by OTX2 binding to Prdm1 enhancer elements (Mills et al., 2017; Wang et al., 2014). By investigating ChIP-Seq data of OTX2 (Samuel et al., 2014) and VSX2 (this article) we found that OTX2 and VSX2 co-occupy the Prdm1 regulatory region that is important for its expression in mouse and human adult retinae (Fig. 6D). Transfection of Otx2 led to a significant increase in Prdm1 enhancer activities in luciferase assays, consistent with its role in activating Prdm1 expression (Fig. 6E) (Wang et al., 2014). However, co-transfection of Vsx2 and Otx2 led to a significant decrease in the ability of OTX2 to activate the Prdm1 enhancer, suggesting that VSX2 can repress OTX2-mediated activation of Prdm1 (Fig. 6E). Finally, we tested whether OTX2 and VSX2 physically interact. Co-IP experiments performed at P7 and in the adult mouse retina confirmed this hypothesis (Fig. 6F). Taken together, these data underscore the involvement of VSX2 in positive and negative transcriptional regulatory loops, whereby VSX2 governs the activities of its enhancer, as well as regulatory elements of genes driving proliferation and cell-type acquisitions, in association with known transcription factors important for retinal development.

Recent data indicate that conserved enhancer elements that regulate genes crucial for animal survival and function are usually under evolutionary constraints to limit non-coding mutations that are detrimental to the expression of their respective genes (Zhen and Andolfatto, 2012). A recent study on subterranean moles, in which vision is largely non-essential, found a statistically significant (accelerated) mutation rate in non-coding sequences near to genes essential for vision, thus providing an indirect assay to measure the functionality of non-coding regulatory DNA associated with retinal genes based on sequence divergence (Partha et al., 2017). With this in mind, we used this evolutionary approach to further understand the functional importance of the VSX2-bound regions in the Vsx2 enhancer landscape. We scanned the mole acceleration profile (Partha et al., 2017) in the Vsx2 genomic landscape and discovered several significantly divergent sequences, most of which lie within the Vsx2 super-enhancer, suggesting that evolutionary constraints that safeguard Vsx2 enhancer sequences have been relaxed in subterranean moles (Fig. 7A). The only exception was the EN2 region, where no evidence for sequence deterioration was observed (Fig. 7A).

By examining the mole acceleration data, we also noticed the presence of a Vsx2 intronic domain that remained highly conserved and occupied by CTCF, a highly conserved transcription factor with ubiquitous expression and diverse roles in eukaryotes (Fig. 7A; blue bar) (Kim et al., 2015). Importantly, CTCF regulates the 3D chromatin architecture, in part by establishing enhancer-promoter contacts, suggesting a possible regulatory role within the Vsx2 enhancer landscape (Daghsni and Aldiri, 2021; Merkenschlager and Nora, 2016). To test the functional significance of this CTCF-bound region, we deleted this CTCF-binding site using CRISPR/Cas9 technology (Fig. S4C), and analyzed Vsx2 expression and retinal structure and function in two independent mouse sublines that carry the deletion. Our data indicated that this CTCF-bound site was completely dispensable for Vsx2 expression and overall retinal development, as no evidence for a change in Vsx2 expression levels, bipolar cell genesis or retinal morphology was found (Fig. S4D-F). Furthermore, ERG analysis indicated that the Vsx2-CTCF^−/− retina remained fully functional (Fig. S4G,H). Thus, although this CTCF-bound region is highly conserved, its functional significance remains unknown.

To further test whether the Vsx2 enhancer autoregulatory module is conserved in humans, we performed ChIP-Seq for VSX2 on adult human retinae, identifying 14,000 peaks (Fig. 7B-D; Table S5). Motif analysis of sequences in peak summits was consistent with the mouse data in which Q50 homeodomain-containing transcription factors were over-represented (Fig. 7C). The VSX2 genome-wide profile in the human retina also revealed extensive binding of photoreceptor and bipolar genes [58.3% (81/139) for rods, 137/255 (53.7%) cones and 48/102 (47.1%) for bipolar gene signatures], in agreement with mouse VSX2 ChIP-Seq data (Table S5). Examples of genes bound by VSX2 are NEUROD4 (bipolar), CRX (photoreceptors), THRB (cones) and PRDM1 (rod precursors) Fig. 7D). Importantly, VSX2 ChIP-Seq in the human retina indicated that the VSX2-binding profile near to the VSX2 locus is conserved (Fig. 7D), with VSX2 peaks (except EN2) exhibiting significant deterioration in mole (Fig. 7A).

Transcriptional enhancers function as platforms for recruiting transcription factors to mediate spatial and temporal regulation of gene expression (Buecker and Wysocka, 2012). Vsx2 plays a crucial role during retinogenesis, controlling multiple levels of retinal proliferation and differentiation, but much remains unknown about how it exerts its function and influences the retinal genome. Our genome-wide profiling of VSX2 occupancy during retinal development provides insight into the gene networks regulated by VSX2, and functional investigations of its regulatory elements reveal distinct regulatory modules that operate at the chromatin level. One of these loops involves a synergistic association with PAX6 in the proliferative retina to positively regulate Vsx2 expression itself. The key importance of this coordination is evidenced by genetic studies in humans and mice, illuminating analogous proliferation defects associated with VSX2 and PAX6 gene-coding mutations (Ferda Percin et al., 2000; Slavotinek, 2011).

Whereas our data point towards the involvement of an autoregulatory module in the maintenance and/or amplification of Vsx2 expression, the mechanisms by which Vsx2 expression is initiated in the developing retina remain unclear. Additionally, our luciferase assays indicate that the Vsx2 R200Q mutation, although showing a significant reduction in its ability to transactivate Vsx2 enhancers, retains significant residual activity when combined with Pax6 (Fig. 2B). This is consistent with in vivo evidence demonstrating that mouse lines bearing the R200Q mutation still express Vsx2 and underscores the complexity of Vsx2 regulation during retinogenesis (Zou and Levine, 2012).

We speculate that the coordination between VSX2 and PAX6 in RPCs is not limited to Vsx2 regulatory elements, that it may involve other transcription factors and chromatin regulating proteins, and that this coordination is likely to exist in a group of proliferation-associated genes. Furthermore, the interaction between PAX6 and VSX2 is limited to RPCs, given that the expression of Pax6 and Vsx2 becomes mutually exclusive in the adult retina.

One of the noteworthy features of the Vsx2 enhancer is that it remains active and is extensively bound by VSX2 in retinal bipolar cells; current evidence implicates OTX2 in regulating Vsx2 expression in the postmitotic retina (Goodson et al., 2020a; Kim et al., 2008; West and Cepko, 2022). Indeed, our co-IP results revealed a robust physical interaction in the adult retina. Thus, it will be interesting to functionally examine whether VSX2 and OTX2 act to regulate elements of the Vsx2 enhancer landscape and other factors required for retinal bipolar differentiation.

Vsx2 expression in Müller glia in the VSX2-SE^−/− retina has not been lost, mirroring its status in RPCs and underscoring shared properties between RPCs and Müller glia (Jadhav et al., 2009). We propose a model in which the Vsx2 regulatory landscape contains elements that function in a complementary non-redundant manner to promote robust Vsx2 expression and ensure proper progression of retinal proliferation and differentiation. Interestingly, preliminary evidence suggests that Vsx2 expression in the developing ventral spinal cord is maintained in our Vsx2 enhancer knockouts, adding a layer of complexity to the regulation of Vsx2 during neurogenesis (Fig. S4I-K). It is likely that Vsx2 expression in RPCs, Müller glia and bipolar cells can be temporally and spatially uncoupled, thus allowing us to predict the presence of additional regulatory elements that are essential for Vsx2 expression.

As this article was in preparation, a report described several deletions in the mouse Vsx2 non-coding regulatory elements, including VSX2-EN1 region (Honnell et al., 2022). Overall, the analysis reported by Honnell et al. is consistent with ours, showing a strong effect on retinal proliferation upon loss of VSX2-EN1 element. Interestingly, although a deletion in VSX2-EN2 seems to have a mild phenotype in retinal development, targeting both EN1 and EN2 leads to microphthalmia, with eye size smaller than that observed after single deletions, suggesting an additive effect of both enhancer constituents (Honnell et al., 2022). Honnell and co-authors also confirmed previous work demonstrating the presence of a regulatory element 17 kb from Vsx2 that is required for the bipolar cell genesis (Goodson et al., 2020a); our ChIP-Seq data indicate that this region is occupied by VSX2 in mouse and human retina (Figs 2A and 7A). Taking all these studies into account, we conclude that the Vsx2 enhancer landscape contains elements that function prominently during specific developmental windows to control aspects of retinal proliferation and bipolar cell differentiation.

The VSX2 binding of non-bipolar cell signature genes is in agreement with the hypothesis that VSX2 maintains retinal bipolar cell identity by suppressing inappropriate temporal and spatial transcriptional programs, particularly those related to rod photoreceptors (Clark et al., 2008; Dorval et al., 2006; Livne-Bar et al., 2006; Murphy et al., 2019). Our motif analysis of VSX2 peaks indicates a significant enrichment of Q50 motifs, which are favorably recognized by homeodomain transcription factors with known repressive and activating functions, including VSX2 (Clark et al., 2008; Green et al., 2003; Hughes et al., 2018; Murphy et al., 2019). The current model suggests that Q50 sequences bound by VSX2 are predominantly associated with repressive functions, primarily targeting rod genes (Murphy et al., 2019). Thus, it is likely that the analysis reflects the widespread occupancy of VSX2 on photoreceptor-related regulatory elements in bipolar cells. The transcriptional regulation by Q50 proteins can be altered by binding co-factors (Slattery et al., 2011). As such, VSX2 can promote the expression of bipolar genes and the suppression of rod photoreceptors in bipolar cells based on selective recruitments of co-activators or co-repressors.

Notably, Prdm1 is a target of VSX2 in mouse and human adult retina, and its enhancer activities mediated by OTX2 can be suppressed by VSX2 in vitro. Our data complement a proposed gene regulatory network that involves VSX2, OTX2 and PRDM1, whereby VSX2 is suggested to suppress rod differentiation by inhibiting Prdm1 expression during the bipolar cell specification (Goodson et al., 2020a; Wang et al., 2014; West and Cepko, 2022). Consistent with this model, studies indicate that VSX2 can function as a transcriptional repressor and earlier work has demonstrated that loss of Vsx2 in the postmitotic retina leads to an increase in photoreceptor cells (Goodson et al., 2020a; Livne-Bar et al., 2006). Although our data further support this modulatory regulatory network, more work is required to explore whether VSX2 directly suppresses Prdm1 expression in bipolar cell precursors in vivo.

Our evolutionary analysis involving mole acceleration data of the Vsx2 regulatory landscape provides support to our functional interrogation. Murine Vsx2 enhancer deletions generated by our group and others indicate that non-regulatory elements with a significant mutation rate in the mole are functionally important for retinal development (Fig. 7A) (Goodson et al., 2020a; Honnell et al., 2022; Norrie et al., 2019). No significant changes in mutation rate were detected in the EN2 region, which is in agreement with results from the knockout mice showing only a mild effect on retinal development upon loss of EN2 element (Honnell et al., 2022). On the other hand, removing a highly conserved CTCF binding site in the Vsx2 intronic region had no obvious effect on retinal development, implying that this site is not involved in Vsx2 promoter-enhancer contacts or that those contacts are dispensable for Vsx2 expression. The mole acceleration data predict the presence of additional sites that might be functionally relevant (Fig. 7A), but whether these sites are biologically important remains to be tested. Overall, our analysis provides the basis for an in-depth understanding of the gene regulatory network controlled by VSX2, revealing the diverse roles it plays during retinogenesis, and underscoring how the chromatin landscape influences essential aspects of retinal cell fate determination.

All mice were maintained in accordance with the guidelines set forth by the Institutional Animal Care and Use Committee of the University of Pittsburgh. C57 BL/6J animals were purchased from Jackson Laboratory.

B6D2 (C57 BL/6J×DBA/2J F1) females (6-8 weeks old) were stimulated by administration of 5 IU of PMSG through intraperitoneal injection at 3:30 pm on day 1, 5 IU of hCG 48 h after PMSG injection, then mated to B6D2 F1 stud males. Embryos were collected from oviducts the following morning and were cultured in KSOM medium in 5% CO₂ at 37°C until electroporation. Embryos were electroporated with 100 ng of Cas9 protein and 200 ng/μl of gRNAs each in an Opti-MEM medium. gRNAs were tested prior to embryo injection for activity in mouse N2A cells using targeted next-generation sequencing as previously described (Sentmanat et al., 2018). Electroporation was performed using the Super electroporator NEPA21 type II and CUY 501-1-1.5 electrode (NEPA GENE) under the following conditions: poring pulse (voltage 40 V, pulse length 2.5 ms, pulse interval 50 ms, number of pulses 4, decay rate 10%, polarity +); and transfer pulse (voltage 7 V, pulse length 50 ms, pulse interval 50 ms, number of pulses 5, decay rate 40%, polarity +/=). After electroporation, the embryos were washed twice in KSOM medium, then cultured in KSOM medium overnight in 5% CO₂ at 37°C. On the following day, the two-cell stage embryos were transferred to the oviducts of pseudopregnant CD1 females (0.5 dpc). The resulting founder animals (P0) were genotyped using PCR and Illumina Mi-Seq to identify mice with desired deletions. Adult animals positive for deletion were backcrossed to C57 BL/6J to generate wild-type, heterozygous and homozygous animals. Mouse genotyping was determined by PCR and then confirmed with Sanger sequencing. The following guides were used to generate the knockout lines: VSX2-SE-KO 5′ guide, 5′-GCAGGCCATGTGCTCGTCGANGG-3′; VSX2-SE-KO 3′ guide, 5′-CAGGGTGCAGGCTGACAACGNGG-3′; VSX2-EN1-KO 5′ guide, 5′-GTTAGACCTAGTCAGAACTCNGG-3′; VSX2-EN1-KO 3′ guide, 5′-TACTTCAGACTCTGGCTCCANGG-3′; VSX2-CTCF-KO 5′ guide, 5′-GAGTCTCTAAAAGTAGGAAANGG-3′; VSX2-CTCF-KO 3′ guide, 5′-ACCTAATCAAGCAATCAAGTNGG-3′. Details of the VSX2, VSX2-EN1 and VSX2-CTCF primers are provided in Table S6.

pCS2FA-transposase was used for in vitro transcription of capped Tol2 transposase mRNA. Selected enhancers were PCR amplified and cloned in the pDestTol2pA2 vector, upstream of the eGFP. pCS2FA plasmid DNA was digested with Not1 and purified with Qiagen kit. Linearized DNA was used as a template for in vitro transcription using a mMACHINE SP6 Transcription Kit. Purified mRNA was adjusted to a concentration of 250 ng/μl and aliquoted for microinjection. The following primers were used in cloning Vsx2 mouse and human enhancer fragments (restriction enzyme sites are underlined): mVsx2_En1_BamHI forward, TAAAGGATCCCCTTCCCACCTTTTGTCTGGTA; mVsx2_En1_BamHI reverse, GAAAGGATCCTGTGCTCGTCGAGGGTAGACT; mVsx2_En2_BamHI forward, TAAAGGATCCCCCGGGCTTAGAGAGCATTT; mVsx2_En2_BamHI reverse, GAAAGGATCCCAGGCACTTCCCAAGGAGAG; hVsx2_En1_BamHI forward, TAAAGGATCCTTGTCTATGCCTAAAGAGGCCA; hVsx2_En1_BamHI reverse, GAAAGGATCCTTCTAGGGCAGAATTCACTCACG; hVsx2_En2_HindIII forward, TAAAAAGCTTGAGAAGCCTGAGGTTTAGCGATT; hVsx2_En2_HindIII reverse, GAAAAAGCTTCTCCACAAGCATGATCCTACTGA.

Constructs containing mouse and human VSX2-EN1 and VSX2-EN2 were injected along with transposase mRNA into one-cell stage AB wild-type zebrafish embryos. The injection volume was calibrated to 1 nl. 25 pg and 50 pg of transposase mRNA were injected per construct. Injected embryos were transferred to a petri dish and incubated at 28.5°C. After 24 h, the embryos were examined for EGFP expression under a fluorescent dissecting microscope (Axio Zoom V.16) to determine the stereotypic expression pattern conferred by the enhancer. The total number of embryos injected with the construct and the number of embryos with the stereotypical EGFP pattern were determined to calculate the frequency of the pattern. Injection experiments were repeated three times, into embryos derived from different AB parents, for biological replicates.

Embryos were dechorionated and imaged using a Leica M165FC stereoscope digital camera. Sectioning was carried out according to the protocol of Uribe and Gross (2007). Briefly, embryos were fixed in 4% paraformaldehyde in PBS (pH 7.4) for 24 h at 4°C. After fixation, the eyes were sucrose-protected and embedded in tissue-freezing medium (TBS, Waltham, MA, USA). The eyes were sectioned at 12 μm on a Leica CM1850 cryostat and the sections were collected on Superfrost/Plus slides (Thermo Fisher Scientific), dried and stored at−20°C.

cDNA of human and mouse Vsx2 and Pax6 were subcloned into pcDNA3, with CMV promoter replaced by EF1α. Mouse and human enhancer fragments were cloned in Nanluc plasmid containing minimal promoter pNL3.1[Nluc/minP] (Promega N1031) using standard cloning techniques. Deletion of VSX2 predicted binding sites was achieved by overlap extension PCR, which has been described before (Lee et al., 2004). Primer sequences are provided in Table S7.

HEK293T cells (ATCC CRL-3216) were cultured in DMEM (Thermo Fisher Scientific, 10569044)+10% FBS (Thermo Fisher Scientific, 16000036)+100 U/ml penicillin, 100 µg/ml streptomycin (Thermo Fisher Scientific, 10378016). Cells were plated in 24-well plates seeded at 30,000 cells per well and allowed to adhere overnight. For Vsx2 enhancer elements, 500 ng of Vsx2 and/or Pax6 constructs were co-transfected with 200 ng of Vsx2 enhancer vectors. For the Prdm1 enhancer, 100 ng of Otx2 and/or Vsx2 vectors were transfected with 200 ng of Prdm1 enhancer. In all cases, 10 ng of firefly luciferase plasmid (pGL4.53[luc2/PGK]; Promega E5011) were delivered.

Transfection was performed using Lipofectamine 3000 (Thermo Fisher Scientific, L3000008) in each well. Empty vectors co-transfected with each of the enhancer constructs and firefly luciferase plasmid (pGL4.53[luc2/PGK]) were used as a control. Cells were lysed and assayed 24 h after transfection using Nano-Glo Dual-Luciferase Reporter (NanoDLR) assay system (Promega N1610) according to the manufacturer's protocol. Luminescence measurements were taken as singlets using a SpectraMax M5 (Molecular Devices) plate reader. Experiments were performed in quadruplicate.

Retina samples were lysed in 100 μl RIPA buffer [1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 0.15 M NaCl, 50 mM Tris (pH 7.5), 1 mM EDTA] and protease inhibitor cocktail (Complete Mini-EDTA free, 11836170001, Roche). One retina from adult mouse or two or three retinas from pups were pooled for each lysate preparation. Protein concentrations were determined by BCA kit (23225, Thermo Fisher Scientific). 20 μg of total protein was loaded per well onto an SDS-PAGE gel and transferred to PVDF membrane. After blocking with 5% nonfat dried milk in TBS-Tween, the membranes were incubated with the appropriate primary antibodies. Immunoblots were quantified by the ImageJ program.

HEK-293T cells were transfected with DNA constructs using Lipofectamine 3000 (Invitrogen) in a six-well plate. Two days after transfection, cells were lysed with high-salt lysis buffer [10% glycerol, 50 mM Tris (pH 7.5), 300 mM NaCl+1% NP-40, 2 mM MgCl₂] with a protease inhibitor cocktail and rotated for 30 min at 4°C. An equal volume of no-salt lysis buffer [10% glycerol, 50 mM Tris (pH 7.5), 1% NP-40+2 mM MgCl₂] was added to cell lysates to reduce salt concentration. 500 μg of protein was incubated with the appropriate antibodies overnight. For endogenous immunoprecipitation, non-denatured native protein lysates from C57 BL/6J E14.5 or adult retinae were prepared using COIP buffer (PBS+10% glycerol +0.6% Triton x-100) with protease inhibitors. On average, 30 retinae from E14.5 or five retinae from adult mice were pooled for each lysate preparation. Equivalent volumes of cell lysates were incubated with the appropriate antibodies or IgG overnight. The next day, protein A/G magnetic beads (88802, Thermo Fisher Scientific) were added and incubated for 3 h, washed with low-salt lysis buffer [10% Glycerol, 50 mM Tris (pH 7.5), 150 mM NaCl+1% NP-40, 2 mM MgCl₂) five times. Beads were eluted with SDS/PAGE loading buffer and subjected to SDS/PAGE and western blotting. Different species of secondary antibodies were used for detection to avoid heavy chain contamination (∼50 kDa). Antibodies used in western blots and immunoprecipitation are listed in Table S8.

Electroretinograms (ERGs) were performed using the Celeris system (Diagnosys), as described by Liu et al. (2020). Before each test, 6-week-old mice were dark-adapted overnight. Mice were anesthetized by an intraperitoneal injection of ketamine/xylazine (80 mg/kg bw/7 mg/kg bw), and eyes were then treated with 1% tropicamide (Akorn) and 2.5% phenylephrine hydrochloride. Eyes were lubricated using a 0.3% hypromellose eye gel (Alcon) before the electrodes are applied. Scotopic ERG responses of dark-adapted eyes to seven flashes from 0.0001 cd·s/m² to 100 cd·s/m² were recorded and averaged from ten sweeps per flash intensity with inter-sweep intervals of 10 to 30 s. After exposure to 10 cd/m² illumination for 10 min, the photopic ERG responses were recorded from the light-adapted eyes in response to flashes from 0.01 cd·s/m² to 100 cd·s/m² in addition to a 10 cd/m² background light. Each group contained five mice. A-wave and B-wave amplitudes of each ERG recording were measured and averaged from five animals.

Noninvasive ultra-high-resolution spectral domain optical coherence tomography (SD-OCT) (Bioptigen) was performed for in vivo imaging of mouse retinae (Chen et al., 2018). Briefly, mice at 6 weeks of age were anesthetized using an intraperitoneal injection of ketamine/xylazine (80 mg/kg bw/7 mg/kg bw). Pupils were dilated with 1% tropicamide (Akorn) and 2.5% phenylephrine hydrochloride prior to SD-OCT imaging. Five frames of OCT images were acquired in the B-scan mode, and averaged for image presentation and analysis. The thickness of the INL, IPL, ONL and whole retina was measured along with the scanned SD-OCT image at six points from the nasal to temporal of the retina. Each group had five mice, and graphs of INL, IPL, ONL and whole retina thicknesses were plotted to obtain the mean±s.e.m. of the five animals.

Eyes were isolated and fixed in 4% PFA overnight at 4°C. After dissection, retinae were equilibrated in 5%, 10%, 20% and 40% sucrose in PBS, and then embedded in a mixture of 40% sucrose in PBS and OCT at a 1:1 volume ratio and frozen. Sections were cut at 17 µm. Retinae were washed three times with PBS with 0.1% Triton and incubated in block solution for 1 h at room temperature and then in primary antibody in block solution overnight at 4°C. The sections were washed three times in PBS and then incubated in secondary antibody for 1 h at room temperature. After washing, sections were incubated with 2 µg/ml Hoechst 33342 (Invitrogen H3570) for 10 min, washed with PBS, then mounted in ProLong Glass Antifade (Invitrogen P36980). Images were acquired with an Olympus Fluoview Confocal microscope (v 4.2). Antibodies used and their dilutions are provided in Table S9.

The EN1 and EN2 fragments were amplified from the mouse genomic DNA using the following primers (restriction enzyme sites are underlined): mVsx2_En1_Stagia3 forward, GCCCTCGAGCCTTCCCACCTTTTGTCTGGTA; mVsx2_En1_Stagia3 reverse, GCCGAATTCTGTGCTCGTCGAGGGTAGACT; mVsx2_En2_Stagia3 forward, GCCGAATTCCCCGGGCTTAGAGAGCATTT; mVsx2_En2_Stagia3 reverse, GCCGAATTCCAGGCACTTCCCAAGGAGAG. Then these fragments were cloned into the Stagia3 vector using restriction enzyme-based methods. The CAG-mCherry plasmid and Stagia3 vector were kind gifts from Dr da Silva's lab (University of Pittsburgh, PA, USA). The plasmids used for electroporation were prepared with a final concentration of 700 ng/μl per plasmid. For electroporation, five pulses at 25 V of 950 ms off and50 ms on were applied to the dissected E15 retinas. Retinas were then carefully transferred to the 12 well plates and cultured for 2 days. The electroporated retinas were then washed with PBS and fixed with 4% PFA for 30 min at room temperature. Before proceeding with AP staining, fixed retinas were checked for the presence of mCherry signal (electroporation efficiency control), as fluorescence intensity becomes very weak after AP staining. The retinas were incubated at 65°C for 1.5 h then with AP buffer [100 mM Trizma, 100 mM NaCl, 5 mM MgCl₂ (pH 9.5)] for 15 min at room temperature before being changed to developer buffer (one NBT/BCIP tablet dissolved in 10 ml ultrapure distilled H₂O). AP signal was checked after incubation for at least 20 min at room temperature. The reaction was stopped by washing once with PBS and images for stained retinas were taken using Leica M165 Stereo Microscope. The retinal sections were prepared as follows. The electroporated retinas were washed with PBS and fixed with 4% PFA for 20 min at room temperature, cryoprotected in 5%, 10%, 20% and 40% sucrose in PBS, embedded in a mixture of 40% sucrose in PBS and OCT at 1:1 volume ratio then frozen. Sections were cut at 14 µm and AP staining carried out as described above.

Click-iT EdU Cell Proliferation Kit (Thermo Fisher Scientific, C10337) was used. Timed pregnant mice (E14.5) were intraperitoneally injected with EdU with a dose of 50 mg/kg body weight. Eyes from the embryos were then collected after 1 h of injection and fixed in 4% PFA for 2-3 h at 4°C. EdU staining was performed on cryosectioned retinas following the manufacturer's protocol.

Eyes from 3-month-old littermates were isolated and fixed in 4% PFA overnight at 4°C. Retinal cryosections were prepared as described in the immunostaining protocol and stained with cell-specific markers: OTX2 for rod photoreceptors; cone arrestin for cone photoreceptors; PAX6 for amacrine cells; calbindin for horizontal cells; SOX9 for Müller glia; and RBPMS for retinal ganglion cells. Sections (three nonoverlapping sections per mouse) demarcating the central retina were identified by the presence of the optic nerve head and subsequently used in the counting. Areas 400 μm away from the optic nerve were used for cell number counts. Three mice per genotype were used.

Statistical analyses and the number of samples (n) are described in detail for each figure panel. A two-tailed unpaired Student's t-test was performed using Excel. All reported P-values were considered significant at P<0.05. Data are presented as mean±s.e.m.

Freshly dissected E14.5 retinas were placed in TRIReagent (Zymo Research R2050-1-50) and RNA was extracted according to Direct-zol RNA MiniPrep kit protocol (Zymo Research R2050). RNA integrity was determined using 4200 Agilent Tapestation. Five retinas were used for each biological replicate except for VSX2-EN1 mice, where 15 retinas were used (owing to their small size). Stranded total RNA libraries were prepared following the manufacturer's instructions (Illumina TruSeq Stranded RNA; Ribo-Zero) with the following modifications. Total RNA input was depleted for rRNA and fragmented. Random primers initiated first- and second-strand cDNA synthesis. Adenylation of 3′ ends was followed by adapter ligation and library amplification with indexing. The sequencing was run on a high-output 150-cycle flowcell with read length 2×75 using the Sequencing Platform NextSeq500.

Raw sequencing reads first went through the pipeline of quality control using Trimmomatic. After pre-processing, surviving reads were aligned to mouse reference genome mm10 by STAR aligner and quantified by read count using HTSeq tool.

Differential expression analyses comparing wild-type and knockout samples were performed by R package DESeq2. Significant changes in gene expression were defined by adjusted P-value lower than 0.05 (FDR=5%) and a fold change greater than 1.5. All the bioinformatics and biostatistical analysis were performed by default parameters.

In brief, retinae dissected from C57/BL6 mice were flash-frozen and stored at −80°C for subsequent treatment. Retinae were thawed on ice and incubated with 1% formaldehyde in PBS at room temperature for 15 min. Fixation was stopped by the addition of 0.125 M glycine (final). Chromatin was isolated by the addition of lysis buffer and was sheared to an average length of 300-500 bp. Sheared chromatin was precleared with protein G agarose beads (Invitrogen). Chromatin immunoprecipitation (ChIP) was performed on 30-40 μg of sheared chromatin using 6 μl of VSX2 antibody (Exalpha, X1180P).

Complexes were washed, eluted from the beads with SDS buffer, and subjected to RNase and proteinase K treatment to collect DNA. Crosslinking was reversed by incubation overnight at 65°C, and DNA was purified by phenol-chloroform extraction and ethanol precipitation. Prior to Illumina library preparation, qPCR was performed to confirm the significant enrichment of target genes. Two ChIP-Seq biological replicas were performed for each stage.

Eye globes procured post-mortem from two de-identified donors (a 46-year female and a 57-year-old male) without previous medical histories of ophthalmological conditions or interventions were used. The human retinae were dissected within less than 6 h of reported death time, cryopreserved and processed for ChIP protocol as described in the mouse section. The donated organs were obtained via the Center for Organ and Recovery & Education (Pittsburgh, PA, USA). The procurement protocol was approved by the University of Pittsburgh Committee for Oversight of Research and Clinical Training involving Decedents (CORID).

Illumina sequencing libraries were prepared from the ChIP and input DNAs by the standard consecutive enzymatic steps of end-polishing, dA-addition and adaptor ligation. Steps were performed on an automated system (Apollo 342, Wafergen Biosystems/Takara). After a final PCR amplification step, the resulting DNA libraries were quantified and sequenced on Illumina's NextSeq 500 (75 nt reads, single-end).

Sequencing reads were aligned to mouse (mm10) or human (hg38) reference genomes by Burrows-Wheeler Aligner (BWA) algorithm. Low-quality alignment and duplicate reads were removed by SAMtools. Standard MACS2 algorithm was used to call peaks, comparing matched treated and input samples, with a cutoff of P-value=1E-7. Peaks that were on the ENCODE blacklist of known false ChIP-Seq peaks were removed. Peaks called from multiple replicates were merged and only common peaks were used subsequently. Seqplots (Stempor and Ahringer, 2016) was used to generate heatmaps. Peaks were visualized by Integrative Genomics Viewer (IGV).

Differential binding events analysis between E14.5 and adult samples were statistically determined by DESeq2 (Love et al., 2014) using individual BAM files. The program normalizes the counts between samples using the median of ratio methods and calculates log2 fold-changes, shrunken-log2FC, P-value and adjusted P-value for each merged region. Genes associated with Vsx2 peaks were called if one or more peaks were present within 10 kb upstream or downstream of a given gene.

HOMER enrichment motif analysis was run with standard settings with repeat masking on. The 200 bp surrounding the summit of peaks were analyzed. GREAT analysis was carried out using mm10 (mouse) or hg38 (human) with default parameters.

We are grateful to Susana da Silva for sharing Stagia3 plasmids and the AP protocol, and to Nathan Clark for sharing the mole acceleration track. We acknowledge the Health Sciences Genomics Research Cores at the University of Pittsburgh and the Magee-Women's Research Institute's Genome Editing, Transgenic, and Virus (GETV) Core Facility. We thank the Genome Engineering and iPSC Center (GEiC) at Washington University in St Louis for gRNA validation services. This research was also supported in part by the University of Pittsburgh Center for Research Computing through the resources provided and by the National Institutes of Health (core grant P30 EY08098 to the Department of Ophthalmology).