Initiation of Otx2 expression in the developing mouse retina requires a unique enhancer and either Ascl1 or Neurog2 activity

The retina is a complex multi-layered tissue that comprises seven major cell types, six neuronal and one glial. These retinal cell types arise from a common pool of undifferentiated progenitors in an order relative to the timing of permanent cell cycle exit, that is, their birthdate (Turner and Cepko, 1987; Turner et al., 1990). The early-born cell types include ganglion cells, amacrines, horizontals and cone photoreceptors, whereas rod photoreceptors, bipolar cells and Müller glia are primarily born late in development (Carter-Dawson and LaVail, 1979; la Vail et al., 1991; Rapaport and Vietri, 1991; Wong and Rapaport, 2009; Young, 1985). This commitment of cell-type identities is a crucial early event that establishes unique gene expression networks that underlie cell type-specific functions. The gene regulatory networks that control retinal cell fate commitment events are only partially understood.

Transcription factors play a major role in entrenching cell-fate choices within the retina. The homeobox transcription factor Otx2 is highly involved in retinal neurogenesis. Otx2 expression begins around embryonic day (E) 12, correlating with the earliest birthdates of photoreceptors (Nishida et al., 2003). Otx2 becomes upregulated in a large subset of progenitors as they exit the cell cycle (Muranishi et al., 2011). These Otx2+ cells appear to give rise to five neuronal types: horizontal, amacrine and bipolar interneurons, and rod and cone photoreceptors (Baas et al., 2000; Brzezinski and Reh, 2015; Brzezinski et al., 2013; Emerson et al., 2013; Fossat et al., 2007; Nishida et al., 2003; Sato et al., 2007). Otx2 is quickly lost in amacrines and horizontals, but is maintained by photoreceptors and bipolar cells into adulthood (Emerson et al., 2013; Koike et al., 2007; Nishida et al., 2003). Loss-of-function studies show that Otx2 regulates fate choice in the retina. Otx2 mutants lack photoreceptors and bipolar cells, but instead generate excess amacrine interneurons (Ghinia Tegla et al., 2020; Nishida et al., 2003; Sato et al., 2007; Yamamoto et al., 2020). Thus, the decision to express Otx2 greatly influences the cell fate composition of the retina. Given that this decision occurs in the last cell cycle, the gene regulatory events upstream of Otx2 occur in retinal progenitors. Progenitors express a wide variety of transcription factors, including several from the basic helix-loop-helix (bHLH) class that control fate decisions throughout the central nervous system (Hatakeyama and Kageyama, 2004; Hatakeyama et al., 2001; Wang and Harris, 2005). Which of these transcription factors, if any, controls Otx2 activation during retinal development is unknown.

A major component of gene regulatory networks is noncoding cis-regulatory DNA sequences known as enhancers. Enhancers provide spatial and temporal specificity to gene expression programs throughout development (Benoist and Chambon, 1981; Gillies et al., 1983; Long et al., 2016). To reduce the number of potential Otx2 regulators, we searched for enhancers that coincide with Otx2 activation during retinal development. We previously used DNase I hypersensitivity site (DHS) sequencing to identify candidates within the mouse retina (Wilken et al., 2015). Three DHS sequence elements were able to drive reporter gene expression in OTX2+ retinal cells. Of note was the DHS-4 element. Compared with the other elements, DHS-4 expression was highest at postnatal day (P) 0 and became severely restricted by P7 (Wilken et al., 2015). Based on this pattern, we hypothesized that DHS-4 acts by transiently initiating Otx2 expression as progenitors exit the cell cycle. By determining which sequences were necessary for DHS-4 expression in the retina, we reasoned that the repertoire of transcription factors that activate Otx2 could be identified.

We tested the role of the DHS-4 enhancer by: (1) examining its expression pattern in retinal development; (2) dissecting its necessary sequence components; (3) testing whether it was required for OTX2 activation; and (4) identifying its upstream regulators. Consistent with our prediction, DHS-4 was transiently expressed during retinal development and became active as progenitors exited the cell cycle. We showed that DHS-4 is necessary for OTX2 expression and that this enhancer requires several bHLH transcription factor-binding sites for its activity. Using single cell RNA-sequencing, we identified the progenitor bHLH transcription factors Ascl1 and Neurog2 as potential upstream regulators of DHS-4 and Otx2. Loss-of-function analysis of Ascl1 and Neurog2 using a CRISPR/Cas9 approach showed that targeting these genes simultaneously, but not singly, reduced OTX2 expression. Our findings suggest a mechanism whereby bHLH transcription factors act redundantly or in a compensatory fashion at the DHS-4 enhancer to activate Otx2 expression as retinal progenitors exit the cell cycle. Interestingly, removing DHS-4 from the postnatal retina had only a modest effect on OTX2 expression. This suggests that there is also redundancy or compensation at the level of enhancer usage. Together, our findings show that the gene regulatory network that controls Otx2 activation is remarkably flexible and resistant to perturbation. Such regulatory network robustness may be necessary to ensure that the retina forms a sufficient number of OTX2+ rods, cones and bipolar cells.

Otx2 becomes activated in the terminal cell cycle during retinal development (Muranishi et al., 2011). We hypothesized that a single enhancer is required for Otx2 activation. Based on its early and potentially transient activity in the developing retina, we focused on the enhancer sequence identified as DHS-4 (Wilken et al., 2015). DHS-4 is a 1489 bp highly conserved region located ∼50 kb downstream of the Otx2 translational initiation codon within exon 3 (Fig. 1A). To examine the spatial and temporal expression pattern of DHS-4, we used a reporter assay in which the enhancer sequence was cloned adjacent to a minimal TATA promoter (Mills et al., 2017; Wilken et al., 2015). Together, these two elements drive expression of a nuclear-localized GFP cassette. To control for electroporation efficiency, we used a ubiquitously expressed EF1α nuclear mCherry (RFP) plasmid (Mills et al., 2017; Wilken et al., 2015). GFP and mCherry plasmids were co-electroporated into P0 mouse retinal explants and cultured for 2 days. Using immunohistochemistry, we observed numerous GFP+ cells that highly, yet incompletely, overlapped with OTX2 (Fig. 1B). This matched the previously observed DHS-4 pattern (Wilken et al., 2015).

We next investigated which sequences in the DHS-4 enhancer were crucial for its regulation. Given the relatively long (>1.4 kb) length of this sequence, we first sought to identify the smallest sub-element that recapitulated the DHS-4 expression pattern. To do this, we cloned sub-elements (Fig. 1C) of the parent DHS-4 sequence into the same TATA-GFP reporter backbone and electroporated P0 retinal explants. After 2 days, explants were stained for GFP, mCherry (RFP) and OTX2. Ubiquitous mCherry control plasmid electroporated cells co-expressed OTX2 32.6% of the time (Fig. 1D,E). Empty plasmid (TATA-only) controls showed similar co-expression with OTX2, but few GFP+ cells were seen (Fig. 1D,E). The DHS-4 parent construct had 67.4% overlap with OTX2 (Fig. 1B,E), significantly higher than TATA-only and ubiquitous mCherry controls (one-way ANOVA, P<0.001 and P<0.0001, respectively). Sub-elements DHS-4A and 4B bisected the 1489 bp region (Fig. 1C). Sub-element 4A had GFP reporter activity and maintained overlap with OTX2 expression that was similar to that in the DHS-4 parent, whereas sub-element 4B had weak nonspecific expression from only a few GFP+ cells (Fig. 1D,E).

We further dissected sub-element 4A and tested a sequence from the distal-most end, sub-element 4C (Fig. 1C). This highly overlapped with OTX2, but had weak reporter activity from only a few GFP+ cells (Fig. 1D,E). Next, we isolated the sequence of sub-element 4A that was not included in sub-element 4C and added adjacent sequence from 4B (Fig. 1C). This sub-element, designated 4D, displayed both robust expression and high overlap (61.6%) with OTX2, which was significantly higher than controls and indistinguishable from the DHS-4 parent (one-way ANOVA, P<0.0001 versus mCherry, P<0.05 versus TATA and P=0.1797 for DHS-4) (Fig. 1D,E). We then bisected the 4D sequence into D.1 and D.2 sub-elements (Fig. 1C). Both of these sub-elements were highly expressed, similar to sub-element 4D; however, intriguingly, they lost their specificity with OTX2, having only 32.6% and 25.9% overlap, respectively, and were indistinguishable from mCherry or TATA-only controls (one-way ANOVA, D.1 versus mCherry P=0.9852, D.1 versus TATA P=0.5468, D.2 versus mCherry P=0.1035, D.2 versus TATA P=0.8598) (Fig. 1C-E). Based on these findings, we considered 4D to represent the minimal enhancer sequence that recapitulated the expression pattern of the full-length DHS-4 sequence (Fig. 1F) (Wilken et al., 2015). The 518 bp DHS-4D sequence was used in downstream experiments.

The DHS-4 enhancer shows activity during embryonic and early postnatal retinal development, but never completely overlaps with OTX2 (Fig. 1) (Wilken et al., 2015). This raises the possibility that DHS-4 is made by proliferative cells as they activate Otx2 and that the stable GFP signal persists in some postmitotic cells (e.g. amacrines and horizontals) that subsequently lose OTX2 expression. To further understand the cells marked by the DHS-4D reporter plasmid, we examined additional markers of early-born cell fates. We quantified the overlap of DHS-4D with AP2B (horizontals and amacrines), ONECUT1 (cones, horizontals and ganglion cells) and RXRG (cones and ganglion cells) (Fig. S1) (Bassett et al., 2012; Emerson et al., 2013; Roberts et al., 2005; Sapkota et al., 2014; Wu et al., 2012, 2013). We identified modest overlap with each of these markers, consistent with transient marking of cones, amacrines and horizontals by DHS-4D (Fig. S1). No DHS-4D expression was seen in the ganglion cell layer, suggesting that DHS-4D does not label ganglion cells. Given that Otx2 expression is activated during the terminal cell cycle (Muranishi et al., 2011), we expected to see DHS-4D overlap with progenitor markers. P0 mouse retinas were electroporated, as before, with either a ubiquitously expressed control plasmid or the DHS-4D reporter and cultured for 2 days. Sections were stained with antibodies against Ki67, a marker associated with cell proliferation, and the overlap with GFP and mCherry was quantified (Fig. 2A). We found that DHS-4D overlapped with Ki67 30.8% of the time, consistent with the enhancer becoming active as progenitors upregulate OTX2 and exit the cell cycle (Fig. 2A). mCherry+ control cells were significantly less likely to co-express Ki67 (13%, unpaired t-test, P<0.0001) (Fig. 2A), further suggesting that DHS-4D is preferentially expressed during the terminal cell cycle. Next, we incorporated 5-ethynyl-2′-deoxyuridine (EdU) 30 min before collecting explants to mark S-phase cells. We saw few GFP+/EdU+ cells, suggesting that DHS-4D is primarily active following S-phase, similar to OTX2 (Fig. 2B) (Muranishi et al., 2011). There were no significant differences in EdU incorporation between GFP and mCherry control cells (Fig. 2B).

DHS-4 enhancer activity appeared to decline with age (Wilken et al., 2015). This suggests that this enhancer is only transiently expressed or marks a restricted cell population in the mature retina. To test this, we conducted in vivo retinal electroporations in newborn mice. We electroporated the DHS-4 and DHS-4D reporters and examined their expression patterns at P25. Neither the parent DHS-4 nor the DHS-4D sub-element displayed expression at P25 (Fig. S2A). To further track this loss of expression, we modified a version of our enhancer reporter plasmid to express both GFP and Cre recombinase. This plasmid was electroporated into newborn ROSA-RFP reporter mouse retinas (Madisen et al., 2010). In this lineage-tracing system, cells that have enhancer activity at any timepoint would permanently express RFP, whereas the cells currently expressing DHS-4D would also express GFP (Fig. S2B). Using this lineage-tracing strategy with DHS-4D, we found, as early as P7, that there were many RFP+ cells, but none was GFP+ (Fig. S2B). These RFP+ cells were primarily photoreceptors and bipolar cells (OTX2+), but we observed some amacrine cells and the occasional ganglion or Müller glial cell based upon marker co-expression, morphology and laminar position (Fig. S2C). This is consistent with Otx2 (and DHS-4D) being expressed by cells that can adopt an amacrine fate in addition to photoreceptors and bipolar cells that maintain OTX2 expression. Given that the lineage system is sensitive, this pattern may also reflect some nonspecific Cre recombinase activity. Taken together, our data argue that the DHS-4D enhancer is expressed in a transient fashion as progenitors exit the cell cycle and is not maintained in mature OTX2+ photoreceptors or bipolar cells.

Given its spatial and temporal pattern, we hypothesized that DHS-4D directly regulates Otx2 initiation during development. To test this, we used a CRISPR/Cas9 strategy to delete the DHS-4D sequence by nonhomologous end joining (Goodson et al., 2020a). We showed that this was effective at perturbing enhancers both ex vivo and in vivo (Goodson et al., 2020a). We acquired the PX458 plasmid, which drives both single guide (sg) RNA expression from a U6 promoter and Cas9 expression from a CMV promoter (Ran et al., 2013). We modified this plasmid to replace the CMV promoter with a ubiquitously expressed EF1α sequence that performs better in the retina (Goodson et al., 2020a; Ran et al., 2013; Semple-Rowland et al., 2010). In addition, we generated new plasmid variants with nuclear-localized versions of GFP and mCherry via fusion to histone H2B. This resulted in an all-in-one plasmid that expresses a targeting sgRNA, ubiquitous Cas9 and persistent nuclear reporters for cell tracking. Three guide plasmids were designed to remove the DHS-4D sequence (Fig. S3) and these were pooled and electroporated into retinal explants. Nontargeting guide plasmids were used as a negative control and guides targeting coding exons of Otx2 were used as positive controls. Plasmids were electroporated into E14.5 explants and cultured for 2 days. Histological sections were stained for OTX2 and fluorescent protein reporter overlap was quantified. Nontargeting guide plasmid (GFP+) overlapped with OTX2 26.9% of the time (Fig. 2C). Positive-control Otx2-targeting guides reduced this to ∼11.5%, whereas targeting the DHS-4D enhancer reduced the overlap to 14.9% (Fig. 2C). Both Otx2-coding and DHS-4D targeting guides significantly reduced overlap with OTX2 compared with nontargeting controls (unpaired t-tests, P<0.0001 and P<0.0001, respectively) (Fig. 2C). Targeting either the DHS-4D- or Otx2-coding sequences was equally efficient at reducing OTX2 expression (Fig. 2C), suggesting that the DHS-4D enhancer is necessary for OTX2 expression in the embryonic retina.

Given that the DHS-4D enhancer was necessary for embryonic OTX2 expression, we reasoned that its essential transcription factor-binding sites would predict the regulatory network that activates Otx2. A cursory examination of the DHS-4D sequences revealed several enhancer box (E-box) response elements (Fig. S4). The E-box-binding domain is characterized by the palindromic CANNTG motif sequence, which can act as a binding site for bHLH transcription factors (Murre et al., 1994). These transcription factors typically bind in a heterodimeric fashion to the E-box, facilitating transcription (Ellenberger et al., 1994; Henke et al., 2009). We identified a total of six canonical E-box sites within DHS-4D (Fig. 3A; Fig. S4).

To test the necessity of these E-boxes for DHS-4D expression, we systematically mutated the sites. To inactivate these sites with the least disruption of the overall DHS-4D sequence, we used a previously reported strategy in which the CANN′TG′ motif is mutated to CANN′AT′ (Fig. 3B). This was shown to prevent activity of bHLH factors because of a lack of DNA-binding specificity (Henke et al., 2009). We co-electroporated E-box mutated reporter constructs with ubiquitous mCherry control plasmids into E14.5 retinal explants and cultured them for 2 days (Fig. 3B). We first mutated single E-box sites, as numbered in Fig. 3A. Explants were stained for GFP, RFP and OTX2, as described above. We found that E-boxes 3, 4 and 6 were each essential for DHS-4 expression because these explants lacked GFP+ cells (Fig. 3C,D). In contrast, mutating E-boxes 1, 2 or 5 resulted in continued robust GFP expression and OTX2 overlap (Fig. 3C,D).

The six E-box sites were not evenly distributed in DHS-4D but formed two clusters (1-2 and 3-5) (Fig. 3A). To test whether there are interactions between the E-box sites, we mutated multiple E-boxes simultaneously and electroporated retinas as described above. This resulted in an additional ten mutant constructs (Fig. 3C). As expected, mutating all six E-box motifs to CANN′AT′ completely prevented enhancer activity (Fig. S5). Constructs that contained mutated E-boxes 3, 4 or 6 were always inactive, replicating the single mutagenesis experiments (Fig. S5). Surprisingly, a construct with the three nonessential sites mutated (1, 2 and 5) was also devoid of reporter activity (Fig. 3C; Fig. S5). This indicated that the essential E-box sites (3, 4 and 6) were not sufficient to drive enhancer expression, highlighting a cooperative requirement beyond the essential sites.

The factors that regulate DHS-4D expression are genetically upstream of Otx2. Our mutagenesis experiments argue that enhancer and Otx2 expression is regulated by bHLH transcription factors. We previously showed that disruption of Notch signaling through small molecule γ-secretase inhibitors in murine E14.5 retinal explants resulted in an increase in Otx2 and cone photoreceptor formation (Kaufman et al., 2019). Transcriptionally profiling these explants over time revealed that expression of several bHLH transcription factors, including Ascl1, Neurog2 and Olig2, preceded the onset of Otx2 expression (Kaufman et al., 2019). To improve our temporal resolution and bypass potential issues from perturbed development, we used a single cell RNA-sequencing approach to identify which bHLH genes are expressed immediately upstream of Otx2.

To gain temporal information upstream of Otx2 activation, we profiled progenitors as they became OTX2+ cells. E14.5 mouse retinas were electroporated with both a DHS-4D-GFP reporter and a ubiquitous mCherry control plasmid (Fig. 4A). Retinas were cultured for up to 2 days and then subjected to fluorescence-activated cell sorting (FACS). The DHS-4D-GFP+ population represented cells that had activated Otx2. Cells that only expressed mCherry represented Otx2− cells, such as progenitors and ganglion cells. To obtain a representation of developing cells along a progenitor to Otx2+ state, we mixed GFP+ (±mCherry) and mCherry-only cells to be 40% and 60% of the total, respectively (Fig. 4A). These cells were subjected to 10X Genomics Chromium single cell capture, labeling and high-throughput sequencing (Fig. 4A). We captured 1843 cells and our read depth was ∼250K per cell, allowing us to map potentially under-represented RNAs. Following quality control, we identified 13 clusters by Uniform Manifold Approximation and Projection (UMAP) dimension reduction using Leiden-based clustering in the Scanpy Python package (Fig. 4B) (McInnes et al., 2018; Wolf et al., 2018). Comparing DHS-4D-GFP and Otx2 mRNA expression on the UMAP representation showed that the GFP signal marks a subset of the Otx2+ cells, consistent with the transient expression pattern of the enhancer (Fig. 4B).

We next performed cluster analysis and used known markers to classify progenitors (Fgf15, Vsx2, Sox2, Pax6), photoreceptors (Otx2, Crx, Prdm1, Rxrg, Thrb, Pdc, Gnat2, Gngt2, Nrl), amacrines/horizontals (Ptf1a, Prdm13, Tfap2a, Tfap2b, Onecut1, Onecut2, Lhx1, Prox1, Ebf1) and ganglion cells (Isl1, Rbpms, Pou4f1, Pou4f2) manually (Fig. 4C). These cell-type annotations were then mapped back onto the UMAP cluster number they most represented (Fig. S6). Consistent with their embryonic age, the clusters represent progenitors (0, 1, 2, 5, 6, 11 and 13), cone photoreceptors (clusters 4 and 12), amacrines and horizontals (clusters 7 and 10) and ganglion cells (cluster 8) (Fig. S6). Clusters 3 and 9 did not fit definitively into a specific group, but were likely progenitors transitioning into a neuronal state (Fig. 4C) (Clark et al., 2019). Comparing these progenitor clusters with the previously classified populations described (Clark et al., 2019), we found that they fell into two broad groups. Clusters 0, 1 and 5 likely represent early/intermediate progenitors because of their relatively high levels of Fgf15, Vsx2 and Sox9. Clusters 2, 6 and 11 can be classified as neurogenic or late neurogenic progenitors based on their expression of Otx2, Ascl1, Neurog2 and Olig2 (Fig. S6). Differences in the stage of the cell cycle likely underlie the relatively large number of progenitor clusters we identified (Fig. S6). DHS-4D-GFP expression was most highly detectable in clusters 2, 3, 6 and 11 (Fig. 4C). This expression pattern overlapped with Otx2 and the clusters representing progenitors and other cells transitioning into cones, amacrines and horizontal interneurons (Fig. 4C).

Trajectory analysis to identify gene expression events along developmental pathways was performed using the diffusion pseudotime function in the Scanpy Python package (Haghverdi et al., 2016). Cells were aligned according to their pseudotime score and plotted in ForceAtlas2 component space, with the original Leiden clusters represented (Jacomy et al., 2014) (Fig. 4D). To identify factors upstream of Otx2, we isolated a pseudotime pathway that ended in the cone photoreceptor cluster (12), a state in which Otx2 becomes permanently maintained (Fig. 4D) (Koike et al., 2007; Nishida et al., 2003). This pathway includes clusters 2, 3, 6 and 11, which have the highest-detectable levels of DHS-4D-GFP (Fig. 4C). Expression of candidate bHLH factors and known markers was plotted along this trajectory (Fig. 4E). Recapitulating the pattern seen from reporter and lineage-tracing experiments (Figs 1, 2; Figs S1, S2), DHS-4D-GFP tracked a subset of cells in a transitory state overlapping both Sox2+ progenitors and Otx2+/Crx+ photoreceptors (Fig. 4E). Analysis of transcription factors overlapping in pseudotime with the DHS-4D-GFP+ cells revealed several bHLH candidates, including Neurod1, Bhlhe22, Ascl1, Neurog2 and Olig2 (Fig. 4E). These candidates were similar to those we observed upon inhibiting Notch signaling in E14.5 explants (Kaufman et al., 2019). The expression of the candidate bHLH factors over pseudotime was not equivalent (Fig. 4E). Some factors showed robust transient expression (Ascl1, Olig2 and Neurog2), others appeared to be upregulated late (Neurod1 and Neurod4) and Bhlhe22 had only modest transient upregulation (Fig. 4E). Based on their co-expression, timing of upregulation and their transient nature, we considered Ascl1, Neurog2 and Olig2 to be the most-likely regulators of DHS-4D and Otx2 activation.

Results from our mutagenesis and single cell RNA-sequencing experiments supported the hypothesis that bHLH transcription factors made by progenitors bind the E-boxes of the DHS-4D enhancer to activate Otx2 expression. This was also supported by strong JASPAR binding predictions at these sites (Fig. S7). Furthermore, we predicted that removing Ascl1, Neurog2 and/or Olig2 would prevent OTX2 expression, similar to CRISPR/Cas9 targeting of the DHS-4D enhancer sequence (Fig. 2C). To test this, we used our CRISPR/Cas9 approach to target the bHLH transcription factors. Guides were made to target early coding exons of each bHLH gene. These were electroporated into E14.5 explants and the retinas were screened after 2 days of culture (Fig. 5A). We first examined changes in bHLH factor expression compared with nontargeting guide controls. We found that targeting Ascl1, Olig2 and Neurog2 resulted in strong reductions in their protein expression compared with nontargeting control (unpaired t-tests, P<0.0001, P<0.0001 and P<0.0001, respectively) (Figs S8, S9). Similar to targeting Otx2 or DHS-4D (Fig. 2C), this did not reach 100% efficiency for Ascl1, Olig2 or Neurog2.

We then tested whether targeting these bHLH factors affected OTX2 expression. Singly targeting Ascl1, Neurog2 or Olig2 did not have a significant effect on OTX2 expression compared with nontargeting guides or previous Otx2 targeting (unpaired t-tests, P=0.853, P=0.164 and P=0.415 compared with nontargeting guides, respectively) (Fig. 2C, Fig. 5A; Fig. S9). This was consistent with the mild phenotypes previously reported from single mutations of these genes (Akagi et al., 2004; Brzezinski et al., 2011; Hafler et al., 2012; Hufnagel et al., 2010; Tomita et al., 1996). The DHS-4D element contains multiple E-boxes, none of which was sufficient on its own for enhancer activity (Fig. 3). Our single cell analysis showed that there were several bHLH factors expressed at the right time and place to regulate DHS-4D (Fig. 4). Thus, we reasoned that multiple bHLH factors can activate DHS-4D and Otx2 expression in a redundant or compensatory fashion. To test this hypothesis, we used additional reporter fluorescent proteins to visualize combinational effects of multiplexed gene targeting (Fig. 5A). These were then combined into a single pooled mixture and electroporated into E14.5 explants, an approach we previously showed to be highly efficient (Goodson et al., 2020a). The percentage of electroporated cells that co-expressed OTX2 was significantly reduced in Ascl1/Neurog2 double-targeted cells compared with nontargeting and single-targeted controls (P<0.0001) (Fig. 5B). This reduction was similar in magnitude to Otx2 and DHS-4D targeting (Fig. 2). We also targeted Ascl1, Neurog2 and Olig2 simultaneously, but we did not observe an additive effect on OTX2 expression (Fig. S9). This was unlikely due to dilution effects, because the triple-targeting condition was equally effective at reducing OLIG2 expression as singly targeting Olig2 (Fig. S9). To assess whether bHLH removal perturbed the cell cycle and indirectly influenced OTX2 expression, we quantified the double Ascl1/Neurog2-targeted electroporations with Ki67 and EdU labeling. No differences were found between controls and knockouts (unpaired t-test, Ki67 P=0.3577 and EdU P=0.9858) (Fig. S10). Next, we targeted Ascl1 and Neurog2 at the P0 timepoint in explants along with negative control and Otx2-targeting guides, and quantified retinas after 3 days of culture (Fig. 5C). Whereas both Otx2 and Ascl1/Neurog2 targeting significantly reduced OTX2 co-expression (unpaired t-tests, P<0.0001 and P<0.0001, respectively), double bHLH targeting only reduced overlap to 36.1% compared with 12.6% in direct Otx2 targeting (unpaired t-test, P=0.0001) (Fig. 5D).

Together, our findings suggest that Ascl1 and Neurog2 regulate Otx2 expression via the DHS-4D enhancer. Given that only targeting Ascl1 and Neurog2 simultaneously affected OTX2, our results highlight the possibility that compensatory or redundant mechanisms are at play. Interestingly, these effects were diminished at later developmental timepoints, raising the possibility that other enhancers or transcription factors can activate Otx2 postnatally.

Perturbing the DHS-4D sequence with CRISPR/Cas9 severely reduced OTX2 expression in E14.5 explants. We reasoned that, if the DHS-4D enhancer was essential for Otx2 activation throughout retinal development, then targeting DHS-4D at P0 should block OTX2 expression and change fate choice in the retina. Alternatively, if another enhancer can take the place of DHS-4D later during development, targeting DHS-4D would have a modest effect on OTX2. To discriminate between these possibilities, we conducted in vivo electroporation with our DHS-4D-targeting CRISPR/Cas9 plasmids in newborn mice and harvested eyes at P7 (Fig. 6A). As before (Fig. 2C), we electroporated guides against DHS-4D and Otx2, or used nontargeting guides as a negative control. The retinas from P7 eyes were immunostained for their fluorescent reporter and OTX2 (Fig. 6B). As with embryonic timepoints, Otx2-targeted cells co-expressed OTX2 at approximately one-third the rate of nontargeting guide controls (unpaired t-test, P<0.0001) (Fig. 6C). Many of the Otx2-targeted cells lacked OTX2 co-expression and resided in the inner portion of the inner nuclear layer, consistent with an amacrine fate. This fate change is consistent with the phenotype of conditional Otx2-mutant mice and targeted knockouts (Ghinia Tegla et al., 2020; Nishida et al., 2003; Sato et al., 2007; Yamamoto et al., 2020). Compared with controls, fewer DHS-4D-targeted cells co-expressed OTX2 (unpaired t-test, P<0.0001) and more electroporated cells were observed in the amacrine area of the inner retina (Fig. 6B,C). Nonetheless, targeting DHS-4D had a significantly weaker effect on OTX2 expression than did targeting the Otx2-coding region (one-way ANOVA, P<0.0001) (Fig. 6C). This suggests that another enhancer can contribute to Otx2 activation when DHS-4D is perturbed in the postnatal retina. A similar diminished effect on OTX2 expression was seen when both Ascl1 and Neurog2 were targeted in P0 explants cultured for 3 days (Fig. 5C). Correspondingly, simultaneously targeting Ascl1 and Neurog2 at P0 had no effect on OTX2 co-expression at P7 (Fig. 6B,C). Taken together, our findings suggest that additional enhancers and transcription factors activate or sustain OTX2 expression in the postnatal retina.

Retinal development is governed by complex gene regulatory networks that are only partially understood. To dissect the Otx2 gene regulatory network, we characterized the function of a minimal enhancer sequence, termed DHS-4D, and demonstrated that it recapitulated the pattern of Otx2 expression initiation during development. This enhancer was necessary for OTX2 expression and its activity depended on E-box bHLH transcription factor-binding sites. Correspondingly, perturbing the bHLH factors Ascl1 and Neurog2 simultaneously prevented OTX2 expression. Our data suggest that bHLH factors made by progenitors during their terminal cell cycle activate the DHS-4D enhancer to initiate Otx2 expression. Initiating Otx2 expression appears to have redundant and/or compensatory features, both in terms of which transcription factors are involved and the cis-regulatory enhancers used. The gene regulatory network controlling Otx2 is complex and flexible. This may make gene expression more robust to ensure that normal fate decisions occur during retinal development.

During retinal neurogenesis, Otx2 is activated by a large subset of progenitors in their terminal division (Fig. 7) (Muranishi et al., 2011). Most of these cells will maintain Otx2 expression and become photoreceptors and bipolar cells (Fig. 7B). A smaller fraction will lose Otx2 expression and adopt amacrine and horizontal fates (Fig. 7B). The DHS-4D enhancer was enriched in OTX2+ cells, but showed transient activity in the retina (Fig. 7A). At embryonic timepoints, the enhancer marked OTX2+ cells (e.g. cones), progenitors, amacrines and horizontal cells (Fig. S1). Our postnatal DHS-4D lineage tracing showed that this enhancer drives transient expression, but contributes to photoreceptor, bipolar and amacrine cell fates (Fig. S2). This transient expression was also seen in our single cell RNA-sequencing analysis, in which DHS-4D-driven GFP mRNA was expressed by progenitors as they activated Otx2 and was quickly downregulated as cells became photoreceptors (Fig. 4). Our findings suggest that the DHS-4D enhancer acts by initiating Otx2 expression as progenitors exit the cell cycle (Fig. 7B). This initiator role was supported by the requirement for DHS-4D during early development and by the necessity for Ascl1 and Neurog2 for OTX2 expression. These two bHLH factors are likely to act only as Otx2 initiators because their expression tapers off sharply as progenitors become postmitotic cells (Fig. 4C) (Brzezinski et al., 2011; Hufnagel et al., 2010; Maurer et al., 2018).

We reduced the DHS-4 enhancer down to a 518 bp DHS-4D sequence, which retained the spatial and temporal features of the parent element. Conspicuous within the DHS-4D sequence were six conserved E-box motifs. E-boxes 3, 4 and 6 were required for DHS-4D activity, arguing that bHLH transcription factors are necessary for enhancer and Otx2 expression. CRISPR/Cas9 targeting of Ascl1 and Neurog2 reduced OTX2 expression, strongly implying that these two factors bind one or more E-boxes within the DHS-4D enhancer. This is supported by recently published ASCL1 chromatin immunoprecipitation (ChIP) sequencing data from P0 mouse retinas, which shows an ASCL1 binding peak within DHS-4D (Fig. S11) (Jorstad et al., 2020). Lack of detailed resolution limited the identification of which particular E-box was bound, but the peak was strongest around E-boxes 3, 4 and 5. Similar data for NEUROG2 were not available and our attempts at ChIP were unsuccessful. Thus, we were unable to determine whether NEUROG2 directly regulates DHS-4D under normal circumstances. However, DHS-4D-GFP expression substantially overlapped with ASCL1 (53.64%) and NEUROG2 (48.57%) embryonically (Fig. S12).

Although bHLH transcription factors regulate DHS-4D, other factors are likely involved in establishing its spatial and temporal pattern. Dividing DHS-4D into sub-elements (D.1 and D.2) resulted in constructs that lacked spatial specificity. This suggests that the spatial features of expression depend on other binding sites located within both the D.1 and D.2 halves of DHS-4D. One possibility is that the necessary E-box sites in each half synergize to drive the correct spatial activity. Alternatively, non-E-box sites may be required to establish specificity. This possibility is supported by recent work from Chan and colleagues, in which several sequences were screened for Otx2 enhancer activity (Chan et al., 2020). They identified a 1.5 kb region, termed O9, which fully encompasses DHS-4D (Fig. S13) (Chan et al., 2020). The O9 element had transient activity in the developing retina, similar to DHS-4D. A 116 bp sub-element (O9-12) fully contained within the DHS-4D region also had retinal expression, although it was unclear whether it retained specificity with OTX2. Chan and colleagues showed that the O9-12 sub-element required a Sox2-binding site for retinal activity and that reducing Sox2 expression with short hairpin (sh)RNA reduced the O9 parent enhancer activity (Chan et al., 2020). These data suggest that SOX2, made by progenitors, contributes to Otx2 initiation via regulation of the DHS-4D enhancer. There are other conserved sequences and predicted binding sites throughout the enhancer, raising the possibility that additional transcription factors are essential for DHS-4D expression. It is likely that the combination of SOX2, bHLH and other transcription factors controls the spatial and temporal specificity of DHS-4D expression.

We observed that the initiation of Otx2 expression during retinal development had redundant or compensatory features. This occurred with transcriptional regulation and at the level of enhancer usage. Although CRISPR/Cas9 targeting of DHS-4D at embryonic times blocked OTX2 expression similarly to targeting its coding region, this was not the case in the postnatal retina. Late targeting resulted in only a partial loss of OTX2 expression. This closely paralleled targeting Ascl1 and Neurog2 postnatally, which only moderately reduced OTX2 expression. The concordance of these results suggests that another enhancer is partially redundant with DHS-4D or can incompletely compensate for its loss. In principle, the regulation of such an Otx2 enhancer could be bHLH dependent. If so, the enhancer activity would not require Ascl1 or Neurog2. It is unclear why this redundancy and compensation are apparent only in the postnatal retina. This effect could be due to changes in accessible chromatin flanking Otx2, leading to the addition/removal of enhancer sites, or in the complement of transcription factors expressed by progenitors over time. Another possibility is that, as the cell cycle length increases over development (Alexiades and Cepko, 1996), there is progressively more time for alternative pathways to compensate for the loss of the DHS-4D enhancer. The use of DHS-4D-knockout mice will help determine additional mechanisms of activating Otx2 expression in the postnatal retina.

The loss of Ascl1 and Neurog2 in combination, but not alone, resulted in reduced OTX2 expression during embryogenesis. This further suggests that redundant or compensatory mechanisms are used to regulate DHS-4D and Otx2 expression. Such mechanisms are consistent with the mild phenotypes seen in single-gene knockout experiments. Loss of Neurog2 moderately affects ganglion cell differentiation and modestly alters OTX2+ cell populations (Hufnagel et al., 2010; Kowalchuk et al., 2018), whereas loss of Ascl1 results in a modest depletion of progenitors and corresponding reductions in late-born cell types (Brzezinski et al., 2011; Nelson et al., 2009; Tomita et al., 1996). There are situations in which bHLH factors can compensate for each other to varying degrees. For example, Neurod1 knock-in to the Atoh7 locus was able to partially rescue the retinal ganglion genesis deficit that occurs in Atoh7 null mice (Mao et al., 2008). In contrast, inserting Ascl1 into the Atoh7 locus did not rescue ganglion cell genesis (Hufnagel et al., 2010). Inserting Neurog2 into the Ascl1 locus can partially rescue some of the Ascl1 mutant phenotypes in the developing brain (Parras et al., 2002). Other Ascl1 and Neurog2 mutagenesis experiments suggest a model in which specific bHLH factors are required in narrow temporal windows such that, when mutants delayed development, progression could be compensated for by a different bHLH factor expressed during the next developmental window (Hufnagel et al., 2010; Kowalchuk et al., 2018; Maurer et al., 2018; Pattyn et al., 2006). Indeed, delayed retinal development is a feature of both Ascl1 and Neurog2 mutants (Brown et al., 1998; Tomita et al., 1996). Lastly, mutating either Ascl1 or Neurog2 can result in the upregulation of the other factor in the developing retina (Akagi et al., 2004; Hufnagel et al., 2010; Kowalchuk et al., 2018). Together, these examples suggest that Ascl1 or Neurog2 can directly compensate for each other to activate Otx2 expression.

It is also possible that Ascl1 and Neurog2 act in a redundant fashion during retinal development. Unlike some other areas of the developing nervous system, Ascl1, Olig2 and Neurog2 can be co-expressed within retinal progenitor cells (Fig. 4C) (Brzezinski et al., 2011; Hufnagel et al., 2010; Kaufman et al., 2019). Given their colocalization within retinal progenitor cells, redundancy could be enabled by the mere availability of each of these bHLH factors at the same time. In addition, although we observed that E-boxes 3, 4 and 6 were required for DHS-4D enhancer activity, these three sites were not sufficient to drive expression. This argues that bHLH factors bound at E-boxes 3, 4 and 6 interact with bHLH factors bound to sites 1, 2 and/or 5 to activate enhancer activity. This could reflect the requirement for a diverse set of bHLH transcription factors to be bound over multiple sites of an enhancer. This type of redundancy model was recently examined in embryonic stem cells, in which it was shown that a diverse assortment of transcription factors could drive enhancer expression far better than any single factor or binding site (Singh et al., 2021). The mutagenesis of multiple E-boxes may have limited the variety of bHLH factors that can occupy these sites, preventing the robust expression of DHS-4D. Similarly, it is possible that a threshold amount of bHLH factor binding at DHS-4D is sufficient to drive expression nonspecifically. In this model, single knockouts would not be sufficient to reduce the levels of bHLH factors below a threshold that generates a phenotype. By removing multiple bHLHs simultaneously (i.e. Ascl1/Neurog2), the overall amount of bHLH proteins may be reduced enough to cause a phenotype. This type of nonspecific threshold model is supported by reports that have shown a dose-dependent requirement of bHLHs to regulate or activate downstream function (Brown et al., 1998; Kataoka et al., 2000). Nonetheless, triple targeting of Ascl1, Neurog2 and Olig2 did not have a more-severe phenotype than double-targeting conditions. Thus, bHLH transcription factors may play more specifically defined roles in regulating Otx2 expression.

Our findings cannot distinguish whether Ascl1 and Neurog2 act redundantly or compensate for each other during retinal development. Although ChIP-seq data show that ASCL1 binds DHS-4D (Jorstad et al., 2020), it is unclear whether NEUROG2 binds under normal conditions or only upon the loss of Ascl1. The former outcome would support a model of redundancy in which either bHLH factor could activate Otx2, whereas the latter result would argue that Neurog2 directly compensates for the absence of Ascl1. It is also possible that both mechanisms are at play, but at different stages of development. ASCL1 and NEUROG2 ChIP experiments at multiple stages of retinal development will help discriminate between compensatory and redundant mechanisms.

Regardless of whether Ascl1 and Neurog2 act in a redundant or a compensatory fashion, we were initially surprised that simultaneously targeting these factors blocked OTX2 expression in the embryonic retina. This was because Ascl1/Neurog2/Neurod4 triple-mutant retinal explants cultured for 2 weeks form a laminated retina that contains photoreceptors (Akagi et al., 2004). Given that targeting Ascl1 and Neurog2 at birth had modest effects on OTX2, it is possible that only the early-born OTX2+ cells (cones and a small fraction of rods) are affected in double mutants. Cones were not specifically assayed in the triple-knockout retinas and could be severely reduced because such a deficiency could easily be masked by the formation of more-numerous late-born rod photoreceptors (Akagi et al., 2004). Another possibility is that combined loss of Ascl1 and Neurog2 results in delayed formation of OTX2+ neurons, a phenotype seen with ganglion and late-born OTX2+ cells in Neurog2 mutants (Hufnagel et al., 2010; Kowalchuk et al., 2018; Maurer et al., 2018). Assaying the timing and extent of rod and cone photoreceptor genesis in Ascl1/Neurog2 double-targeted cells would distinguish between these mechanisms.

Photoreceptor and bipolar cells maintain OTX2 expression into adulthood (Fig. 7) (Beby and Lamonerie, 2013; Fossat et al., 2007; Nishida et al., 2003). Given the transient nature of the DHS-4D activity, other enhancers are needed to maintain Otx2 in the adult retina. These likely contribute to the resilience and flexibility of OTX2 expression. Several potential Otx2 enhancers have been described for the retina and other developing systems (Chan et al., 2020; Emerson and Cepko, 2011; Kurokawa et al., 2004, 2014; Muranishi et al., 2011; Perez-Cervantes et al., 2020; Wilken et al., 2015). For example, the DHS-2 and DHS-15 enhancers appear to mark OTX2+ cells during late development and in the maturing retina (Wilken et al., 2015). These enhancers overlap with elements O5 and O20, respectively, described by Chan and colleagues (Chan et al., 2020), who also introduce a previously undescribed enhancer, O7, which shows a similar expression pattern in the postnatal retina (Chan et al., 2020). It is likely that enhancers O7, DHS-2, DHS-15 and other elements regulate Otx2 expression in photoreceptors and bipolar cells. Similar to Otx2 initiation, this may involve redundant or compensatory mechanisms. We propose that retinal Otx2 expression occurs in two phases: initiation and maintenance (Fig. 7). Once Otx2 is initiated by DHS-4, OTX2 can then act in an autoregulatory fashion at other enhancers to maintain its expression through cell type-specific mechanisms in photoreceptors and bipolar cells. In support of this, OTX2 ChIP-sequencing shows OTX2 binding at the DHS-2, DHS-15 and O7 enhancers (Bunt et al., 2011; Samuel et al., 2014). Consistent with a role in initiation and not maintenance, no OTX2 binding was seen at the DHS-4D sequence (Fig. S13). The necessity of DHS-2, DHS-15, O7 and other putative Otx2 maintenance enhancers and how they are regulated remains to be elucidated.

All animal experiments were conducted with approval from the University of Colorado Denver Institutional Animal Care and Use Committee. Male and female mice were used equally. Retinas were collected at either E14.5 or P0 from wild-type CD1 (strain 022, Charles River Laboratories) or ROSA-RFP mice [B6.Cg-Gt(ROSA)26Sor^tm14(CAG-tdTomato)Hze/J] (strain 007914, Jackson Laboratories) (Madisen et al., 2010).

To screen enhancer activity, E14.5 or P0 retinas were dissected, electroporated and cultured as ex vivo explants as previously described (Mills et al., 2017). Briefly, eyes were dissected in cold Hank's Balanced Salt Solution (HBSS with Ca²⁺ and Mg²⁺) (Corning) containing 6 mg/ml glucose and 0.05 M HEPES. P0 retinas were dissected free, whereas E14.5 eyes were dissected until only the retina and lens remained. Next, 1.5 µl of electroporation mix (45 µg of plasmid DNA dissolved in 9 µl of H₂O with 6 µl of methyl green in glycerol) was pipetted onto the photoreceptor side of the retinas. The retinas were electroporated with five square wave pulses of 50 V for 50 ms with 250 ms intervals (Bio-Rad Gene Pulser Xcell). The electroporated explants were then placed in growth media [Neurobasal Medium containing 1× N2 supplement, 1× L-glutamine, 1× penicillin/streptomycin and 1% fetal bovine serum (FBS)] (Gibco/Thermo Fisher Scientific) in a 12-well plate. Explants were cultured under 5% CO₂ in a 37°C incubator using a nutator (at 12 RPM) to oxygenate them gently for up to 72 h.

For in vivo electroporation, newborn CD1 or ROSA-RFP mice were placed in a clear plastic bag on ice and cryo-anesthetized for 3-5 min. Once fully anesthetized and unresponsive to physical contact, mice were transferred to a solid ice pack, where they remained throughout the procedure. A 31G needle was used to create an incision lengthwise along the eyelid to expose the eye. A second 31G needle was inserted along the sclera nasally adjacent to the cornea to make a hole. A 32 g blunt tip Hamilton syringe, loaded with 0.5 µl of 2 µg/µl DNA in sterile H₂O, was inserted in the hole and the DNA was slowly injected into the subretinal space of the central retina. The positive electrode of a Tweezertrode (BTX) was placed over the eye containing the inserted DNA, whereas the negative end was placed on the opposing side of the head. Mice were electroporated with five 80 V square wave pulses for 50 ms with a 950 ms delay between each pulse using a Bio-Rad Gene Pulser Xcell. After electroporation, Neosporin was applied to the exterior of the eye and Buprenex (0.1 mg/kg) was given subcutaneously to minimize pain. Pups were then placed on a circulating water heating pad until recovered, and returned to their home cage.

Following culturing, explants destined for histology were fixed in 2% paraformaldehyde for 20 min at room temperature. Following in vivo electroporations, eyes were punctured with a 30 g needle in the cornea and fixed in 2% paraformaldehyde at room temperature for 1-2 h. After fixation, explants and eyes were cryopreserved through an increasing sucrose gradient (10%, 20%, 30%), frozen in OCT medium (Sakura) and stored at −80°C until they were cryosectioned.

Retinal tissues were cryosectioned at 10 µm on a Thermo Fisher Scientific MICROM HM 550 cryostat and immediately placed on Colorfrost Plus microscope slides (Thermo Fisher Scientific). Slides were incubated at room temperature for 1 h in milk block solution (the supernatant of 5% milk and 0.5% Triton X-100 in PBS) (Brzezinski et al., 2010). Primary antibodies were diluted in milk block and the slides incubated overnight (12-18 h) at room temperature. The slides were washed in PBS and incubated for 1 h at room temperature with fluorescently conjugated secondary antibodies (Jackson ImmunoResearch) diluted 1:500 in milk block solution. After another round of PBS washes, the slides were mounted. Primary antibodies used in this study were: mouse anti-ASCL1 (1:250; 556604, BD Biosciences) or rabbit monoclonal anti-ASCL1 (1:250; ab211327, Abcam), rabbit anti-AP2B (1:250; sc-8976, Santa Cruz Biotechnology), mouse anti-BRN3a (1:100; sc-8429, Santa Cruz Biotechnology), mouse anti-Ki67 (1:250; 550609, BD Biosciences), mouse anti-NEUROG2 (1:250; MAB3314, Santa Cruz Biotechnology), rabbit anti-OLIG2 (1:250; AB9610, Merck Millipore), goat anti-OTX2 (1:250; BAF1979, R&D Systems), rabbit anti-ONECUT1 (1:250; sc-13050, Santa Cruz Biotechnology), rabbit anti-PAX6 (1:250; 901301, BioLegend), mouse anti-RXRG (1:250; sc-514134, Santa Cruz Biotechnology), goat anti-SOX2 (1:250; sc-17320, Santa Cruz Biotechnology) and mouse anti-VSX2 (1:100; sc-374151, Santa Cruz Biotechnology). For visualizing plasmid-expressed fluorescent reporters, the GFP signal was enhanced by staining with chicken anti-GFP (1:1000; ab13970, Abcam) and mCherry was detected with mouse anti-RFP (1:500; ab65856, Abcam). For explant EdU labeling, 1 µl of 10 mM EdU was added to 1 ml of warm media 30 min prior to fixation. Visualization of EdU incorporation was performed using the Click-iT™ EdU Imaging Kit (C10337, Thermo Fisher Scientific), after the secondary step, as described previously (Brzezinski et al., 2013; Goodson et al., 2020a; Kaufman et al., 2019).

Images of retinal immunostaining were acquired using a Nikon C2 laser scanning confocal microscope. Images were captured as a 1024×1024 pixel area using each laser sequentially to generate the final image. Three to five z-stacks (1-1.5 μm per slice) were captured, maximum-intensity z-projected with ImageJ (Schneider et al., 2012) and minimally processed using Adobe Photoshop. Quantification of images was performed using the following workflow: Nikon ND2 stacks were imported into ImageJ, z-stacks were flattened and color channel levels were normalized. Cell counts were acquired manually, whereby each channel was visually inspected and labeled cell numbers were recorded in Microsoft Excel. All cells were identified in image regions of interest normalized to 250 µm of retinal length or per 200× field. For each experiment involving cell quantification, a minimum of three biological samples of mouse retinal explants or in vivo retinas were used. Two-tailed unpaired t-tests or one-way ANOVA were performed using Excel or GraphPad Prism 8; P<0.05 was considered to be significant.

The enhancer sequence for the full-length DHS-4 was obtained from Wilken et al. (2015). For all enhancer cloning, primers were designed against mouse mm10 GRCm38 genomic sequences and amplified from C57BL/6J mouse DNA (strain 664, Jackson Laboratories). Enhancer sequences were cloned into the pMin-nGFP plasmid, upstream of a minimal TATA promoter to drive the expression of nuclear GFP (Wilken et al., 2015). The plasmid was digested with EcoRI and KpnI to linearize, followed by In-Fusion HD-based Gibson Assembly of various enhancer PCR fragments into the final plasmid product following the manufacturer's protocol (Takara Bio USA). Site-directed mutagenesis of E-box sites within the DHS-4 sub-element D (DHS-4D) enhancer was performed using inverse PCR amplification of the parent plasmid with primers incorporating the CANNTG to CANNAT mutation (Mills et al., 2017). A version of DHS-4D with all six E-boxes replaced by mutant CANNAT sequences was ordered as a gene block (Integrated DNA Technologies; IDT) to generate a fully mutant variant. In some cases, to speed construction, we used site-directed mutagenesis to revert mutated sites back to wild type. To generate lineage-tracing constructs, the pMin-nGFP vector (Wilken et al., 2015) was modified to co-express Cre recombinase (Goodson et al., 2020b). Briefly, the nuclear-localized GFP cassette was modified to include a P2A self-cleavage peptide sequence and a Cre recombinase cassette. As with the other constructs, In-Fusion cloning was used to insert Otx2 enhancers into the EcoRI and KpnI sites, generating DHS4-GFP-2A-Cre and DHS4D-GFP-2A-Cre plasmids. All enhancers and primer sequences are listed in Table S1. Base pMin plasmids driven by the whole DHS-4 and minimal DHS-4D elements were deposited in the Brzezinski Lab repository with Addgene (https://www.addgene.org/Joseph_Brzezinski/).

The pSpCas9(BB)-2A-GFP (PX458) plasmid was acquired through Addgene (plasmid #48138) as a gift from Dr Feng Zhang (Massachusetts Institute of Technology, Cambridge, USA) (Ran et al., 2013). This plasmid was first modified by digesting with KpnI and AgeI and replacing the Cbh promoter with an EF1α promoter sequence PCR amplified from pLVX-mCherry (632561, Takara Bio USA). Second, the GFP reporter sequence was replaced with mCherry to establish a second reporter color. Finally, Histone H2B nuclear localization peptide was then added to each reporter variant. This resulted in both cytoplasmic (EF1α-pSpCas9-2A-GFP and EF1α-pSpCas9-2A-mCherry) and permanently nuclear-localizing (EF1α-pSpCas9-2A-H2B-GFP and EF1α-pSpCas9-2A-H2B-mCherry) versions of each variant. These newly generated plasmid backbones and nuclear reporter variants have been deposited with Addgene (#159654 and #159655 at https://www.addgene.org/Joseph_Brzezinski/).

To induce loss-of-function phenotypes, guide RNAs were designed to target protein-coding exon sequences for direct gene disruption or to flank crucial enhancer sequences. All guides were designed using the CRISPR Design and Analyze Guides tool, available through Benchling.com. On-target scores were ranked in descending order and guides with scores greater than 70 and off-target scores less than 50 were considered for use. Guides were submitted to BLAST and screened to identify any potential off-target effects on coding sequences. Three guides per target were selected and ordered as paired oligos from IDT. All guide sequences are listed in Table S1. CRISPR/Cas9 plasmid all-in-one reporter backbones created as described above were digested with BbsI and used for Golden Gate cloning to insert annealed and phosphorylated guide oligos as detailed previously (Ran et al., 2013). Electroporations that targeted enhancer regions were always performed with all three guide plasmids. In the case of targeting coding regions, we found that two guides were sufficient. Therefore, we used two all-in-one CRISPR/Cas9 plasmids for these electroporations. For the in vivo Ascl1/Neurog2 multiplexed experiment (Fig. 6) and the triple-gene knockout (Ascl1/Neurog2/Olig2)-targeting work (Fig. S9), we developed a new strategy to deliver two guides on a single plasmid. Using our modified PX458 backbone, we ligated a second U6 guide cassette into the XbaI and KpnI restriction sites. A donor plasmid containing the U6 guide cassette was engineered to have compatible flanking XbaI and KpnI restriction ends. This enabled a single all-in-one plasmid to target a coding region or enhancer with two flanking guides. Enhancer- and gene-targeting knockout constructs have been deposited with Addgene (#159656, #159657, #159658, #159659, #159660, #159661, #171101, #171098, #159662, #159663, #159664 and #171102 at https://www.addgene.org/Joseph_Brzezinski/).

Retinas from E14.5 CD-1 mice were harvested as above and electroporated with DHS-4D in pMin-nGFP and a ubiquitously expressed pLVX-H2B-mCherry plasmid construct. Retinas were placed in growth media and incubated as above until 16, 20 and 48 h post-electroporation. Retinas were dissociated with the Worthington Papain Dissociation System with some modifications (Worthington Biochemical Corporation). Briefly, up to five mouse embryonic retinas were added per activated Papain tube and incubated on a nutator at 37°C for 20-45 min. The dissociation status of the tissue was checked every 10 min and the cells gently triturated two to three times using a 1 ml pipette. Immediately prior to achieving a final single cell suspension, warmed DNase solution was added dropwise, gently inverted to mix and then incubated for 5 min on rotator at 37°C. Next, prewarmed ovomucoid solution was slowly added dropwise and again gently inverted to mix. Cells were spun down for 5 min at 300 g and the supernatant carefully aspirated. Cells were then resuspended and washed with 1% FBS+PBS (without calcium or magnesium). After washing, cells were spun for 5 min at 300 g and resuspended with 1% FBS+PBS for flow cytometry. FACS was performed with assistance from the Flow Cytometry Shared Resource (University of Colorado Cancer Center). GFP+ (± mCherry) and mCherry-only cells were separately collected in 10% FBS+PBS (without calcium or magnesium), counted on a hemocytometer and resuspended in 10% FBS+PBS. GFP+ (±mCherry) and mCherry-only cells were mixed at a 40/60 ratio and cell suspensions were then submitted to the University of Colorado Denver Anschutz Medical Campus Genomics and Microarray Core for droplet barcoding on the 10X Chromium using the Chromium Single Cell 3′ Reagent Kit V2 (10X Genomics).

Following barcoding, constructed libraries were sequenced on an Illumina NovaSEQ6000. Initial demultiplexing of raw base calls was conducted with Cell Ranger 2.2 (10X Genomics). The sequence for GFP was added to the mouse reference genome to enable alignment and detection of reads from the DHS-4D-GFP plasmid. Using cellranger mkfastq and cellranger counts, we completed alignment (mm10), filtering, barcode counting and unique molecular identifier (UMI) counting. Based on UMIs and following removal of doublets, we captured 1843 cells. Read depth was estimated to be ∼250K reads per cell. Further quality control, clustering and trajectory analyses were performed using the Scanpy 1.4.6 (https://github.com/theislab/scanpy) (Wolf et al., 2018) python package running in a JupyterLab 2.1.0 notebook. After quality control (removal of high mitochondrial representation and low number of genes detected), we used UMAP dimension reduction by Leiden-based clustering in the Scanpy Python package to identify 13 clusters using default parameters (McInnes et al., 2018; Wolf et al., 2018). The workflow pipeline was created by following and integrating steps within the pbmc3k, visualizing-marker-genes and paga-paul15 tutorials found within the Scanpy documentation site. We did not use regression to account for cell cycle or other state-like properties. Clusters were analyzed for, and visualized with, known cell type-specific marker signatures using the sc.tl.rank_genes_groups and sc.pl.rank_genes_groups_dotplot functions. Trajectory analysis was performed using the diffusion pseudotime (sc.tl.dpt) method in Scanpy (Haghverdi et al., 2016). Partition-based graph abstraction (sc.tl.paga) was used to align cells along a pseudotime path. Cells were then aligned according to their pseudotime score and plotted in ForceAtlas2 component space (sc.tl.draw_graph) and visualized with their original Leiden clusters (Jacomy et al., 2014). To identify candidate transcription factors, we isolated the pseudotime pathway that ends with cluster 12, which includes Otx2+ photoreceptors. Expression values of candidates were extracted from the temporally mapped diffusion pseudotime trajectories. Raw expression patterns along these trajectories for known genes and candidate bHLH genes were exported and plotted as a heatmap using Morpheus (https://software.broadinstitute.org/morpheus/). RNA-sequencing data are available at GEO under accession number GSE172101.

The JASPAR 2020 database online tool (http://jaspar.genereg.net/) was used to identify potential transcription factor-binding sites in the DHS-4 sequence. bHLH transcription factor motifs (Ascl1, Neurog2, Olig2, Neurod1, Neurod4 and Bhlhe22) were used for the scan. The analysis was run with a minimum score threshold of 80%. Binding-site maps were created using the BioPython package. A SeqRecord object was created in Genbank format using the enhancer DNA sequence and annotated with motif sites and scores from JASPAR. Colors represent percentiles above the 80% score threshold, with pure black being the highest score range (>95%). Output was saved as a Genbank file (.gb) and was imported into SnapGene Viewer (SnapGene) for visualization and creation of figure images. Custom code is available at https://github.com/MLKaufman/tfbs_annotate.

Genomic tracks were visualized using either the UCSC Genome Browser or the Integrative Genomics Viewer (Kent et al., 2002; Robinson et al., 2011). P0 mouse whole-retina ASCL1 ChIP-seq (Jorstad et al., 2020) data were used in Fig. S11. The following datasets were used in Fig. S13: 4- to 5-week-old mouse whole-retina ChIP-seq (Samuel et al., 2014), 8-week-old mouse whole-retina CRX ChIP-seq (Corbo et al., 2010) and E14, P0 and P21 mouse whole-retina ATAC-seq (Norrie et al., 2019).

The authors thank Tom Reh and Matt Wilken for helpful discussion about the roles of Otx2 enhancers in the retina, and Ian Purvis and Omar Ochoa Olmos for critically reading the manuscript. The authors thank Katrina Diener and the CU Denver Genomics and Microarray Core for helpful discussions and technical assistance regarding 10X Genomics single cell RNA barcoding and high-throughput sequencing. The authors thank Dmitry Baturin and Lester Acosta of the CU Cancer Center Flow Cytometry for technical assistance with cell sorting. For guidance with analysis of the single cell RNA-sequencing data, we thank G. Devon Trahan.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199399