Ikaros family proteins redundantly regulate temporal patterning in the developing mouse retina

The generation of cell diversity in the central nervous system (CNS) is a highly controlled and regulated process. Neural progenitors alter their potential to generate specific neurons and glia using both spatial and temporal patterning cues (Sagner and Briscoe, 2019). In the Drosophila nervous system, temporal patterning is regulated by the expression of transcription factors referred to as ‘temporal identity’ factors, which control the developmental competence of neural progenitor cells, allowing cell-type production to change as development proceeds (Brody and Odenwald, 2000; Cleary and Doe, 2006; Erclik et al., 2017; Grosskortenhaus et al., 2005, 2006; Isshiki et al., 2001; Kambadur et al., 1998; Li et al., 2013; Novotny et al., 2002; Pearson and Doe, 2003). A classic example of temporal patterning in vertebrates is the developing mouse retina, where seven broad cell types are formed in a sequential but overlapping manner from multipotent retinal progenitor cells (RPCs) (Rapaport et al., 2004; Young, 1985a,b). Retinal ganglion cells (RGCs), amacrine cells, cone photoreceptors and horizontal cells are mostly generated during the embryonic period of retinogenesis, whereas rod photoreceptors, bipolar cells and Müller glia are primarily generated during the rodent postnatal period (Carter-Dawson and LaVail, 1979a,b; Rapaport et al., 2004; Turner et al., 1990; Young, 1985a,b).

How exactly RPCs change competence over time to control retinal histogenesis remains poorly understood, although some progress was made in recent years (Davis et al., 2011; Decembrini et al., 2009; Dupacova et al., 2021; Elliott et al., 2008; Georgi and Reh, 2010; Gordon et al., 2013; Iida et al., 2011; Javed et al., 2020; La Torre et al., 2013; Mattar et al., 2015; Yang et al., 2003; Zibetti et al., 2019). Of note, homologs of Drosophila temporal identity factors were found to regulate temporal patterning in mouse RPCs (Elliott et al., 2008; Javed et al., 2020; Mattar et al., 2015). Specifically, inactivation of Ikzf1 (a homolog of Drosophila hunchback), a member of the Ikaros family of zinc-finger transcription factors, leads to reduced numbers of early born cell types in the retina, except cone photoreceptors, which remain unchanged (Elliott et al., 2008). Together with data showing that stochastic mechanisms contribute to cell diversification in the developing retina (Gomes et al., 2011; He et al., 2012), this work has led to a model in which temporal identity factors function permissively to bias the probability of generating specific combinations of cell types at different times (Mattar and Cayouette, 2015). Thus, loss of Ikzf1 reduces the chance of generating most early-born cell types, rather than completely blocking their production. However, the mechanisms regulating the timely generation of cones are still unclear. Recently, the homolog of Drosophila pdm/nub, Pou2f1, was shown to play a part in the temporal regulation of cone photoreceptor production by upregulating Pou2f2, which in turn represses the rod determinant Nrl in photoreceptor precursors, thereby favoring the cone fate (Javed et al., 2020). Intriguingly, Ikzf1 was found to upregulate Pou2f1 expression, but because Ikzf1 knockout retinas have normal numbers of cones (Elliott et al., 2008), these results suggest that unidentified factor(s) might function with Ikzf1 to confer competence in RPCs to generate cones.

Mechanisms regulating the production of late-born cell types also remain incompletely understood. The homolog of Drosophila castor, Casz1, another zinc-finger transcription factor, confers competence to generate mid/late-born rod and bipolar cells in the mouse retina, while suppressing the production of Müller glia, the latest-born cell type (Mattar et al., 2015). Foxn4 was found to operate downstream of Ikzf1 and upstream of Casz1 to regulate mid-late temporal identity (Liu et al., 2020), but it is not involved in Müller glia production. Instead, the temporal regulation of Müller glia production in the developing mouse retina appears largely regulated by the Nfi family of transcription factors (Clark et al., 2019), which promote expression of glial differentiation genes, even in early-stage RPCs that do not normally generate glia (Lyu et al., 2021 preprint). It remains unclear, however, how exactly Nfi factors are upregulated during late stages of retinogenesis to control glia production. Additionally, Sox9, Vsx2 and Sox2 regulate Müller glia development but, as they are broadly expressed throughout retinal development, it is unclear how their pro-glial activities are temporally gated (Lin et al., 2009; Livne-bar et al., 2006; Poché et al., 2008; Surzenko et al., 2013; Taranova et al., 2006).

There are five members of the Ikaros family of transcription factors in mice, four of which are expressed in the developing retina (Elliott et al., 2008). Aside from Ikzf1, the role of Ikaros family members in the mammalian CNS has not been explored. We therefore wondered whether other Ikaros family members might contribute to regulating RPC competence during mouse retinogenesis. We show here that Ikzf4 is required together with Ikzf1 to control cone production during early stages of retinogenesis. At late stages, we report that Ikzf4 is required for Müller glia production. Using CUT&RUN and transcriptional assays, we show that Ikzf4 binds to cis-regulatory elements of genes involved in repressing the rod photoreceptor fate at early stages, and then switches to bind genes involved in Müller glia production at late stages. Together with previous observations, these results suggest combinatorial roles for Ikzf1 and Ikzf4 during retinogenesis to regulate cell fate in a context- and temporal-dependent manner.

We have previously reported expression of Ikzf4 transcripts in the retina using in situ hybridization (Elliott et al., 2008), but the protein expression pattern remained unknown. To fill this knowledge gap, we characterized a commercially available anti-Ikzf4 antibody. First, we electroporated CAG:GFP or CAG:Ikzf4-IRES-GFP vectors in E17 mouse retinas and immunostained retinal sections 2 days later using the Ikzf4 antibody. As expected, we found that the antibody recognizes overexpressed Ikzf4 proteins (Fig. S1A-C′). Second, to ensure that the antibody is specific for Ikzf4, we immunostained retinas isolated from Ikzf4^+/+ and Ikzf4^−/− (RIKEN Bioresource https://scicrunch.org/resolver/IMSR_RBRC06808) mouse embryos at E12, E15 and P9. We found that Ikzf4 is expressed in virtually all retinal cells at E12 and E15, whereas it is restricted to a few retinal cell subtypes at P9 (Fig. S1D-H). No immunostaining signal was detected in the Ikzf4^−/− retinas, indicating that the antibody specifically recognizes Ikzf4 (Fig. S1D-H′). We hypothesized that Ikzf4 was expressed in RPCs based on its expression pattern. Accordingly, we found that virtually all proliferating Ki67^+ve RPCs also stained for Ikzf4 at embryonic stages (Fig. 1A-C), and many Ki67^+ve cells also co-labeled with Ikzf4 at P0 and P2 (Fig. 1D-F), indicating that Ikzf4 is expressed in RPCs throughout development.

We next compared our immunostaining data for Ikzf4 with published scRNA-seq datasets in the developing mouse and human fetal retinas (Clark et al., 2019; Lu et al., 2020). Focusing on the RPC population, we generated UMAP plots by sub-setting the RPC populations at various developmental stages. We found that Ikzf4/IKZF4 mRNA is expressed in both early and late RPCs (Fig. S1I,J), consistent with our immunostaining results. In addition to RPCs and neurogenic progenitors, we found Ikzf4/IKZF4 expression in differentiated retinal cell type clusters such as rods, amacrine, horizontal and retinal ganglion cells (Fig. S2A,B).

To assess expression of Ikzf4 in mature cells, we used cell-type specific antibodies at P7. We detected expression of Ikzf4 in early-born cell types such as Lim1^+ve horizontal cells, Brn3a^+ve RGCs, Pax6^+ve amacrine cells and S-opsin⁺ cone photoreceptors (Fig. 1G-H″″). We could also detect weak Ikzf4 immunostaining in S-opsin⁻ cells in the ONL, suggesting some expression in rod photoreceptors (Fig. 1H′,H″). These results are consistent with our Ikzf4/IKZF4 expression analysis in scRNA-seq datasets (Clark et al., 2019; Lu et al., 2020) (Fig. S2A,B). Interestingly, we also found Ikzf4 expression in Nfia⁺ and/or Nifb⁺ cells that are either Chx10⁻ or Chx10⁺ (Fig. 1I-I‴), likely representing Müller glia and bipolar cells, respectively. Taken together, these data indicate that Ikzf4 is detected in all mature early-born cell types and in some late-born cell types.

To investigate Ikzf4 function in the developing mouse retina, we genetically manipulated late-stage RPCs by misexpressing Ikzf4. We first infected P0 retinal explants with retroviral vectors expressing Venus or Ikzf4-IRES-Venus and analyzed individual clone compositions 14 days later using cell type-specific markers, cell morphology and nuclear layer position, as we did previously (Elliott et al., 2008; Javed et al., 2020). In the Ikzf4-IRES-Venus-infected clones, we found many Venus⁺ cells located in the ONL that stained for Rxrg, a marker of cone photoreceptors in this layer (Fig. 2A-B″), as well as many cells with a Müller glia morphology (Fig. 2C). Quantifications revealed a significant increase in the proportion of Rxrg⁺ cone-like cells and Müller glia in the Venus⁺ clones, at the expense of rods (Fig. 2D). The proportion of one-cell clones was also slightly increased after Ikzf4 expression (Fig. 2E), suggesting precocious cell cycle exit or cell death. To distinguish between these possibilities, we electroporated CAG:GFP or CAG:Ikzf4-IRES-GFP vectors in P0 retinas. After culturing retinal explants for 2 days, we added EdU for 2 h then fixed and stained the explants for GFP and EdU. We observed a significant decrease in EdU⁺GFP⁺ cells after expression of Ikzf4 (Fig. 2F-H). As we found no change in the number of cells staining for cleaved caspase 3 (Fig. S3A-C), we conclude that Ikzf4 expression reduces clone size by promoting early cell cycle exit, rather than cell death. Finally, we also found that most one- and two-cell clones in the Ikzf4 condition contain either cone-like cells or Müller glia, whereas control clones contain mostly rods (Fig. S3D,E). Together, these results suggest that Ikzf4 promotes cone-like cells and Müller glia fate at the expense of rods when expressed in late RPCs.

To support the retroviral lineage-tracing results and provide more in-depth analysis of the cell types produced, we electroporated retinas in vivo at P0 with either CAG:GFP or CAG:Ikzf4-IRES-GFP and determined the identity of GFP⁺ cells 14 days later using immunostaining for cell type-specific markers. As observed with retroviral vectors, we found an increase in Rxrg⁺GFP⁺ cells following Ikzf4 electroporation (Fig. 2I-K). Although endogenous cones expressed both Rxrg and S-opsin, as expected, GFP⁺ cells in the Ikzf4 condition expressed Rxrg, but not S-opsin or other mature cone markers such as M-opsin and PNA (Fig. S3F-G‴). The GFP⁺Rxrg⁺ cells we found in the ONL were not mis-localized retinal ganglion cells (RGCs), which also express Rxrg (Mori et al., 2001), as they did not express markers of RGCs, such as Brn3a or Brn3b (Fig. S3H-I‴). Instead, Ikzf4-expressing GFP⁺ cells in the ONL co-labeled with Crx/Otx2, which specifically labels photoreceptor cells in this layer and bipolar cells in the INL at the equivalent age of P14 (Fig. S3J-K″). Thus, consistent with our observations in explants infected with retroviral vectors, expression of Ikzf4 in postnatal RPCs induces the generation of cone-like cells. Also consistent with our retroviral experiments in explants, we observed GFP⁺Sox2⁺, GFP⁺Hes1⁺Chx10⁻ and GFP⁺Lhx2⁺Nfia/b⁺ Müller glia after electroporation of Ikzf4 in P0 RPCs in vivo and ex vivo (Fig. S3L-O″). These cells had typical Müller glia morphology with apical and basal processes extending through the entire retina, indicating that expression of Ikzf4 in P0 RPCs also promotes the glial fate.

As photoreceptor precursors normally give rise to rods at postnatal stages, we postulated that the immature cones generated might be the result of Ikzf4 inhibiting rod production. Consistent with this hypothesis, we observed a reduction in transcript levels of Nrl and Nr2e3, two rod-specific genes, and an increase in Rxrg expression in the GFP⁺ cell population 6 days after electroporation of CAG:Ikzf4-IRES-GFP in P0 retinal explants (Fig. 2L). The number of GFP⁺ cells in the ONL that stain for Nrl and Nr2e3 was also reduced after Ikzf4 expression (Fig. 2M-O, Fig. S3P-Q″). Thus, we conclude that expression of Ikzf4 in late-stage RPCs promotes glia and represses the rod photoreceptor fate, which lead to photoreceptor precursors turning on Rxrg expression. We postulate that the lack of a transcriptional network promoting cone differentiation at P0 prevents the full maturation of Ikzf4-induced cone-like cells.

Based on the above results, we hypothesized that Ikzf4 may be required for cone and Müller glia development. We therefore analyzed retinas from Ikzf4^+/+, Ikzf4^+/− and Ikzf4^−/− mice, and quantified the various retinal cell types using specific markers at P10, a stage when cell genesis is largely complete. Although we found no change in the number of RGCs (Brn3a⁺), amacrine cells (Pax6⁺), horizontal cells (Lim1⁺) or bipolar cells (Otx2⁺), we observed a reduction in Sox2⁺ and Lhx2⁺ Müller cells in the INL of Ikzf4^−/− retinas compared with Ikzf4^+/+ and Ikzf4^+/− retinas (Fig. 3A-C). Unexpectedly, we found no difference in the number of Rxrg⁺ cells in the ONL in Ikzf4^−/− retinas compared with control Ikzf4^+/+ (Fig. 3C). These results indicate that Ikzf4 is required for Müller glia development, but not for cone photoreceptors.

We have previously shown that inactivation of Ikzf1 decreases early-born cell type production, with the exception of cone photoreceptors, which are unaltered in Ikzf1 KO mice (Elliott et al., 2008). As Ikzf1 and Ikzf4 are expressed in the same cells as early as E11 (Fig. S4A-D), we wondered whether Ikzf1 and Ikzf4 might genetically interact to regulate cone development. To test this idea, we generated Ikzf1 and Ikzf4 double knockout mice. Interestingly, when both alleles of Ikzf1 were knocked out together with one or two alleles of Ikzf4 (Ikzf1^−/−;Ikzf4^+/− and Ikzf1^−/−;Ikzf4^−/−), embryos were paler and exhibited reduced liver size compared with Ikzf1^+/−;Ikzf4^−/− and Ikzf1^+/−;Ikzf4^+/− embryos (Fig. S4E-H). Moreover, we found that Ikzf1^−/−;Ikzf4^+/− and Ikzf1^−/−;Ikzf4^−/− mice die at early perinatal stages, usually before P2, whereas single Ikzf1^−/− or Ikzf4^−/− mice are viable, showing functional redundancy for survival. Therefore, we had to focus our analysis of the retina at embryonic stages.

We first quantified cone numbers using Rxrg immunostaining at E15, when cone genesis is at its peak. Although we found no difference in cone photoreceptor numbers between most genotypes analyzed, we found a significant reduction in cone numbers in Ikzf1^−/−;Ikzf4^−/− double knockout animals (Fig. 3D-F), indicating an epistatic relationship between Ikzf1 and Ikzf4 to control cone photoreceptor development, at least partially. We also investigated whether Ikzf4 might function with Ikzf1 to regulate production of other early-born cell types, which are only mildly decreased in Ikzf1 KO retinas (Elliott et al., 2008). As the expression of specific amacrine and horizontal cell markers generally starts at postnatal stages (Clark et al., 2019), when double knockout animals are lethal, we focused our attention on RGC production, as Brn3b is robustly expressed at E15.5 (Xiang, 1998). We found that the number of RGCs is reduced in Ikzf1^−/−;Ikzf4^+/− (Fig. 3G), consistent with our previously-published data (Elliott et al., 2008), but this was not enhanced in Ikzf1^−/−;Ikzf4^−/− double knockout animals. Thus, unlike our observations for cones, Ikzf4 does not act redundantly with Ikzf1 to regulate RGC development.

We have previously reported that expression of Ikzf1 in late RPCs is sufficient to confer competence to generate early-born horizontal, amacrine and retinal ganglion cells (Elliott et al., 2008). To explore how Ikzf1 might temporally reprogram late-stage RPCs, we misexpressed Ikzf1 in P0 retinas, and carried out CUT&RUN and RNA-sequencing (Fig. S5A). We first curated the top differentially expressed genes by comparing the RNA-seq datasets from the CAG-GFP and CAG-Ikzf1-transfected RPCs using DESeq2 (Fig. S5A, Table S1). We performed gene ontology (GO) enrichment using GONet to find the top biological processes affected by Ikzf1 misexpression (Pomaznoy et al., 2018). In the upregulated list, we found many genes associated with retinal ganglion cells and amacrine cell development, such as Bhlhe22, Tfap2b, Nr4a2 and Pax6 under the GO biological process ‘nervous system development’, as predicted given that Ikzf1 promotes these early-born cell fates (Fig. S5B, Table S2) (Elliott et al., 2008). Conversely, in the downregulated list, we found many genes associated with bipolar and Müller glia development such as Hes1, Irx3, Notch3, Rbpj, Sox2 and Sox9 under the GO biological process ‘regulation of neuron differentiation’ (Fig. S5B, Table S3). We also found the late temporal factor Casz1 in the same GO biological process, consistent with our previously published results that Ikzf1 represses Casz1 (Mattar et al., 2015). We next assessed the CUT&RUN feature distribution using ChIPSeeker to annotate the MACS2 called peaks (Yu et al., 2015), and found that Ikzf1 binding is mostly observed away from the promoters in intergenic and intronic regions (Fig. S5C), as previously reported in B cells (Schwickert et al., 2014). Using Hypergeometric Optimization of Motif EnRichment (HOMER) to discover transcription factor motifs (Heinz et al., 2010), we found the canonical ‘GGAA’ Ikzf-binding motif (Molnár and Georgopoulos, 1994) or the complementary sequence in the dataset (Fig. S5D), as predicted. These results suggest that Ikzf1 binds to GGAA motifs in distal and intronic cis regulatory elements (CRE). To narrow down the list of potential target genes, we next compared the significantly altered transcripts in Ikzf1 misexpression RNA-seq with that of bound regions in Ikzf1 CUT&RUN using Genomic Regions Enrichment of Annotations Tool (GREAT) (Table S4) (McLean et al., 2010). We found a total of around 1000 upregulated and downregulated transcripts that are bound by Ikzf1 (Fig. S5E, Table S5). Out of these, retinal ganglion cell and amacrine cell development genes were enriched in the upregulated category, such as Bhlhe22, Nr4a2 and Tfap2b, whereas bipolar cell and Müller glia development genes, such as Casz1, Hes1, Irx3, Irx5, Irx6, Nfib, Rbpj and Sox2, were enriched in the downregulated category (Fig. S5F,G, Table S4). These results are consistent with Ikzf1 functioning as an early temporal identity factor through the positive and negative regulation of multiple genes involved in early- and late-born cell type production, respectively.

To provide molecular insights into Ikzf4 function in retinal development, we first carried out CUT&RUN on E14 and P0 retinal extracts using a validated anti-Ikzf4 antibody (S1A-H′). As the expression of Ikzf4 in E14 and P0 mouse retinas is mostly restricted to RPCs, with only a small number of differentiated cells also expressing Ikzf4, we postulated that this experiment would give a reliable read-out of Ikzf4-binding targets in RPCs. Using MACS2 peak calling on the CUT&RUN datasets, we found 2472 Ikzf4 peaks at E14 and 2130 peaks at P0 (Fig. S6A). When we intersected the peaks between the Ikzf1 at P0 and Ikzf4 E14 and P0 datasets using bedtools (Quinlan and Hall, 2010), we found only a few overlapping peaks (Fig. S6A), suggesting that Ikzf1 and Ikzf4 do not bind to the same cis-regulatory elements (CREs) in RPCs. Using ChIPSeeker and HOMER (Heinz et al., 2010; Yu et al., 2015), we found that Ikzf4 primarily binds to the ‘GGAA’ motif or the complementary sequence in promoters at both E14 and P0 (Fig. S6B,C).

Next, we carried out GO classification using GREAT and annotated the top GO biological processes at E14 (Table S6) (McLean et al., 2010). We found that ‘Positive regulation of chromatin organization’ was one of the top biological processes (Fig. 4A). Among the list of bound genes associated with this GO term, we found the DNA methyltransferase Dnmt1, which has previously been shown to regulate photoreceptor differentiation (Fig. 4A) (Rhee et al., 2012), suggesting a role for Ikzf4 in the regulation of chromatin organizers during early retinogenesis. We next focused our analysis on known regulators of cone or rod cell fate. We found that Ikzf4 binds to many genes involved in suppressing the rod fate and promoting the cone fate, such as Sall3, Thrb, Onecut1, Onecut2 and Pou2f2 (de Melo et al., 2011; Emerson et al., 2013; Javed et al., 2020; Jean-Charles et al., 2018; Ng et al., 2001; Sapkota et al., 2014) (Fig. 4B).

Of particular interest to us in the E14 Ikzf4 CUT&RUN data was the peak observed in a region of open chromatin 47 kb upstream of the Pou2f2 promoter that scored as a significant Pou2f2-associated peak using GREAT (Fig. 4B). Indeed, we have previously reported that Pou2f2 represses the rod-promoting transcription factor Nrl to favor cone development. Along with our results showing that Ikzf4 misexpression promotes cone-like cell production and that Ikzf4 is redundantly required with Ikzf1 to promote the cone fate, we hypothesized that Ikzf4 might regulate Pou2f2 expression to induce cone production. To explore this possibility, we electroporated P0 retinas with either CAG:GFP or CAG:Ikzf4-IRES-GFP, and sorted GFP⁺ cells 18 h later for RT-qPCR. We found that Ikzf4 upregulates transcript levels of Pou2f2, whereas there is no change in the expression of temporal identity factors such as Casz1a, Casz1b and Foxn4, which are regulated by Ikzf1 (Fig. 4C) (Liu et al., 2020; Mattar et al., 2015). These results suggest that Ikzf4 promotes cone development, at least in part, by binding and upregulating Pou2f2 gene expression. Consistently, we found that the number of GFP⁺Rxrg⁺ cells in the ONL was significantly reduced when Pou2f2 was knocked down concomitantly with Ikzf4 expression (Fig. 4D-F), indicating that Ikzf4 requires Pou2f2 to promote cone development.

To gain insights into how Ikzf4 might promote Müller glia development, we classified GO terms using GREAT and annotated the top biological processes from the P0 CUT&RUN data (Table S7) (McLean et al., 2010). Among these, we found an enrichment in Notch signaling-related GO terms (Fig. 5A), containing Notch signaling genes such as Rbpj, Notch, Hes1 and Hes5, which were previously reported to promote the Müller glia fate (Furukawa et al., 2000; Hojo et al., 2000; Nelson et al., 2011; Zheng et al., 2009). Next, we assessed genes enriched in the Müller glia cluster in P14 scRNA-seq data (Clark et al., 2019). Remarkably, we found Ikzf4-binding peaks in all the top eight genes enriched in the Müller glia cluster, such as Nlgn1, p27kip1 (Cdkn1b), Tuba1b, Trpm3, Sorcs1, Notch1, Slc1a3 and Vim (Fig. S7A). However, as late RPCs and Müller glia have similar transcriptomes (Blackshaw et al., 2004; Roesch et al., 2008), these genes are also expressed in RPCs, complicating the interpretation of these results. Therefore, we focused our analysis on genes that are required for Müller glia development. We found Ikzf4-binding signal at genomic regions around Sox8, Sox9, Lhx2, Hes1, Nfia, Nfib and Nfix gene bodies, some of which were called as peaks by MACS2 (Fig. 5B) (Clark et al., 2019; de Melo et al., 2018, 2016; Muto et al., 2009; Poché et al., 2008; Zibetti et al., 2019). In contrast, we did not find Ikzf4 binding at other intronic open chromatin regions in gene bodies of Nfib, Nfix and Nrl, which are highly expressed at P0, showing the specificity of the identified Ikzf4-binding sites (Fig. S7B).

Next, we performed transcription factor occupancy prediction by investigation of ATAC-seq signal (TOBIAS) on open chromatin regions from previously-published datasets (Aldiri et al., 2017) that were bound by Ikzf4 at E14 and P0, to assess motif footprints of transcription factors that could be binding to the same regions as part of a gene regulatory network (Bentsen et al., 2020). We found that, out of the 849 motifs analyzed from the JASPAR database (Fornes et al., 2020), motifs of transcription factors associated with Müller glia differentiation, such as Lhx family, Sox family and Nfib were differentially enriched in P0 open chromatin regions with Ikzf4 binding compared with E14 (Table S8, Fig. S7C,D). We also found that, in the P0 retina, Ikzf4 binds to open chromatin regions that are active or poised enhancers, as suggested by the enrichment of H3K27ac and H3K4me3 histone marks, respectively (Creyghton et al., 2010; Orford et al., 2008), and depletion of the H3K27me3 histone mark (Aldiri et al., 2017; Cao et al., 2002) (Fig. S7E). Therefore, these data suggest that Ikzf4 binds to CREs that are important for Müller glia differentiation at late stages of retinal development.

We then assessed whether Ikzf4 regulates transcripts of genes involved in Müller glia development by electroporating P0 retinas with either CAG:GFP or CAG:Ikzf4-IRES-GFP, and carried out RT-qPCR on sorted GFP⁺ cells 18 and 72 h later. We found that Ikzf4 promotes expression of Sox8 and Sox9 18 h post-electroporation. Although we found no change in Lhx2, Rnf12, Nfia, Nfib, Nfix or Ldb1 at 18 h (Fig. 5C), their expression was increased 72 h after electroporation (Fig. 5D). Conversely, we found a reduction in many Müller glia specification genes in P6 Ikzf4^−/− retinas compared with Ikzf4^+/−, including Sox8, Lhx2, Hes5 and Rbpj (Fig. 5E). As the Nfi family of proteins are crucial to confer RPC competence to generate Müller glia, we next validated the increase in mRNA levels of Nfib by electroporating P0 retinal explants and stained sections for Nfia and Nfib 72 h later. As predicted, we found a significant increase in the number of GFP⁺ cells stained for Nfia and Nfib after Ikzf4 overexpression (Fig. 5F-H). These results suggest that Ikzf4 is sufficient to induce Müller glia production by binding to and upregulating genes involved in glial cell specification, including Notch signaling genes and Nfi temporal identity factors.

It has previously been shown that Hes1 expression oscillates during progenitor proliferation (Shimojo et al., 2008). Interestingly, Hes1 expression decreases when the RPCs exit the cell cycle and is repressed in cells fated to become neurons, whereas it is maintained in cells fated to become glia (Furukawa et al., 2000; Imayoshi et al., 2013). As our CUT&RUN data indicate that Hes1 locus is significantly bound by Ikzf4 (Fig. 5B), we postulated that Ikzf4 might regulate Müller glia production at least partially by regulating Hes1, and perhaps also the other hairy and enhancer-of-split homolog Hes5. To study the dynamics of Ikzf4 binding at the Hes1 and Hes5 promoter, we co-electroporated reporter constructs pHes1-dsRed or pHes5-dsRed together with either CAG:GFP or CAG:Ikzf4-IRES-GFP in P0 retinas. We tracked the electroporation patches over time to assess the dynamics of dsRed expression, which reports activity of the promoters. We observed that Ikzf4 promotes the activity of the Hes1 promoter compared with the control GFP at 48 and 72 h (Fig. S8A-D″). Even 6 days after electroporation, Ikzf4-electroporated retinas maintained high expression of dsRed, whereas expression of dsRed was reduced in the control GFP condition (Fig. S8E-F″). In contrast, we did not observe similar dynamics in dsRed expression for the Hes5 promoter at 48 h or 6 days after Ikzf4 electroporation compared with the GFP control (Fig. S8G-J″). Of note, Ikzf4-mediated upregulation of the Hes1 promoter activity appears to be specific to retinal cells, as Ikzf4 had no effect on pHes1-dsRed activity in HEK293 cells (Fig. S8K-L″). We next assessed whether the increase in Hes1 promoter activity leads to an increase in Hes1 protein expression. To test this, we electroporated P0 retinas with either CAG:GFP or CAG:Ikzf4-IRES-GFP and analyzed the number of GFP⁺ cells staining for Hes1 in retinal sections 44 h later. As predicted, we observed an increase in the number of GFP⁺Hes1⁺ cells after Ikzf4 expression (Fig. 6A-C). Together, these results indicate that Ikzf4 activates the Hes1 promoter, which leads to elevation of Hes1 protein expression.

We then sought to identify the Ikzf4-binding sites required for the regulation of Hes1. When we analyzed the Hes1 promoter region containing the Ikzf4 peaks, we found three ‘GGAA’ Ikzf-binding motifs (Fig. 6D). To study the functional requirement of these motifs, we mutated each one and assessed how this affected the ability of Ikzf4 to regulate the activity of the Hes1 promoter. We co-electroporated CAG:GFP or CAG:Ikzf4-IRES-GFP with either the wild-type or mutated Hes1 promoter constructs (Fig. 6E). When we mutated binding sites 1 (mut1), 2 (mut2) or 3 (mut3) and electroporated with CAG:GFP, we observed a reduction in dsRed signal compared with the wild-type construct (Fig. 6F-Q), confirming the importance of these sites for the activity of the Hes1 promoter. However, co-expression of Ikzf4 was sufficient to promote dsRed expression, albeit less so for mut2 (Fig. 6F′-Q′). This suggests that none of these sites alone is required for Ikzf4 binding and activation of the Hes1 promoter (Fig. 6F-Q′). Interestingly, however, we found that, when we mutated sites 2 and 3 together (mut2+mut3), Ikzf4 was no longer able to activate the Hes1 promoter (Fig. 6R-T′). Taken together, these results show that Ikzf4 binds at two ‘GGAA’ sites that are required for the sustained expression of Hes1 in post-mitotic cells.

As Ikzf4 also bound regions associated with other Notch signaling genes, we finally wondered whether the regulation of Hes1 promoter activity is dependent on Notch signaling. To test this, we electroporated P0 retinas with vectors expressing either GFP or Ikzf4-IRES-GFP, cultured the retinal explants with either DMSO as a control or the Notch signaling inhibitor DAPT, and fixed the explants 6 days later to assess dsRed expression. We found that Ikzf4 overexpression increases the number of dsRed⁺GFP⁺ cells in the DMSO condition, whereas this effect is abrogated with the addition of DAPT (Fig. 6U-X), suggesting that Ikzf4 regulates Hes1 promoter activity in a Notch signaling-dependent manner.

Neural diversity in the CNS is generated by a combination of spatial and temporal factors working in concert to establish the vast repertoire of neurons and glia. Although factors regulating fate decisions upon cell cycle exit have been extensively studied, much less is known about how neural progenitors alter their developmental output over time. In this study, we show that Ikzf4 functions epistatically with the previously identified early temporal identity factor Ikzf1 to control the production of early-born cone photoreceptors. We also show that Ikzf4 is required for Müller glia production at late stages of retinogenesis. Mechanistically, we report that, during early stages of development, Ikzf4 generally binds and regulates genes involved in repressing the rod cell fate, including Pou2f2, thereby favoring cone production. At late stages, Ikzf4 binds to genes involved in Müller glia development, including Sox8, Sox9, Lhx2, Nfi and the Notch signaling effector Hes1. Ikzf4 is sufficient to induce sustained expression of Hes1 during glia differentiation via two ‘GGAA’ Ikzf-binding sites in the promoter in a Notch signaling-dependent manner. Taken together, this study identifies redundancy between Ikaros family proteins and dynamic regulation of target gene selection as key mechanisms underlying temporal patterning in retinal progenitors (Fig. 7).

Single knockout retinas of Ikzf1 or Ikzf4 have normal number of cones (Fig. 3C) (Elliott et al., 2008), but here we report that double knockouts of Ikzf1 and Ikzf4 have reduced cone numbers, suggesting redundant function for Ikzf1 and Ikzf4 in cone production. Alternatively, loss of Ikzf1 and Ikzf4 may only delay cone production, but the perinatal lethality of Ikzf1/Ikzf4 double knockouts prevented us from addressing this question. Future conditional inactivation should help circumvent this problem. Although we find that Ikzf1 and Ikzf4 bind different CREs, we also find co-binding at the same genes, suggesting that inactivation of one factor could be compensated by the other. It will be interesting to assess whether Ikzf1 switches binding profile upon loss of Ikzf4, and vice versa.

Our data indicate that Ikzf4 binds to and induces expression of Pou2f2, which suppresses the rod fate at early developmental stages, thereby favoring the cone fate. We observe an increase in Rxrg⁺ cells and decrease of Nrl/Nr2e3⁺ cells when Ikzf4 is overexpressed in late RPCs, suggesting that Ikzf4 is sufficient to reopen a window of competence for cone genesis that is normally lost at this stage. However, overexpression of Ikzf4 in late RPCs does not lead to the production of mature cones, unlike Pou2f1 (Javed et al., 2020), which suggests that it likely suppresses rod differentiation rather than promoting cone differentiation. We postulate that Ikzf4 might partially promote cone genesis, but additional factors required to promote full cone maturation are likely missing at late stages of retinogenesis. Ikzf4 also promotes early cell cycle exit, which is uncharacteristic of temporal identity factors such as Ikzf1, Pou2f1 and Casz1, but not Foxn4 (Elliott et al., 2008; Javed et al., 2020; Li et al., 2004; Mattar et al., 2015). Altogether, our results suggest that Ikzf4 is part of a gene regulatory network that confers RPCs with the competence to generate cones, possibly by repressing the rod fate at early stages.

Our data indicate that Ikzf4 switches transcriptional targets at late stages of retinogenesis to control Müller glia production. One possible mechanism to explain this switch is that Ikzf4 is co-expressed with Ikzf1 during early stages of retinal development, but not at late stages. The absence of Ikzf1 at late stages of development may redirect Ikzf4 from photoreceptor genes to glial genes, although such activity would have to be indirect because Ikzf1 and Ikzf4 have divergent genome occupancy. The current model of gliogenesis in the retina proposes that Nfia, Nfib and Nfix confer late-stage temporal identity to RPCs to generate Müller glia and bipolar cells (Clark et al., 2019; Lyu et al., 2021 preprint). In addition to Nfia, Nfib and Nfix, Lhx2 interacts with Rnf12 in late RPCs to promote gliogenesis by activating the expression of Sox8 and/or Sox9 and Notch target genes in postmitotic precursors destined to become Müller glia (de Melo et al., 2016, 2018; Jadhav et al., 2006; Muto et al., 2009; Nelson et al., 2011; Poché et al., 2008; Zhu et al., 2013; Zibetti et al., 2019). Lhx2 also dynamically alters its DNA-binding profile at early and late stages of retinal development to switch from promoting neurogenesis to promoting gliogenesis (Zibetti et al., 2019), as we observe here with Ikzf4. So how does Ikzf4 fit in the above proposed model of gliogenesis? As Lhx2, Nfia, Nfib and Nfix cKOs show no significant change in Ikzf4 transcript levels (Clark et al., 2019; de Melo et al., 2016), and our data show that Ikzf4 binds to and induces expression of Lhx2 and Nfi, we propose that Ikzf4 may be upstream of these factors in the gliogenic gene regulatory network. Given the weak expression levels of Ikzf4 and the low sensitivity of the RNA-seq analysis, however, it is possible that previous studies failed to detect changes in Ikzf4 expression in the Lhx2 and Nfi cKOs. In any case, our data suggest a model wherein Ikzf4 favors the emergence of Sox8⁺ and/or Sox9⁺ precursors with sustained Notch signaling, as observed with the increase in Hes1⁺ cells after overexpression of Ikzf4 in RPCs, which then go on to become Müller glia.

We report that Ikzf4 binds to regions close to many Notch signaling gene bodies, including multiple sites at the promoter of Hes1. Although we found an increase in Hes1 mRNA levels after Ikzf4 overexpression, the reduction of Hes1 expression observed in Ikzf4^−/− retinas was not statistically significant compared with Ikzf4^+/− (Fig. 5E), suggesting that Ikzf4 is sufficient but not completely necessary for Hes1 expression. One possibility is that other Ikaros family members, such as Ikzf2 and Ikzf5 (Elliott et al., 2008), which are both expressed in the late developing mouse retina, are compensating to regulate Hes1 expression. Another possibility is that Ikzf4 not only regulates Hes1 but also several other Notch signaling factors to promote Müller glia. Consistently, we observe binding of Ikzf4 to other Notch signaling genes at P0, such as Notch1 and Rbpj (Fig. 5A). Both Hes5 and Rbpj mRNA level are significantly reduced in Ikzf4 KO retinas (Fig. 5E), suggesting a role for Ikzf4 in the regulation of several Notch signaling factors, which together may promote Müller glia development. As Hes1 promoter activation by Ikzf4 is impaired by blocking Notch signaling (Fig. 6X), the activation of the Notch pathway appears necessary for Ikzf4 to promote expression of Hes1 and ultimately Müller glia development.

Previous studies have elucidated the role of Ikzf4 in the immune system. Most notably, Ikzf4 function in T-cell differentiation varies considerably depending on CD4⁺ T-cell subtype, highlighting the dynamic role of Ikzf4 function based on the cellular context (Liu et al., 2014; Pan et al., 2009; Powell et al., 2019; Read et al., 2017; Rieder et al., 2015; Sekiya et al., 2015; Sharma et al., 2013), similar to what we report here. Interestingly, widespread expression of Ikzf4 has been reported in different organs, including the CNS (Perdomo et al., 2000). One study has detailed the role of Ikzf4 in regulating expression of PSD-95 in adult cochlear afferent neurons (Bao et al., 2004), but the function of Ikzf4 in the developing CNS remains largely elusive. The findings reported here suggest that Ikzf4 may have widespread regulatory functions in the CNS to control cell fate specification. Future studies on conditional knockouts of Ikzf4 alone or in combination with other Ikzf genes will be interesting to address this issue.

All experiments were carried out in accordance with the Canadian Council on Animal guidelines and approved by the IRCM animal care committee. Ikzf1 (Wang et al., 1996) and Ikzf4 (International Mouse Phenotyping Consortium and RIKEN Bioresource RRID: IMSR_RBRC06808) knockout mice were raised in the C57BL/6J background (Mus musculus). All other mouse experiments were performed on wild-type CD1 mice (Mus musculus, Charles River Laboratories). Animals of either sex were used in this study.

Retroviruses were designed, produced and concentrated as previously described (Cayouette et al., 2003). Retinal explants were cultured as previously outlined (Cayouette et al., 2001). Retroviral infections of retinal explants and analyses of the retroviral clones were carried out as previously stated (Javed et al., 2020). Retinal explants were fixed either 2, 3, 6, 12 or 14 days after electroporation, as required for the experiment, and processed for immunostaining. For the Notch signaling inhibition experiment, DMSO or DAPT was added to the medium as previously described at concentration of 10 μM for the first 3 days of culture, and explants fixed after 6 days in vitro (Nelson et al., 2011).

Ikzf1 and Ikzf4 cDNAs were cloned into a pCIG2-IRES-GFP and pCLE-venus vector using restriction sites previously outlined (Gaiano et al., 2000; Hand et al., 2005). pHes1-dsRed vector was mutated with Infusion HD Cloning Plus kit from Takara using primers listed in Table S1 (Matsuda and Cepko, 2007).

P0 or P1 eyes were injected with DNA plasmid at 3 μg/μl concentration containing 0.5% Fast Green and electroporated as previously described (de Melo and Blackshaw, 2011).

RNA extraction and qPCR quantitation were performed as previously described (Javed et al., 2020). Primers used are listed in Table S9 (Batsché et al., 2005; Ouimette et al., 2010).

P0 CD1 retinas were dissociated using Accutase and transfected using the Amaxa Neural Stem Cell Kit according to the manufacturer’s protocol. Cells were seeded into six-well plates coated with PLL/Laminin, and cultured as previously described (Gomes et al., 2011). After 9 h, cells were harvested and dissociated with Accutase followed by FAC-sorting for GFP. Control GFP condition contained 4×10⁵ cells per biological replicate (n=4) and Ikzf1 condition contained 3×10⁵ cells per biological replicate (n=4). Cells were sorted directly into lysis buffer and purified using the RNeasy micro kit from Qiagen. Sample quantity (RNA integrity number=RIN) and quality were assessed using the Agilent RNA 6000 Pico Kit (Agilent 5067-1513) on the Bioanalyzer 2100. rRNA was depleted using Ribo-Zero Magnetic Gold Kit for rRNA depletion (Human/Mouse/Rat) (Epicentre for Illumina, MRZG12324) according to the manufacturer’s guidelines. Library preparations were made out using SMARTer Stranded RNA-Seq Kit (Clontech, 634836 or 634837) according to the manufacturer's guidelines. Libraries were diluted and pooled equimolarly, and then sequenced in pair end 50 cycles (PE50) on a v4 flowcell (Illumina HiSeq PE Cluster Kit v4 cBot, PE-401-4001) of the Illumina HiSeq 2500 System.

The age of mouse embryos was calculated from pregnant females with the day of vaginal plug considered as day 0 (E0) and collected at E11, E14, E15, E16, E17, P0, P2 and P7 for spatiotemporal analyses. For endogenous Ikzf1 and Ikzf4 immunostaining, the retinas were dissected and fixed for 2 min in 4%PFA/PBS followed by immersion in 20% sucrose/PBS for 1 h. Retinas were then embedded in OCT, frozen in liquid nitrogen, sectioned at 25 μm using a cryostat and immunostained on the same day. For all other antibodies and Ikzf4 immunostaining against overexpressed Ikzf4 protein, the decapitated heads from embryos or eyes from postnatal pups were fixed for 15 min in 4%PFA/PBS and immersed in 20% sucrose/PBS for 2 h.

Immunofluorescence was performed as previously described (Javed et al., 2020). A list of primary antibodies can be found in Table S10, including DHSB antibodies (Venkataraman et al., 2018).

EdU (30 μM) was added to the culture medium for 2 h before collection and fixation. Click-iT EdU Alexa Fluor 647 was used to label cells that incorporated EdU.

Statistical tests were performed for each experiment in this study, as indicated in the figure legends. All quantifications in the bar graphs of this study are represented as mean±standard error of the mean (s.e.m.), whereas n number and individual values on the graphs represent biological replicates. Statistics for the retroviral clonal analyses were performed as previously outlined (Pounds and Dyer, 2008). Retinal explants containing disorganized layers and poor immunostainings were discarded and analysis was limited to the well-organized regions of the retinal explants. All experiments were repeated at least three times.

Ikzf4 knockouts were analyzed as follows. Two sections of P10 central retinas oriented temporo-nasally were examined with quantifications of 200 μm (Pax6), 400 μm (Otx2, Sox2 and Lhx2 staining in the INL, Rxrg in the ONL) and 800 μm (Brn3a and Lim-1) lengths of the imaged section. The investigator was blinded to the genotype of animals. An ImageJ analysis macro was written to count cells automatically in a section. Analyze particles in ImageJ was used after defining a region of interest and setting threshold for each antibody to auto-count cell numbers of Pax6⁺, Otx2⁺, Sox2^+ve, Lhx2⁺ and Brn3a⁺ cells, whereas Rxrg⁺ and Lim-1⁺ cells were manually counted. Ikzf1/4 double knockouts were analyzed as follows. Three sections of E15 central retinas in embryonic heads oriented dorso-ventrally were analyzed with quantification of 200 μm for Rxrg⁺ cells and 400 μm for Brn3a⁺ cells. Rxrg⁺ cells were manually counted due to background signal, whereas Brn3a⁺ cells were counted using the ImageJ analysis macro as described above. All cell quantifications were performed by blinding the investigators to the genotype of the animals.

CUT&RUN was performed as previously described (Skene et al., 2018), with a few added modifications. For Ikzf1 CUT&RUN, CAG-Ikzf1 electroporated P0 retinal explants were cultured and 50,000 GFP⁺ cells were sorted directly into the buffer with Concanavalin A beads. For Ikzf4 CUT&RUN, E14 and P0 retinas were dissected, and 1,500,000 cells were used for each stage. The entire procedure was carried out in 200 μl PCR tubes. Digitonin at 0.01% concentration was used and pAG-MNase digestion was performed for 30 min on ice.

Libraries were prepared with the KAPA DNA HyperPrep Kit (Roche 07962363001 - KK8504). This protocol includes an End-Repair/A-tailing step and an adapter ligation step followed by a PCR amplification (enrichment) of ligated fragments. The adapters used for ligation were IDT for Illumina TruSeq UD Indexes (Illumina - 20022371). The final enriched product (library, after PCR) was purified using KAPA purification beads (Roche 07983298001 - KK8002) and a dual-SPRI size selection was performed (with KAPA beads) to select fragments between 180-500 bp. Libraries were then quantified using a Nanodrop microvolume spectrophotometer (ng/μl) and quality was assessed using the Agilent High Sensitivity DNA Kit (Agilent - 5067-4626) on a Bioanalyzer 2100. The libraries were then quantified by q-PCR to obtain their nanomolar (nM) concentration. Libraries were diluted, pooled equimolar and sequenced in pair end 50 cycles (PE50) on a S1 flowcell (Illumina - 20012863) of the Illumina NovaSeq 6000 System.

scRNA-seq analyses of previously published datasets was performed as follows. Fastq raw reads were aligned and counted to generate matrices for each individual stage from Mouse retina atlas (GEO: GSE118614) (Clark et al., 2019) and Human fetal retina atlas (GEO: GSE116106, GSE122970 and GSE138002) (Lu et al., 2020) using Cellranger 4.0 (10x Genomics). Velocyto (La Manno et al., 2018) with run10x function was used on cellranger output folders to generate loom files for each individual stages. Seurat (Butler et al., 2018) was used to analyze the mouse and human retina loom files and subset RPC clusters for Ikzf4/IKZF4 expression analyses using markers previously described (Clark et al., 2019; Lu et al., 2020). Scanpy (Wolf et al., 2018) was used to analyze the same loom files and generate Ikzf4/IKZF4 expression UMAPs along with cell type markers (Fig. S2).

RNA-seq fastq raw reads were analyzed using the salmon quantification package (Patro et al., 2017). Differential gene expression analyses between CAG-GFP and CAG-Ikzf1 was performed using DESeq2 on Salmon quant output files (Love et al., 2014). R intersect() function was used to compare differentially expressed gene lists with Ikzf1 CUT&RUN peaks.

CUT&RUN analyses were performed by aligning the raw fastq reads with the mouse mm9 genome using default parameters of bowtie2 on Galaxy platform (Afgan et al., 2016; Langmead and Salzberg, 2012). Bam files generated for Ikzf4 and IgG CUT&RUN at each stage were used to call peaks using 0.5 FDR parameter on MACS2 (Feng et al., 2012). Bedtools was used to find overlapping peaks between the two stages (Quinlan and Hall, 2010). Bigwig files were generated using deeptools2 (Ramírez et al., 2016). Integrative Genomics Viewer (IGV) was used to visualize bigwig files.

TOBIAS footprinting analysis was performed by first intersecting ATAC-seq peaks at E14 and P0 from previously published datasets (Aldiri et al., 2017), with Ikzf4 E14 and P0 regions using bedtools (Bentsen et al., 2020; Quinlan and Hall, 2010). TOBIAS ATACorrect and ScoreBigwig was performed on ATAC-seq E14 and P0 bam files separately to generate corrected ATAC-seq bigwig files. The two E14 and P0 regions were merged using bedtools, and BINDetect was used to estimate differentially bound motifs based on scores, sequence and motifs. The JASPAR non-redundant vertebrate motifs package was used for motif annotation on the mm9 genome (Fornes et al., 2020).

Volcano plots were generated using the EnhancedVolcanoplot github package (https://github.com/kevinblighe/EnhancedVolcano), whereas Upset plots were generated using Upsetplot shiny web tool (Conway et al., 2017). Information on each software version is listed in Table S10.

We thank Christine Jolicoeur for technical assistance; Jessica Barthe and Androne Constantin from the IRCM animal core facility; Odile Neyret-Djossou from the IRCM molecular biology platform; and Éric Massicotte and Julie Lord from the IRCM flow cytometry platform. We thank Seth Blackshaw for critical comments on the manuscript. We thank Aurelie Huang-Sung and Nicole Francis for providing the pAG-Mnase for the CUT&RUN experiments. We thank Benoit Boulan for writing the ImageJ macro. We also thank all members of the Cayouette lab and Shahrzad Bahrampour for their comments and support.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.200436.reviewer-comments.pdf