Regulation of retinal amacrine cell generation by miR-216b and Foxn3

Cellular diversity is a fundamental feature of the nervous system: networks comprising different types of neurons and glia form neural circuits with distinct functional properties. The mouse retina has six major classes of neurons that can be further divided into over 100 subtypes (Clark et al., 2019; Macosko et al., 2015; Shekhar et al., 2016; Yan et al., 2020). This complex mixture of cell types performs initial visual processing. During retinal development in vertebrates, specific cell types are generated in a broadly consistent order, and individual retinal progenitor cells can give rise to combinations of differentiated cell types (reviewed by Cepko, 2014). Multiple transcriptional regulators and signaling pathways control the potential of retinal progenitor cells to generate specific cell types at different developmental times (Clark et al., 2019; Konstantinides et al., 2015; Mattar et al., 2015).

Amacrine cells are a diverse subset of retinal interneurons that participate in local processing of visual information (Masland, 2012). Amacrine cell subtypes can differ by neurotransmitters, morphology and other properties, with over 60 subtypes detected in the mouse by single-cell analysis of gene expression (Yan et al., 2020). During embryonic and early postnatal development, a series of transcription factors regulates the formation of amacrine cells, including the Forkhead protein Foxn4, the basic-helix-loop-helix (bHLH) proteins Ptf1a, Neurod1, Neurod4 and Ascl1, the Rorb nuclear receptor, the Meis1 homeobox, and the zinc-finger transcription factor Plagl1/Zac1 (Fujitani et al., 2006; Inoue et al., 2002; Islam et al., 2013; Li et al., 2004; Liu et al., 2013, 2020; Ma et al., 2007; Nakhai et al., 2007). Several of these factors also regulate the formation of horizontal cells, another class of interneuron, and/or other retinal neurons.

Noncoding RNAs also have been implicated as regulators of retinal development, including microRNAs (miRNAs). Mature miRNAs are ∼22 nt small RNAs that, when loaded onto Argonaute proteins, function as negative regulators of mRNA stability and translation (reviewed by Bartel, 2018). Target-miRNA interactions usually involve base-pairing between the miRNA seed sequence (positions 2-8 of the mature miRNA) and a complementary sequence in the target mRNA, often in 3′ untranslated region (3′ UTR). Loss of mature miRNAs after retina-specific disruption of the miRNA processing enzyme Dicer leads to altered production of retinal cell types, including fewer amacrine cells, with prolonged generation of early cell types (Davis et al., 2011; Georgi and Reh, 2010) and/or widespread cell death (Damiani et al., 2008), depending on the timing of Dicer disruption. miRNAs have been implicated in the differentiation or maintenance of retinal neurons and Müller glia, as well as in the neurogenic potential of Müller glia (Kara et al., 2019; Reh and Hindges, 2018; Walker and Harland, 2009; Wohl et al., 2017, 2019; Xiang et al., 2017).

Here, we show that ectopic expression of the miR-216b miRNA during mouse retinal development leads to an increase in amacrine cells and a decrease in bipolar cells. miR-216b is normally expressed in differentiated amacrine cells, starting during retinal development. We identify the 3′ UTR of the mRNA encoding the Foxn3 transcription factor as a target of miR-216b. Inhibition of Foxn3 by RNAi or CRISPR disruption in developing retinas leads to an increase in amacrine cells, whereas overexpression of Foxn3 reduces amacrine cell number and co-expression of Foxn3 with miR-216b can partially reverse the effects of miR-216b. Our observations identify Foxn3 as a novel regulator of retinal interneuron formation and suggest that miR-216b is a regulator of gene expression in amacrine cells.

To identify miRNAs enriched in retinal amacrine cells, we used small RNA sequencing to compare miRNA levels in embryonic day 18.5 (E18.5) retinas isolated from mice heterozygous or homozygous for disruption of the gene encoding Ptf1a, a transcription factor required for formation of most amacrine and horizontal cells in the retina (Fujitani et al., 2006; Nakhai et al., 2007). In retinas without Ptf1a, several miRNAs were reduced relative to the retinas from Ptf1a heterozygotes, with miR-216b-5p the most downregulated miRNA (Fig. 1A, Table S1). Mir216b is part of a genomic miRNA cluster with Mir216a and Mir217; miR-216a-5p/3p and miR-217-5p also were decreased in Ptf1a knockout retinas. Real-time quantitative reverse-transcription PCR (qRT-PCR) confirmed downregulation of the predominant mature miRNAs from these miRNA genes in E18.5 Ptf1a knockout retinas, relative to retinas from heterozygotes (Fig. 1B). A developmental time course of retinal expression by qRT-PCR showed that all three mature miRNAs decreased from postnatal day 0 (P0) to P7 but were still expressed in P12 and adult retinas. In contrast, miR-183-5p, an unrelated miRNA expressed in photoreceptors and bipolar cells (Xu et al., 2007; Zhuang et al., 2020), increased between P0 and adult (Fig. 1C). miRNA in situ hybridization for mature miR-216b-5p revealed expression in the inner retina at P0 and in the inner nuclear layer (INL) and in the ganglion cell layer (GCL) at P12, with the strongest signal at P12 in the amacrine layers of the inner INL (Fig. 1D). Expression was substantially lower or absent in the neuroblast layer at P0 and the ONL at P12. Cells labeled by miR-216b-5p in situ hybridization overlapped with the amacrine cell-specific transcription factor AP2α (Tfap2a) (Bassett et al., 2007) at P0 or P12 (Fig. 1E-J). miR-216b-5p also overlaps with AP2α at E18.5 in retinas, but miR-216b-5p was reduced and AP2α was absent in Ptf1a knockout retinas (Fig. S1). In situ hybridization for miR-216b-5p, miR-216a-5p and miR-217-5p in retinas at different ages revealed expression in INL and GCL (Fig. 1D and Fig. S2). At P0, expression was detected in inner retina, but the miRNAs were present at low levels or not expressed in the neuroblast layer. Reduced expression in Ptf1a mutant retinas and localization to the inner retina are consistent with expression of all three miRNAs in postmitotic amacrine cells.

The miR-216a-5p and miR-216b-5p miRNAs have similar seed sequences and potentially could bind to overlapping sets of target sites, whereas miR-217-5p has a distinct seed sequence and is expected to bind to different target sites (Fig. 2A). To determine whether miR-216a/miR-216b/miR-217 could influence retinal development, we ectopically expressed these miRNAs in neonatal mouse retinas. DNA plasmid vectors that express GFP with pre-miR-216b, pre-miR-216a or pre-miR-217 expressed from the intron of the same transcription unit (Fig. 2A), either together or individually, were introduced into P0 mouse retinas by in vivo electroporation (Matsuda and Cepko, 2004; Zhang et al., 2012). GFP expression was used to identify electroporated cells. Ectopic miR-216b expression was confirmed by in situ hybridization (Fig. S3). A plasmid vector expressing GFP and a pre-miR-155-based shRNA against luciferase was used as a control [a functional synthetic miRNA with no detectable effect on the retina (Zhang et al., 2012)]. Forced expression of miR-216a, miR-216b or all three miRNAs together led to a significant increase in the frequency of amacrine cells among GFP-labeled cells at P7, relative to control, as assessed by immunodetection of AP2α (Fig. 2B-G). Introduction of miR-216b or miR-216a individually yielded similar increases in amacrine cells among the GFP-labeled cells, whereas introduction of miR-217 by itself did not (Fig. 2E,G). Although control retinas had few or no GFP-labeled displaced amacrine cells in the GCL, we observed occasional GFP-labeled amacrine cells in the GCL after electroporation with miR-216b, miR-216a or miR-216a/miR-216b/miR-217 together. In addition, AP2α-negative GFP-labeled cells in the INL decreased after expression of miR-216b or all three miRNAs together (Fig. 2D,F,H), suggesting a decrease in bipolar neurons among the electroporated cells. Although antibody markers specific for bipolar cells did not work well at P7, additional experiments described below show that miR-216b decreases bipolar cell frequency among the electroporated cells when scored at P12.

Homozygous disruption of Ptf1a leads to nearly complete loss of amacrine cells (Fujitani et al., 2006). As miR-216a/b expression depends in part on Ptf1a, we investigated whether expression of miR-216b would be sufficient to direct formation of amacrine cells in retinas without Ptf1a. Expression vectors for miR-216b or the Luc shRNA control were introduced by electroporation into E16.5 retinas from mice heterozygous or homozygous for disruption of Ptf1a. As Ptf1a knockout mice are not viable after birth, retinas were maintained as explants in vitro and assessed for AP2α expression at 8 days in vitro (DIV). The miR-216b expression vector failed to generate AP2α-labeled cells in the absence of Ptf1a (Fig. S4), indicating that miR-216b cannot direct formation of amacrine cells without Ptf1a.

We used Argonaute PAR-CLIP (Hafner et al., 2010) to identify potential endogenous miRNA targets in developing retina, including targets of miR-216a/b. P0 mouse retinas were cultured as explants in 4-thiouridine (4SU) containing media for one DIV. After UV crosslinking and lysis, endogenous Argonaute 1-4 proteins with crosslinked RNA were immunoprecipitated and cDNA libraries were prepared from the RNA fragments using a modified PAR-CLIP method, then analyzed by high-throughput sequencing (see Materials and Methods). We generated five independent libraries from Argonaute immunoprecipitations and two control libraries from IgG immunoprecipitations. PAR-CLIP with 4SU frequently generates T to C substitutions at crosslinks during reverse transcription, allowing single base resolution of crosslink positions (Hafner et al., 2010). To distinguish crosslinks from random sequencing errors, we identified mapped sequences with consistent T to C substitutions at the same genomic position in two or more Argonaute PAR-CLIP libraries, without a T to C substitution at the same position in control libraries. For multiple closely spaced crosslinks, we identified the most frequent crosslink (±15 nt around each crosslink) to create a set of 14,054 most frequent, consistent crosslink sites, which we refer to as predominant crosslinks. Most predominant crosslink sites were in 3′ untranslated regions (UTRs) of mRNAs, with coding regions the next most frequent location (Fig. 3A), similar to prior PAR-CLIP analyses of Argonaute (Hafner et al., 2010; Lipchina et al., 2011). As lysates included both cytoplasm and nuclei, nuclear Argonaute likely contributed to crosslinked sites in introns and ncRNAs (Sarshad et al., 2018).

Argonaute PAR-CLIP frequently generates crosslinks in target RNAs within a few nucleotides of a bound miRNA seed sequence (Hafner et al., 2010). We looked at the distribution of sequences complementary to miRNA seed sequences within ±15 nt windows centered on a subset of predominant crosslink sites that mapped to 3′ UTRs. Seed matches for 15 miRNA seeds present in retinas at P0 were strongly enriched near crosslink sites (Fig. S5). Based on these results, and additional results described below, we conclude that Argonaute PAR-CLIP identified retinal miRNA-binding sites. Here, we analyze potential miR-216a/b sites in the Argonaute PAR-CLIP data; analysis of other retinal miRNA binding sites will be described elsewhere.

Seed matches were identified for miR-216a/b-5p near 404 predominant crosslink positions (Tables S3 and S4). miR-216a-5p and miR-216b-5p differ at position 8 (last base of the seed, Fig. 2A). As both miRNAs function similarly when expressed in retinas, we focused on sites with a match to positions 2-7 of both miRNAs, without a match to a more abundant miRNA seed or a seed match located closer to the crosslink site. We identified 57 predominant crosslink sites that met these criteria (Table S3). The site with the largest number of crosslinks near a miR-216a/b-5p 2-7 seed match was in the 3′ UTR of the Foxn3 transcription factor mRNA; crosslinks at this site were present in all five Argonaute PAR-CLIP libraries (Fig. 3B,C). We also assessed evolutionary conservation of predicted seed sites and the miR-216a/b-5p seed match in the Foxn3 3′ UTR is conserved to chickens and fish (Fig. 3D, Table S3).

We constructed a Luciferase reporter with the first 266 nt of 3′ UTR from Foxn3 (Fig. 4A). The Foxn3 3′ UTR fragment included both the frequently crosslinked miR-216a/b-5p site, as well as a second potential miR-216a/b-5p site near the stop codon. The second site had T to C substitutions in two distinct reads in one retinal PAR-CLIP library and therefore was not identified as a consistent crosslink site. In HEK293 cells in culture, the Foxn3 3′ UTR 1-266 reporter was inhibited by co-transfection of the miR-216b expression vector but not the miR-216b-mut expression vector, which expresses a version of pre-miR-216b with a seed sequence mutation in miR-216b-5p (and compensating changes in the 3′ end of miR-216b-3p to allow folding of the precursor). A reporter in which both seed matches in the Foxn3 3′ UTR fragment were mutated was not inhibited by the miR-216b vector. However, the miR-216b-mut vector downregulated the reporters with mutated miR-216a/b-5p sites, as we used compensatory site mutations in the Foxn3 3′ UTR fragment, complementary to the seed sequence in miR-216b-5p-mut (Fig. 4A,B). We also constructed a Luciferase reporter with the full-length Foxn3 3′ UTR (6192 nt) and polyadenylation site (Fig. 4A,C). The full-length Foxn3 3′ UTR reporter was inhibited by co-transfection of the miR-216b expression vector, but not the miR-216b-mut expression vector, in HEK293 cells. Mutation of the two miR-216a/b-5p sites in the full length 3′ UTR prevented significant repression by miR-216b, but allowed the full UTR reporter to be repressed by the matching miR-216b-mut. These observations indicate that the Foxn3 3′ UTR is a miR-216b-5p target.

Intriguingly, Foxn3 is closely related to Foxn4, a transcription factor required for the initial steps of amacrine cell formation (Li et al., 2004). However, although Foxn4 is a transcriptional activator (Lelièvre et al., 2012; Liu et al., 2013), Foxn3 is a transcriptional repressor (Scott and Plon, 2005). This difference suggested that Foxn3 might inhibit amacrine cell formation, so that miR-216a/b inhibition of Foxn3 could allow increased amacrine cell formation. Foxn3 is expressed in the head and eye in vertebrates, and reduced Foxn3 function leads to a small eye phenotype in Xenopus and mice, as well as head neural crest defects (Samaan et al., 2010; Schuff et al., 2007). However, a role for Foxn3 in the generation of specific retinal cell types has not been described. We therefore further analyzed expression of Foxn3 during mouse retinal development. At E16.5 or P0, we detected widespread expression of Foxn3 mRNA in mouse retina by in situ hybridization, both in the neuroblast layer and in some differentiating cells of the inner retina (Fig. 4D,F). We compared Foxn3 expression with Foxn4, which is expressed in retinal precursor cells that can generate amacrine interneurons. Foxn4 mRNA was restricted to cells in the neuroblast layer at E16.5 or P0, consistent with previous reports (Li et al., 2004), and a subset of cells expressed both Foxn3 and Foxn4 mRNAs at P0 or E16.5 (Fig. 4D-H, Fig. S6). In P7, P12 and adult retinas, Foxn3 expression was restricted to INL and GCL, with most expressing cells in the outer INL (Fig. 4I-K) at P7 and P12. Foxn3 mRNA expression in INL overlaps with markers of bipolar cells and a small subset of amacrine cells at P12 (Fig. 4L,M; Fig. S7). Based on recent single cell sequencing data for amacrine cells (Yan et al., 2020), Foxn3 mRNA expression was correlated with genes expressed in cholinergic amacrine cells, such as Sox2, and the highest level of Foxn3 mRNA was present in cholinergic amacrine cells (Table S5).

To determine whether reduced Foxn3 expression affects amacrine cell formation, we inhibited Foxn3 with shRNAs in developing retinas. We used RNA polymerase II-driven shRNA vectors that co-express GFP and miR-155-based shRNAs (Chung et al., 2006). Because an antibody suitable for immunodetection of mouse Foxn3 is not available, we confirmed the ability of each shRNA to reduce endogenous Foxn3 mRNA in a mouse cell line using qRT-PCR (Fig. S8), although the shRNA 635 reduced Foxn3 mRNA more effectively. The shRNA vectors were delivered into P0 retinas by in vivo electroporation. In parallel, we electroporated the control Luc shRNA vector, the expression vector for miR-216b or the miR-216b-mut expression vector. We observed a significantly increased proportion of AP2α-positive amacrine cells among the GFP-labeled cells in retinas collected at P12 with Foxn3 shRNA 635, the more effective shRNA, relative to the control shRNA vector, whereas a smaller increase with Foxn3 shRNA 195 was not significant (Fig. 5A-C,K). miR-216b expression also increased the fraction of amacrine cells among the GFP-labeled cells at P12, whereas miR-216b-mut expression did not (Fig. 5D,E,K). The fraction of Vsx2 (also known as Chx10) -positive bipolar cells among the GFP-labeled cells decreased with Foxn3 shRNA 195, Foxn3 shRNA 635 and miR-216b (Fig. 5F-I,L). Expression of miR-216b-mut did not significantly decrease the fraction of bipolar cells among GFP-labeled cells (Fig. 5J,L). These results indicate that the miR-216b vector likely alters retinal development via target repression through the miR-216b-5p strand/seed, and that inhibition of Foxn3 in the developing retina yields changes in retinal cell types similar to changes seen after overexpression of miR-216b.

As an independent test of Foxn3 function, we used CRISPR/Cas9 (Sarin et al., 2018; Wang et al., 2014) to disrupt the Foxn3 gene in embryonic retina explants in culture, then assessed changes in amacrine cell number. Two different Cas9/sgRNA expression vectors, each expressing GFP and two independent sgRNAs against Foxn3, or a control Cas9/GFP vector were electroporated into mouse retinas at E16.5, prior to the peak of amacrine cell generation, and the retinas were maintained as explants for 8 DIV. We observed an increased fraction of AP2α⁺ cells among the GFP-labeled cells in explants with either of the CRISPR vectors targeting Foxn3, relative to the control vector (Fig. 6A-F). To confirm Cas9 disruption of the Foxn3 gene in retinal cells after electroporation with CRISPR vectors at E16.5, we amplified the four genomic targets by PCR from GFP-positive retina explants at 2 DIV, then detected indels by Illumina sequencing of the PCR products. For each vector targeting Foxn3, one of the two sgRNAs was effective (Fig. S9, Table S6). The increase in amacrine cell fraction after CRISPR disruption of Foxn3 is similar to the results observed after postnatal inhibition of Foxn3 by RNAi. These results indicate that reduced Foxn3 function in cells of the developing retina permits increased amacrine cell formation.

To assess the consequences of Foxn3 overexpression in developing retinas, we introduced expression vectors for a Foxn3 cDNA (without its 3′ UTR) and GFP into E16.5 retinas by electroporation, and then maintained the retinas in explant culture for 2 or 8 DIV. Forced expression of Foxn3 reduced the fraction of AP2α-positive cells among GFP-labeled cells at 8 DIV, relative to a control expression vector (Fig. 7A,B,I). Ptf1a is transiently expressed in newly generated postmitotic amacrine cells (Fujitani et al., 2006). The Foxn3 expression vector reduced the fraction of Ptf1a-positive cells among the GFP-labeled cells at 2 DIV, relative to the control (Fig. 7E-H,K). We also scored cell proliferation/DNA synthesis by incorporation of 5-ethynyl-2′-deoxyuridine (EdU) in the explants at 2 DIV (Zeng et al., 2010). The fraction of GFP-labeled cells that incorporated EdU was significantly increased with the Foxn3 expression vector relative to control (Fig. 7C-D,J). These data indicate that Foxn3 can inhibit amacrine cell formation at an early step, prior to Ptf1a expression and that expression of Foxn3 likely reduces cell cycle exit of retinal progenitor cells.

We tested whether co-expression of the Foxn3 protein could reverse the effects of forced miR-216b expression in the developing postnatal retina. We performed in vivo electroporation at P0 to co-deliver the miR-216b and Foxn3 expression vectors individually or together, as well as control vectors, into the retina. We assessed amacrine and bipolar cell markers in GFP-labeled cells at P12 (Fig. 8A-D,I,J). Expression of miR-216b increased the fraction of AP2α amacrine cells and decreased the fraction of Vsx2⁺ bipolar cells among GFP-labeled cells at P12, relative to the Luc shRNA control vector. Co-expression of Foxn3 with miR-216b partially reversed both the increase in amacrine cells and the decrease in bipolar cells among GFP-labeled cells that were observed with expression of miR-216b alone. Expression of Foxn3 by itself in the postnatal retina slightly decreased Vsx2-positive cells among the GFP-labeled cells at P12 but did not significantly decrease AP2α-positive cells. However, we observed an increase in GFP-labeled cells within the INL that were not labeled by either Vsx2 or AP2α when Foxn3 was expressed, with or without miR-216b (Fig. 8K). We scored GFP-labeled cells in the ONL, likely rod photoreceptors based on morphology and lack of Vsx2 or AP2α expression (Fig. 8L). Expression of miR-216b did not significantly change the frequency of GFP-labeled cells in the ONL, although co-expression of Foxn3 and miR-216b slightly decreased the frequency of these cells. Finally, the frequency of Müller glial cells (Sox2 positive/AP2α negative) among the GFP-labeled cells in the INL (Fig. 8E-H,M) did not change significantly with miR-216b, with or without Foxn3. The ability of Foxn3 to partially reverse the effects of miR-216b are consistent with inhibition of Foxn3 as a component of the mechanism by which ectopic expression of miR-216b alters cell type generation in the developing retina.

Expression of miR-216b-5p overlaps with differentiated amacrine cells in the mouse retina. miR-216b forms a genomic miRNA cluster with miR-216a and miR-217, two miRNAs expressed from the same primary transcript (Kato et al., 2009). All three miRNAs are expressed in mouse inner retina, consistent with previous observations of miR-216a/miR-217 expression in zebrafish (Olena et al., 2015; Wienholds et al., 2005), and these miRNAs were downregulated in retinas without Ptf1a, a bHLH transcription factor required for the differentiation of nearly all horizontal and amacrine cells. As Ptf1a is transiently upregulated only after amacrine cells become postmitotic (Fujitani et al., 2006), all three miRNAs are preferentially expressed in postmitotic amacrine cells. The lack of high-level expression of these miRNAs in the neuroblast layer of the developing retina, as well as the dependence on Ptf1a, suggest that miR-216a/b are unlikely to regulate initial commitment or cell cycle exit of amacrine cells during retinal development. Nonetheless, ectopic expression of pre-miR-216a or pre-miR-216b in the developing mouse retina led to a significant increase in amacrine cells. The increase in amacrine cells with miR-216b depended on the miR-216b-5p seed sequence, indicating that seed-based target repression by mature miR-216b-5p is likely to drive the increase in amacrine cells. However, expression of miR-216b in retinas in which the Ptf1a gene was disrupted failed to generate amacrine cells, indicating that expression of miR-216b cannot replace Ptf1a. We propose that the normal role of miR-216b and miR-216a is likely to be reducing the expression of repressors of amacrine cell differentiation (and possibly other genes) in postmitotic, differentiating amacrine cells, and that ectopic expression of these miRNAs in retinal progenitor cells leads to premature inhibition of their targets and a consequent increase in amacrine cell formation (Fig. 8N,O). A functional requirement for the miR-216a/miR-216b/miR-217 genomic cluster in retinal development has not been evaluated, as the homozygous deletion of the gene cluster in mouse embryos is lethal prior to retina formation (Azevedo-Pouly et al., 2017). However, mice lacking the individual miRNAs are viable, suggesting that miR-216a/b may have redundant roles in embryonic development (Sutaria et al., 2019). A possible association between miR-216a/b sequence variants and inherited retinal disease in humans has been reported (Huang et al., 2018), and miR-216b is upregulated in human retinoblastoma (Ravishankar et al., 2020). In zebrafish, miR-216a regulates sorting nexin 5 (SNX5) during retinal development (Olena et al., 2015), but the miR-216a target sites are not conserved in mouse Snx5. Zebrafish miR-216a also is involved in Müller cell reprogramming during retinal regeneration (Kara et al., 2019).

We identified Foxn3 as a target of miR-216a/b-5p, based on crosslinking of the Argonaute protein by PAR-CLIP near an evolutionarily conserved miR-216a/b-5p seed match in the Foxn3 3′ UTR. We confirmed that miR-216b-5p could repress reporters based on the Foxn3 3′ UTR via two miR-216a/b-5p seed matches. Foxn3 is widely expressed in the early retina, and, strikingly, inhibition of Foxn3 expression in postnatal retina by RNAi or disruption of the Foxn3 gene by CRISPR/Cas9 in embryonic retinal explants led to an increase in amacrine cell formation, similar to that observed with miR-216b overexpression. Furthermore, ectopic expression of Foxn3 in embryonic retinal explants reduced the frequency of differentiated amacrine cells at 8 DIV. Ectopic Foxn3 also reduced expression of Ptf1a and modestly increased the fraction of proliferating cells in explants 2 days after electroporation, consistent with inhibition of amacrine cell formation at an early step, prior to cell cycle exit. Taken together, these data indicate that Foxn3 is a negative regulator of amacrine cell formation in the developing retina, likely limiting the initial formation of amacrine cells. Co-expression of Foxn3 with miR-216b partially reversed the increase in amacrine cell frequency seen with miR-216b expression, suggesting that inhibition of Foxn3 likely contributes to the effects of ectopic miR-216b. However, additional miR-216a/b targets also may contribute to the phenotypes observed after ectopic expression of miR-216a/b. In addition, regulation of Foxn3 by other miRNAs has been described in non-neural cells, including miR-182 (Zhang et al., 2021), a miRNA highly expressed in retina (Xu et al., 2007; Zhuang et al., 2020), and miR-135b (Han et al., 2019), via a site near miR-216a/b site 1 in the Foxn3 3′ UTR. miR-135b-5p is expressed in retina (Table S2), so combinatorial or redundant miRNA regulation of retinal Foxn3 is possible.

Foxn3 has been linked to multiple biological processes, including cell proliferation, epithelial-mesenchymal transition and regulation of glucose metabolism (Huot et al., 2014; Karanth et al., 2018; Li et al., 2017). Reduced Foxn3 function in frogs or mice leads to defects in head neural crest and a small eye phenotype (Samaan et al., 2010; Schuff et al., 2007). We focused on Foxn3 both because a conserved miR-216a/b-5p seed match in the Foxn3 3′ UTR had the most crosslinks among potential miR-216a/b-5p sites in the retina Argonaute PAR-CLIP data, and because Foxn3 is closely related to Foxn4. Foxn4 is a transcription factor expressed in proliferating retinal progenitor cells that is required for amacrine and horizontal cell formation (Li et al., 2004), and this role is evolutionarily conserved (Boije et al., 2013). Foxn4 is a transcriptional activator that directly activates transcription of Ptf1a and other genes (Del Barrio et al., 2007; Liu et al., 2020; Luo et al., 2012). Foxn3 and Foxn4 have been reported to bind to similar DNA target sequences, although Foxn3 also recognizes an additional class of DNA target sequences that Foxn4 does not (Nakagawa et al., 2013; Rogers et al., 2019). As Foxn3 is a transcriptional repressor (Scott and Plon, 2005), we speculate that Foxn3 could bind to and repress direct Foxn4 target genes, thus reducing amacrine cell formation. Consistent with this model, overexpression of Foxn3 reduced the frequency of Ptf1a-positive cells in the developing retina. Alternately, the ability of Foxn3 to recognize binding sites structurally distinct from sites recognized by Foxn4 could allow Foxn3 to regulate other target genes.

Ectopic expression of miR-216b or inhibition of Foxn3 by RNAi significantly reduced the number of retinal bipolar cells among the electroporated retinal cells, whereas co-expressed Foxn3 partially reversed the reduction of bipolar cells by miR-216b. Although clones of mouse retinal cells derived from a single progenitor cell can contain both amacrine and bipolar cells (Turner et al., 1990), the majority of these two cell types are generated at different times during development, with peak amacrine formation near E17 and peak bipolar generation at P1-P3 (Voinescu et al., 2009). Ectopic expression of Foxn4 during retinal development reduces bipolar cell formation (Li et al., 2004). Lineage tracing in Dll4-Cre mice indicates that most or all bipolar cells are generated from retinal progenitor cells that did not express Dll4 (Zou et al., 2015), a direct target of Foxn4 (Luo et al., 2012), suggesting that bipolar cells may arise from progenitor cells in which Foxn4 target genes were not activated. These observations suggest that both increased amacrine cell formation and decreased bipolar cell formation could arise from elevated Foxn4 activity, consistent with a model in which Foxn3 limits Foxn4 function, and miR-216b-5p inhibits Foxn3 expression. We also observed that Foxn3 mRNA is expressed in bipolar cells in the P12 retina, consistent with a role for Foxn3 in bipolar cells. However, overexpression of Foxn3 by itself in postnatal retinas reduced the fraction of bipolar cells relative to control, albeit less than miR-216b, suggesting that Foxn3 can have other effects, directly or indirectly, on bipolar cell formation.

Forced expression of Foxn3 in embryonic retinal progenitor cells did not prevent all amacrine cell formation among electroporated cells. Foxn3 interacts with other proteins (Li et al., 2017; Scott and Plon, 2005), so Foxn3 function could be limited by absence of a co-factor in some cells. Function or stability of Foxn3 protein also could be modulated by post-translational modification, as reported for the related Foxn2 protein (Ma et al., 2018). However, we cannot rule out the possibility that Foxn3 was expressed at an insufficient level in some cells, or that some electroporated cells committed to the amacrine cell fate prior to ectopic Foxn3 expression.

Morphological and molecular analyses have identified numerous subtypes of amacrine cells in mouse retina (MacNeil and Masland, 1998; Macosko et al., 2015; Yan et al., 2020). Based on recent single cell sequencing (Yan et al., 2020), cholinergic amacrine cells express Foxn3 mRNA. It is possible that miR-216a/b limits expression of the Foxn3 protein in cholinergic amacrine cells. As cholinergic amacrine cells are generated prior to E17 (Voinescu et al., 2009), RNAi and CRISPR experiments presented here were too late to reduce Foxn3 expression in those cells.

Transcription of the miR-216a/217 miRNA primary transcript in diabetic kidney cells is activated by TGFβ signaling and bHLH transcription factors (Kato et al., 2009). Intriguingly, amacrine cells express TGFβ2 and the Tgfbr2 receptor (Ma et al., 2007; Tachibana et al., 2016), which could potentially contribute to miR-216a/b/miR-217 upregulation via autocrine signaling. In addition to Ptf1a, the Ascl1, Neurod1 and Neurod4 bHLH proteins are redundantly required for amacrine cell formation (Akagi et al., 2004; Inoue et al., 2002), suggesting that those proteins could regulate transcription of miR-216b/a/miR-217 in the retina. A genome-wide analysis of Ptf1a and Ascl1-binding sites identified ChIP-Seq peaks for Ptf1a and Ascl1, including in the region upstream of the miR-216b miRNA precursor on Chr.11, in cells from mouse spinal cord (Borromeo et al., 2014). In addition, miR-216a/b and miR-217 are highly expressed in mouse pancreas (Azevedo-Pouly et al., 2017; Sutaria et al., 2019), a tissue dependent on Ptf1a function during development, indicating a link between Ptf1a and expression of these miRNA genes in multiple tissues.

Previous analyses of miRNA function in retina have revealed roles in developmental timing. Loss of all miRNAs after disruption of the Dicer processing enzyme can lead to prolonged production of early retinal cell types and loss of later cell types, including amacrine cells (Georgi and Reh, 2010). Other miRNAs also have been implicated in amacrine cell development. We identified miR-181b as another miRNA with elevated expression in mouse amacrine cells that depends on Ptf1a function (Zhuang et al., 2020). In Medaka, retinal miR-181a/b expression depends on TGF-β signaling, and miR-181a/b have been implicated in amacrine and ganglion cell process maturation (Carrella et al., 2015a,b).

Our results identify Foxn3 as a novel regulator of amacrine cell and bipolar formation during mouse retinal development, and they suggest a role for miR-216b in gene regulation in amacrine cells. Both are expected to function as repressors, suggesting that their roles are likely to be to limit expression of other regulators. Multiple repressive factors, such as Prdm1 and Vsx2, regulate bipolar cell and amacrine cell formation (Goodson et al., 2020), as well as additional, unidentified transcriptional repressors in bipolar or Müller cells (Chan et al., 2020). Identification of targets of Foxn3 in retina should help to decipher regulatory networks that control cell diversification in the developing retina.

Plasmids were constructed using standard techniques. Oligonucleotides and gBlocks used for plasmid construction are listed in Table S7. The pUS2, pUS2-MT, pUS2-eGFP, pUS2-Puro, pUS2-Luc, pUI4-GFP-SIBR, pUI4 Luc-shRNA (Chung et al., 2006) and pUS2-Tol2 transposase (Gupta et al., 2018) have been described previously. The pUS6, pUI6-GFP-SIBR and pUI6 Luc-shRNA plasmids are variants of pUS2, pUI4-GFP-SIBR and pUI4 Luc-shRNA that have ITRs from the Tol2 transposon flanking the expression cassettes to allow for integration in the presence of Tol2 transposase (Balciunas et al., 2006). Partial primary transcript sequences (pre-miRNAs and some flanking sequences) for the mouse miR-216a, miR-216b and miR-217 genes were amplified by PCR from mouse genomic DNA and cloned into pUI4-GFP-SIBR and pUI6-GFP-SIBR using primers with the indicated restriction sites. Pre-miR-216b-mut, with a 3 nt mutation in the seed sequence of miR-216b-5p (AATCTCT to AATAGAT) and a compensating change in miR-216b-3p to maintain base pairing in the pre-miRNA, was synthesized as a gBlock (IDT) and inserted into pUI6-GFP-SIBR. The mouse Foxn3- and Foxn4-coding regions were isolated by RT-PCR from embryonic mouse retina RNA and inserted into pUS2 and pUS6. Foxn3 shRNA constructs pUI6-GFP-SIBR Foxn3-195×2 and pUI6-GFP-SIBR Foxn3-635×2 contain two tandem miR-155-based SIBR cassettes (Chung et al., 2006), which target the coding region of the mouse Foxn3 mRNA. The NanoLuc coding region was PCR amplified from plasmid pNL.1.1 (Promega) and inserted into pUS2-MT to generate pUS2-MT-Nanoluc. The Foxn3 3′UTR 1-266 fragment (first base of Foxn3 stop codon is position 1) containing wild-type or mutated miR-216a/b target sites (mutated site sequences altered to match the seed sequence of miR-216b-mut) were generated by RT-PCR from mouse retina RNA, then inserted downstream of firefly Luciferase in pUS2-Luc. The remaining Foxn3 3′ UTR and presumptive endogenous polyadenylation site were amplified from mouse genomic DNA, then inserted into the Foxn3 3′ UTR 1-266 luciferase reporter plasmid DNA using the NEBuilder HiFi DNA Assembly Cloning Kit (New England Biolabs). pUBC458 was made by replacing CBH promoter in pX458 (AddGene #48138) (Ran et al., 2013) with the human UBC promoter and first intron from pUI4-GFP-SIBR (Chung et al., 2006), followed by replacement of the original sgRNA scaffold with an optimized sgRNA scaffold (Dang et al., 2015) to make pUBC458op, both plasmids retaining the U6 promoter and Bbs1 cloning sites from pX458. Oligonucleotides for Foxn3 sgRNAs were designed using E-CRISP (Heigwer et al., 2014) and inserted into pUBC458 (Foxn3 CRISPR1) or pUBC458op (Foxn3 CRISPR2). A second U6 sgRNA expression cassette was inserted adjacent to the existing U6-sgRNA cassette to express two sgRNAs.

All mouse (Mus musculus) experiments were approved by the Institutional Animal Care & Use Committee at the University of Michigan. Wild-type retinas were isolated from CD-1 mice of both sexes (Charles River). Ptf1a mutant mice have the Ptf1a-coding region replaced with the Cre-coding region (Kawaguchi et al., 2002). Eyes were fixed with 2% paraformaldehyde (ThermoFisher) in phosphate-buffered saline (PBS, Sigma) for 30 min at room temperature, retinas were dissected from the eyes, then cryoprotected in PBS containing 30% sucrose (Sigma) for 2 h at 4°C, and embedded in OCT (Sakura), prior to preparation of cryosections (16 µm) using a cryostat. All the retinas were processed similarly to obtain tissue for either in situ hybridization or antibody staining.

1 µl of plasmid DNA vectors were injected into the subretinal space of P0 mouse pups and introduced into cells by in vivo electroporation as described previously (Zhang et al., 2012). The concentration and combination of DNA vectors are listed below. 2 µg/µl pUI4 miRNA expression vectors or pUI4 Luc-shRNA control (Fig. 2), or 2 µg/µl pUI6 miR-216b expression vector, 2 µg/µl pUI6 miR-216b-mut, 2 µg/µl pUI6 Luc-shRNA or 2 µg/µl pUI6 Foxn3 RNAi vectors, and 200 ng/µl pUS2-Tol2 were injected (Fig. 5). 2 µg/µl pUI6 miR-216b expression vector or 2 µg/µl pUI6 Luc-shRNA, and 1 µg/µl pUS6-Foxn3 expression vector or 1 µg/µl pUS6 control, and 300 ng/µl pUS2-Tol2 were injected in Fig. 8. After DNA injection, five square 80 V pulses of 50 ms duration with 950 ms intervals were applied by using an ECM830 (BTX) with forcep-type electrodes. Electroporated retinas were collected at P7 (Fig. 2) or P12 (Figs 5 and 8).

For retinal explants, heads from E16 mouse embryos were split in half along the sagittal suture and the cut surface of the tissue was placed on a square plate electrode. Plasmids were introduced into retinal cells by in vitro electroporation. 0.5 µl of plasmid DNA vectors were injected into the subretinal space of each eye. 2 µg/µl of pUBC458/pUBC458op Foxn3 CRISPR vectors or pUBC458/pUBC458op control (Fig. 6), 2 µg/µl pUI6 miR-216b or pUI6 Luc shRNA expression vectors and 200 ng/µl pUS2-Tol2 (Fig. S4), or 1 µg/µl pUS6 or pUS6-Foxn3 overexpression vector combined with 1 µg/µl pUS6-GFP expression vector and 200 ng/µl pUS2-Tol2 (Fig. 7) were injected. After DNA injection, five square 30-V pulses of 50 ms duration with 950 ms intervals were applied by using an ECM830 (BTX) with a platinum needle electrode above the eye in combination with the plate electrode underneath the embryonic tissue (Matsuda and Cepko, 2004). Electroporated retinas were dissected from the eyes, placed in six-well plates on Transwells with polyester membrane (Corning, pore size 0.4 µm), the ganglion cell layer facing upwards, and cultured in 1.2 ml of high-glucose Dulbecco's modified Eagle's medium (DMEM) supplemented with 20% horse serum and 1% penicillin-streptomycin (all Thermofisher). Retinas were collected 2 or 8 days after electroporation.

miRNAs in situ hybridization with alkaline-phosphatase detection by BCIP/NBT was performed as described previously (Deo et al., 2006; Zhuang et al., 2020). RNA oligonucleotide probes were synthesized with both 5′ and 3′ fluorescein modifications (Sigma) and gel purified. Sequences of oligoribonucleotide probes are in Table S7. For fluorescent detection of miRNA in situ hybridization, the same method was used until the alkaline-phosphatase detection step. After the alkaline-phosphatase conjugated anti-fluorescein Fab2 antibody (Roche, 11426338910) was removed, slides were washed twice in 1× Tris-buffered saline (Thermofisher) (Zhuang et al., 2020) for 10 min each. A Fast Red tablet (Sigma) was reconstituted in water according to the manufacturer's instructions and 200 µl Fast Red solution was added to each slide at room temperature. Slides were monitored closely, and staining was stopped by washing the slides in PBS.

mRNA in situ hybridization with RNA probes was performed as described previously (Deo et al., 2006; Zhang et al., 2012). Restriction enzyme linearized plasmid pUS2-Foxn3 or pUS2-Foxn4 was used as template for the synthesis of UTP-fluorescein or UTP-digoxigenin (Roche) labeled riboprobes by in vitro transcription. For split initiator mRNA in situ hybridization chain reaction (HCR) (Choi et al., 2018), Foxn3 or Foxn4 mRNA was hybridized with a pool of 20 probe pairs (IDT), and detected using matching 72 nt hairpin HCR amplifiers (Molecular Technologies). Detailed protocols for mRNA in situ HCR were described previously (Zhuang et al., 2020). Sequences of in situ HCR probes are in Table S7.

Indirect immunofluorescence analysis was performed on retinas without in situ hybridization or after in situ hybridization. For AP2α antibody staining, retina sections were blocked with 1:400 unlabeled affinity-purified Fab fragment donkey anti-mouse IgG (H+L) (Jackson Immunoresearch, 715-007-003) and 5% donkey serum (Thermofisher) in PBS with 0.05% Triton X-100 (Thermofisher) at 4°C overnight, incubated with mouse anti-AP2α (1:1000, Developmental Studies Hybridoma Bank, 3B5), then incubated with Alexa Fluor 647-conjugated or Alexa Fluor 546-conjugated anti-mouse secondary antibody (both 1:1000, Jackson Immunoresearch, 715-605-051) for 1 h at room temperature. For Vsx2 (Chx10), Ptf1a, Sox2 and GFP staining, tissue sections were blocked with 5% donkey serum, and incubated with sheep anti-Chx10 (1:100, Millipore, AB9016), rabbit anti-Ptf1a (1:800, Li and Edlund, 2001), rabbit anti-Sox2 (1:1000, Millipore, AB5603) or chick anti-GFP (1:500, Aves Labs, GFP-1020) at 4°C overnight, followed by incubation with Alexa 594-conjugated anti-sheep secondary antibody, Alexa 594-conjugated anti-rabbit secondary antibody or Alexa 488-conjugated anti-chick secondary antibody (1:1000, Jackson Immunoresearch, 713-585-147, 711-585-152 and 703-545-155). After antibody staining, Hoechst 33342 (Roche) was used to stain nuclei. For EdU labeling, retinal explants were incubated with EdU for 12 h prior to collection, then the Click-iT EdU Cell Proliferation kit (ThermoFisher) was used to detect EdU, as directed by the manufacturer.

BCIP/NBT staining was imaged with a Leica DSMIRB microscope with a SPOT-RT digital camera or a Zeiss Axiovert microscope with an AmScope MU-2003-BI camera. Fluorescent antibody staining with GFP or in situ hybridization/HCR were imaged using an Olympus FV1000 confocal microscope. Images were analyzed and colors were assigned with FV10-ASW 3.1 Viewer or FIJI/ImageJ. Representative images were cropped, assembled and labeled using Adobe Photoshop.

GFP⁺/antibody-positive cells were analyzed from at least three retinas for each condition. Cells were counted manually from images, and analysis was not masked because, for most experiments, altered cell numbers among the GFP-labeled cells in specific retinal layers were discernible by eye prior to counting. Cell counts were totaled across several sections for each sample. The range for the number of GFP+ cells counted per vector (all samples combined) in each figure: 1516-2587 (Fig. 2), 1898-2511 (Fig. 5), 1529-3029 (Fig. 6), 1962-2070 (Fig. 7I), 648-738 (Fig. 7J), 1527-1606 (Fig. 7K), 1937-3570 (Fig. 8I-L) and 2345-3356 (Fig. 8M). Error bars show standard deviation for all graphs. For qRT-PCR in Fig. 1B, one-tailed t-tests were used. For cell count graphs, two tailed t-tests were used for pairwise comparisons, whereas multiple comparisons were analyzed by one-way ANOVA with Dunnett's multiple comparisons test or Holm-Šídák's multiple comparisons test. Luciferase reporter assays in Fig. 4 were analyzed by two-way ANOVA with Tukey's multiple comparisons test. Graph generation and statistical analysis were performed with GraphPad Prism 9 or Microsoft Excel. The Fig. 3A pie chart was made with DataGraph (Visual Data Tools).

PCR primers are listed in Table S7. RNA was extracted from retinas using TRIzol (ThermoFisher) and treated with RNase-free DNaseI (Promega). For miRNA qRT-PCR, 1 µg of RNA was polyadenylated using a Poly (A) Tailing Kit (Thermofisher). After re-purification with TRIzol reagent, total RNA was reverse-transcribed with 0.5 μg of poly (T) adaptor containing a 5′-end universal tag sequence and SuperScript II Reverse Transcriptase (Thermofisher). Specific miRNA primers, in combination with a universal reverse primer that binds to the universal tag sequence, were used to amplify specific miRNA sequences (Thomson et al., 2006). For mRNA qRT-PCR, 1 µg of RNA was reverse transcribed using Superscript II and random 12mer primers. 2.5 ng of cDNA for miRNA assay, or 20 ng of cDNA for mRNA assay was used per real-time PCR with SYBR Green (Bio-Rad) to detect DNA, using a BioRad iCycler with MyiQ light module. Expression of Foxn3 mRNA was normalized to GAPDH mRNA. Data presented are mean of three RNA samples for each condition, each assayed using triplicate PCR replicates. The 2^−ΔΔCT method was used to calculate fold change.

HEK293 (Takara Bio) and P19 cells tested negative for mycoplasma. In 24-well plates, HEK293 cells were grown in DMEM with 10% fetal bovine serum (FBS), 100 units/ml penicillin and 100 µg/ml streptomycin (all ThermoFisher). For 266 nt UTR reporter assay, 50 ng pUS2-Luc, pUS2-Luc-Foxn3-UTR-WT, pUS2-Luc-Foxn3-UTR-3′-216b-mutant, pUS2-Luc-Foxn3-UTR-5′-216b-mutant or pUS2-Luc-Foxn3-UTR-5′+3′-216b-mutant was co-transfected with 600 ng of pUI6, pUI6-miR-216b or pUI6-miR-216b-mutant for each well. For full-length UTR reporter assay, 50 ng pUS2-Luc, 100 ng pUS2-Luc-Foxn3-UTR-WT-full length or 100 ng pUS2-Luc-Foxn3-UTR-5′+3′-216b-mutant-full length was co-transfected with 250 ng of pUI6, pUI6-miR-216b or pUI6-miR-216b-mutant for each well. 25 ng of pUS2-MT-Nanoluc was included for each well as an internal control for transfection efficiency. Transfections were performed with Lipofectamine 2000 (ThermoFisher) according to the manufacturer's instructions. Reporter activity was assayed 48 h after transfection using the Bright-Glo and Nano-Glo Luciferase Assay kits (Promega). Firefly Luciferase activity was normalized to Nanoluciferase activity to control for transfection efficiency. A FLUOstar OPTIMA plate reader was used to measure luminescence intensity. Relative expression data shown in Fig. 4 are from six independent transfections. Each transfection sample was analyzed by duplicate (technical replicate) luciferase assays: graph data points show mean of the assay replicates.

Mouse P19 cells were grown in MEM-α with 2.5% FBS, 7.5% calf serum and 1:100 penicillin-streptomycin (all ThermoFisher) (Farah et al., 2000). To determine the effectiveness of Foxn3 RNAi constructs in P19 cells, 3 µg of Foxn3 RNAi vectors and 1 µg of pUS2-Puro were co-transfected into one well of a 6-well plate. Transfections were performed with TransIT-LT1 Transfection Reagent (Mirus) according to the manufacturer's instructions. The transfected cells were shifted to medium with 12.5 µg/ml puromycin (Sigma) 8 h after transfection and selected for 16 h. Total RNA was extracted by using TRIzol 48 h after transfection.

Argonaute PAR-CLIP libraries were prepared based on the method described by Hafner et al. (2010), with modifications based in part on tissue processing used for HITS-CLIP (Chi et al., 2009). P0 retinas from CD-1 mice were isolated and grown as explants for 24 h in high-glucose DMEM supplemented with 20% horse serum and 0.1 mM 4-thiouridine (Sigma). Six retinas were used to make one library. Live, intact retina explants were irradiated with 365 nm UV light (∼300 mJ/cm²) using a Stratalinker (Stratagene), then lysed in 300 μl of low-salt lysis buffer [50 mM HEPES (pH 7.5), 50 mM KCl, 0.5% sodium deoxycholate, 0.1% SDS, 0.5% NP-40, 0.5 mM DTT and Protease inhibitor cocktail (Sigma)] with 6 μl of RNaseout (Life Technologies). The lysates were sequentially treated with 12.6 μl of RNase-free DNase (Promega) and 7.5 μl of 1:1330 diluted micrococcal nuclease (11.3 gel units, New England Biolabs: NEB) for 5 min at 37°C. 8.4 μl of 0.5 M EGTA (Sigma) was added to stop the micrococcal nuclease digestion. Lysates were cleared by centrifugation at 16,000 g for 10 min at 4°C. Cleared lysates were incubated with 10 μg of 2A8 anti-Argonaute1-4 antibody (EMD-Millipore MABE56) (Nelson et al., 2007) or G3A1 mouse IgG1 isotype control antibody (Cell Signaling Technologies #5415) bound to Protein G Dynabeads (Life Technologies) and rotated at 4°C overnight. Beads were washed three times in IP wash buffer [50 mM HEPES (pH 7.5), 300 mM KCl, 0.5% NP-40, 0.5 mM DTT and Protease inhibitor cocktail], three times in high-salt wash buffer [50 mM HEPES (pH 7.5), 500 mM KCl, 0.5% NP-40, 0.5 mM DTT and Protease inhibitor cocktail] and resuspended in dephosphorylation buffer [100 mM NaCl, 50 mM Tris-HCl (pH 7.9) and 10 mM MgCl₂]. 4 μl of calf intestinal alkaline phosphatase (NEB) was added to dephosphorylate the RNA for 15 min at 37°C. Beads were washed twice in phosphatase wash buffer [50 mM Tris-HCl (pH 7.5), 20 mM EGTA, 0.5% NP-40] and twice in polynucleotide kinase (PNK) buffer [50 mM Tris-HCl (pH 7.5), 50 mM NaCl, 10 mM MgCl₂]. Beads were resuspended in 80 µl of PNK buffer with 5 mM DTT, and incubated with 8 µl of T4 Polynucleotide Kinase (NEB) and 2 µl of 10 mM ATP for 30 min at 37°C. The beads were washed five times in PNK buffer. The protein-RNA complexes were eluted in 30 µl of 1×Novex Tris-glycine SDS sample buffer (Life Technologies) and separated by 10% SDS-PAGE. A gel slice corresponding to 100 to 150 kDa was isolated (based on size markers in an adjacent lane; Precision Plus Protein Dual Color Standards from Bio-Rad). The gel slice was transferred to a D-Tube Dialyzer Midi Tube (EMD-Millipore). The protein-RNA complexes were electroeluted in 1× NuPage MOPS SDS running buffer (Life Technologies) at 100 V for 2.5 h. The eluate was digested in 50 mM Tris-HCl (pH 7.5), 72 mM NaCl, 6 mM EDTA, 1 mM SDS and 1.2 mg/ml Proteinase K (Sigma) at 55°C for 30 min. RNA was recovered by acidic phenol/chloroform (Life Technologies) extraction and ethanol (Thermofisher) precipitation. The recovered RNA was used to prepare a cDNA library using a TruSeq Small RNA Library Preparation Kit (Illumina) according to the manufacturer's directions, with the following modification: after reverse transcription, the reaction was denatured, separated on a 10% TBE urea gel (Life Technologies); a gel slice corresponding to 80 to 150 nt was isolated (based on size markers in an adjacent lane: low range ssRNA ladder, NEB). The reverse transcribed product in the gel slice was eluted, precipitated by adding 2-propanol, suspended in Tris-HCl buffer, then used in subsequent steps. PAR-CLIP libraries were sequenced on an Illumina HiSeq 2500 (University of Michigan Advanced Genomics Core).

Single-end read PAR-CLIP cDNA sequences were mapped to the mouse mm10 genome using STAR (Dobin and Gingeras, 2016). Sequences mapping to ChrM or rRNA sequences were removed. Multiple reads with identical sequence and length were considered duplicates and only a single read was retained. We considered sequence as well as length to avoid discarding potentially identical cDNA fragments with crosslinks at different positions, although this meant that duplicate reads with distinct errors introduced during the PCR amplification of the library would be considered as separate reads. T to C sequence substitutions relative to the genome were identified with Bambino (Edmonson et al., 2011). Reads with low base quality mismatches or near the ends of reads were filtered, so the count of T to C substitutions and coverage at a specific position is conservative (both crosslink counts and read coverage at crosslink sites exclude some input sequences due to filters). T to C substitutions present at the same genomic position in at least two of the five Argonaute PAR-CLIP libraries, but not detected in the two IgG control libraries, were identified, combined and filtered using scripts, BEDtools (Quinlan, 2014) and an in-house program. Reads with small deletions at the same position as a consistent T to C substitution also were considered crosslinked reads. Consistent crosslinks within 15 nt of a position with a greater number of reads with a consistent crosslink were removed, to create the set of predominant crosslinks. Predominant crosslink sites were annotated using GENCODE VM16 gene/exon annotations (Frankish et al., 2019) and BEDTools. 6 or 7 nt miRNA seed matches with the 5′ base of the match within ±15 nt of a predominant crosslink site (31 nt windows) were identified, based on seeds from miRNAs detected in P0 retinas (Table S2). To identify evolutionary conservation of miRNA seed sites, PhastCons (Siepel et al., 2005) scores were determined for both the 31 nt window and individual seed matches at each predominant crosslink, based on the mm10 MAF60 alignment. In addition, sequences corresponding to the seed sequences from several selected species were retrieved via hg19 MAF100 multiple alignments, using Galaxy (Afgan et al., 2018) at the public server usegalaxy.org. The subset of predominant crosslinks that had a 6 or 7 nt match to the miR-216a and/or miR-216b seed sequences in the genome were selected for further analysis. Plots of read coverage were generated using IGV (Thorvaldsdottir et al., 2013).

To evaluate miRNA seed enrichment near predominant crosslinks (Fig. S5), 3′ UTR exons with crosslinks were identified. Overlapping exons on the same strand were merged, and crosslinks near the ends of exons were filtered to avoid seed matches outside the exons. All seed matches within ±15 nt were counted. If overlapping matches were possible for the same seed match, only the longest match was counted. To determine the expected distribution of miRNA seeds near crosslink positions, a Monte-Carlo method was used. Randomly selected Ts from the same set of exons were used as pseudo-crosslinks (same number of pseudo-crosslinks selected per exon as predominant crosslinks identified for that exon, with same filters on position). Seed matches were counted within ±15 nt of the pseudo-crosslinks, for 100,000 trials. Bonferonni correction was used to adjust P-values for multiple testing.

Retinas from individual E18.5 embryos (Ptf1a knockout or heterozygous) were isolated and pooled. The genotype of embryos was determined by PCR using the primers: Ptf1a-F, CGAGGACAACGTCAGCTATTG; Ptf1a-R, TCTCGCACATTAGCGGCTTG; Cre-F, CATGCTTCATCGTCGGTCC; and Cre-R, GATCATCAGCTACACCAGAG. Total RNA was purified by using TRIzol, then denatured and separated on a 15% TBE urea gel (Life Technologies). Small RNA was recovered from a gel slice corresponding to ∼18-30 nt (based on size markers in an adjacent lane: 20/100 Ladder, IDT). The gel purified small RNA from 1 µg of total RNA was used to prepare a small RNA library using a NEB Next Multiplex Small RNA Library Prep Kit (New England Biolabs), according to the manufacturer's directions, with the following modification: after reverse transcription, the reaction was denatured, separated on a 10% TBE urea gel; a gel slice corresponding to 60 to 80 nt was isolated (based on size markers in an adjacent lane: low range ssRNA ladder, NEB). The reverse transcribed product in the gel slice was eluted, precipitated by adding 2-propanol (Thermofisher), suspended in Tris-HCl buffer and then used in subsequent steps. Small RNA libraries were sequenced on an Illumina HiSeq 2500 (University of Michigan Advanced Genomics Core). Insert sequences had adaptor sequences trimmed and inserts <18 nt or >30 nt were removed. Sequences were then mapped to mature miRNA sequences in miRBase 22 (www.mirbase.org) (Kozomara et al., 2019) using a custom Python script (https://github.com/daimh/microRNA). Unmapped sequences were discarded. Counts for sequences with equivalent mappings to multiple miRNA sequences were divided among the matches. Counts for each mature miRNA were normalized to input library sizes. We also determined the most frequent sequence for each mature miRNA. Small RNA libraries from P0 CD-1 mouse retinas were made and analyzed by a similar protocol, except that the TruSeq Small RNA Library Preparation Kit (Illumina) was used for library preparation and miRNA sequences were mapped to miRBase 21.

Published amacrine cell single cell RNA-seq data (Yan et al., 2020) was downloaded from NCBI GEO (GSE149715) and used to generate a gene expression table with normalization and batch correction using the Seurat package (https://satijalab.org/seurat/) as described previously (Yan et al., 2020). Pearson correlation of Foxn3 expression with the other 18,204 genes passing Seurat filtering (Table S5) was performed using R cor.test.

Genomic DNA was prepared from GFP-positive CD-1 E16 retina explants 48 h after electroporation with either a Foxn3 CRISPR vector or the control vector. Genomic DNA from two pooled explants was purified by using DNeasy Blood & Tissue Kit (Qiagen) for each sample. Q5 high-fidelity DNA polymerase (NEB) was used to PCR amplify genomic DNA at each Cas9 target site. PCR primers are listed in Table S7. PCR products were gel purified, then analyzed by Illumina sequencing at the Massachusetts General Hospital CCIB DNA Core. Sequence reads were analyzed using the online version of CRISPResso2 (Clement et al., 2019).

We thank colleagues who generously provided reagents: Marina Pasca di Magliano and Chris Wright for Ptf1a mutant mice, Helena Edlund for anti-Ptf1a antibody, and Mike Uhler for plasmids. We thank Tom Glaser, Brenda Bass, David Chen, Robert Thompson, Anne Vojtek, Beth Rousseau, Robert Davis and Mike Uhler for helpful discussions. This work used the Vision Research Core at the University of Michigan Kellogg Eye Center, supported by the National Institutes of Health (P30EY007003).

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199484.