De novo genesis of retinal ganglion cells by targeted expression of Klf4 in vivo

The most prevalent cause of irreversible blindness worldwide is degeneration of the neurons that connect the retina to the brain, the retinal ganglion cells (RGCs) (Pascolini and Mariotti, 2012). Thus, restoration of vision in advanced states of diseases such as glaucoma and other optic neuropathies requires the replacement of damaged RGCs. The retina of adult rodents is permissive to transplanted RGCs, which were reported to integrate into the host retinal circuitry and extend axons to expected target areas in the brain (Venugopalan et al., 2016). Although promising, low integration efficiency and rejection by the host pose significant challenges to cell therapy. Concern has also been raised about direct transfer of proteins between donor and host cells that could have misled the identification of donor cell integration, although no material transfer has yet been detected in RGCs (Boudreau-Pinsonneault and Cayouette, 2018; Nickerson et al., 2018; Pearson et al., 2016; Santos-Ferreira et al., 2016; Singh et al., 2016). An alternative approach is to promote de novo genesis of RGCs in adult retina via in vivo reprogramming (Vetter and Hitchcock, 2017).

Although mammalian retinas have only a restricted regeneration potential (Karl and Reh, 2012; Löffler et al., 2015), zebrafish and other teleosts regenerate retinal neurons by Müller glia reprogramming (Goldman, 2014). Recovery relies on the capacity of Müller cells to sense damage, re-enter the cell cycle and produce multipotent progenitors that give birth to all retinal cell types, including RGCs. Although much effort has been directed at strategies to enhance cellular activation and proliferation, the restricted neurogenic potential of Müller glia cells is not as well understood (Jorstad et al., 2017; Pollak et al., 2013; Ueki et al., 2012; Wohl and Reh, 2016; Yao et al., 2018). Investigation of factors capable of modulating the activation of genes necessary for RGC generation could, therefore, help in devising novel strategies for the restoration of vision.

During retinal development, RGC is the first cell type to be generated from a single pool of multipotent progenitors, which give birth to all retinal cell types through continuous restriction of differentiation potential (Rapaport et al., 2004; Turner et al., 1990; Young, 1985). Regulators of RGC competence such as Ikzf1/Ikaros, Atoh7 and microRNAs 125, 9 and let-7 are not sufficient to override fate restriction and promote de novo genesis of RGCs in vivo outside of their developmental window (Elliott et al., 2008; La Torre et al., 2013). In the present study, we investigated the potential of Krüppel-like factor 4 (Klf4) to determine ganglion cell potency in vivo. Klf4 is a key transcriptional regulator of differentiation potential, well known for its abilities as a reprogramming factor (Soufi et al., 2015; Wei et al., 2013). Notably, Klf4 is expressed in RGCs of developing rodents (Moore et al., 2009; Njaine et al., 2014) and is upregulated during endogenous reprograming of chicken Müller glia cells (Todd and Fischer, 2015). However, although Klf4 has been described as an inhibitor of RGC axonogenesis (Moore et al., 2009), its role in the regulation of RGC competence and specification remains poorly explored.

Here, loss-of-function in zebrafish and mice showed that Klf4 is not required for the generation of RGCs during retinal development. Notwithstanding, targeted expression of Klf4 in the retina of neonatal rats was sufficient to reprogram the identity of late retinal progenitor cells (late RPCs) and confer RGC potency. Analysis of gene expression further disclosed the reactivation of molecular pathways involved in RGC migration and differentiation upon Klf4 overexpression. Accordingly, induced RGCs (iRGCs) located at the proper layer in the mature retina and projected axons towards the optic nerve. Thus, our study shows that RGC potency is inducible in late progenitor cells by a single factor and identifies Klf4 as a promising candidate to drive RGC regeneration.

Klf4 is expressed in the developing and mature retina of rat (Njaine et al., 2014), mouse (Fig. S1A) (Moore et al., 2009) and fish (Li et al., 2011), and its expression in RGCs is developmentally regulated. In the rat, Klf4 was detected from embryonic day 17 (E17) to postnatal day 21 (P21) but the content increases at birth, as shown by microarray in acutely purified RGCs (Moore et al., 2009). It has been previously suggested that Klf4 is not essential for RGC development and survival (Fang et al., 2016). Nevertheless, the function of Klf4 in the development of retinal ganglion cells remained largely unexplored; therefore, we first tested for an endogenous role of Klf4 in the generation of RGCs through an early deletion of Klf4.

Using a Cre-loxP system, Klf4 was deleted selectively in progenitor cells of the peripheral retina before the onset of RGC genesis by combining two transgenic mouse lines, α-Cre and Klf4 floxed allele (Fig. 1A). The deletion was confirmed by analysis of gene expression in whole extracts of P0 retinas; qRT-PCR showed a consistent decrease in the expression of Klf4 (Fig. S1B). Klf4^α-Cre animals did not exhibit defects in ocular growth (Fig. S1C-E), retinal lamination (Fig. S1F,G) or optic nerve thickness (Fig. S1H,I). No differences between knockout and wild-type animals were found at P60 in whole-mounted retinae labeled for either a ubiquitous RGC marker (BRN3a; Fig. 1B,C,H) or for a marker of intrinsically photosensitive retinal ganglion cells (OPN4/melanopsin; Fig. 1D,E,H). RGC nerve fascicles were also unaffected by Klf4 loss, as shown by beta-III tubulin (TUBB3) whole-mount staining. (Fig. 1F,G). We extended this analysis by knocking out Klf4 in the whole retina after the onset of ganglion cell genesis (Klf4^Nes-Cre) and testing for spatial vision acuity. Optomotor responses of Klf4^Nes-Cre mice in a virtual-reality test were detected at a visual acuity slightly different from that of control mice (Fig. 1I).

In parallel, we generated a zebrafish mutant line for Klf4 using the CRISPR genome-editing system (Fig. 1J). Deletion of the transcriptional regulatory domains (Fig. S1L) and formation of a premature stop codon (Fig. 1L) were validated by PCR and sequencing of the genomic DNA (Fig. 1K; Fig. S1L) and cDNA (data not shown). Klf4^del/del fish were healthy and fertile (Fig. S1J,K), and developed RGCs that integrated into the ganglion cell layer (GCL) similar to control fish, as shown by labeling whole-mounted 48 hpf embryos for both the RGC marker ZN5 and F-actin (Fig. 1M,N).

The data therefore indicate that Klf4 is not required for the generation of RGCs during normal development in either mouse or zebrafish. Compensation or redundancy by other members of the Klf family (Moore et al., 2009; Njaine et al., 2014) could be responsible for the lack of impact in retinal development and specifically in ganglion cell generation.

The loss-of-function experiments suggested that Klf4 is not necessary for the establishment of RGC potency during development. We next tested whether Klf4 reprogramming abilities could be used to promote de novo genesis of RGCs in vivo. We used late RPCs as a model for progenitors with limited neurogenic potential. Klf4 was cloned from rat retina cDNA under the control of a strong ubiquitous promoter (human ubiquitin C promoter) (Fig. 2A). Neonatal (day of birth, P0) retinas were co-electroporated with pGFP (reporter plasmid) plus either pCTR (empty vector) or pKLF4 (overexpression plasmid) (Fig. 2A). Robust increase in nuclear KLF4 was detected at 39 h after in vitro electroporation (Fig. 2C,D). Among a sample of 285 cells, all GFP⁺ cells located within the neuroblastic layer (NBL) were also positive for KLF4, thus validating the use of the reporter gene encoding GFP to identify the cells submitted to Klf4 overexpression (Fig. S2A). At day 4 after in vitro electroporation (4D), we found GFP⁺ cells at the RGC layer in the pKLF4 group, whereas in the control condition GFP⁺ cells remained confined either to the photoreceptor layer (ONL, outer nuclear layer) or the interneuron layer (INL, inner nuclear layer) (Fig. 2E,F). The finding of GFP⁺ cells at the RGC layer suggested that Klf4 overexpression induced a shift in cell fate compared with controls.

We next electroporated the retinae of P0 rats in vivo following sub-retinal injections of vectors (Fig. 2B) and examined the location and morphology of Klf4-overexpressing cells at later postnatal stages. In control retinas at 10 days post-electroporation (10D), GFP⁺ cell bodies were found only within the inner and outer nuclear layers, whereas the sparse GFP labeling observed at the GCL was associated with extensions from cells located at the INL, most probably Müller glia. The location and morphology of GFP⁺ cells suggest that RPCs transfected in the controls gave birth to photoreceptors, Müller glia, bipolar cells and a few amacrine cells (Fig. 2G-I), consistent with the known differentiation potential of postnatal RPCs. In sharp contrast, approximately 57.4% (±8.4%) of Klf4-overexpressing cells were located at the GCL, IPL and nerve fiber layer, whereas only a few were found in the ONL (Fig. 2H,I). These data are consistent with the induction of amacrine cells and/or RGC fate.

Electroporation of the apical side of the retina at P0 preferentially introduces DNA into both RPCs and precursors of rod photoreceptors, the most numerous neurons of the retina (Matsuda and Cepko, 2004). To distinguish these two populations as the source of the neurons observed mostly at the basal side of the tissue, we targeted Klf4 expression specifically to postmitotic photoreceptor precursors using Cre/loxP-mediated inducible expression under the control of the rhodopsin promoter (Fig. 2J). At 14 days after electroporation, we detected expression of KLF4 and of a Cre-induced RFP reporter (Fig. S2B,C). As expected, in control retinas, most RFP⁺ cell bodies were located in the photoreceptor layer. The few observed in the INL (Fig. 2K; Fig. S2B,C) were most likely ectopic or migrating photoreceptors (Günhan et al., 2003). Similarly, the vast majority of photoreceptor precursors overexpressing Klf4 remained confined to the photoreceptor layer (Fig. 2K). In contrast, overexpression in late RPCs using the Cre/loxP system under the control of the ubiquitous promoter CAG (Fig. S2D-F) reproduced the phenotype of the expression of Klf4 under the control of the ubiquitin C promoter. Therefore, Klf4 overexpression in RPCs generates basal neurons regardless of the promoter, whereas expression of Klf4 in committed rod photoreceptor precursors (rhodopsin promoter-driven Cre) did not shift cell fate (Fig. 2A-I).

These results indicate that progenitor cells overexpressing Klf4 generate neurons located in the basal region of the retina, thus strengthening the hypothesis of a change in the fate of restricted RPCs.

In adult rodents, both regular and displaced ganglion and amacrine cells are found in the GCL and in the inner part of the INL (Linden, 1987; Nadal-Nicolas et al., 2014; Perry and Walker, 1980). Therefore, to investigate the identity of the GFP⁺ cells located at the basal side, we examined a panel of cell type markers among Klf4-induced neurons. Klf4-overexpressing cells located from the INL to the GCL and nerve fiber layer did not express rhodopsin (rod photoreceptor marker) and there was in fact a robust decrease in the number of rhodopsin⁺/GFP⁺ cells (Fig. 3A,B,G). This makes it unlikely that the cells in basal layers are displaced photoreceptors. In addition, the GFP⁺ cells in pKLF4 electroporated retinas were CHX10 (BP marker) negative, meaning that they are unlikely to be bipolar cells (Fig. 3C,D). On the other hand, although Klf4 overexpression did not change the number of calbindin⁺ (AC and HC marker) cells, the majority of calbindin⁺/GFP⁺ cells were located at the basal side of retina where amacrine cells reside (Fig. 3E-H).

Strikingly, there was a large increase in the number of TUBB3⁺ (RGC and amacrine marker) cells among Klf4-overexpressing cells compared with control cells. About half the number of TUBB3⁺ cells were immunolabeled for another marker of RGCs and amacrine cells, NEUN (Rbfox3) (Fig. 4A-D,H-I,K), and/or RBPMS, a specific marker of RGCs (Fig. 4E,F,J,K), which indicates that at least half of the TUBB3⁺/GFP⁺ population are RGCs. Notably, transversal section of the central retina showed that the progeny of Klf4-overexpressing progenitors project TUBB3⁺ axons towards the optic nerve head (Fig. 4G), a unique feature of RGCs. These results strongly suggest that Klf4 overexpression is sufficient to promote the generation of RGCs outside of their developmental window.

Next, we explored how expression of Klf4 in late RPCs promotes the generation of iRGCs. It has been previously reported that overexpression of Klf4 leads cortical progenitors to exit the cell cycle prematurely (Qin et al., 2011; Qin and Zhang, 2012). To assess the proliferation rate of RPCs upon Klf4 overexpression, we pulse-labeled in vitro electroporated retinal explants with bromodeoxyuridine (BrdU) (Fig. 5A). At both 39 h and 48 h of culture, the ratio of BrdU⁺ cells among the GFP⁺ cells was lower in the retinas transfected with pKLF4 than in controls (Fig. 5B-E). In contrast, conditional knockout retinas of neonatal mice did not exhibit changes in BrdU labeling (Fig. S3A-C) nor in the expression of cell cycle regulators (Fig. S3D). These results suggest that RPCs exit the cell cycle prematurely in response to Klf4 overexpression, although the proliferation rate is not affected in the absence of Klf4.

To gain mechanistic insight into the generation of iRGCs by Klf4, we profiled the transcriptome of RPCs overexpressing Klf4. Retinal explants were dissociated 39 h after in vitro electroporation and the transfected population (GFP⁺) was isolated by fluorescence-activated cell sorting (FACS). The transcriptome of the sorted GFP cells from pCTR and pKLF4 groups were examined by microarray (Fig. 5F). The mRNA content was increased in 692 genes and decreased in 96 genes following overexpression of Klf4, as compared with control RPCs (Fig. 5G). In accordance with the characterization of the effects of Klf4 overexpression (Figs 2–4), gene ontology analyses showed enrichment of biological processes such as cell cycle, differentiation, cytoskeleton organization, migration and cell survival (Fig. 5H). We then asked whether the transcriptional profile of Klf4-overexpressing cells resembles any specific retinal cell type. Comparison with the cell type signatures identified previously by single-cell RNA-seq (Macosko et al., 2015) showed that RPCs overexpressing Klf4 were enriched only for ganglion cell-specific genes (Fig. 5I). In fact, validation by qRT-PCR of several genes defined as specific for ganglion cells (Macosko et al., 2015) as well as genes described as regulated in progenitors competent to generate RGCs (Gao et al., 2014) confirmed the upregulation of many factors considered relevant for commitment, specification or terminal differentiation of ganglion cells (Fig. 5J). Certain targets, such as Prdm16, a factor recently described as a marker of an RGC subtype (Groman-Lupa et al., 2017), were detected only upon Klf4 overexpression (data not shown). We also found induction of p21^Cip1 and p57^Kip2 cyclin-dependent kinase inhibitors (Fig. 5J), in agreement with the evidence for premature cell cycle exit in response to Klf4 expression (Fig. 5A-E). In addition, we detected a downregulation of photoreceptor specification genes NrI and Crx (Fig. 5J), consistent with the decrease in generation of rod photoreceptors upon Klf4 expression. Notably, we detected a very strong induction of Atoh7 (Fig. 5J), a master regulator of ganglion cell identity during development (Brown et al., 2001; Brzezinski et al., 2012; Gao et al., 2014; Wang et al., 2001; Yang et al., 2003) and this was accompanied by induction of some of its downstream targets such as Eya2 (Gao et al., 2014). These findings suggest that overexpression of Klf4 suppresses the differentiation program for photoreceptors and confers ganglion cell potency through reactivation of the gene-regulatory network of RGC development.

Although most Klf4-overexpressing late RPCs acquire the molecular and morphological features of RGCs, we found no evidence of the expression of either Brn3a or Brn3b, two important regulators of RGC maturation and survival (Fig. 5G, Fig. 6A-D). This result was surprising because Atoh7 and Eya2, direct regulators of Brn3b (Wang et al., 2001; Yang et al., 2003), were strongly induced by Klf4 (Fig. 5G,J). Notwithstanding, the examination of whole-mount staining of retinas at 30 days after electroporation showed that iRGC cells survived in the absence of Brn3a and Brn3b expression. Moreover, those cells projected axons aligned with endogenous fascicles and extended multiple dendrite-like neurites (Fig. 6E,F). Interestingly, our transcriptome analysis detected the upregulation of 5 of the 36 genes (enrichment of 4.6-fold and P value of 0.004) that require Brn3b to be expressed in RGCs (Mu et al., 2004). In addition, combined overexpression of Klf4 and Brn3b did not increase the efficiency of RGC generation (data not shown). Together, these results suggest that Klf4 overexpression might bypass the requirement of Brn3b to promote ganglion cell survival.

The results of the current study show that although Klf4 is not essential for RGC generation during retinal development in either mouse or zebrafish retinas, it is sufficient to induce de novo genesis of RGCs in vivo outside their developmental window. Late RPCs overexpressing Klf4 exit the cell cycle prematurely, reside mostly in the ganglion cell and inner plexiform layers, contain molecular signatures of RGCs and project axons towards the initial segment of the optic nerve. Notably, cell cycle exit was accompanied by strong upregulation of Atoh7, a master regulator of the transcription network for RGC differentiation. Even though Klf4-induced RGCs did not express Brn3b or Brn3a, described as key regulators of RGC maturation (Badea et al., 2009; Wang et al., 2000), they survived for up to 30 days and their axons integrated in fascicles that reach the optic nerve head. The diagram in Fig. 7 summarizes our main findings.

Klf4 activated a program for differentiation of ganglion cells that included the expression of Atoh7, a transcription factor with a key role in the commitment of progenitor cells to the ganglion cell fate (Brown et al., 2001; Brzezinski et al., 2012; Gao et al., 2014; Wang et al., 2001; Yang et al., 2003). However, other important transcription factors for terminal differentiation were missing. Atoh7 and Eya2, both of which were upregulated upon Klf4 overexpression, have been described as necessary for Brn3b expression (Gao et al., 2014; Wang et al., 2001) and the latter together with Isl1 participates in essential steps of ganglion cell differentiation, such as specification, maturation, axonal projection and survival (Gan et al., 1999; Mu et al., 2008; Qiu et al., 2008; Wang et al., 2000; Wu et al., 2015). Recently, Dlx1 and Dlx2 were also described as crucial for the differentiation of RGCs downstream of Atoh7, both in parallel and cooperating with Brn3b (Zhang et al., 2017). Neither of those transcription factors were upregulated in response to Klf4 overexpression. Our results thus suggest that either alternative programs are activated after Klf4-induced change in the cell fate of RPCs or that iRGCs might not fully mature. The latter hypothesis could explain why GFP⁺/TUBB3⁺ axons were detected only up to the initial segment of the optic nerve. However, an alternative explanation relates to previously described evidence that Klf4 and other members of the Klf family are inhibitors of axon growth (Moore et al., 2011, 2009).

Expression of Atoh7 is crucial but not sufficient for RGC generation in retinogenesis. Likewise, retroviral Atoh7/Math5 overexpression in retinal explants does not bias progenitors towards the RGC fate or induce cell cycle exit (Prasov and Glaser, 2012). Klf4 could potentially interfere in additional machineries that, in combination with Atoh7, succeed in reprogramming restricted progenitors. The role of Klf4 as a pioneer factor might interfere in epigenetic modifications crucial to the progression of retinogenesis, which normally impedes the differentiation program of RGCs at late stages. For example, the ability to interact with methylated DNA was described as crucial for the function of Klf4 (Wan et al., 2017), and one possibility is that Klf4 remodels the chromatin enabling Atoh7 transcriptional activity. In addition, the analysis of histone modifications in retinal cells at various stages of development suggests that RGC genes are silenced by histone modifications as retinogenesis proceeds (Ueno et al., 2016). Therefore, to reprogram the potency of late progenitors, and thus of Müller glia cells or other endogenous cell sources, ideally one would need an individual factor capable of remodeling the state of activation of genes necessary for RGC differentiation, a role that Klf4 might play in the molecular mechanism responsible for the generation of iRGCs in the rat retina (Iwafuchi-Doi and Zaret, 2014; Soufi et al., 2015).

Curiously, although Klf4 is sufficient to induce ganglion cell generation and differentiation, Klf4-deficient retinas develop normally. In addition, we did not detect the previously reported (Fang et al., 2016) changes in the thickness of axon bundles in the nerve fiber layer. This discrepancy is probably a result of differences in the expression pattern of the Cre recombinases used to knock out Klf4 (Marquardt et al., 2001; Rowan and Cepko, 2004). Furthermore, it is possible that other members of the same family, already described in the retina (Moore et al., 2009; Njaine et al., 2014), could have redundant or compensatory roles in the absence of Klf4. In fact, it has been shown that Klf4 is crucial for self-renewal and pluripotency of stem cells, but Klf2 and Klf5 are redundant in these roles (Jiang et al., 2008).

The design of new strategies to increase the proliferative and neurogenic potential of mammalian Müller glia is one of the areas of interest in the development of regenerative therapies for retinal degenerative diseases (Vetter and Hitchcock, 2017; Yao et al., 2018). Attempts to apply in vivo the transcription factor Ascl1, either alone or in combination with other factors such as miRNA or drugs that impact chromatin organization, interfered with the neurogenic potential of Müller glia. In these previous studies, Müller glia generated mostly bipolar cells, both through transdifferentiation as well as from Müller glia-derived progenitors, but did not generate ganglion cells (Jorstad et al., 2017; Pollak et al., 2013; Ueki et al., 2015). Another recent study also derived rod photoreceptors from Müller glia (Yao et al., 2018). Late RPCs and Müller glia in the rodent retina exhibit a similar gene expression profile and limited neurogenic potential (Cepko, 1999; Blackshaw et al., 2004; Ooto et al., 2004; Jadhav et al., 2009; Nelson et al., 2012; He et al., 2012; Karl and Reh, 2012; Löffler et al., 2015). Thus, our finding that Klf4 is sufficient to induce de novo genesis of retinal ganglion cells from late RPCs encourages further investigation of its reprogramming abilities in Müller glia.

In conclusion, we found that a single transcription factor can interfere with the differentiation program of late RPCs towards the generation of ganglion cells. These findings raise further questions related to mechanisms activated during the change in cell fate: (1) How does Klf4 activate a molecular program for the generation of RGCs outside of their developmental window? (2) How can iRGCs differentiate, survive and project axons without Brn3b? (3) Is there any step missing for the iRGCs to project to superior targets, or does continuous expression of Klf4 represses the further outgrowth of iRGC axons? Nonetheless, we suggest that Klf4 is a relevant candidate for the investigation of new strategies to replenish retinal ganglion cells that are affected in severe retinal degeneration such as glaucoma.

All experiments with rodents were planned according to international rules and were approved by the Ethics Committee on Animal Experimentation of the Health Sciences Center of the Federal University of Rio de Janeiro (CEUA/CCS/UFRJ, protocol 065/15). Mice and rats of either sex were used in the study. Previous evidence indicates that retinal development is very similar in these two species (Martins and Pearson, 2008; Rapaport et al., 2004; Turner et al., 1990). Neonatal (P0) Lister hooded rats (RRID:RGD_2312466) were used for in vitro and in vivo electroporation because of the advantage of easier access for in vivo DNA injection in the eyes of rats rather than mice at the same developmental stage. The Klf4-loxP (MGI:2183917; RRID:MGI:2664982), α-Cre (MGI:3052661; RRID:IMSR_EM:00756) and Nestin-cre (MGI:2176173; RRID:IMSR_JAX:003771) transgenic mice have been described previously (Katz et al., 2002; Marquardt et al., 2001; Nakamura et al., 2006; Tronche et al., 1999). Mouse strains were maintained on a C57Bl/6 background (RRID:IMSR_JAX:000664). Klf4loxP/loxP mice were crossed with Nestin::Cre or α::Cre mice to produce control (Klf4 loxP/loxP:Cre^−/−, referred to as Klf4^CTR) and conditional knockout (Klf4loxP/loxP:Cre^+/−, referred to as Klf4^αCre or Klf4^NesCre) littermates. Cre transgenics were maintained in heterozygosity. All animals were housed in a temperature controlled (21-23°C) environment under a 12 h light/dark cycle.

Wild-type AB (RRID:ZIRC_ZL1) zebrafish were maintained and bred at 26°C. Embryos were raised in E3 medium at 28°C as previously described (Kimmel et al., 1995). These experiments were carried out exclusively in the laboratory of Dr Caren Norden (MPI-CBG, Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany) and approved by EU directive 2011/63/EU as well as the German Animal Welfare Act. ZiFiT Targeter software was used to design two guide RNAs (gRNAs) to target exon 3 of the Klf4 gene: (A) GGAGAAACGTCGCTGATATC (chromosome 21, 477167-477148) and (B) GGCTCAATCTTGGCCATGAC (chromosome 21, 477362 - 477343). Single-stranded sense (TAGGN18-20) and antisense (AAACN18-20) oligonucleotides complementary to these gRNAs were synthesized (Integrated DNA Technologies) and inserted into an in vitro transcription vector (DR274) containing the structural portion of gRNA that interacts with the Cas9 nuclease (Hwang et al., 2013). The linearized pT3TS-nlsCas9nls vector (Jao et al., 2013) and gRNA DNAs were purified with the PCR Clean-Up System Protocol (Promega). The gRNAs and Cas9 nuclease were transcribed in vitro with MEGAshortscript kit (Ambion) and mMESSAGE mMACHINE kit (Ambion), respectively. The transcribed gRNAs were purified using mirVana microRNA isolation kit (Ambion). For generation of the Klf4 knockout zebrafish line, the two gRNAs (A and B, 50 µg each) were injected together with 150 µg Cas9 RNA and 50 µg GFP RNA in the yolk sac of embryos at one cell stage. Identification of zebrafish with deletion in the germ line was done by genotyping outcrossed offspring with oligonucleotides (sense 5′-GCGTGCATCCTTTAATACTGGC-3′ and antisense 5′-ATGCTCATCCGGAGACAGTG-3′). Adult animals were identified through caudal fin genotyping. Heterozygous animals with a 367 bp deletion of exon 3 and the consequent frameshift in the Klf4 gene were mated for the generation of homozygous knockout embryos.

Plasmids CAG::LNL-DsRed (Addgene: #13769), CAG::CreERT2 (Addgene: #14797) and Rho::Cre (Addgene #13779) (Matsuda and Cepko, 2007) were purchased. Plasmids Ub::CST and Ub::GFP (Matsuda and Cepko, 2004) were kindly provided by Dr Michael A. Dyer (St Jude Hospital, Tennessee, USA). KLF4 cDNA was amplified by PCR with the primers 5′-GAATTCACCATGGCAAGGCAGCCACCTGGCGAGTCT-3′ and 5′-GCGGCCGCTTAAAAGTGCCTCTTCATGTGTAA-3′. These fragments were inserted into the Ub::CST and CAG::LNL-DsRed vectors to generate the Ub::KLF4 and CAG::LNL-KLF4 plasmids, respectively.

Electroporation in vitro and in vivo was carried out in rats as described previously (Donovan et al., 2006; Matsuda and Cepko, 2007). Briefly, for in vitro electroporation, P0 Lister hooded rat retinas were dissected and transferred to an electroporation chamber (Nepagene, model CUY520P5) filled with DNA solution (1 μg/μl in Hanks' balanced salt solution). Five 25 V square pulses were delivered for 50 ms at 950 ms intervals (Nepagene, CUY21SC). Electroporated retinas were cut into 1 mm² pieces (total of eight explants per P0 retina) and cultured at 37°C in Erlenmeyer flasks under orbital agitation with Dulbecco’s modified Eagle medium (DMEM, Gibco) with 20 mM HEPES (Sigma) and supplemented with 5% fetal calf serum (Cultilab), 2 mM L-glutamine (Sigma) and 10 U/ml penicillin, 100 μg/ml streptomycin (Gibco) at pH 7.4 for a maximum of 4 days. Littermates were used as control for in vitro electroporation. For in vivo electroporation, P0 Lister hooded rats were anesthetized on ice. DNA mix (1 μl of a 5.0-6.5 μg/μl solution) containing 0.1% Fast Green dye (Sigma) was injected into the subretinal space with a Hamilton syringe equipped with a 33 G blunt end needle. Five 99 V pulses were delivered for 50 ms at 950 ms intervals, using a forceps-type electrode (Nepagene, CUY650P7) with Neurgel (Spes Medica).

Eyes from adult mice were processed and measured as described (Martins et al., 2008). Eyes were fixed in 4% phosphate-buffered paraformaldehyde. The axial length and two coronal axes (dorso-ventral and medial-lateral) of each eye were measured with a digital pachymeter and the volume of the eye was calculated by applying the formula 4/3π*x*y*z. Representative images were captured with an AxioCamERc 5s camera attached to a Zeiss stereoscope.

Visuomotor reflex was assessed as described previously (Prusky et al., 2004). Briefly, the mouse was positioned on the suspended platform of an optomotry apparatus (OptoMotry; Cerebral Mechanics). Vertical lines moving at 12° per second were projected onto the screens and the spatial frequency of 0.1-0.492 c/d was adjusted using the OptoMotry software (Cerebral Mechanics). The visuomotor response was defined as a reflex movement of the head in the same direction and velocity as the moving grating. The mean of the maximum spatial frequency attained for either eye was taken as the visual threshold of each mouse.

For the assessment of cell proliferation in vivo, newborn Klf4^CTR and Klf4^αCre mice received an intraperitoneal injection of 150 mg/kg BrdU (Sigma). The retinas were fixed 2 h after injection, sectioned and subjected to immunofluorescence as described below. To assess proliferation in vitro, electroporated retinal explants from P0 rats were incubated with BrdU at a final concentration of 10 µM in the last 2 h of culture. The explants cultured for 39 h were treated with 0.05 mg/ml of trypsin (Gibco) and 0.1 mg/ml of DNAse (Sigma) at 37°C, followed by mechanical dissociation. The dissociated cells were plated on glass slides treated with poly-L-lysine (200 µg/ml, Sigma) before fixation with 4% paraformaldehyde in 0.2 M PBS (Sigma). The cells were immediately subjected to immunofluorescence analysis as described previously (Njaine et al., 2010). The explants cultured for 48 h were fixed, sectioned and subjected to immunofluorescence as described below.

For immunostaining of sections, retinal explants and whole eyes were fixed by immersion in 4% paraformaldehyde in PBS for 2 h and 16 h, respectively. Serial transversal sections of cryoprotected material (10 μm) were mounted on microscope slides treated with either poly-L-lysine (300 μg/ml) or silane (6%, Sigma). In gain-of-function experiments, regions with electroporated GFP⁺ cells were immunolabeled as reported previously (Cavalheiro et al., 2017). All images were acquired using structural illumination (Imager M2 Apotome, Zeiss). The microscopes were operated with the Zen Blue (Zeiss) software. Apotome images were acquired with a Plan-Apochromat 20×/0.8 or EC Plan-Neofluar 40×/1.3 oil DIC objective and AxioCam MRm camera. In gain-of-function experiments, image acquisition was aimed at recording the strong Klf4 labeling, and in such conditions endogenous expression is not detected. In loss-of-function experiments, the most peripheral region of the retina was analyzed; in gain-of-function experiments, the electroporated region was examined. GFP⁺ cells from at least three transversal sections were quantified for each retinal explant analyzed. Two lines were drawn perpendicular to the outer nuclear layer to define the electroporated region to be quantified.

Immunostaining of flat-mounted retinas was carried out as reported previously (Petrs-Silva et al., 2011). Blocking solution was 5% horse serum (Vector Laboratories) in 0.5% Triton X-100; primary antibodies were incubated for at least 2 days. All images were acquired in a confocal microscope (LSM 510 META, Zeiss, using a Plan-Neofluar 20× or 40×/1.3 oil DIC objective and photomultiplier tubes. The system was operated with the Zen Black (Zeiss) software. All primary antibodies used in immunofluorescence, and basic information on the protocols are described in Table S1.

Whole zebrafish embryos fixed in 4% paraformaldehyde in PBS at pH 7.4 were permeabilized with 0.25% trypsin in PBS, immersed in a blocking solution (10% normal goat serum, 1% bovine serum albumin and 0.2% Triton X-100 in 10 mM PBS) overnight at 4°C and incubated with primary antibody for 2 days at 4°C (see Table S1). The embryos were then incubated for 2 days at 4°C with the appropriate secondary antibody and 4-6-diamidino-2-phenylindole (DAPI; Thermo Fisher). Phalloidin-Alexa Fluor 488 (200 U/ml; Thermo Fisher) was used at 1:50 dilution. Images were acquired using spinning disk confocal SD4 (SD4-Andor Revolution WD Borealis Mosaic) composed by scan head Yokogawa CSU-W1 (4000 rpm, 50 µm pinhole disc) (Andor Technology). The Olympus UPLSAPO objective 60×/1.3 SIL and Andor iXon 888 Ultra with Fringe suppression were used. The microscope was operated through the Andor iQ software version 3.4.1.

Retinal explants cultured for 39 h after in vitro electroporation were dissociated by incubation with 0.05 mg/ml trypsin and 0.1 mg/ml DNAse and by mechanical homogenization at 37°C for 15 min. Dissociated cells were filtered using Cell Strainer 70 µm (Falcon). Transfected (GFP-positive cells) were separated from non-transfected cells (GFP-negative cells) through FACS (Moflo, DakoCytomation) at the Flow Cytometry Facility (Instituto de Microbiologia Paulo de Góes, UFRJ, Brazil). The sorted cells were lysed with Trizol (Invitrogen) and RNA was extracted with a Trizol-Rneasy microkit (Qiagen) hybrid protocol. RNA quality and concentration were analyzed using the Agilent Bioanalyzer (RNA Nano 6000 chip, Agilent Technologies) and only samples with RIN above 8 were used. For qRT-PCR analyses, cDNA was synthetized using 50 ng of RNA with Quantitect Reverse Transcription Kit (Qiagen) according to the manufacturer's protocol.

For analysis of global gene expression, RNA extracts from cells isolated by FACS were processed at the Gene Expression Facility of the Max-Planck Institute for Molecular Cell Biology and Genetics according to standard Agilent protocols (SurePrint G3 Rat GE 8×60K, Agilent One-Color Microarray-Based Gene Expression Analysis). Four samples from independent experiments were used for each experimental group. The microarray raw data files were normalized by the quantile method using GeneSpring 13.0 software. The normalized data was filtered by percentile (20-100). Statistical analyses of the raw data were performed in the GeneSpring 13.0 software. Comparison between groups was done by unpaired t-test; values of P≤0.05 from probes with fold changes greater than 1.5 were considered significant. The Benjamin–Hochberg false discovery rate correction was applied for the control of false positives. Gene ontology analysis was performed with Qiagen's Ingenuity Pathway Analysis (IPA) using the SurePrint G3 Rat GE 8×60K microarray as the reference data set. The list of significantly upregulated genes in the pKlf4 group was used for comparison with the molecular signature of each retinal cell type identified by single-cell RNA-seq (Macosko et al., 2015). The enrichment rate (r) was calculated by the ratio of Ke (overlap) to the expected ratio (K) based on the reference list (rat genome). Enrichment for cell-type specific genes over the total transcriptome was analyzed by hypergeometric test; P values below 0.05 were considered significant. Raw microarray data with statistical analysis have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO series accession number GSE123297 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE123297).

Whole retinas from mice were lysed using Trizol (Invitrogen) and extraction of RNA was carried out according to the manufacturer's protocol. Genomic DNA was eliminated with a DNA-free Kit (Ambion). Nanodrop was used to define the purity and amount of RNA in samples. A First-Strand cDNA Synthesis Kit (GE Healthcare) was used to synthetize cDNA.

qRT-PCR of cDNA samples from whole mouse retinas and from cells isolated by FACS were carried out in triplicate in an ABI7500 machine (Applied Biosystems) using TaqMan probes synthesized with 5′-FAM and 3′-BHQ or Power SYBR Green PCR Master Mix (Applied Biosystems), respectively (Tables S2 and S3). The primers were designed using Primer Quest (Integrated DNA Technologies SciTools) or Primer Blast software. PCR product identity was confirmed by melting-point analysis. The Δ-ΔCt method was used to determine relative expression normalized to Gapdh or Erk2 mRNA levels. The ΔCt mean of all groups was defined as the calibrator (Hellemans et al., 2007; Rocha-Martins et al., 2012). For qRT-PCR of GFP⁺ cells isolated by FACS, three to four biological replicates were analyzed in technical triplicates.

For conditional knockout mice experiments, at least three histological sections from each animal or an area of at least 1.20 mm² from the periphery of each flat-mounted retina were used for quantification. The number of biological replicates (n) was at least three. For optomotor reflex experiments, the number of mice required to achieve significance was not calculate beforehand and was based on numbers used in similar behavioral studies. For in vitro electroporation experiments, retinal explants from different animals were pooled prior to cultivation. GFP⁺ cells from at least three transversal sections were quantified from each retinal explant analyzed, a total of 300-1000 GFP⁺ cells per explant. The number of experimental replicates (n) was at least three. For the in vivo electroporation experiments, 200-1000 GFP⁺ cells were counted for each animal analyzed.

All the results were represented as mean±s.e.m. Statistical comparisons between two experimental groups were performed using Student's t-test (two-tailed, unpaired); P values below 0.05 were considered significant. Computations assumed that populations exhibit the same scatter (s.d.) and Gaussian distribution. Multiple t-tests were corrected using the Holm–Sidak method. For all analyses, Prism software version 6.0 (GraphPad Software) was used.

The authors thank Dr Jeffrey Goldberg and the University of Miami for providing the Klf4 floxed mice. We also thank Sylvia Kaufmann from the Max Planck Institute of Molecular Cell Biology and Genetics who helped M.R.-M. with CRISPR; Franciane de Queiroz Ferreira, Mariana Anjo Barbosa, Matheus Sampaio Moreira, Thais Porto Marinho, José Nilson dos Santos, Gildo de Brito Souza and José Francisco Tibúrcio for technical support at the Neurogenesis lab at Instituto de Biofísica Carlos Chagas Filho, Universidade Federal do Rio de Janeiro; and the technicians Leonardo de Araújo Leal and Bruno Maia Santos from the Flow Cytometry Facility at the Instituto de Microbiologia Paulo de Góes, Universidade Federal do Rio de Janeiro.