KDM3A-mediated demethylation of histone H3 lysine 9 facilitates the chromatin binding of Neurog2 during neurogenesis

Neurogenesis is a complex process through which a variety of neurons are generated to populate the central and peripheral nervous systems. Extensive studies in recent decades have revealed that proneural factors in the bHLH family are evolutionarily conserved regulators of neurogenesis in invertebrates and vertebrates (Bertrand et al., 2002). One of the bHLH factors that plays pivotal roles in neurogenesis is neurogenin 2 (Neurog2). Neurog2, a vertebrate homolog of the Drosophila atonal gene, is expressed in various domains of the central and peripheral nervous systems (Simmons et al., 2001; Sommer et al., 1996). Neurog2 is a master regulator of neurogenesis in multiple locations, including the neocortex, retina and spinal cord (Heng et al., 2008; Lee and Pfaff, 2003; Mattar et al., 2008; Mizuguchi et al., 2001; Novitch et al., 2001). A common theme for the function of Neurog2 is that it regulates formation of neuronal precursors and initiates the expression of a panel of genes, including NeuroD1, that are necessary for neuronal commitment (Castro and Guillemot, 2011; Wilkinson et al., 2013). The Xenopus proneural gene neurogenin-related 1 (Xngnr1) is orthologous to mammalian Neurog2 and plays a hierarchical role in initiating primary neuron formation through multiple neuronal genes, including Neurod1 and Myt1 (Bellefroid et al., 1996; Ma et al., 1996). It is well established that Neurog2, via its basic domain, recognizes its targets through binding the E-box DNA regulatory elements with certain degrees of preference. The HLH domain of Neurog2 has been shown to directly or indirectly associate with multiple nuclear factors, controlling context-dependent functions of Neurog2 during neuronal fate specification and diversification (Castro and Guillemot, 2011; Lee and Pfaff, 2003; Wilkinson et al., 2013). Recently, it has been shown that the phosphorylation state of a conserved serine/threonine residue in the second helical domain regulates DNA-binding activities of Neurog2 and Ascl1 (Quan et al., 2016).

The crucial role of Neurog2 in neuronal development has also been demonstrated in direct neuronal reprogramming of non-neural cells in vivo and in vitro (Aravantinou-Fatorou et al., 2015; Blanchard et al., 2015; Grande et al., 2013; Liu et al., 2016, 2013; Masserdotti et al., 2015; Thoma et al., 2012). It is therefore of central importance to understand how Neurog2 transactivates its neuronal target genes.

Neuronal specification and differentiation during embryonic development involves multiple steps of cell fate transition, during which histone post-translational modifications (PTMs) are of crucial importance (Maze et al., 2013). The ability of Neurog2 to transdifferentiate terminally differentiated non-neural cells into neurons raises the issue of how Neurog2 reinitiates the expression of neuronal genes that are silenced, at least in part, by repressive histone PTMs (Vierbuchen and Wernig, 2012). Previous studies have indicated that Neurog2 is able to recruit the p300/CBP histone acetyltransferase during neurogenesis in the spinal cord, promoting the deposition of active histone marks at neuronal genes (Koyano-Nakagawa et al., 1999; Lee et al., 2009). Neurog2 has also been shown to cooperate with the Brg1-containing chromatin-remodeling complex during retinal neurogenesis (Seo et al., 2005). These findings highlight the ability of Neurog2 to modulate chromatin states suitable for its target activation. However, it remains unclear whether Neurog2 is able to actively remove repressive histone PTMs.

Both di- and tri-methylated histone H3 lysine 9 (H3K9me2/3) are associated with heterochromatin at pericentromeric and telomeric regions, and repress developmental genes (Black et al., 2012; Greer and Shi, 2012). The methylation state of H3K9 is controlled by two groups of enzymes with antagonistic activities. Of interest, lysine demethylases (KDMs) in the Jmjd1 subfamily, including KDM3A (Jmjd1A/JHDM2A), KDM3B (Jmjd1B/JHDM2B) and KDM3C (Jmjd1C), have been characterized to specifically remove dimethylation of H3K9 (Loh et al., 2007; Yamane et al., 2006). Genetic studies have uncovered crucial roles for KDM3A in several physiological processes, such as spermatogenesis, sexual determination and glucose metabolism (Cheng et al., 2014; Herzog et al., 2012; Kuroki et al., 2013; Liu et al., 2010; Okada et al., 2007; Tateishi et al., 2009). Multiple lines of evidence indicate that removal of H3K9 dimethylation is involved in neuronal differentiation, in addition to many aspects of cellular function. For example, KDM3C (JMJD1C) has been reported to repress neural differentiation of human embryonic stem cells (Wang et al., 2014a) and is required to sustain the expression of miR-302, a crucial epigenetic regulator of pluripotency and neural differentiation (Wang et al., 2014a). In addition, a novel splice variant of LSD1/KDM1A has been recently found to specifically demethylate H3K9me1/2 (Laurent et al., 2015). Depletion of this LSD1 isoform increases H3K9me2 levels at its target promoters and impairs neuronal differentiation in vitro (Laurent et al., 2015). Furthermore, KDM7A and PHF8 (aka KDM7B), both of which can demethylate H3K9me1/2 and inhibit other types of histone methylation, have also been implicated in the neural fate control (Huang et al., 2010a,b; Qi et al., 2010; Tsukada et al., 2010).

Studies using the Xenopus laevis model have advanced the understanding of mechanisms underlying neural fate specification and determination. The molecular function of Spemann's organizer in neural induction was first uncovered in Xenopus (Hemmati-Brivanlou et al., 1994; Sasai et al., 1994; Smith and Harland, 1992). NeuroD1 was initially demonstrated to be able to convert the Xenopus naïve ectoderm into neurons (Lee et al., 1995). In the present study, we have performed loss-of-function studies for the Xenopus KDM3A and have found evidence that KDM3A is essential for the chromatin accessibility of Neurog2 during neuronal transcription. The Neurog2-regulated H3K9 demethylation via association with KDM3A ensures proper neuronal transcription. Our current findings also reveal that KDM3A is selectively required for Neurog2- but not for Ascl1-initiated neuronal transcription.

To delineate the function of KDM3A in early vertebrate development, we investigated the expression pattern of kdm3a mRNA in early Xenopus embryos. Whole-mount in situ hybridization indicated that kdm3a mRNA was expressed throughout early Xenopus development. During the cleavage to gastrula stages, kdm3a was detected in the animal hemisphere (Fig. 1A,B), from which the neural ectoderm emerges. During the early neurula stage, kdm3a was ubiquitously expressed with a slight enrichment in the CNS (Fig. 1C). At later stages of development, kdm3a was more obviously enriched in the CNS (Fig. 1D,E), particularly in the anterior parts of the CNS, including the eye fields. These data suggest that KDM3A is expressed in the right place at the right time to regulate neural development in Xenopus.

To study the function of KDM3A in neural development, we analyzed the effects of KDM3A depletion using a translation-blocking morpholino oligo (MO) that recognizes the start codon-spanning sequences of kdm3a mRNA (Fig. S1A). A 5 bp mismatch mutant of this MO (cMO) was used as a control for MO specificity (Fig. S1A). The efficacy of KDM3A MO (3A MO) was verified using western blot analysis with an antibody directed against KDM3A. Microinjection of 3A MO (80 ng) into both cells at the two-cell stage resulted in a marked decrease of KDM3A protein from the mid-gastrula (stage 11) to the neural tube closure stage (stage 18) (Fig. 1F), confirming that the KDM3A MO is an efficient tool for assessing the function of KDM3A during crucial stages of neural development in Xenopus. By contrast, microinjection of the same dose of cMO did not affect KDM3A expression (Fig. 1G). To determine whether KDM3A is involved in neural development, we took advantage of the animal cap assay. The naïve ectoderm dissected from the animal pole of mid-blastula stage embryos is pluripotent and can be directed into neural stem cells or neuronal fates, depending on signals provided. We compared the responsiveness of 80 ng cMO- or 3A MO-injected animal caps with the BMP antagonist noggin and other neural inducers. Using qPCR, we found that depletion of KDM3A did not discernably affect the ability of noggin to induce the expression of neural stem cell marker genes in the SoxB1 class (Fig. S1B), suggesting that KDM3A is dispensable for the BMP antagonist-mediated neural induction. However KDM3A depletion diminished the ability of FGF8a to activate neuronal genes, including neurog2, neurod1, ebf2 and tubb2b (Fig. 1H). Interestingly, the ability of FGF8a to induce expression of the neural stem cell marker sox2 and the pan-mesodermal gene brachyury (t) was not discernably affected (Fig. 1H). Additional analysis indicated that KDM3A was not required for the activin-induced mesoderm differentiation (Fig. S1C). These data collectively suggest that KDM3A is specifically required for the neurogenic potential of naïve ectoderm but not the neural induction per se. To elaborate on this, we analyzed the responsiveness of the naïve ectoderm to the overexpressed Neurog2 and Ascl1. First, we verified that ectopic expression of Myc-tagged Neurog2 or Ascl1 (MT-Neurog2, MT-Ascl1) was not affected by co-injection of either cMO or KDM3A MO (Fig. S1D,E). Intriguingly, using qPCR, we found that depletion of KDM3A differentially affected abilities of Neurog2 and Ascl1 to induce target gene expression. Ectopic Ascl1 exhibited comparable activities in cMO- or 3A MO-injected explants to stimulate the expression of several neuronal genes (Fig. 1I). By contrast, Neurog2 failed to activate neuronal genes, including neurod1, neurod4, ebf2 and tubb2b when KDM3A was depleted (Fig. 1J), suggesting that KDM3A is specifically necessary for the neuronal transcription by Neurog2 but not for Ascl1-induced neuronal gene expression. Injecting the MO-resistant kdm3a mRNA restored the expression of KDM3A protein (Fig. 1G) and the ability of FGF8a and Neurog2 to induce target gene expression (Fig. 1H,J), suggesting that the KDM3A MO-mediated depletion effects are specific.

Primary neurogenesis in Xenopus begins as soon as the neural plate has emerged, lasting up to the larva stages (Thuret et al., 2015). Early-born neurons from primary neurogenesis are essential for adoptive behaviors of the larva stage animals. Primary neurons can be conveniently identified by virtue of characteristic expression of neuronal marker genes, including neurod1, myt1 and tubb2b (also known as neural tubulin), among others during neural fold to neural tube stages (Bellefroid et al., 1996; Ma et al., 1996; Moody et al., 1996; Talikka et al., 2002). To verify whether KDM3A plays a role in primary neurogenesis, we sequentially injected 40 ng cMO or 3A MO and mRNA encoding KDM3A (300 pg), together with 200 pg β-galactosidase (β-gal) mRNA into one cell at the two-cell stage (Fig. 2A). Embryos at stage 24/25 were then subjected to red-Gal staining followed by whole-mount in situ hybridization for tubb2b (Fig. 2B). We found that the expression of tubb2b in KDM3A morphants was markedly reduced (Fig. 2B,B′). Injecting the MO-resistant kdm3a mRNA partially rescued the tubb2b expression in KDM3A morphants (Fig. 2B,B′), indicating that KDM3A is required for the primary neuron formation. Supporting this notion, KDM3A depletion also diminished the expression of other neuronal differentiation-related genes, including neurod1, myt1 (a pan-neuronal marker), HB9 (a marker of motor neurons), islet1 (a marker of both motor and sensory neurons) and xHox11L2 (a marker of Rohon-Beard neurons) when examined at the neural tube-closing stage (Fig. S2A). In line with the notion that KDM3A is dispensable for the transition of naïve ectoderm to neural stem/progenitor cells, we found that expression of the neural stem/progenitor markers such as sox3 and pax6 remained nearly normal in KDM3A morphants at the neural tube closing stage (Fig. S2B). Additional examination of marker genes at the late neurula stage (stage 18) indicated that KDM3A depletion did not discernibly affect the retina field formation or the anteroposterior pattern formation of neural tube (Fig. S2B). Moreover, the expression of representative markers for the neural plate border, neural crest specification and epidermis did not seem to be discernibly altered by KDM3A depletion (Fig. S2C). We noticed that expression patterns of msx1, zic1 and xk81b2 appeared somewhat differently in KDM3A morphants (Fig. S2C). This was likely due to the neural tube closure delay caused by KDM3A depletion (Fig. S2D), which could be partially rescued by re-injection of the MO-resistant kdm3a mRNA (Fig. S2D). These data together confirmed the notion that KDM3A is not required for neural induction per se but is necessary for neuronal differentiation in vivo. Importantly, reinjection of the 3A MO-resistant kdm3a mRNA could partially rescue the expression of tubb2b in KDM3A morphants (Fig. 2B,B′) and the neural tube closure (Fig. S2D), strongly suggesting that the observed neuronal defects are specifically due to KDM3A depletion.

To verify whether KDM3A is required for primary neuron induction by proneural factors in the bHLH family, we sequentially injected 40 ng 3A MO and neurog2 or ascl1 mRNA (50 pg for each) together with β-gal (200 pg) into a single cell at the two-cell stage and examined the expression of tubb2b during the neural fold stages (stage 16/17). We observed that KDM3A depletion abolished the ability of Neurog2 to induce ectopic expression of tubb2b (Fig. 2C,C′). By contrast, Ascl1 remained able to induce ectopic neuronal formation in the KDM3A-depleted embryos (Fig. 2C,C′). Furthermore, KDM3A depletion also drastically diminished the abilities of Xenopus neurogenin 1 (Xneurog1), neurogenin3 (Xneurog3) and mouse Neurog2 to induce ectopic neuron formation (Fig S3A and B), indicating that KDM3A is required for all these tested neurogenin factors to induce neuronal formation.

To additionally address the specificity of KDM3A depletion, we designed a second MO (KDM3A MO2) recognizing the kdm3a 5′UTR region (Fig. S3C). Microinjection of KDM3A MO2 (40 and 80 ng) at the two-cell stage also markedly depleted the KDM3A expression in late gastrulae (Fig. S3D). Microinjection of the KDM3A MO2 reproduced the effects of 3A MO, including abolishing the ability of ectopic Neurog2 (50 pg mRNA) to induce expression of several neuronal genes in the naïve ectoderm explants (Fig. S3E), diminishing the expression of tubb2b (Fig. S3F), slowing down neural tube closure and causing a defect in the eye formation (Fig. S3F). Injecting MO-resistant kdm3a mRNA (300 pg) partially rescued the tubb2b expression, the neural tube closure delay and eye formation in the 3A MO2-injected embryos (Fig. S3F), reaffirming that KDM3A is specifically required for neurogenesis in Xenopus. Taken together, we conclude that KDM3A is essential for primary neurogenesis in Xenopus.

It has been established that Neurog2 plays a hierarchical role in initiating the formation of neuronal precursors and subsequent differentiation (Ma et al., 1996). We therefore investigate the expression of Neurog2 and its targets in KDM3A morphants. Both whole-mount in situ hybridization and qPCR analyses revealed that Neurog2 was activated in nearly normal levels in KDM3A-depleted embryos at late neurula stages (Fig. 3A,B), suggesting that KDM3A depletion mainly affects the differentiation capacity of neuronal precursors. Consistent with this notion, we found that ectopic Neurog2 was able to induce ectopic expression of neurod1 (Fig. 3C) and myt1 (Fig. S4A) in 80 ng cMO-injected embryos, but failed to do so in 80 ng KDM3A MO-injected embryos (Fig. 3C, Fig. S4A). Moreover, KDM3A depletion also markedly blocked the ability of murine Neurog2 to induce the expression of tubb2b (Fig. S3A) and neurod1 (Fig. S3B). Because Neurog2 has been shown to integrate multiple environmental signals during neuronal differentiation (Bellefroid et al., 1996; Ma et al., 1996), we also tested depletion effects of KDM3A on retinoic acid (RA) and FGF8. We found that the activities of RA and FGF8 to induce ectopic neuron formation were abolished in embryos depleted of KDM3A (Fig. S4B; data not shown). These data together indicate that KDM3A is required for primary neurogenesis through Neurog2-mediated neuronal differentiation programs.

It has been shown that, during neurogenesis in the mouse spinal cord, Neurog2 is able to recruit CBP and establish the acetylated H3 marks at neuronal genes (Lee et al., 2009). Neurog2 has also been shown to promote H3K4me3 marks at neuronal genes (Lee et al., 2009). In order to determine whether KDM3A depletion affects H3K27ac or H3K4me3 deposition at the Neurog2-responsive neuronal genes, we performed anti-H3K27ac and anti-H3K4me3 ChIP followed by qPCR analyses. To do this, we sequentially injected 80 ng cMO or KDM3A MO at the one-cell stage and then 100 pg neurog2 mRNA into both cells at the two-cell stage. ChIP-qPCR analyses were performed when embryos developed to the late neurula stage (stage 18). We verified that ectopic Neurog2 significantly increased the deposition of H3K27ac (Fig. 3D) and H3K4me3 (Fig. 3E) at the neurod1 and tubb2b promoters. KDM3A depletion alone did not discernably alter H3K27ac or H3K4me3 marks at the neurod1 and tubb2b promoters (Fig. 3D,E). However, KDM3A depletion markedly reduced the ectopic Neurog2-promoted deposition of H3K27ac (Fig. 3D) and H3K4me3 (Fig. 3E) at the examined neuronal loci. These data together confirm that KDM3A is required for Neurog2 to transactivate its target genes.

Our analyses thus far suggest that KDM3A might function together with Neurog2 to promote neuronal differentiation. To determine whether Xenopus Neurog2 interacts with KDM3A, we performed co-immunoprecipitation using embryonic lysates overexpressing epitope tagged-Neurog2 and KDM3A, or their deletion mutants (Fig. 4A). We found that 6MT-KDM3A could be detected in the immunocomplex precipitated by anti-HA-Neurog2 antibody (Fig. 4B), indicating that Neurog2 is able to associate with KDM3A. The C-terminal domain deletion mutant of Neurog2 (Neurog2-ΔC) failed to precipitate KDM3A; however, the N-terminal domain deletion mutant of Neurog2 (Neurog2-ΔN) remained able to bind KDM3A (Fig. 4C), revealing that the C-terminal domain of Neurog2 is necessary for the interaction. Furthermore, we found that deletion of the N-terminal domain of KDM3A (KDM3A-ΔN) abolished the interaction with Neurog2 (Fig. 4D), indicating that the N-terminal domain of KDM3A is involved in the interaction. For comparison, we also investigated whether Ascl1 is able to associate with KDM3A. Using the same experimental designs, we did not find evidence that Ascl1 interacts with KDM3A (data not shown). We then asked whether KDM3A interacts with mammalian Neurog2. To answer this, we used the neuronal cells derived from mouse NE-4C cells treated with retinoic acid (RA) and the neural stem cells derived from the H9 human stem cell line (see the supplementary Materials and Methods for detailed experimental procedures). We found that endogenous KDM3A and Neurog2 were detected in the same immunocomplex brought down by anti-KDM3A antibody (Fig. S4C), suggesting KDM3A interacts with mammalian Neurog2. These data collective indicate that KDM3A is a candidate co-factor for the proneural factors in the neurogenin subfamily.

The association between Neurog2 and KDM3A suggest that Neurog2 and KDM3A might cooperate to activate neuronal genes. To test this notion, we depleted Neurog2 using a previously characterized Neurog2 MO (Klisch et al., 2006) with or without additional depletion of KDM3A. Injection of Neurog2 MO alone at a lower dose (30 ng) only moderately reduced the expression of myt1 and/or tubb2b (Fig. 4E). However, co-injection of 30 ng Neurog2 MO and 30 ng KDM3A MO together could significantly inhibit the expression of these neuronal genes (Fig. 4E), validating the notion that Neurog2 collaborates with KDM3A during neuronal gene activation. Together, we conclude that KDM3A physically and functionally interacts with Neurog2 during primary neurogenesis in Xenopus.

Recent progress has provided evidence that Neurog2 and Ascl1 initiate non-overlapping genetic programs during the conversion of non-neural cells into functional neurons (Masserdotti et al., 2015; Raposo et al., 2015; Wapinski et al., 2013). Our current findings suggest that KDM3A is selectively required for Neurog2 but not for Ascl1 to initiate neuronal programs. NeuroD1, a crucial regulator of neurogenesis, is a direct target of Neurog2 (Mattar et al., 2004; Miyata et al., 1999; Ozen et al., 2007; Sommer et al., 1996; Talikka et al., 2002). We exploited this fact further for better understanding how KDM3A is specifically involved in Neurog2-regulated neuronal differentiation. While overexpression of Neurog2 but not Ascl1 was able to induce neurod1 expression in the naïve ectoderm (Fig. S5A,B), depletion of Ascl1, using a previously characterized splice-blocking MO (Gao et al., 2016), markedly inhibited the expression of neuronal genes, including tubb2b, myt1 and neurod1 (Fig. S5C,C′). This is in accordance with our previous finding that Ascl1 is a competence factor for neurogenesis in Xenopus (Gao et al., 2016). Interestingly, our ChIP-qPCR analyses revealed that overexpressed Ascl1, like Neurog2 (Fig. 5A), could also bind the upstream regulatory region of neurod1 (Fig. 5B). Ectopic Ascl1 only slightly enhanced the KMD3A binding around the −263 bp position relative to the transcription start site (TSS) of neurod1 promoter (Fig. 5C). However, ectopic Ascl1 failed to reduce H3K9me2 levels from the neurod1 promoter region that has been examined (Fig. 5D). By contrast, Neurog2 markedly enhanced the KDM3A binding at all examined locations (Fig. 5C) and concurrently reduced the deposition of H3K9me2 at these locations of the neurod1 promoter (Fig. 5D), demonstrating that Neurog2 specifically facilitates recruitment of KDM3A to neurod1. Conversely, depletion of Neurog2 reduced the binding of KDM3A (Fig. 5E) and concomitantly increased the expression of H3K9me2 at neurod1 (Fig. 5F). By contrast, depletion of Ascl1 did not discernably change levels of the H3K9me2 marks at the neurod1 promoter (Fig. S5D). We noticed that Ascl1 depletion resulted in a noticeable but insignificant increase of H3K9me2 at the −36 position relative to the TSS of tubb2b promoter (Fig. S5E). Results from additional ChIP-qPCR experiments indicated that overexpressed Ascl1 did not enhance the recruitment of KDM3A to the −36 position of tubb2b promoter (Fig. S5F). It is thus possible that other H3K9 demethylase(s) are involved in the Ascl1-induced expression of tubb2b. These analyses indicate that KDM3A is a chromatin regulator specifically used by Neurog2 during the activation of neurod1 expression. To further strengthen this notion, we examined overexpression effects of Neurog2 and Ascl1 on the promoter region of myt1, which is a common target of Neurog2 and Ascl1 (Seo et al., 2007; Vasconcelos et al., 2016). We verified through ChIP-qPCR that both ectopic Neurog2 and Ascl1 were able to bind the promoter sequence of myt1 (Fig. S5G). However, only overexpressed Neurog2 was able to increase the binding of KDM3A on the promoter region of myt1 (Fig. S5H) and concomitantly decrease the H3K9me2 marks on the promoter region of myt1, but Ascl1 failed to do so (Fig. S5I). These data collectively reaffirmed the conclusion that KDM3A is specifically used by Neurog2 to decrease the H3K9me2 marks during neuronal activation.

We additionally verified that Neurog2 could also recruit KDM3A to the neuronal differentiation gene tubb2b. First, we confirmed that Neurog2 was able to directly transactivate tubb2b through dual-luciferase reporter assays (Fig S6A,B). ChIP-qPCR revealed that overexpressed Neurog2 bound the promoter region of tubb2b (Fig. S6C). Importantly, ectopic Neurog2 significantly decreased the deposition of H3K9me2 at the promoter region of tubb2b in cMO-injected embryos; however, it failed to do so in the KDM3A MO-injected embryos (Fig. S6D). Conversely, depletion of Neurog2 significantly reduced the binding of KDM3A (Fig. S6E) and increased the expression of H3K9me2 at the promoter region of tubb2b (Fig. S6F), affirming that Neurog2 is able to recruit KDM3A and demethylate H3K9me2 at its targets that are important for neuronal differentiation. Taken together, we conclude that Neurog2 is able to recruit KDM3A to its targets and promote H3K9me2 demethylation around its binding sites.

Having established a role for Neurog2 in recruitment of KDM3A to its target neuronal genes, we then sought to investigate how KDM3A regulates the transactivity of Neurog2. Consistent with reports from others (Yamane et al., 2006; Loh et al., 2007), we found that overexpression of KDM3A by injecting a high dose of kdm3a mRNA (2 ng) in early Xenopus embryos decreased the expression level of soluble H3K9me2 and to some extent that of H3K9me1 (Fig. 6A). Interestingly, we observed through a series of ChIP-qPCR analyses that ectopic expression of KDM3A, but not of an enzymatically dead mutant (KDM3A H1130Y) (Yamane et al., 2006), reduced levels of the H3K9me2 marks at the neurod1 and tubb2b promoters (Fig. S7A,B), suggesting that KDM3A overexpression is able to modulate H3K9me2 at these neuronal genes. However, simple overexpression of KDM3A was not sufficient for the establishment of H3K27ac at neurod1 or tubb2b (Fig. S7C,D). Consistently, whole-mount in situ hybridization revealed that ectopic KDM3A did not discernably change the expression of tubb2b or neurod1 (Fig. 6B), suggesting KDM3A alone is not sufficient to activate neuronal genes. These observations are instead consistent with the notion that Neurog2 is rate limiting in the activation of these neuronal genes. Depletion of KDM3A resulted in a detectable increase of chromatin H3K9me2, as revealed by western blot analysis using a high-salt buffer with chloric acid extracted-chromatin protein from cMO- and 3A MO-injected embryos at the late neurula stage (Fig. 6C) (Tien et al., 2015). This was further supported by our ChIP-qPCR analyses of the promoter regions of neurod1 (Fig. 6D) and tubb2b (Fig. S7E). Moreover, KDM3A depletion drastically reduced the ability of ectopic Neurog2 to decrease the expression of H3K9me2 at the promoters of neurod1 (Fig. 6E) and tubb2b (Fig. S6D). To further determine how KDM3A regulates the activity of Neurog2, we investigated using ChIP-qPCR the binding of Neurog2 at its targets upon KDM3A depletion. We found that KDM3A depletion diminished the binding of Neurog2 at promoters of neurod1 (Fig. 6F) and tubb2b (Fig. S7F). Taken together, we conclude that KDM3A-mediated demethylation of H3K9me2 is necessary for effective binding of Neurog2 at its target sites.

KDM3A, a well-characterized H3K9me2/1 demethylase, has been implicated in spermatogenesis and metabolism (Cheng et al., 2014; Herzog et al., 2012; Kuroki et al., 2013; Liu et al., 2010; Okada et al., 2007; Tateishi et al., 2009). In this study, we provided evidence that KDM3A is a novel co-factor for Neurog2 during neuronal precursor formation and differentiation in Xenopus. KDM3A interacts with Neurog2; the interaction involves the N-terminal sequence of KDM3A and the C-terminal domain of Neurog2. Depletion of KDM3A diminishes the abilities of Neurog2 to reduce the repressive H3K9me2 and to increase active histone marks in promoter regions of neuronal genes. Our study reveals that KDM3A is essential for primary neurogenesis in Xenopus. Our findings advance the understanding of the molecular mechanisms underlying proneural factor-initiated cell fate commitment and differentiation.

Primary neurogenesis in Xenopus provides an excellent in vivo system for delineating the molecular mechanisms underlying neural fate specification and determination (Ma et al., 1996; Thuret et al., 2015). Expression of the proneural genes in the neurogenin subfamily during CNS development is regulated by environmental signals, including Shh, RA, FGF and the Notch/Delta lateral inhibition system (Chitnis, 1999; Koyano-Nakagawa et al., 1999; Moody et al., 1996). Neurog2 appears to be activated on time and its mediolateral expression pattern remains nearly normal in KDM3A morphants (Fig. 3A), suggesting that KDM3A is not required for the initial selection of neuronal precursors. Instead, KDM3A depletion diminishes the abilities of Neurog2, FGF8 and RA to induce neuronal gene expression (Fig. 1H,J, Fig. S4B), suggesting that the KDM3A-mediated histone H3K9me2/1 demethylation is essential for neuronal differentiation downstream of Neurog2 and other environmental cues. By analyzing the activation of several representative neuronal genes, our primary goal was to understand how Neurog2 and KDM3A jointly initiate neuronal precursor differentiation. To achieve this, Ascl1, a proneural factor in the achaete-scute subfamily, was compared in several respects. Interestingly, we found that Ascl1 remains capable of inducing expression of neurod4 and tubb2b when KDM3A is depleted (Figs 1I and 2C), suggesting that KDM3A is selectively required for Neurog2-initiated neuronal differentiation. Supporting this notion, we provided evidence that Neurog2 interacts with KDM3A. Under the same experimental conditions, an association between Ascl1 and KDM3A was not detected (data not shown). Furthermore, we demonstrated that Neurog2 is required for KDM3A recruitment at neuronal gene loci. Neurog2 promotes H3K9me2 demethylation depending on the presence of KDM3A. Conversely, ectopic Ascl1 fails to modulate H3K9me2 marks at the promoters of neurod1 or myt1, despite its detectable binding to these neuronal promoters. It has been demonstrated that Neurog2 and Ascl1 trigger in vitro neuronal transdifferentiation through non-overlapping genetic programs (Masserdotti et al., 2015; Raposo et al., 2015; Wapinski et al., 2013). Our current findings provide at least one additional angle to account the different transactivities of Neurog2 and Ascl1, i.e. Neurog2 but not Ascl1 is able to overcome the repressive effects of H3K9me2 on neuronal genes due to a specific interaction between Neurog2 and KDM3A. More-detailed structure-function analyses in the future will help to better understand how Neurog2 selectively uses KDM3A and modulates chromatin environment on site. We have previously shown that the pre-neurula stage expression of Ascl1 is also required for neurogenic potential and FGF8a-induced neuronal differentiation (Gao et al., 2016). Our current analyses indicate that Ascl1 transactivates its target genes likely in a KDM3A-independent manner. Because functional neurons induced by Neurog2 and/or Ascl1 hold great promise for regenerative medicine (Chen et al., 2015), further investigations on comparing the difference between Neurog2 and Ascl1 using the Xenopus model will continue to provide developmental principle-inspired improvements in neuronal transdifferentiation protocols (Ang and Wernig, 2014; Pang et al., 2011; Yang et al., 2011). Interestingly, our preliminary investigation indicates that KDM3A is upregulated during induced neuronal differentiation of human induced pluripotent stem cells.

We found that KDM3A depletion increases H3K9me2 marks and reduces the binding of Neurog2 to its targets, suggesting that Neurog2 might not be able to stably bind closed chromatin. This points to an additional difference between Neurog2 and Ascl1, because Ascl1 has been shown to function as a pioneer transcription factor that is able to access closed chromatin (Vierbuchen et al., 2010; Wapinski et al., 2013). How does Neurog2 access and transactivate its target neuronal genes? We found that both H3K9me2 in the soluble pool and H3K9me2 associated with the neuronal gene loci are inversely correlated with KDM3A expression levels, implying that KDM3A may function through both Neurog2-independent and Neurog2-dependent fashion (Fig. 6G). The activity of KDM3A in both of these respects might be balanced by the H3K9-specific methylases such as G9a/GLP (Fuks et al., 2003; Smallwood et al., 2007; Snowden et al., 2002), allowing a suboptimal recruitment of Neurog2 at its targets. Recently, it has been proposed that histone turnover at chromatin is essential for neuronal transcription (Lee et al., 2009; Maze et al., 2013, 2015; Seo et al., 2005). Therefore, the enriched expression of KDM3A in the developing CNS may be involved in maintaining neuronal gene loci in primed states (Fig. 6G, lower panel), which facilitates the access to chromatin by Neurog2. The association of Neurog2 with KDM3A at regulatory regions of neuronal genes enhances the Neurog2 residence at its targets (Fig. 6G, upper panel), engendering the determinate activation of neuronal genes. This hypothesis is consistent with our data demonstrating that KDM3A overexpression alone is not sufficient to activate neuronal genes, and the notion that Neurog2 is rate limiting for neuronal gene activation. The full activation of neuronal genes by Neurog2 likely also needs the recruitment of other types of histone-modifying enzymes, such as CBP and/or chromatin remodeling complexes (Lee et al., 2009; Maze et al., 2013; Seo et al., 2005). While KDM3A depletion alone does not affect the basal levels of H3K4me3 or H3K27ac at neuronal genes, we found that depletion of KDM3A compromises the ability of Neurog2 to establish H3K27ac and H3K4me3 at neuronal gene loci (Fig. 3D,E). Further studies are needed to determine in greater detail how Neurog2 coordinates removal of repressive histone marks on one hand and decoration of active histone marks on the other. Our current findings open a novel venue for deeper understanding the molecular mechanisms underlying the neuronal precursor determination triggered by proneural factors in the bHLH family.

H3K9 methylation (H3K9me2/3) is not only involved in safeguarding stable terminal differentiation, but also in regulating developmental gene expression in the right place at right time (Black et al., 2012; Soshnev et al., 2016). Our current analyses posit a specific role for KDM3A in neuronal differentiation. However, this does not exclude the possibility that H3K9me2 may function in broader developmental contexts. KDM3A is a maternal gene, the function of which remains unclear. Interestingly, it has been recently reported from others that chromatic H3K9me2/3 is not discernably detected until mid-gastrula stages (Wang et al., 2014b). Injection of KDM3A MO at the two-cell stage could reduce KDM3A protein with a concomitant increase of H3K9me2 from the mid-gastrula stage onwards (Fig. 1F; data not shown), implying that KDM3A is required for maintaining a proper level of expression of H3K9me2 during the early phases of embryonic cell fate allocation in Xenopus. However, the expression of neural stem cell markers did not seem to be affected in KDM3A morphants or in the naïve ectoderm explants treated with Noggin. Besides, the anteroposterior pattern formation of the CNS was retained in KDM3A morphants. Hence, we conclude that KDM3A and H3K9me2 are dispensable for the transition of naïve ectoderm to neural ectoderm. Instead, H3K9me2 might play a part in establishing chromatin states that subject neuronal genes to precise regulation by spatiotemporally controlled developmental cue (Black et al., 2012; Maze et al., 2013). It will be necessary to determine in future the genome-wide occupancy of KDM3A and/or H3K9me2 for better understanding the roles for KDM3A and H3K9me2 in regulating neural development. Moreover, different H3K9 demethylases in the JmjD1 (KDM3) subfamily may play different roles in neural development. Supporting this notion, KDM3C has been reported to repress neural differentiation of human embryonic stem cells (Wang et al., 2014a), suggesting that different KDM3 members might be used by different transcription factors during neural fate specification and differentiation. Our current findings strongly suggest that the joint actions of Neurog2 and KDM3A are essential for lowering the barrier of H3K9me2 against gene activation in the neuronal precursors (Fig. 6G).

Last but not least, KDM4A (also known as JmjD2A), a demethylase for H3K9me3, has been implicated in the neural crest specification in chick and frogs (Matsukawa et al., 2015; Strobl-Mazzulla et al., 2010). We found that KDM3A depletion does not discernably affect the expression of representative neural plate border genes and neural crest specification factors, indicating that KDM3A and perhaps H3K9me2 demethylation are dispensable for neural crest specification. Little attention has been paid to the difference between H3K9me2 and H3K9me3 in cell fate specification in literature (Black et al., 2012; Jamieson et al., 2016). Our current findings and the findings for KDM4A from others imply that the valence of H3K9 methylation may be differentially implicated in cell fate specification and determination.

The Xenopus laevis frogs were purchased from NASCO (Wisconsin) and reared in a temperature-controlled facility. Embryos were obtained through artificial insemination following standard procedures (Sive et al., 2000). For microinjection, fertilized eggs were dejellied at the one-cell stage using 2% cysteine (pH 8.0) followed by thorough washing in 0.1×MMR (Marc's Modified Ranger) and then transferred into 4% Ficoll/0.3×MMR. The injection scheme of MOs and mRNAs is indicated in figures and/or figure legends where appropriated. Animal caps were dissected embryos with desired treatments at the stage 8.5 and cultured to desired stages, as specified in the relevant figure legends. For RT-PCR analyses, 10 animal caps and/or two whole embryos for each treatment group were prepared and lysed with Tris-HCl/SDS/Proteinase K (pH 7.4) buffer supplemented with RNase inhibitors. The total RNA was then extracted using acidic phenol:chloroform, precipitated using 70% isopropanol and resuspended with 20 µl nuclease-free water. cDNA was made with 1 µl of RNA for each sample using SuperScript III reverse transcriptase Kit (Invitrogen #18080093). qPCR analyses were performed using the LightCycler 2.0 (Roche) following the manufacturer's instructions. All qPCR primers were optimized to obtain melting curves that can separate reverse transcriptase (RT)-containing samples from RT-minus and H₂O control samples following procedures provided by the manufacturer's instructions for the LightCycler (Roche) and described previously (Sive et al., 2000; Xanthos et al., 2002). All RT-qPCR data presented in Figs 1H-J, 3B and 4E were obtained through combining two independent biological repeats (i.e. repeated with embryos from two different females) with three technical replicates for each repeat and shown as mean±s.e.m. (standard error of mean). The two-tailed Student's t-test was performed to assess significance.

Embryos at desired stages were fixed with MEMFA followed by the standard procedures of whole-mount in situ hybridization (Sive et al., 2000). In cases where β-gal was co-injected for lineage tracing, as shown in Figs 2, 3C and 6B, the red-Gal staining was carried out before the whole-mount in situ hybridization procedures. All in situ hybridized embryos were then bleached with H₂O₂ for 12 h under fluorescent lights. Information of all cDNA constructs used for in situ probe labeling can be found in Table S1. The dual-luciferase reporter assay was performed following procedures provided in the supplementary Materials and Methods.

Embryos overexpressing Xenopus Neurog2 and/or KDM3A were lysed in RIPA buffer followed by co-immunoprepicitation (CoIP) procedures as described previously (Ding et al., 2013; Zhang et al., 2014). The immunocomplex was then differentiated in denaturing PAGE followed by western blotting analyses as described previously (Ding et al., 2013; Zhang et al., 2014). Chromatic histone H3 detection shown in Fig. 6C was performed by extracting the nuclear fraction of histone from cMO or KDM3A MO-injected embryos using trichloric acid. Titration of antibodies was as follows: 0.5 µg desired antibodies were used all CoIP experiments; anti-histone antibodies were used at 1:500 dilution; anti-Myc and anti-HA were used at 1:1000 dilution. All Co-IP and western blot experiments were repeated at least twice using embryos from different females. Information about all antibodies used in this study can be found in Table S2. Details of the experimental procedures for the CoIP of endogenous Neurog2 and KDM3A as shown in Fig. S4C can be found in the supplementary Materials and Methods.

ChIP-qPCRs were carried out following protocols described previously (Blythe et al., 2009; Gao et al., 2016). Detailed procedures can be found in the supplementary Materials and Methods. Fifty embryos were used for each sample for the ChIP procedures. All ChIP-qPCR experiments were repeated twice using different batches of embryos. Three technical replicates were made in each independent repeat. Results from six repeats were then combined and assessed for significance by performing the two-tailed Student's t-test. The results were then presented as mean±s.e.m. Information about all antibodies used for ChIP can be found in Table S2. For ChIP experiments, 1 µg of anti-Myc, anti-KDM3A, anti-H3K9me2, anti-H3K4me3 and anti-H3K27ac was used.

Sequence information for morpholino antisense oligos, PCR primers, ChIP-qPCR primers can be found in Table S3. All PCR-related raw data can be found in the Table S4.

We thank Dr B. Y. Mao (Kunming Institute of Zoology, Chinese Academy of Sciences, Kunming, China) for providing cDNA constructs.