Co-option of the PRDM14-CBFA2T complex from motor neurons to pluripotent cells during vertebrate evolution

Germ cell specification in multicellular organisms is subdivided into two modes: preformation and epigenesis (Johnson et al., 2003; Extavour and Akam, 2003). In different species, germ cell specification can occur via either of two modes. During preformation, cytoplasmic germ cell determinants, known as the germ plasm, are asymmetrically localised in oocytes and predetermine the cellular competency for germ cell formation, e.g. in flies, worms, teleost fishes and frogs. In contrast, germ cells are segregated from pluripotent cells, which can differentiate into both germ cells and somatic cells, by receiving induction signals from surrounding tissues in epigenesis, e.g. in salamanders, mice and humans (Chatfield et al., 2014; Ohinata et al., 2009; Sasaki et al., 2015). The phylogenetic distribution of preformation and epigenesis in metazoans implies that epigenesis is an ancestral mode of germ cell formation and that preformation evolved repeatedly and independently (Johnson et al., 2003). Because the establishment of pluripotent cells in early embryos is essential for germ cell development during epigenesis, understanding the emergence and diversity of gene regulatory networks (GRNs) needed for pluripotency is important for determining the ancestral mechanisms of germ cell formation.

Pluripotent mouse stem cells, otherwise known as mouse embryonic stem cells (mESCs), were first established by culturing the inner cell mass of a blastocyst at embryonic day (E) 3.5 (Evans and Kaufman, 1981), and have served as a useful model for characterising the pluripotency of GRNs. The self-renewal of mESCs is regulated by a core circuitry of transcription factors: POU5F1 (also known as OCT3 or OCT4), NANOG, SOX2 and the KLF family of proteins (Niwa, 2014). Sox2 and the Klf family of genes are essential for maintaining ESC pluripotency; however, these factors are also expressed in somatic lineages in mice (Avilion et al., 2003; Jiang et al., 2008; Kuo et al., 1997). In contrast, Pou5f1 has been shown to act as a master regulator of pluripotency, and its expression is only associated with pluripotent cells and germline cells in mice (Yeom et al., 1996; Yamaguchi et al., 2005). Pou5f1 and its paralogue Pou5f3 might have emerged from a common ancestor of gnathostomes, while Pou5f1 was eliminated from teleosts, frogs and birds (Frankenberg et al., 2014). In contrast, the genome of the axolotl encodes both Pou5f1 and Pou5f3, and these genes are associated with embryonic pluripotent cells and germ cells (Tapia et al., 2012). Conditional knockout (KO) of Pou5f1 leads to apoptosis of primordial germ cells (PGCs) in mice (Kehler et al., 2004), whereas knock down of Pou5f3.3 expands the early number of PGCs in Xenopus embryos (Butler et al., 2018). These studies suggest that Pou5f1 is required for germ cell formation in species exploiting epigenesis, whereas Pou5f1 can be abrogated in species exploiting preformation (Johnson and Alberio, 2015).

Previously, we have shown that PR domain-containing 14 (PRDM14) is expressed in pluripotent cells and germ cells in mice, and is required for the establishment of primordial germ cells (Yamaji et al., 2008). Prdm14-deficient mESCs expand only in the presence of inhibitors of the extracellular signal-regulated kinase (ERK) and glycogen synthase kinase 3 β (GSK3β) (together known as ‘2i’) plus leukaemia inhibitory factor (LIF), but not in serum plus LIF (Yamaji et al., 2013). ESCs maintained in 2i plus LIF are transcriptionally similar to peri-implantation epiblast cells at around E4.5 and are called naïve ground state pluripotent stem cells (Nichols and Smith, 2009). The autocrine of fibroblast growth factor 4 (FGF4) activates ERK signalling and destabilises the transcriptional networks required for pluripotency (Kunath et al., 2007). PRDM14 represses Fgfr1/2 expression to maintain ESC pluripotency under the serum-plus-LIF condition (Yamaji et al., 2013). Prdm14 expression is closely linked to naïve pluripotent cells, the inner cell mass (ICM) and ESCs in mice, and its expression is shut off during the early transition period from naïve to primed pluripotency (Yamaji et al., 2008). Epiblast-like cells (EpiLCs), which are derived from ESCs, are transcriptionally similar to post-implantation epiblast cells at around E5.5, and are commonly called primed pluripotent cells (Hayashi et al., 2011). Prolonged Prdm14 expression disturbs the spontaneous differentiation of mouse and human ESCs induced by embryoid body formation (Okashita et al., 2015; Tsuneyoshi et al., 2008). Furthermore, Prdm14 overexpression in epiblast-like cells with primed pluripotency induces their conversion into ESCs with naïve pluripotency mediated by the re-establishment of transcriptional networks for naïve pluripotency (Okashita et al., 2016). Contrary to the PRDM14 functions in pluripotent cells in mammals, Prdm14 is expressed in primary motor neurons but not in pluripotent cells, and its expression is required for primary motor neuron maturation in zebrafish embryos (Liu et al., 2012). Here, we have compared the functions and expression patterns of PRDM14 orthologues in order to understand their evolutionary conservation and the functional diversity among different deuterostome lineages.

The phylogenetic distribution of Prdm genes has been well documented in a previous study (Vervoort et al., 2016). To construct a phylogenetic tree of Prdm14 orthologues among metazoa, we first searched for Prdm14 orthologues in the genomes of diverse animals using the web tool aLeaves (Kuraku et al., 2013). Unexpectedly, we identified Prdm14 orthologues in the genome of the sea anemone Nematostella vectensis, which belongs to the non-bilaterian phylum Cnidaria. In the superphylum Lophotrochozoa, we identified Prdm14 orthologues in the genomes of Capitella teleta, Helobdella robusta, Lottia gigantea and Schistosoma mansoni, but not in the planarian Schmidtea mediterranea. In contrast, we did not identify Prdm14 orthologues in Caenorhabditis elegans and Drosophila melanogaster that belonged to the superphylum Ecdysozoa. Recent phylogenetic analysis has suggested that the Priapulid worm (Priapulida phylum) is the earliest branching ecdysozoan that, together with Kinorhyncha and Loricifera phyla, forms a sister group to nematodes (C. elegans) and arthropods (Drosophila) (Martín-Durán and Hejnol, 2015). Interestingly, the Priapulus caudatus genome was found to encode a putative Prdm14 orthologue, implying that Prdm14 might have been lost during evolution within the ecdysozoans. Among the deuterostomes, sea urchins in Echinodermata and acorn worms in Hemichordata have Prdm14 orthologues in their genomes, but Prdm14 orthologues were not identified in currently available Ciona intestinalis genome assemblies. Next, we generated a phylogenetic tree of Prdm14 orthologues (Fig. 1A and Fig. S1) and analysed the synteny of genomic regions containing the identified Prdm14 orthologues (Fig. 1B). The genomic localisation of Prdm14 between Ncoa2 and Slco5a1 is conserved among vertebrates, except for zebrafish – a teleost fish. Interestingly, in the genomes of medaka and spotted gar (which belong to a different subgroup of Teleostei), Prdm14 is located between Ncoa2 and Slco5a1, as described above for non-teleost vertebrates, suggesting that the gene order in the genomic region downstream of Prdm14 might have been rearranged recently in the otocephalan lineage leading to zebrafish. In the amphioxus genome, Prdm14 and Slco5a1 are localised in different scaffolds of the genome.

Mouse Prdm14 (mPrdm14) is indispensable for the self-renewal of ESCs cultured with serum plus LIF (Yamaji et al., 2013). Two inhibitors of MAPK/ERK kinase and GSK3β (referred to as ‘2i’) in combination with LIF overcome the loss of self-renewal activity in mPrdm14 KO ESCs. These findings indicate that mPRDM14 is required for the suppression of ESC differentiation in the presence of serum plus LIF. To assess the capacity of PRDM14 orthologues to sustain the self-renewal of ESCs cultured with serum plus LIF, we established four cell lines: mPrdm14 KO ESCs harbouring an empty vector, sea urchin PRDM14 (sPRDM14), zebrafish PRDM14 (zPRDM14) or mPRDM14 under the 2i plus LIF condition (Fig. 2A,B). We first compared the expression levels of the pluripotency-associated genes Dnmt3a, Dnmt3b and Dnmt3l, and the differentiation-associated genes using quantitative RT-PCR (qRT-PCR) analysis in wild-type ESCs and in mPrdm14 KO ESCs expressing sPrdm14, Prdm14 or mPrdm14, or in mPrdm14 KO ESCs with the empty vector (Fig. 2C). Among the pluripotency-associated genes, only Tcl1, which has been identified as a direct target of PRDM14 (Yamaji et al., 2013), was consistently downregulated in mPrdm14 KO ESCs expressing the empty vector or sPRDM14 compared with wild-type ESCs. Tcl1 downregulation in mPrdm14 KO ESCs was reversed by the exogenous expression of zPRDM14 or mPRDM14. It has been shown that Dnmt3a, Dnmt3b and Dnmt3l repression by PRDM14 is essential for sustaining global hypomethylation in mouse ESCs (Yamaji et al., 2013). In mPrdm14 KO ESCs expressing the empty vector or sPRDM14, Dnmt3a, Dnmt3b and Dnmt3l expression levels were consistently upregulated versus those in wild-type ESCs, whereas the exogenous expression of zPRDM14 or mPRDM14 completely reversed Dnmt3a, Dnmt3b and Dnmt3l repression in mPrdm14 KO ESCs. Comparing the transcriptional activities of PRDM14 orthologues suggested that zPRDM14, but not sPRDM14, has the similar transcriptional activity to mPRDM14. Aligning the primary amino acid sequences of sPRDM14, zPRDM14 and mPRDM14 demonstrated that the PR domain and zinc-finger domain are relatively conserved, but the N-terminal region of PRDM14 shows low conservation among these orthologues (Fig. S1B). Because the zinc-finger domain of PRDM14 binds specific DNA sequences (Chia et al., 2010), we analysed the binding capacity of sPRDM14, zPRDM14 and mPRDM14 at target genes in mESCs using chromatin immunoprecipitation (ChIP)-qPCR. Although we did not detect the transcriptional activity of sPRDM14, ChIP-qPCR analysis clearly demonstrated that sPRDM14 bound to target genes similar to zPRDM14 and mPRDM14 in mESCs, except for Fgfr1 (Fig. 2D). The reduction of PRDM14 binding at Fgfr1 might be one of the causes for the defect of ESC self-renewal by sPRDM14 in serum plus LIF condition. On the other hand, these findings suggest the possibility that sPRDM14 cannot recruit complexes involving transcriptional regulation at target genes in mESCs.

We next explored the growth rate and self-renewal capacity of mPrdm14 KO ESCs expressing the empty vector, sPRDM14, zPRDM14 or mPRDM14 in growth medium containing LIF and either 2i or serum. Growth rates were comparable among these ESCs in medium containing 2i plus LIF (Fig. 3A). However, when the cells were grown in medium containing serum plus LIF, although mPrdm14 KO ESCs expressing zPRDM14 or mPRDM14 continued to proliferate beyond 8 days, mPrdm14 KO ESCs expressing the empty vector or sPRDM14 did not (Fig. 3A). We compared the expression of pluripotency markers in ESCs grown in medium containing serum plus LIF or 2i plus LIF. Pluripotency markers were maintained by exogenous zPRDM14 or mPRDM14 expression, but not by sPRDM14 or the empty vector (Fig. 3B). To determine the self-renewal activity of mPrdm14 KO ESCs expressing each PRDM14 orthologue, we performed alkaline phosphatase (AP) staining of these cells cultured with serum plus LIF for 9 days (4 passages) or 14 days (7 passages) (Fig. 3C,D). Although we observed a small number of mPrdm14 KO ESCs expressing sPRDM14 with AP activity in the presence of serum plus LIF up through 9 days, the cells did not expand after four passages. In contrast, we could expand mPrdm14 KO ESCs expressing zPRDM14 or mPRDM14 with AP activity for at least 2 weeks (7 passages). These cells retained pluripotency markers, which were expressed at slightly higher levels than in wild-type cells (Fig. 3E).

Our group and another group have previously shown that mPRDM14 is involved in DNA hypomethylation through the repression of Dnmt3a, Dnmt3b and Dnmt3l, and the recruitment of ten-eleven translocation (TET) proteins, which oxidise 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) in ESCs (Yamaji et al., 2013; Okashita et al., 2014). First, we compared the protein expression levels of DNMT3A, DNMT3B and DNMT3L in wild-type ESCs and mPrdm14 KO ESCs harbouring the empty vector or expressing sPRDM14, zPRDM14 or mPRDM14 (Fig. 4A). The expression levels of DNMT3A, DNMT3B and DNMT3L were consistently reduced in mPrdm14 KO ESCs expressing zPRDM14 or mPRDM14 (but not sPRDM14) compared with those in mPrdm14 KO ESCs expressing empty vector, in agreement with the observed mRNA expression levels (Fig. 2C). Next, to assess the transcriptional repression activity of PRDM14 orthologues at the Dnmt3b promoter, we performed luciferase assays using an 8.6 kb, upstream region of Dnmt3b. Consistent with the western blot results described above, zPRDM14 and mPRDM14, but not sPRDM14, repressed Dnmt3b promoter activity (Fig. 4B). Because mPRDM14 can activate the pluripotency marker Tcl1 and the germline-specific genes Dazl, Sycp3 and Asz1 by inducing DNA demethylation (Okashita et al., 2014), we analysed the expression and DNA methylation of these genes in wild-type ESCs and in mPrdm14 KO ESCs expressing the empty vector or the sPRDM14-, zPRDM14-, or mPRDM14-expressing vector (Fig. 4C-E). Expression of these genes was consistently reduced in mPrdm14 KO ESCs harbouring the empty vector compared with wild-type ESCs, with elevated DNA methylation around the transcription start sites of these genes. Consistent with the compensation assay data for ESC self-renewal and for DNMT3A, DNMT3B and DNMT3L repression (shown above), zPRDM14 and mPRDM14, but not sPRDM14, drove upregulation and demethylation of these genes.

mPRDM14 can interact with TET1/2 and CBFA2T2 to induce DNA demethylation and transcriptional regulation, respectively (Okashita et al., 2014; Nady et al., 2015). Therefore, we analysed the interaction of sPRDM14, zPRDM14 and mPRDM14 with mouse TET2 (mTET2) and mouse CBFA2T2 (mCBFA2T2) by co-immunoprecipitation (co-IP) analysis. zPRDM14 and mPRDM14 interacted with endogenous mouse TET2 and CBFA2T2, whereas sPRDM14 interacted only with TET2 (Fig. 5A). We have previously shown that PRDM14 actively demethylates DNA at pluripotency-marker genes (including Esrrb and Tfcp2l1) through the TET-BER pathway to enhance POU5F1 recruitment during the transition from primed cells to naive pluripotency (Okashita et al., 2016). To compare the capacities of sPRDM14, zPRDM14 and mPRDM14 to recruit TET proteins to Esrrb and Tfcp2l1, we performed ChIP with antibodies against FLAG (FLAG-PRDM14), TET1 and TET2 at Esrrb and Tfcp2l1 in mPrdm14 KO ESCs expressing the empty vector, sPRDM14, zPRDM14 or mPRDM14. Consistent with the co-IP experiments, we observed that sPRDM14, zPRDM14 and mPRDM14 enhanced TET1 and TET2 binding at Esrrb and Tfcp2l1 (Fig. 5B). PRDM14 can repress the transcription of target genes, including Fgfr1 and Dnmt3, by recruiting polycomb repressive complex 2 (PRC2) (Yamaji et al., 2013). Therefore, we investigated the recruitment of SUZ12, which is an essential component of PRC2, and the enrichment of H3K27me3 at Fgfr1 and Dnmt3b in mPrdm14 KO ESCs expressing each PRMD14 orthologue (Fig. 5C). SUZ12 recruitment and H3K27me3 at these genes were enriched in mPrdm14 KO ESCs expressing mPRDM14 or zPRDM14 compared with mPrdm14 KO ESCs-expressing the empty vector or sPRDM14. These findings suggest that PRDM14 acquired the capacity to bind CBFA2T2, which might be essential for the recruitment of the PRC2 complex, during the evolutionary route to Chordata after the splitting from the common ancestor of Chordata and Echinodermata.

As shown above, PRDM14 derived from the teleost zebrafish completely compensated for mPRDM14 functions in DNA demethylation and pluripotency maintenance. Next, we examined the functional conservation of the Prdm14 gene isolated from the Cephalochordate amphioxus (amPRDM14) (Fig. 6A). Surprisingly, amPRDM14-overexpressing mPrdm14 KO ESCs retained AP activity for 9 days when grown in medium containing serum plus LIF (Fig. 6B). To examine the conservation of its function in transcriptional regulation, we compared global gene expression changes between control (empty), sPRDM14-, amPRDM14- and mPRDM14-overexpressing mPrdm14 KO ESCs by microarray analysis. We plotted the relative expression levels of genes that were upregulated (n=273) and downregulated (n=288) by mPRDM14 in sPRDM14- and amPRDM14-overexpressing mPrdm14 KO ESCs (Fig. 6C,D). Genes upregulated by mPRDM14 tended to be upregulated by sPRDM14 and amPRDM14, which is consistent with the data showing sPRDM14 and amPRDM14 can interact with TET2 (Fig. 5A and data not shown). Interestingly, a scatter plot comparing genes downregulated by mPRDM14 showed a weak negative correlation in sPRDM14-overexpressing mPrdm14 KO ESCs (Fig. 6C). sPRDM14 could bind to target genes, but not to mCBFA2T2 (Figs 2E and 5A), suggesting that sPRDM14 overexpression exerted a dominant-negative effect on transcriptional repression by endogenous CBFA2T2 in mPrdm14 KO ESCs. In contrast, supporting the self-renewal assay results of amPRDM14-expressing ESCs, genes downregulated by mPRDM14 tended to also be downregulated by amPRDM14 (Fig. 6D). qRT-PCR analysis confirmed the consistent upregulation or downregulation of mPRDM14 target genes by axPRDM14 (Fig. 6E) and co-IP analysis clearly showed that amPRDM14 interacted with CBFA2T2 (Fig. 6F). These findings suggest that the acquisition of transcriptional regulation activity of PRDM14 orthologues tightly correlates with the CBFA2T2-binding capacity.

A previous report has clearly revealed that mPRDM14 interacts with mCBFA2T2 through the pre-PR domain and PR domain (Nady et al., 2015). To identify the responsible amino acids for the incompatibility of sPRDM14 with mCBFA2T2, we reconstructed a model structure of sea urchin and mouse PRDM14 orthologues that form a complex with mouse CBFA2T2 using SWISS-MODEL, based on the previous crystal structure data for mPRDM14 (pre-PR and PR domain) and the mCBFA2T2 complex (Nady et al., 2015) (Fig. 7A,B). Interestingly, the density and strength of the negative charge on the interface of the PRDM14 pocket interacting with mCBFA2T2 were significantly weaker in sPRDM14 compared with that in mPRDM14. Previous mutation analysis of mPRDM14 and mCBFA2T2 have clearly shown that the basic amino acids, Arg105 and Lys109 (located at the N-terminal region of mCBFA2T2) bind to a mPRDM14 pocket formed by the pre-PR and PR domain through electrostatic interaction (Nady et al., 2015). These findings suggest that the weak negative charge on the surface of the sPRDM14 pocket might be responsible for the incompatibility of sPRDM14 with mCBFA2T2. To identify the contribution of the pre-PR domain and PR domain in the binding capacity of PRDM14 with mCBFA2T2 and the maintenance of ESC pluripotency, we performed compensation assays for mouse PRDM14 using two chimeric proteins consisting of the N-terminal region of sPRDM14 (including the pre-PR domain) and C-terminal region of mPRDM14 (including the PR domain), and vice versa (Fig. 7C,E). We generated a model structure of the chimeric protein consisting of the two PRDM14 orthologues (Fig. 7D). Interestingly, the negative charge on the surface of the interacting pocket with CBFA2T2 depended on the pre-PR domain rather than the PR domain (Fig. 7C). To examine whether the model structure of the chimeric protein supports the function in the maintenance of ESC pluripotency, we established chimeric PRDM14-overexpressing mPrdm14 KO ESCs (Fig. 7D). Consistent with the model structure of the chimeric protein (Fig. 7C), both self-renewal activity and interaction capacity of PRDM14 with mCBFA2T2 strongly depended on the pre-PR domain (Fig. 7E,F). Furthermore, s/mPRDM14-overexpressing mPrdm14 KO ESCs partially retained AP activity, suggesting that the PR domain is also involved in the self-renewal activity of mPRDM14.

sPRDM14 could not interact with endogenous mCBFA2T2 and compensate for mPRDM14 functions in maintaining mESC pluripotency (Figs 3C and 5A). To investigate the possibility of the incompatibility between sPRDM14 and mCBFA2T2, we tried to establish mPrdm14 KO ESCs co-expressing sPRDM14 and sCBFA2T. First, we obtained the sequence of Cbfa2t2 orthologues and paralogues in the genome among metazoans. Cnidaria (sea anemone), protostomes and non-vertebrate deuterostomes possess only one Cbfa2t homologue in their genomes, whereas vertebrate genomes contain Cbfa2t1, Cbfa2t2 and Cbfa2t3. Three Cbfa2t gene families might have emerged from a single Cbfa2t gene by two rounds of whole-genome duplication in the vertebrate lineage after the split of amphioxus from their common ancestor (Dehal and Boore, 2005) (Fig. 8A,B). The nervy homology region 1 (NHR1) domain of mCBFA2T2 is necessary and sufficient for the mPRDM14-mCBFA2T2 interaction (Nady et al., 2015). The alignment of mCBFA2T1/2/3 and sCBFA2T showed that the NHR1 domain is highly conserved, including amino acids involved in hydrogen bonding with PRRDM14 (Figs 7B and 8B). Next, we established Prdm14 KO ESCs co-expressing sPRDM14 and sCBFA2T to examine the self-renewal activity of ESCs (Fig. 8C). Switching from the 2i plus LIF condition to the serum plus LIF condition promoted differentiation of mPrdm14 KO ESCs harbouring the empty vector or expressing sPRDM14 (Fig. 8D). Interestingly, co-expressing sCBFA2T and sPRDM14 in mPrdm14 KO ESCs maintained pluripotency under the serum plus LIF condition, similar to mPrdm14 KO ESCs expressing mPRDM14. As mentioned above, the interaction of PRMD14 orthologues with CBFA2T2 closely correlated with the maintenance of ESCs. Co-IP analysis showed that sPRDM14 interacted with sCBFA2T in ESCs (Fig. 8E). sPRDM14 could not interact with endogenous mCBFA2T2 (Fig. 5A); however, the interaction of sPRDM14 with endogenous mCBFA2T2 was recovered by exogenous expression of sCBFA2T (Fig. 8F). CBFA2T has been shown to oligomerise through its NHR2 domain, which stabilises PRDM14 binding to chromatin in mESCs (Tu et al., 2016), suggesting that sPRDM14 interacts with mCBFA2T2 through sCBFA2T-mCBFA2T2 oligomerisation (Fig. 8G). Indeed, sCBFA2T interacted with both sPRDM14 and endogenous mCBFA2T2 (Fig. 8F). Finally, to investigate whether sCBFA2T can compensate for mCBFA2T2 function, we deleted exon 3 of mCbfa2t2 in sPRDM14 and sCBFA2T-expressing mPrdm14 KO ESCs (Fig. S2). The sPRDM14-sCBFA2T complex compensated for the mPRDM14-mCBFA2T2 complex in the maintenance of ESC pluripotency (Fig. 8H). Supporting the compensation of sPRDM14 with sCBFA2T for mPRDM14 function, key target genes regulated by mPRDM14 were consistently regulated by the combination of sPRDM14 with sCBFA2T in mPrdm14 KO ESCs (Fig. 8I). To compare the global changes in gene expression, we performed microarray analysis. Genes downregulated by mPRDM14 showed no correlation in sPRDM14-expressing mPrdm14 KO ESCs (Figs 6C and 8J). Interestingly, a significant positive correlation of downregulated genes by mPRDM14 was observed in mPrdm14 KO ESCs co-expressing sPRDM14 and sCBFA2T (Fig. 8K), which strongly supports the self-renewal activity of mPrdm14 KO ESCs co-expressing sPRDM14 and sCBFA2T (Fig. 8D). These findings clearly indicate that sPRDM14 can cooperate with sCBFA2T, but not mCBFA2T2, in transcriptional regulation required for the maintenance of ESC pluripotency.

Prdm14 expression is associated with germline and pluripotent cells in mice or motor neurons in zebrafish (Yamaji et al., 2008; Liu et al., 2012). However, the expression pattern of Prdm14 has not been examined thoroughly in other species, especially the invertebrate chordates. To explore the conservation of Prdm14 expression in chordates, we studied the developmental expression of Prdm14 in amphioxus. Using whole-mount in situ hybridisation, amPrdm14 was not detectable at the cleavage and blastula stages, but started to become expressed in the ectoderm at the gastrula stage in amphioxus embryos (Fig. 9A-C). At the neurula stage, we detected amPrdm14 expression in pairs of cells localising in the neural tube (Fig. 9D-F). This expression pattern was reminiscent of that for Mnx/Hb9, which is expressed in developing motor neurons (Fig. 9G) (Ferrier et al., 2001). To confirm the expression of Prdm14 in motor neurons, we performed double labelling of Prdm14 and the primary motor neurons (PMNs) marker Islet, which is expressed in all PMNs (Jackman et al., 2000). Islet was co-expressed with Prdm14 in the anterior cell of the two Prdm14-positive cells (Fig. 9H-M), indicating that Prdm14 is expressed in the motor neurons of amphioxus embryos. In amphioxus, PGCs are marked by Vasa and Nanos expression, and are accumulated at the posterior end of the embryo at the neurula stage (Wu et al., 2011). However, we did not detect Prdm14 expression at the posterior end of the embryo (Fig. 9D,E), suggesting that amphioxus Prdm14 is produced in motor neurons, but not in PGCs. These findings provide evidence that the specific expression of Prdm14 in motor neurons is conserved between amphioxus and zebrafish.

The phylogenetic distribution of Prdm14, Tet and Cbfa2t2 orthologues indicates that these factors arose at least in the common ancestor of bilaterians and that Prdm14 might have been lost during the evolution within the Ecdysozoa (Figs 1A and 8A). Comparative analysis of Prdm14 expression among deuterostomes suggests that PRDM14-TET and PRDM14-CBFA2T2 complexes were co-opted from motor neuron into pluripotent cells around the time when tetrapods emerged. Our data show that all PRDM14 orthologues that we tested can interact with TET2 and we have previously shown that mouse PRDM14 cooperate with TET protein to maintain and acquire pluripotency in mouse pluripotent cells (Okashita et al., 2016). Furthermore, the contribution of TET proteins in early mouse embryo has been intensively investigated by several groups (Dai et al., 2016; Khoueiry et al., 2017). Contrary to the case of mouse, although Prdm14 is also expressed in pluripotent cells of early chick embryo, the level of 5-hydroxymethylcytosine, which is produced from 5-methylcytosine by TET proteins, is quite low in pre-streak chick embryo (Almeida et al., 2012). Therefore, the co-option of PRDM14-TET complex into pluripotent cells might depend on the expression of TET proteins in pluripotent cells during the emergence of mammals. The data presented here show that amPRDM14 and zPRDM14 interacted with endogenous mCBFA2T2 and compensated for mPRDM14 function in maintaining ESC pluripotency. Furthermore, the combination of sPRDM14 and sCBFA2T expression compensated for mPRDM14 and mCBFA2T2 functions, suggesting that PRDM14 cooperated with CBFA2T in transcriptional regulation in the common ancestor of deuterostomes. Sea stars exploit an inductive mechanism for germ cell formation; hence, part of the embryonic gut may maintain multipotency to give rise to PGCs (Wessel et al., 2014). Interestingly, Prdm14 is expressed in the embryonic gut, but not in PGCs in sea star embryos (Fresques et al., 2014). Our study provides clear evidence that sPRDM14 cooperates with sCBFA2T to maintain mESC pluripotency, suggesting that the PRDM14-CBFA2T complex serves essential functions in maintaining embryonic gut multipotency in echinoderm embryos.

Intriguingly, Prdm14 is expressed in the neuronal system in amphioxus, zebrafish and Xenopus embryos (Eguchi et al., 2015), and is required for motor neuron maturation in zebrafish embryos (Liu et al., 2012). The CBFA2T family is also expressed in the neural system in Xenopus and chicken embryos, and its expression is involved in neurogenesis, including motor neurons (Cao et al., 2002; Aaker et al., 2009). Our data provide evidence that the dependence of PRDM14 function on CBFA2T2 is conserved within deuterostomes, suggesting that PRDM14 might also cooperate with CBFA2T to control neurogenesis in non-mammalian deuterostomes. Contrary to the expression pattern of Prdm14 in neuronal systems in non-mammalian vertebrates, Prdm14 expression is lost in the nervous system, and is restricted in pluripotent cells and primordial germ cells in mammals (Yamaji et al., 2008; Tang et al., 2015). These findings suggest that Prdm14 expression shifted from the neuronal system to pluripotent and germ cells en route from a vertebrate common ancestor to mammals during evolution (Fig. 8A). Prdm14 is expressed in primary motor neurons (PMNs) but not in secondary motor neurons (SMNs) in zebrafish embryos (Liu et al., 2012). In contrast, PMNs have not been described in amniotes and it is thought that SMNs in zebrafish embryo closely resemble motor neurons in amniotes, including mouse (Eisen, 1994). These findings suggest that the loss of Prdm14 expression in motor neurons might be linked to the loss of PMNs during the emergence of amniotes.

Previous ChIP-Seq analysis showed that POU5F1 is localised at a distal enhancer of the Prdm14 locus both in mESCs and human ESCs (hESCs) (Okashita et al., 2016; Tsankov et al., 2015). Determining whether POU5F1 positively regulates Prdm14 expression in mouse and human pluripotent stem cells is important for understanding how Prdm14 integrated into the pluripotent network during evolution, which warrants further investigation. We propose two major possibilities to explain the change in the Prdm14 expression pattern during evolution. The first possibility is that the expression pattern of upstream factors of Prdm14 shifted from motor neurons to pluripotent cells and germ cells. Pou5f1 belongs to the PouV class of genes, which is one of several classes in the POU-domain gene family. Based on currently available genomic information, the amphioxus genome contains PouIII genes, but not PouV genes, and PouV-class genes might have evolved from PouIII genes in the common ancestor of gnathostomes (Frankenberg and Renfree, 2013; Onichtchouk, 2016). Interestingly, the amphioxus PouIII genes, Brn1, Brn2 and Brn4 are expressed in the neuroectoderm at the gastrula stage and then in motor neurons at the neurula stage, suggesting that Brn1, Brn2 and Brn4 may control Prdm14 expression in amphioxus motor neurons (Candiani et al., 2002). Therefore, the emergence of POU5F1 from PouIII genes in the vertebrate genome and their restricted expression in pluripotent cells and germ cells might have enabled restricted Prdm14 expression in pluripotent cells and germ cells. Another possibility is that Prdm14 expression changes arising during evolution were caused by the evolution of cis-regulatory elements neighbouring the Prdm14 locus. Although the mouse genome contains POU5F1-recognition sequences in the distal enhancer of Prdm14 (Galonska et al., 2015), the evolutionary origin and conservation of this element have not been investigated in non-mammalian vertebrates. We consider amphibians and reptiles to be key organisms for understanding how Prdm14 was co-opted into the network for pluripotency-associated transcription factors during evolution.

In the early development of amniotes, including reptile and mammals, the pluripotent state of epiblast cells should be maintained for the long term compared with non-amniote vertebrates to ensure the maturation of extra-embryonic tissues. Integration timing of the PRDM14-CBFA2T2 axis from neurogenesis to pluripotent cells in the evolution of deuterostomes is closely associated with the emergence of amniotes. Prdm14 disruption destabilises pluripotency networks (Chia et al., 2010; Yamaji et al., 2013), whereas PRDM14 overexpression stabilises pluripotency networks in mESCs and hESCs (Okashita et al., 2015; Tsuneyoshi et al., 2008), suggesting that co-opting PRDM14 into transcriptional networks for pluripotency in pluripotent cells might have served some roles in the emergence of amniotes during evolution. Therefore, we consider it to be an essential challenge to uncover the de novo mechanisms stabilising pluripotency transcriptional networks during the emergence of mammals. In summary, our work comparing the expression pattern and function of PRDM14 offers a framework for addressing how pluripotency transcriptional networks evolve in deuterostomes.

Amino acid sequences of putative homologues were retrieved from aLeaves (Kuraku et al., 2013), incorporating Ensembl release 84. Multiple-sequence alignment was performed with MAFFT software, version 7.299b (Katoh and Standley, 2013) using the ‘-linsi’ option. The aligned sequence sets were processed using Trimal software, version 1.4, rev15 (Capella-Gutierrez et al., 2009), with the ‘-automated1’ and ‘-nogaps’ options. Molecular phylogenetic trees were inferred with the maximum-likelihood method using the RAxML program, version 8.2.8 (Stamatakis, 2014) with the ‘-m PROTCATWAG -f a -# 1000’ option. Bootstrap probabilities were obtained with 1000 re-samplings.

mPrdm14 KO ESCs (Yamaji et al., 2013) were maintained in N2B27 basal medium with 3 μM CHIR99021 (Sigma) and 0.4 μM PD0325901 (Wako) on mitomycin-treated STO feeder cells. DNA encoding FLAG-tagged sPrdm14, amPrdm14 (Eurofins Genomics) and sCbfa2t (GENEWIZ) were chemically synthesised based on their known sequences and were cloned into the pCAGGS-IRES-Puro vector or the pCAGGS-IRES-Blast vector. Prdm14 KO ESCs were transfected with each expression vector and selected in medium containing 0.5 μg/ml puromycin or 5 μg/ml blasticidin.

Total RNA was extracted using TRIzol (Thermo Fisher Scientific) according to the manufacturer's instructions. cDNA was synthesised using ReverTra Ace qPCR RT Master Mix (TOYOBO), according to the manufacturer's instructions. Subsequently, cDNA was amplified with FastStart SYBR Green Master (Roche) using gene-specific primers (Table S1), and fluorescence was measured using the Light Cycler 96 system (Roche).

Total RNA was purified using the PureLink RNA Mini Kit (Thermo Fisher Scientific) and labelled using the GeneChiP WT Terminal Labeling and Control Kit (Thermo Fisher Scientific). The labelled single-stranded cDNA was hybridised to the GeneChiP Mouse Gene 2.1 ST Array Strip (Thermo Fisher Scientific). The signal intensities of microarray probes were scanned and quantitated using the GeneAtlas Personal Microarray System (Thermo Fisher Scientific). CELL files of Gene ChIP data were normalised using the Robust Multi-array Average program with the default settings in the Affymetrix Expression Console. Microarray data have been deposited in GEO under accession number GSE112904.

ChIP was performed as described previously (Tachibana et al., 2008). Briefly, DNA–protein complexes were crosslinked in phosphate-buffered saline containing 1% formaldehyde for 10 min at room temperature. Fixation was terminated by the addition of 125 mM glycine. The fixed cells were lysed in sodium dodecyl sulphate (SDS) lysis buffer and sonicated with a BioRuptor. The ChIP samples were mixed with antibodies against the FLAG epitope (Sigma), TET1 (Millipore), TET2 (Abcam), SUZ12 (Abcam) and H3K27me3 (Millipore), as well as SureBeads Protein A or G Magnetic Beads (BioRad) overnight at 4°C. DNA-protein-SureBead Protein A or G complexes were collected using a magnetic rack (BioRad). De-fixed DNA was purified by phenol-chloroform extraction, followed by ethanol precipitation. Precipitated DNA was used as a template for qPCR analysis with gene-specific primers (Table S1).

mPrdm14 KO ESCs expressing the empty vector or a PRDM14 orthologue were maintained in N2B27 basal medium with 2i plus LIF. To evaluate the self-renewal activity of ESCs, 5×10⁴ cells were plated in 12-well plates in medium containing serum plus LIF and passaged every 3 days. Nine days after cultivation in medium containing serum plus LIF, the cells were subjected to AP staining, as described previously (Cox and Singer, 1999).

Cells were lysed by boiling in SDS sample buffer (Wako), and proteins were separated on SDS-polyacrylamide gels. Next, the proteins were electroblotted to polyvinylidene difluoride membranes, and the membranes were probed using the following primary antibodies: anti-FLAG (Sigma, F1804; 1/500), anti-DNMT3A (Abcam, ab13888; 1/500), anti-DNMT3B (Abcam, ab176166; 1/500), anti-DNMT3L (Abgent, AP1062a; 1/500), anti-histone (Abcam, ab1791; 1/5000) and anti-TY1 (Sigma, SAB4800032; 1/2000). Following binding of the primary antibody, the membrane was incubated with horseradish peroxidase-coupled secondary antibody (Santa Cruz, sc-2005; 1/2500). Detection was achieved using the ImmunoStar Zeta luminescent reagent (Wako).

For IP analysis, the cells were lysed in lysis buffer [50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 1% NP40, and 0.1% Triton-X]. The lysates were incubated with an anti-FLAG antibody (Sigma; F1804) and SureBeads Protein A or Protein B (BioRad) for 1 h at room temperature. The FLAG-tagged proteins bound to SureBeads Protein A or Protein B were captured using a magnetic rack and the precipitated proteins were eluted with 20 mM glycine.

Genomic DNA was purified from ESCs using the Wizard SV Genomic DNA Purification System (Promega) according to the manufacturer's instructions. Bisulphite treatment was performed using the EpiTect Fast DNA Bisulfite Kit (Qiagen) according to the manufacturer's instructions. Bisulphite-treated DNA was amplified by TAKARA EpiTaq HS (TAKARA) with gene-specific primers (Table S1). Glucosylation of genomic DNA followed by methylation-sensitive qPCR analysis was performed as described previously (Okashita et al., 2016). The sequences of the primers used in this study are shown in Table S1.

The model structure for the sea urchin PRDM14 complex with CBFA2T2 was built by the homology modelling method using the previously reported mouse PRDM14-CBFA2T2 complex structure (Nady et al., 2015) (Protein Data Bank code; 5ECJ) as a modelling template. Homology modelling was performed with SWISS-MODEL (Biasini et al., 2014). The initial model structures were refined by rigid-body refinement, energy minimisation and simulated annealing with no experimental energy terms to remove model bias. The refinement was performed by CNS version 1.3 (Brünger et al., 1998). The figures containing molecular structures and electrostatic molecular surface were prepared with PyMOL. The region corresponding to the disorder region in mouse PRDM14 (amino acid number 217-235) in sea urchin PRDM14 model structure was omitted from the molecular structure.

To synthesise riboprobes, amphioxus cDNA fragments were amplified as templates as previously described (Wu et al., 2011). DIG-labelled Prdm14 and Mnx/Hb9 antisense riboprobes and fluorescein-labelled Islet antisense riboprobes were synthesised using T7 RNA polymerase (Roche). Whole-mount in situ hybridisation was performed on the 4% paraformaldehyde-fixed amphioxus embryos as described previously (Wu et al., 2011). The AP substrates NBT and BCIP (Roche) were used for colour detection. For double fluorescent in situ hybridisation, anti-DIG-POD and anti-fluorescein-POD antibodies (Roche) were used to detect the riboprobes, followed by the amplification of fluorescence signal using the TSA Plus Cyanine 3 and Fluorescein Evaluation Kit (PerkinElmer). 4′,6-diamidino-2-phenylindole (DAPI) (Invitrogen, 1 μg/ml in PBS containing 0.1% Tween) was used for nuclear staining. The results were photographed using either a Zeiss AxioCam MRC camera mounted on a Zeiss Imager A2 microscope or a Zeiss confocal system (Zeiss LSM 880 with Airyscan).

We thank Andrew Johnson for providing genomic information for axolotl Prdm14 and for stimulating discussions. We thank Kentaro Kajiwara for the constructing of plasmids. We thank Mitinori Saitou for providing the mPrdm14 KO ESCs.