Maternal Nanog is required for zebrafish embryo architecture and for cell viability during gastrulation

The transcription factors Oct4 (known as Pou5f1 in mammals and Pou5f3 in zebrafish), Sox2 and Nanog are best known for their prominent functions as regulators of pluripotency in embryonic stem cells (ESCs) (Mitsui et al., 2003; Boyer et al., 2005; Loh et al., 2006), and their ability to reprogram somatic cells to a pluripotent state (Takahashi and Yamanaka, 2006; Silva et al., 2009). In developing embryos, Oct4 and Nanog also play distinct roles during the transition from cell pluripotency to terminal differentiation (Teo et al., 2011; Loh and Lim, 2011; Thomson et al., 2011; Frum et al., 2013; Radzisheuskaya et al., 2013; Le Bin et al., 2014). The question of how the same transcription factors can regulate seemingly opposing processes, such as pluripotency maintenance and cell differentiation, is far from being resolved. Functional studies on zebrafish and Xenopus, in which the earliest developmental stages are experimentally tractable, have large potential to contribute to our understanding of this question. Zebrafish Nanog, which is the functional ortholog to mouse Nanog in cross-species rescue assays, enables efficient reprogramming of mammalian Nanog^−/− somatic cells to full pluripotency (Theunissen et al., 2011). Integration of Nanog into the embryonic early zygotic gene network will be useful to decode how pluripotency, patterning, and cell differentiation are coupled in the vertebrate embryo.

Zebrafish Nanog is abundantly translated during the transcriptionally silent period preceding zygotic genome activation (ZGA; Lee et al., 2013). At zebrafish ZGA, the first major wave of zygotic transcription occurs and a large specific subset of maternal transcripts is cleared by the microRNA miR-430 (Giraldez et al., 2006). Maternal Nanog has been implicated in both processes. Nanog regulates the post-ZGA clearance by directly activating miR-430 (Lee et al., 2013). Together with Pou5f3 and the SoxB1 family, maternal Nanog acts to activate transcription during ZGA. The combined loss of these factors results in the failure of the activation of >75% of zygotically transcribed genes (Lee et al., 2013; Leichsenring et al., 2013). Single loss-of-function phenotypes for SoxB1 group genes (Okuda et al., 2010), Oct4/Pou5f3 (Reim et al., 2004; Lunde et al., 2004) and Nanog differ significantly, suggesting overlapping as well as separate functions in the embryo during gastrulation stages (Onichtchouk and Driever, 2016).

The zebrafish epiboly movement involves the thinning and spreading of a multilayered sheet of cells over the yolk (Solnica-Krezel, 2006; Bruce, 2016). The zebrafish blastula embryo consists of the deep embryonic cells of the blastoderm and three extra-embryonic structures: a large yolk cell surrounded by a microtubule-rich cortical layer (yolk cytoplasmic layer or YCL), the underlying deep cells of the yolk syncytial layer (YSL), and the enveloping layer (EVL). Epiboly initiation commences with doming when the yolk cell bulges upward into the blastoderm (Kimmel et al., 1995), accompanied by the radial intercalation of the deep cells (Warga and Kimmel, 1990). Two different types of cytoskeletal structures forming in the extra-embryonic layers are crucial for epiboly: microtubules of the YCL (Solnica-Krezel and Driever, 1994) and microfilaments of the YSL. The second phase of epiboly depends on deep cell motility and on the F-actin microfilament ring which forms at the border of the YSL and the EVL (Zalik et al., 1999; Cheng et al., 2004; Koppen et al., 2006).

Previous studies indicate that three transcription factors are involved in epiboly control: Pou5f3, Eomesodermin a (eomesa) and Mxtx2. In the maternal-zygotic (MZ) mutant of Pou5f3, MZspg (Spiel-ohne-Grenzen), two crucial components driving epiboly are severely affected: microtubules of the YCL and radial intercalation of deep cells (Reim and Brand, 2006; Lachnit et al., 2008; Song et al., 2013). Oct4/Pou5f3 mutants also show multiple independent patterning defects (Reim et al., 2004; Lunde et al., 2004; Belting et al., 2011), and increased apoptosis at the end of gastrulation (Kotkamp et al., 2014). In contrast, only epiboly movements are critically affected in MZeomesa mutants and in Mxtx2 knockdown (KD) embryos. MZeomesa mutants show a delay in doming and YCL microtubule defects. However, once started, epiboly progresses normally (Du et al., 2012). Mxtx2 is expressed shortly after ZGA and is specifically required for epiboly within the YSL. In Mxtx2 KD embryos, epiboly initiation occurs normally and YCL microtubules are not affected, but epiboly progression is impaired owing to abnormal F-actin ring formation and reduced motility of deep cells (Hirata et al., 2000; Bruce et al., 2005; Wilkins et al., 2008).

A zebrafish Nanog morpholino KD phenotype was previously described in two studies (Schuff et al., 2011; Xu et al., 2012). The main features of the KD phenotype are abnormal morphogenetic movements and embryo death at the end of gastrulation. Xu et al. (2012) demonstrated that in Nanog KD embryos expression of mxtx2 and Nodal pathway genes in the YSL is absent. Epiboly and the normal expression of Nodal pathway genes could be rescued by injection of mxtx2 mRNA into the Nanog KD embryos. These experiments led to the two-step model of action of Nanog via the YSL: Nanog activates mxtx2 in the YSL and Mxtx2 induces the expression of the ligands of Nodal family genes, which send their signals to the embryonic cells.

Poor correlation between morpholino-induced and mutant phenotypes, as observed on a large scale in zebrafish, led to the suggestion that mutant phenotypes should become the standard metric to define gene function in zebrafish (Kok et al., 2015). Thus, to understand the functions of Nanog, complete and reproducible genetic ablation of its gene activity was necessary. Here, we used transcription activator-like effector nucleases (TALENs) to create new mutant alleles to investigate the requirements of Nanog in zebrafish development. We show that both initiation (yolk doming), as well as progression of epiboly is impaired in MZnanog embryos. Embryonic axial structures do not form and massive cell death starts at the end of gastrulation. Our data and published literature indicate that Nanog regulates three independent events that contribute to normal epiboly progression. These are: (1) spatial arrangement of microtubule structures in the yolk cortex, (2) Mxtx2-dependent gene expression in the YSL, and (3) miR-430-dependent degradation of maternal mRNAs in deep embryonic cells. Furthermore, Nanog regulates cell survival. This function can be distinguished from the function Nanog serves during epiboly. Using transplantation experiments, we demonstrate that the survival of early blastula-stage MZnanog^−/− cells is reduced even in the wild-type embryo environment. We observe multiple changes in expression timing and patterns of known developmental regulators. This is compatible with the global role of Nanog in zygotic genome activation.

We introduced a mutation into the first exon of the nanog gene to analyze the function of zebrafish Nanog using the TALEN technique (Cermak et al., 2011). This resulted in a deletion of 10 bp and a frame shift after amino acid 17 with a stop codon after 15 additional amino acids (Fig. 1A). Whereas the zygotic homozygous mutants were viable and fertile, the maternal (M) and maternal-zygotic (MZ) mutants showed severe developmental defects (Fig. 1B, Movie 1). After blastula stage [4 hours post-fertilization (hpf)], MZnanog and Mnanog embryos were delayed in their epiboly movements and a shield-like structure appeared later than in wild-type controls (Fig. 1B, 4-8 hpf). At 8 hpf, epiboly movements were partially restored in Mnanog, whereas MZnanog embryos formed an enlarged shield and underwent developmental arrest (Fig. 1B, 8 hpf). This was often followed by yolk lysis and detachment of blastoderm from the yolk (Fig. 1C, Movie 1). MZnanog embryos that survived until 24 hpf consisted of unstructured cell masses with no recognizable body axis. In Mnanog mutant embryos, a short body axis of anterior type was often present (Fig. 1B, 24 hpf). MZnanog mutants could be rescued by injecting nanog mRNA into 1-cell-stage embryos, resulting in fertile adults (see Fig. 5).

Nanog mRNA and protein are present in the germline of the ovary and testes (Wang et al., 2016). Therefore, Nanog might be required during oogenesis and the earliest developmental stages. We tested whether pre-ZGA cleavages occur normally in MZnanog mutants by comparing three parallel time-lapse recordings of the wild-type and MZnanog embryos (Movie 1). We did not detect any significant differences in cell cleavage rates before ZGA by live observations (Fig. S1A), or in cell division rates at blastula stages by scoring mitotic cells (Fig. S1B-D, Table S1, Materials and Methods). We next addressed whether cell survival is affected in the MZ and Mnanog mutants in respect to the wild type. Using Acridine Orange staining of living embryos (Materials and Methods, Fig. S2), we detected massive cell death in MZnanog mutants after 10 hpf (Fig. 2A). The amount of dead cells in Mnanog mutants was reduced compared with MZnanog (Fig. 2B), but exceeded the that of wild-type controls (Fig. 2C,D; P<10⁻⁶, Student's t-test).

We performed in situ hybridization to compare the expression of zygotic YSL genes and mesendodermal markers as well as maternal eomesa mRNA between wild-type and MZnanog embryos. mxtx2 transcripts were not detectable in MZnanog and Mnanog at any stage examined (Fig. 3A). Expression of the early endoderm-specifying genes sox32 (Alexander et al., 1999), mixer (mixl1) (Henry and Melton, 1998) and gata5 (Reiter et al., 1999) started later and was reduced compared with wild-type embryos. This reduction was not due to a general ZGA delay: the zygotic mesoderm markers ntl (ta) and foxd3 (Schulte-Merker et al., 1994; Chen and Kimelman, 2000; Dougan et al., 2003) started at the appropriate time and were distributed similarly to the wild type in blastula (Fig. 3B,C, Fig. S3). At later stages, ntl was slightly reduced in MZnanog compared with wild type (Fig. 3F). Maternal eomesa mRNA (Du et al., 2012) was abundant in both genotypes at 4 hpf and remained in 6 hpf MZnanog whereas it was cleared from the wild type (Fig. 3B,G). We analyzed the expression of the definitive endodermal marker sox17 (Alexander and Stainier, 1999) to address whether Nodal signaling is active and whether endoderm forms in Mnanog and MZnanog. Variously shaped marginal domains of sox32 and sox17 expression were detectable in MZnanog mutants at 8 hpf. MZspg mutants, which lack sox17 expression (Lunde et al., 2004; Reim et al., 2004), were stained in parallel and are shown for comparison (Fig. 3D). Sox17-positive endodermal cells were distinguishable in MZnanog and Mnanog mutants at 9 hpf, although their numbers were smaller compared with wild-type embryos (Fig. 3E, arrows). We concluded that Nodal signaling is reduced but not absent in MZnanog embryos.

Morphological defects in Nanog-like morphants have been explained by the abnormal extra-embryonic YSL structure (Xu et al., 2012), which prompted us to examine MZnanog embryos for the presence of YSL and its morphology. We compared initial YSL formation in MZnanog and the wild type and found that it was not distinguishable from the controls until sphere stage (Fig. 4A-D). At sphere stage, nuclei within the YSL undergo three rapid divisions and become post-mitotic. As epiboly begins, nuclei align at the periphery to form one row and some nuclei move from the periphery to the center underneath the blastoderm (Kimmel and Law, 1985; Kane et al., 1992; D'Amico and Cooper, 2001). To visualize these movements, we labeled the YSL nuclei by injecting the SYTOX Green vital dye and performed time-lapse recordings of the living embryos (Movie 2 for wild type and Movie 3 for MZnanog, Fig. 4E-L). Although no differences could be detected between genotypes in number or positioning of nuclei at blastula stages (Fig. 4E-H′), in MZnanog the nuclei did not align at the periphery of the YSL (compare Fig. 4I and 4J), centripetal migration of the YSL nuclei did not occur (Fig. 4K,L, boxed areas), and the doming of the yolk did not start (Fig. 4K′,L′, yellow arrows). The absence of yolk doming in MZnanog was confirmed using light-sheet microscopy (Fig. 4M,N). To test whether the F-actin ring is impaired in MZnanog, we visualized microfilament structures in wild type and MZnanog using rhodamine-phalloidin staining. The F-actin ring with a row of endocytic vesicles was readily detectable in the wild type, but was strongly reduced in MZnanog embryos (Fig. 4O-P′, Fig. S4). We next compared yolk cortical microtubules between wild type, MZnanog and MZspg mutants, in which abnormalities of YCL microtubule structure were previously reported (Lachnit et al., 2008) (Fig. 4Q-S′). As can be seen from Fig. 4R,R′, the yolk cell of MZnanog formed a severely altered array of cortical microtubules resembling the MZspg YCL phenotype (Fig. 4S,S′), but with more irregular with asymmetric holes between microtubule bundles. The large patches of microtubule-free areas in MZnanog could be a reason for the yolk lysis observed at gastrulation (compare Fig. 1C and Fig. 4R). Thus, Nanog mutation affects the cytoskeletal architecture of two extra-embryonic yolk structures: microfilaments in the YSL and cortical microtubule arrays in the YCL.

We performed a set of rescue experiments to test whether MZnanog mutant defects can be restored by microinjections of nanog mRNA either in the YSL or in the whole embryo at the 1-cell stage (Fig. 5). Injection of nanog mRNA into MZnanog embryos at the 1-cell stage resulted in restoration of epiboly and complete body axis (Fig. 5B,F,J,P) and rescue of the cell death (Fig. 5Q,R). The rescued embryos could be grown to fertile adults. After nanog mRNA was injected into the YSL, some MZnanog embryos displayed a transient improvement in epiboly, but they could neither complete epiboly nor form a body axis (Fig. 5D,H,L,P). The lack of rescue by nanog mRNA injection into the YSL could be explained by insufficient time for translation of injected nanog mRNA. Indeed, injection of nanog mRNA into MZnanog embryos restored mxtx2 expression when nanog mRNA was injected at the 1-cell stage (Fig. 6F) but not when it was injected into the YSL (data not shown).

Zygotic mxtx2 mRNA is expressed in wild-type embryos after ZGA (Hirata et al., 2000) and, in that case, microinjection of mRNA at the 512-cell stage should suffice to restore the function of Mxtx2 in the YSL of MZnanog embryos. To address whether some or all aspects of the MZnanog phenotype can be rescued by Mxtx2, we injected a range of mxtx2 mRNA concentrations into 1-cell stage or into the YSL of MZnanog embryos (Fig. S5). Surprisingly, microinjection of mxtx2 mRNA in the YSL did not rescue epiboly, but 1-cell-stage injection in MZnanog resulted in complete epiboly rescue (Fig. 6A, Fig. S5).

Severe epiboly delay was seen in the MZdicer mutants, in which the whole RNA degradation pathway is inactive (Giraldez et al., 2005). Therefore, it is possible that aberrant accumulation of maternal miR-430 targets contributes to epiboly defects in MZnanog. We further tested, using the same experimental set-up as above, whether miR-430 contributes to any of the aspects of MZnanog phenotype. Indeed, 1-cell-stage but not YSL microinjection of miR-430 rescued epiboly (Fig. 6A, Fig. S6).

We next investigated whether the cell death phenotype of MZnanog could be rescued by mxtx2 or miR-430. We found that neither injection reduced the number of dying cells in MZnanog embryos (Fig. 6B,D). Formation of embryonic axes of variable length was often observed in mxtx2-injected MZnanog embryos (Fig. 6C, Fig. S5); however, overall survival and axis formation rates varied from experiment to experiment. When directly compared with nanog mRNA-injected embryos of the same MZnanog experimental batch, the survival rates of mxtx2 mRNA- and miR-430-injected embryos were poor (Fig. 6E, see figure legend for details). To investigate whether epiboly rescue by mxtx2 or miR-430 occurs using the same molecular mechanisms, we performed in situ hybridization for Nodal pathway targets sox32 and sox17 on MZnanog embryos injected with nanog, mxtx2 or miR-430 at the 1-cell stage (Fig. 6G,H, Table S2). miR-430 injection did not have any effect on sox32 and sox17, whereas nanog injection resulted in a moderate increase of scattered sox32- and sox17-positive cells compared with MZnanog control. mxtx2 injection to MZnanog resulted in the appearance of supernumerous clustered sox32- and sox17-positive cells (Fig. 6G,H), indicating excessive endoderm formation even when compared with the wild-type control (Fig. S7). The elevated eomesa expression in MZnanog mutants at midgastrula stages (see Fig. 3G) could be rescued by miR-430 but not by mismatch control microRNA injection (Fig. 6I, Table S2). Therefore, we assumed that the delay in maternal mRNA degradation, caused by the absence of miR-430, contributes to the epiboly defect in MZnanog.

Using Nanog KD experiments, Perez-Camps and colleagues have recently suggested a role for Pou5f3-Nanog complexes in patterning the dorsoventral axis (Perez-Camps et al., 2016). We checked whether the effects of Nanog KD on the expression of dorsoventral markers could be reproduced in MZnanog. We took the maternal-zygotic Pou5f3 mutant MZspg^m793 for comparison, in which the deficiency of ventral-specifying genes has been previously documented (Reim and Brand, 2006; Belting et al., 2011). We did not observe expansion of the dorsal genes chordin (chrd) and gsc (Fig. 7A,B,F,G,K,L), but transcription of vox and vent was more prominently downregulated in MZnanog than in MZspg (compare Fig. 7H,I with 7M,N, and Fig. S8A,C,E). Remarkably, bmp2b and vox expression in the ectodermal domain was completely absent in MZnanog, but only reduced in MZspg (Fig. 7E,J,O; see also Fig. S8B,D,F). This result is in agreement with the cooperative action of Pou5f3 and Nanog in ventral mesoderm (Perez-Camps et al., 2016) and suggests an additional role for Nanog in establishing ventral domains of bmp2b and vox in the ectoderm. In support of this idea, BMP-mediated repression of SoxB1 neural genes at the ventral side of the ectoderm (Dee et al., 2007) does not occur in MZnanog (Fig. S9A).

To characterize further the interactions between Pou5f3 and Nanog gene regulatory networks, we investigated the effects of Nanog on Pou5f3 bona fide direct transcriptional regulatory targets. Pou5f3 is absolutely required for early zygotic expression of the transcription factors sox2, her3 and foxd3 (Onichtchouk et al., 2010; Iwafuchi-Doi et al., 2011). Expression of foxd3 (Fig. 3C, Fig. S3) and sox2 (Fig. S9B,C) was not strongly affected in MZnanog mutants. Expression of the strictly zygotic gene her3 first becomes visible in the future neuroectoderm at 4.7 hpf in the wild type (Hans et al., 2004; Bae et al., 2005). Unexpectedly, her3 was strongly increased at 4.7 hpf and expanded throughout the blastoderm of MZnanog (Fig. 7P). We have previously shown that accumulation of her3 transcripts over time follows a biphasic pattern, where low levels of her3 transcription start at ZGA (3 hpf) and increase to high levels prior to shield stage (6 hpf; Onichtchouk et al., 2010). To investigate whether this temporal pattern is affected in the absence of Nanog, we obtained a her3 expression time curve using quantitative RT-PCR on wild-type, MZnanog and MZspg embryos, respectively (Fig. 7Q). Strikingly, already at ZGA the expression level of her3 in MZnanog was ten times greater than the her3 wild-type control level. It further increased sharply and reached a maximum at 5 hpf. Expression time curves of the SoxB1 family neural genes sox2, sox3 and sox19a did not show similar dramatic changes in MZnanog (Fig. S9C). Thus, we concluded that in vivo Nanog acts as a potent direct or indirect repressor of her3. Pou5f3 activates her3 transcription by binding upstream of the her3 promoter (Onichtchouk et al., 2010). In order to investigate whether Nanog repression is mediated by the same genomic region, we made use of the Tg(her3enh1-GFP) transgenic reporter line, in which expression of GFP is regulated by 2.2 kb of her3 upstream sequence. This sequence harbors a conserved Pou5f3-binding regulatory site directly upstream of the TATA box (Onichtchouk et al., 2010). We microinjected Nanog or control mRNA in Tg(her3enh1-GFP) embryos at the 1-cell stage, harvested the embryos at 6 hpf and measured the changes of endogenous her3 RNA and GFP by quantitative RT-PCR. Whereas overexpression of Nanog did not change GFP levels (Fig. 7R), levels of endogenous her3 RNA (Fig. 7S) were significantly downregulated by both concentrations injected. We concluded that Nanog represses her3 by using regulatory sequences outside of the 2.2 kb upstream DNA element.

Cell death, observed in MZnanog mutants at the end of gastrulation, could either result from a cell-autonomous requirement for Nanog or from the absence of important extracellular or extra-embryonic signals in the MZnanog embryo. To determine whether Nanog is cell-autonomously required for cell survival, we performed a series of transplantation experiments of labeled wild-type and MZnanog donor cells to the unlabeled wild-type host. Both donors and host were at the blastula stage (sphere, 4 hpf). At this early stage, cells are considered to be pluripotent (Ho and Kimmel, 1993). First, in a set of preliminary experiments, we transplanted labeled wild-type cells into wild-type recipients and examined them after 24 hpf, to define which cell types are easily recognizable and abundant enough to score (see Materials and Methods). By cell shape and position within the 24 hpf embryo, we could reliably distinguish seven cell type groups: four groups of ectoderm derivatives (skin, neural, placode, floorplate; Fig. 8A-D), two groups of mesoderm derivatives (muscle and notochord; Fig. 8E,F) and one group of undefined origin, which comprised singular bean-shaped cells, positioned directly underneath the embryonic epidermis (Fig. 8G). We then addressed the possible cell-autonomous differences between MZnanog and wild-type cells by co-transplanting cells of both genotypes labeled with different fluorescent colors into an unlabeled wild-type host at the sphere stage (52 host embryos were transplanted in ten experiments). The fluorescent cells in the 24 hpf host embryos that matched to one of the defined seven cell types were scored. The order of transplantation and color of labeling dextran was alternated between wild-type and MZnanog cells to avoid experimental bias (see Materials and Methods and Table S3 for details).

Although MZnanog precursors did contribute to six out of seven cell types, this contribution was much smaller than that of the wild-type precursors (Fig. 8H,I). For all cell types except the undefined bean-shaped ones, more host embryos carried derivatives of wild-type than of MZnanog precursors (Fig. 8J). The total number of cells (885) derived from the wild-type precursors exceeded by more than four times the number of MZnanog derivatives (188). Wild-type cells were more abundant than MZnanog cells within all categories of cell types, except bean-shaped cells (Fig. 8K). The differences between MZnanog- and wild type-derived cells per host embryo were statistically significant in all categories: P-values in paired Student's t-test varied from 5.2E⁻⁷ (neural) to 0.024 (floorplate) and the P-value for the increase in bean-shaped cells in MZnanog was 0.000101 (Table S3). In summary, MZnanog precursors contributed to <20% of mesodermal (muscle and notochord) or ectodermal (skin, neural, placode and floorplate) cell types (Fig. 8L). Finally, we estimated the cell type distribution within all differentiated cells within each genotype (Fig. 8M,N). Whereas non-defined bean-shaped cells were scored as <1% (0.67%) of all wild-type cells, they represented 15% of all MZnanog cell derivatives. We hypothesize here that the bean-shaped cells are simply remnant transplanted blastula cells. They did not divide further (thus stayed big) and also did not differentiate. It is possible that they might become stuck below the EVL/skin because of adhesion problems.

Genomic studies and transcriptome analyses (Xu et al., 2012; Lee et al., 2013) revealed a broad range of Nanog genomic binding sites and transcriptionally regulated targets in zebrafish, suggesting that Nanog contributes to multiple developmental processes. However, the extent to which Nanog might contribute in specific, non-redundant ways to biological regulation was remained controversial.

Here, we investigated the cell-autonomous and non-cell autonomous requirements for Nanog in embryonic (deep cells) and extra-embryonic (YSL and YCL) tissues using maternal-zygotic mutant MZnanog embryos. Taken together, these findings suggest that Nanog influences diverse developmental processes, and interferes with more than one signaling cascade. The pleiotropic action of Nanog resembles that of Oct4/Pou5f3 (summarized in Fig. 9).

The YSL nuclei migration defect and the deficiency of the F-actin ring in MZnanog embryos resemble the phenotype described for Mxtx2 morphants (Wilkins et al., 2008). However, the yolk burst phenotype observed in 100% of Mxtx2 KD embryos was not readily apparent in MZnanog (Fig. 1C, Fig. 4O,P). Additionally, yolk doming (Fig. 4K-N) and YCL microtubule organization (Fig. 4R,S) were abnormal. Injection of mxtx2 mRNA into the YSL at the 512-cell stage did not rescue epiboly in MZnanog (Fig. S5), strongly suggesting that the absence of mxtx2 can only partially explain the severe yolk cell phenotype of MZnanog mutants.

Previous studies revealed that Mxtx2 is strictly required only in the YSL (Bruce et al., 2005; Wilkins et al., 2008). Therefore, the rescue of MZnanog by 1-cell stage injection of mxtx2 mRNA (Fig. 6A-C) is somewhat puzzling. In addition to restoring normal Mxtx2 function in the YSL, Mxtx2 was misexpressed in the deep cells, where it is normally not present. As found previously (Xu et al., 2012), Mxtx2 genomic binding sites largely overlap with Nanog-binding sites. One can therefore speculate that Mxtx2 might directly activate Nanog target genes when overexpressed. Another possible explanation for Mxtx2-mediated rescue of MZnanog is the ability of Mxtx2 to activate Nodal signaling (Hong et al., 2011; Xu et al., 2012). Motile and adhesive endodermal cells migrate to the interface of blastoderm and yolk and spread by random walk movement (Warga and Nüsslein-Volhard, 1999; David and Rosa, 2001). Mxtx2 induces supernumerous endoderm cells in MZnanog (Fig. 6G,H, Fig. S7). Their properties might artificially compensate for MZnanog epiboly defect. In any case, rescue of MZnanog by mxtx2 injection into 1-cell stage embryos is based on an ectopic overexpression artifact. The yolk cell phenotype in MZnanog is likely caused by a combination of two or more independent defects: abnormal microtubule structures, the absence of Mxtx2 and its target genes and, presumably, also excessive maternal mRNA accumulation in the yolk cell.

Nanog directly activates transcription of miR-430 (Lee et al., 2013), which is in charge of mRNA clearance for a large group of maternal transcripts after ZGA (Giraldez et al., 2006). Zebrafish MZdicer mutants, in which the whole microRNA processing pathway is inactive, show strong epiboly delay without affecting axis formation or other gastrulation movements (Giraldez et al., 2005). Accumulation of non-degraded maternal transcripts contributed to epiboly delay in MZnanog, as miR-430 overexpression in the whole embryo significantly improved epiboly (Fig. 6A, Fig. S6). Experiments performed by Gagnon et al. (2018) demonstrate that MZnanog blastoderm transplanted to the wild-type yolk completes epiboly, thus tolerating excessive maternal RNA load in the deep cells. By contrast, our miR-430 rescue shows that just removing maternal mRNA load from intact embryo allows the MZnanog embryo to pass through epiboly in spite of severe yolk cell defects. One cannot exclude the possibility that the absence of miR-430-mediated mRNA degradation in the yolk cell of MZnanog also contributes to the mutant yolk cell phenotype, which would explain the results of both papers. In addition, the morphology of EVL surface cells apparently differs between wild type and MZnanog (Fig. 4O,P, Fig. S4); thus, abnormal EVL in MZnanog might also contribute to doming defects (Morita et al., 2017), although this possibility needs further investigation. From all available data, we can conclude that multiple factors contribute to epiboly failure in MZnanog.

We demonstrate that MZnanog mutants show increased cell death at the end of gastrulation (Fig. 2). We have shown in a cell transplantation assay that survival of selected morphologically defined cell lineages – muscle, neurons, skin and placodes – is reduced in MZnanog compared with wild-type cells, and survival of all cell types seems to be equally diminished (Fig. 8). This cell-autonomous phenotype could be rescued only by supplementing nanog mRNA or by a paternal nanog wild-type allele in Mnanog, but not by miR-430 microinjection or by ectopic overexpression of Mxtx2 (Fig. 6). Transplantation results of GFP transgenic Mnanog heterozygous cells, as performed by Gagnon et al. (2018), taken together with our MZnanog cell transplantation results, demonstrate that nanog is not absolutely required for cell differentiation into specific lineages. Rather, it is needed for cell viability and robustness of lineage specification. The ‘bean-shaped cells’ in our experiments might represent a cell type that appears only upon transplantation (their shape fits that of blastoderm cells). Despite many developmental signals, they remain undetermined and may develop no adhesion profile to fit into tissues. Although it is generally believed that in mammals Nanog is mostly dispensable after pluripotency is established, mouse Nanog mutant ESCs also have reduced viability and differentiation capacity compared with wild-type ESCs (see figure 1F in Chambers et al., 2007), resembling zebrafish MZnanog cells.

MZnanog and Pou5f3 (Oct4 paralogous) MZspg mutant embryos are similar in some aspects of their phenotype, such as epiboly delay, misregulation of ventral genes, and enhanced cell death at the end of gastrulation, but differ in other aspects, such as the ability to form sox17-positive endoderm (illustrated in Fig. 9). Both transcription factors contribute to epiboly by regulating the YCL microtubule cytoskeleton (Fig. 4R-S′). Other known pathways involved in epiboly regulation rely on different target genes and mechanisms downstream of Pou5f3 (Song et al., 2013) and Nanog. In contrast to MZnanog, mxtx2 expression is not changed in MZspg (Onichtchouk et al., 2010), and YSL epiboly occurs relatively normally (Lachnit et al., 2008). Nanog and Oct4/Pou5f3 synergize with the ventral specification pathway via BMP signaling and Vox (Peres-Camps et al., 2016; this study). Pou5f3 regulates cell survival by inhibiting p53 (tp53) and activating mych (Kotkamp et al., 2014). The decrease in cell survival, observed in both Nanog and Oct4/Pou5f3 mutants, could be caused by multiple independent changes due to a widespread role of Nanog and Pou5f3 in gene expression. Although each of these changes is not essential for cell survival, the multiple changes add up, providing an effect which reduces the robustness and viability of cells.

It is known that Nanog and Oct4 often co-occupy the same regulatory sequences to directly activate gene expression (Boyer et al., 2005; Lee et al., 2013). However, this does not appear to be the case for all Nanog/Oct4 target genes. We show regulation of her3 as an example of antagonistic regulation by Pou5f3 and Nanog: her3 is activated by Pou5f3 (Onichtchouk et al., 2010), and strongly repressed by Nanog (Fig. 7P,Q). We further show that Pou5f3 and Nanog use different her3 regulatory regions (Fig. 7R,S).

Taken together, our data suggest that Nanog, similarly to Oct4/Pou5f3, controls expression of a large network of developmentally important signaling molecules and transcription factors.

Fish were raised, maintained and crossed under standard conditions as described by Westerfield (2000). Embryos obtained from crosses were raised in egg water at 28.5°C. Developmental stages of MZspg and MZnanog embryos were indirectly determined by observation of wild-type embryos born at the same time and incubated under identical conditions. For controls, we used the AB-TL wild-type embryos. Staging was performed following the 1995 Kimmel publication (Kimmel et al., 1995). All experiments were implemented in accordance with German Animal Protection Law (TierSchG).

For nanog mutant generation, we followed the published protocol of Cermak et al. (2011). Using Mojo hand on the website www.talendesign.org/ (Neff et al., 2013), we created the TALENs targeting the codon encoding the 12th amino acid of the first exon of the nanog gene (spacer sequence is indicated by square brackets, flanking TAL1 and TAL2):

5′-TTTAACCCATCTTATC[ATGCATATGCATACG]GGCTCATGTACCCGCA-3′ (according to assembly Jul. 2010, zv9/danRer7, chr24:12,738,352-12,738,398).

The targeting site contains the restriction site (5′-CATATG-3′) for the enzyme NdeI (NEB), which cuts between CA and TATG. For assembling the correct RVDs, we used the Golden Gate TALEN and TAL Effector Kit 2.0 (Kit #1000000024, Addgene), which includes all plasmids needed. As a destination vector we chose the pRCIscript-Goldy TALEN (Addgene plasmid #38142, deposited by Daniel Carlson; Carlson et al., 2012). After successfully generating the two TALEN plasmids, we performed RNA transcription with the mMESSAGE mMACHINE T3 Transcription Kit (Ambion) followed by injection of RNA into AB-TL wild-type embryos at the 1-cell stage to the yolk with a concentration of 100 ng/µl for each TALEN-RNA. To check the proper functioning of the TALENs, we extracted genomic DNA from 20 each of 24 hpf wild-type and injected embryos by lysing the cells with lysis buffer (10 mM Tris pH 8, 50 mM KCl, 0.3% Tween20, 0.3% NP-40, 1 mM EDTA) and incubating the embryos at 98°C for 10 min. After cooling down, Proteinase K solution (20 mg/ml, A3830, AppliChem) was added and incubated overnight or at least 2 h at 56°C. Subsequently, Proteinase K was destroyed by heating the samples to 98°C for 10 min. For PCR, we designed primers flanking the targeting site: Nanog-f1 5′- TCTAAACCCGCCCACAACC-3′ and Nanog-r1 5′-GGTCGGGCTCAGTCTTGTTG-3′ resulting in a 523 bp product. After digestion with NdeI (overnight, 37°C) fragments of 284 and 238 bp lengths were generated in the case that no mutation had taken place. In more than 70% of nanog TALEN-injected single G0 embryos an undigested DNA band could be detected, indicating high efficiency of mutagenesis. Eleven out of 19 tested G0 mosaic fish transmitted the mutation to the F1 generation. Two of G0 nanog TALEN-injected mosaic female founders, when crossed to wild-type males, generated abnormal embryos phenotypically resembling morpholino KD Nanog phenotypes (Xu et al., 2012), which we interpreted as a bi-allelic targeting of the gene. We selected a founder carrying the 10 bp deletion (5′-TTTAACCCATCTTATCAT|ACGGGCTCATGTACCCGCA-3′; the line shows the position of the missing base pairs), which caused a frame shift and a premature stop codon after 15 amino acids (Fig. 1A), for further experiments. Resulting F2 homozygous nanog mutants developed into fertile adults, suggesting that Nanog zygotic contribution is not essential for development.

To visualize the expression pattern of some chosen genes, we performed whole-mount in situ hybridization as previously described (Holzschuh et al., 2003), with some modifications. Embryos were fixed at the appropriate stages with 4% paraformaldehyde (PFA) at 4°C overnight. The next day, the fixed embryos were washed thrice in PBST (8 g/l NaCl, 0.2 g/l KCl, 16 mM Na₂HPO₄, 4 mM NaH₂PO₄, 0.1% Tween-20, pH 7.3) for 5 min and were dechorionated. The dechorionated embryos were dehydrated in stepwise increasing concentrations of methanol (25%, 50%, 75% in PBST, finally in 100%) and were stored in 100% methanol at −20°C until further use. The sources of the probes used for in situ hybridization are listed in Table S5. To create the plasmid templates for eomesa, mxtx2 and sox19a, corresponding partial open reading frame sequences were PCR-amplified using cDNA from 4-6 hpf embryos and primers listed in the Table S6. PCR products were then cloned into PCRII TOPO vector (Invitrogen).

The embryos were rehydrated in stepwise decreasing concentrations of methanol (75%, 50% and 25% in PBST), each 5 min on a shaker. Subsequently, they were washed three times in PBST. Afterwards, the embryos were transferred into 300 µl Hyb-Mix (50% formamide, 5× SSC, 50 μg/ml heparin sodium salt, 0.1% Tween-20, 5 mg/ml torula RNA) and incubated for 4 h at 65°C. For hybridization, the Hyb-Mix was replaced by 300 µl Hyb-Mix containing the desired probe (0.5-2.0 µl/100 µl) and incubated overnight at 65°C.

The probes were removed and the embryos were transferred into a 48-well plate with a net as bottom and placed into the in situ robot BioLane HTI using program HL2_3. The robot washed the embryos thrice with 50% formamide in 2× SSCT (20× SSC: 3 M NaCl, 300 mM trisodium citrate; 2× SSCT and 0.2× SSCT contain the corresponding SSC strength together with 0.1% Tween-20) for 20 min, then once with 2× SSCT for 15 min and thereafter thrice with 0.2× SSCT for 20 min. All steps were carried out at 65°C.

The next steps were performed at room temperature (RT). The embryos were washed twice with PBST for 10 min and subsequently incubated for 2 h in blocking solution [2% heat-inactivated goat serum, 2 mg/ml bovine serum albumin in PBST]. After blocking, the samples were incubated with an anti-digoxigenin antibody (1:5000 in PBST; Roche Diagnostics, 11 093 274 910) overnight at 4°C.

Embryos were rinsed once in PBST and then washed six times in PBST for 20 min each wash. Afterwards, they were washed for 10 min in 100 mM Tris-HCl pH 9.5. The embryos were then removed from the robot and transferred into 12-well plates and incubated in staining buffer (100 mM NaCl, 50 mM MgCl₂, 100 mM Tris-HCl pH 9.5, 0.1% Tween-20, 1 mM levamisole) for 15 min. For staining, the staining buffer was replaced with staining solution [per 1 ml staining buffer 3.5 µl BCIP (stock solution 50 mg/ml in 100% DMF) and 3.5 µl NBT (stock solution 50 mg/ml in 100% DMF)] and incubated in the dark on a shaker until a proper coloring developed. Unbound dye was removed by washing six times with stop solution for 20 min or even longer. For storage, the embryos were moved through incubations in solutions with an increasing concentration of glycerol (25%, 50% and 75% in stop solution) until 90% glycerol in stop solution (PBST pH 5.5, 1 mM EDTA), at which point the embryos sank down and then were placed to 4°C storage. Following this, images were recorded with a Leica MZ APO stereo microscope, an AxioCam MRc camera and AxioVS40 V4.8.0.0 software (Zeiss).

miR-430 and mismatch control microRNAs were synthesized as described by Giraldez et al. (2005). The nanog and mxtx2 mRNA for rescue experiments were synthesized using the SP6 Ambion mMessage Machine kit according to the manufacturer's instructions. CS2+nanog and CS2+mxtx2 plasmids were linearized by NotI and used as templates.

The wild-type and MZnanog eggs for all injections were collected within 10 min. YSL injections were performed according to D'Amico and Cooper (2001) at the 512- to 1000-cell stage. For 1-cell stage and YSL injections, in addition to injected mRNA or microRNA we used a dextran conjugated to Alexa Fluor 488 or tetramethylrhodamine (Thermo Fisher Scientific, 12.5 ng per embryo) to visualize the YSL or embryonic cells. Just after YSL injection, the embryos were visually examined under an epifluorescence microscope, and the mis-injected (30-60%) embryos showing fluorescence in the deep cells or in the yolk were excluded from the follow-up experimental analyses. For transplantation experiments, dextran Alexa Fluor 488 or tetramethylrhodamine dextran (Thermo Fisher Scientific) were injected at the 1-cell stage at a concentration of 12.5 ng per embryo. For the rescue of MZnanog embryos, we performed nanog mRNA injections into MZnanog mutant embryos with amounts of 50 pg, 100 pg, 150 pg and 300 pg per embryo. Use of 150 pg per embryo usually resulted in the healthiest embryos when injected at the 1-cell stage. mxtx2 mRNA injections into 1-cell stage MZnanog mutant embryos were performed with amounts of 6 pg, 12 pg, 25 pg, 50 pg, 75 pg and 150 pg per embryo. Use of 12 pg per embryo was optimal for epiboly rescue when injected at the 1-cell stage. As controls for mRNA, we injected unrelated H2B mRNA and dextran in YSL of wild-type and mutant embryos, also at the 512-cell stage. miR-430 or mismatch microRNAs were injected in the range 1.25 to 10 nM per embryo; 10 nM miR-430 per embryo resulted in the best epiboly rescue.

To visualize yolk syncytial nuclei (YSN), we injected 1 nl of 0.25 mM of SYTOX Green (Thermo Fisher Scientific) mixed with 0.2 M KCl into wild-type and MZnanog embryos at the 512-cell stage as described by D'Amico and Cooper (2001). Several embryos were embedded into 1.5% low melting agarose and simultaneously recorded for up to 5 h with a Leica MZ16 FA stereo-fluorescence microscope and a Leica DFC 350 FX camera. Additionally, wild-type and MZnanog embryos were injected with SYTOX Green and tetramethylrhodamine dextran (12.5 ng per embryo), embedded in 1% low melting agarose, placed in a Zeiss capillary and imaged with ZEISS Lightsheet Z.1 equipped with W Plan-Apochromat 10×/0.5NA (water immersion) Working Distance=3.7 mm and W Plan-Apochromat 20×/1.0NA Corr (water immersion) Working Distance=2.4 mm objectives at 4.7 hpf. Four-version z-stack pictures were processed with Zeiss ZEN black 2014 (dual side fusion and multi-view processing). Lightsheet z-stack overview pictures of the lateral halves of the embryos were exported via Arivis Vision4D. Green channel was subtracted from the red channel in the Lightsheet images to increase the visibility of the YSL nuclei (green) within the YSL cytoplasm (red).

To visualize the cell membranes and cell nuclei, embryos were fixed at 3 hpf with 4% PFA overnight at 4°C. Then the embryos were washed in PBST and dechorionated. For permeabilization, the embryos were treated either with a combination of 0.5% Triton X-100 and 0.5% NP40 for 1 h or in a PBSDT solution (1% DMSO and 0.3% Triton X-100 in PBST) overnight. In order to stain the deep cells as well, the embryos were washed thrice in 0.3% Triton X-100 in PBST and afterwards stained in PBSDT. Wild-type and MZnanog embryos were stained with a rhodamine-phalloidin staining solution (6.6 µM; 1:100 in PBSDT) for 2 days. While the embryos were in the staining solution, they were first kept at RT for 4 h and then on a slow shaker at 4°C. The embryos were always kept in the dark. They were washed with PBST to stop the staining and a SYTOX Green staining solution (5 mM; 1:40,000 in PBSDT) was added afterwards. The embryos were stained in the dark for 1 h and 10 min at RT. To stop the staining, the embryos were washed once in PBSDT for 15 min and then once with PBST for 15 min. They were then embedded in 1.5% low-melting agarose and images were taken with a Zeiss LSM 510 META NLO confocal inverted microscope.

SYTOX Green (0.2 mM, Thermo Fisher Scientific) was mixed with 0.2 M KCl and 1 nl was injected into wild-type and MZnanog mutant embryos at the 1-cell stage. Embryos that showed sufficient SYTOX Green fluorescence were selected under the fluorescence stereomicroscope and placed in a furrow of an agarose plate. They were covered with a 24×60 mm cover slide, which was fixed with 2% low melting agarose and overlaid with 0.3× Danieau's buffer to prevent drying out of the embryos. Subsequently, the development of the embryos was documented using a Leica MZ16 FA fluorescence stereomicroscope and Leica Microsystems LAS AF software, which took a brightfield image and a fluorescent image (GFP filter, 395 nm) every 3 min until the next morning. The temperature during documentation was 25°C. Resulting images were analyzed with ImageJ by counting the single images between each cell division. The obtained number of images was multiplied by 3 min to achieve the time span in minutes between the cell divisions.

We used a modified procedure from Kotkamp et al. (2014). Embryos were fixed at 3, 4 and 4.7 hpf in Gard's FG fixative for 2 h and incubated overnight in PBST containing 0.33 M glycine. After manual dechrorionation, embryos were stored in methanol in −20°C. Then 30-50 embryos per stage per genotype were washed five times in PBSTX (1× PBST, 0.3% Triton X-100), stained with SYTOX Green working solution (5 mM; 1:40,000 in PBSTX) for 30 min, washed three times in PBST and transferred into 90% glycerol in PBS. One image per embryo was taken at animal-lateral view with a Leica MZ16 FA stereo-fluorescence microscope and a Leica DFC 350 FX camera (75×, 300 ms, 3.0 gain) in the GFP channel. Images were processed with Adobe Photoshop CS6 (‘Unsharp Mask’ filter, amount: 150%, radius: 5 pixels, threshold: 0), then brightness and contrast were automatically adjusted in ImageJ (open source software). Total nuclei and mitotic nuclei were counted manually (Fig. S1C,D). ‘Mitotic cells’ refer to the cells going through metaphase and anaphase, for which the elongated shape of chromosomes could be easily recognized (Fig. S1D, arrows). Metaphase plates were counted as one, and anaphase as two mitotic cells. The ratio of mitotic to total cells was expressed in mitoses per 100 cells.

For detection of dead (apoptotic) cells we used a modified Acridine Orange staining procedure (Westerfield, 2000). At 9 hpf (when the non-injected wild-type control embryos reached 90% epiboly), living embryos were transferred to 5 µg/ml Acridine Orange (AO) solution in egg water and incubated for 30 min. The embryos were washed in the egg water three times for 15 min on a slowly rotating shaker in the dark. Images (15-30 per condition) were taken in GFP channel using a Leica MZ16FA microscope equipped with a Leica DFC350FX camera. Acquisition of images lasted several hours until wild-type controls reached the 3- to 6-somite stage, therefore care was taken to alternate control and experimental images. Exposure time was kept the same for all the images; no bleaching was observed. Images were then semi-automatically processed to quantify dead cells as illustrated in Fig. S2. Images in TIFF format were batch-processed in Adobe Photoshop (‘Unsharp Mask’ filter, amount: 244, radius: 9.3 pixels, threshold: 0). Total area, number and average size of AO-stained spots was recorded for each image as shown in Fig. S2 in ImageJ using the macro:

• //run(“Threshold…”);

• setThreshold(60, 255);

• run(“Analyze Particles…”, “size=4-100 show=Nothing display exclude summarize”);

• run(“Open Next”).

The macro was run with three different thresholds for each experiment. Statistical differences in cell death (expressed as either total area or number of AO-stained spots per embryo, Fig. S2) between the experimental groups were threshold-independent. The results were expressed as total area of AO-stained spots and normalized to mean MZnanog control value.

Embryos were fixed at the desired stage in 4% PFA overnight at 4°C. After washing thrice in PBST, rhodamine phalloidin (1:500 in PBST) was added and incubated at RT for 4 h in the dark. Embryos were washed twice in PBST for 15 min to remove unbound rhodamine phalloidin. Subsequently, the embryos were mounted on a slide with a hollow in a drop of Dako Faramount Aqueous Mounting Medium and covered with a cover slide. Embryos were documented using a Zeiss Axio Examiner D1 microscope, an AxioCam MRc camera and Zen 2012 blue edition software (Zeiss), using the Alexa 555 emission filter. Images were edited with Zen 2012 SP1 black edition software.

For visualization of the microtubules, embryos were fixed in Gard's FG fixative and anti-β-Tubulin staining (clone KMX-1 at 1:500, Merck, MAB3408) was performed as previously described (Gard, 1991; Solnica-Krezel and Driever, 1994).

Quantitative RT-PCR was performed as described by Onichtchouk et al. (2010). rpl5b was used as a normalization control. Sequences of PCR primers are listed in Table S4.

To obtain a her3enh1-GFP reporter construct, we inserted EGFP downstream of the 2.2 kb her3 regulatory sequence in the PCRII-topo vector (Onichtchouk et al., 2010). The region containing her3GFP sequence was excised, gel-purified and microinjected into F0 AB-TL fish to achieve random integration into the genome. The F1 fish positive for integration were selected at 24 hpf by EGFP fluorescence pattern, recapitulating her3 expression (Hans et al., 2004), and further crossed. The Tg(her3enh1-GFP) was recorded in the internal database under allele number m1430 and is available upon request.

Transplantations were performed as described by Ho and Kimmel (Ho and Kimmel, 1993). The wild-type and MZnanog donor cells were labeled with dextran Alexa Fluor 488 and tetramethylrhodamine dextran, MW 10 KDa (Thermo Fisher Scientific, 12.5 ng per embryo) by injecting into the 1-cell stage. Five to 15 cells were taken from the injected wild-type donor and MZnanog donor at sphere stage (4 hpf) and simultaneously transplanted into the wild-type recipient at the sphere stage, using the same transplantation needle. Pictures were taken from 24 hpf embryos using a Zeiss Axio Examiner D1 microscope and a Zeiss AxioCam MRc camera. The order of transplantations and labeling colors between wild-type and MZnanog donors was swapped in each experiment (ten experiments in total, Table S1). Transplanted cells were scored in 52 embryos in total. Embryos were selected for scoring using the following criteria: at least one recognizable cell type labeled with the first color and any (including unknown) cells labeled with the second color should be visible in the same embryo. Cell numbers and genotypes of transplanted cells in each of 52 embryos analyzed are listed in Table S3.

We thank Wolfgang Driever and Sebastian Arnold for discussion, and Anna Gebhard, Rainer Duden, Lev Yampolsky, Logan Doil and anonymous reviewers for comments on earlier versions of the manuscript. We are grateful to James Gagnon and Alex Schier for sharing their data before publication. We are also grateful to Sabine Goetter for excellent fish care and to Andrea Buderer for administrative support, and to LIC for the demonstration of Lightsheet Microscopy.