AB and mutant zebrafish were reared as described (Westerfield, 2000). For eomesa mutant experiments eggs from homozygous eomesa^fh105/fh105 females were in vitro fertilized with eomesa^+/fh105 sperm yielding a mixture of Meomesa and MZeomesa mutant embryos. Since previous studies have revealed no differences in endodermal or mesodermal expression between Meomesa and MZeomesa mutant embryos we did not distinguish between them in this study (Du et al., 2012; Xu et al., 2014). All zebrafish studies complied fully with the United Kingdom Animals (Scientific Procedures) Act 1986 as implemented by King’s College London, The University of Warwick, or were in accordance with the policies of the University of Toronto Animal Care Committee.

Full length tbx16 and eomesb open reading frames were cloned with C-terminal myc tags into XhoI and XbaI sites in pCS2+ by PCR from zebrafish cDNA using the following primers: tbx16-myc CAT​ACT​CGA​GAT​GCA​GGC​TAT​CAG​AGA​CCT and CGC​GTC​TAG​ACT​ACA​GAT​CCT​CTT​CTG​AGA​TGA​GTT​TTT​GTT​CCC​AGC​ACG​AGT​ATG​AGA​AAA; eomesb-myc ATA​TCT​CGA​GAT​GCC​CGG​AGA​AGG​ATC​CAG and GCG​CTC​TAG​ACT​ACA​GAT​CCT​CTT​CTG​AGA​TGA​GTT​TTT​GTT​CGC​TGC​TGG​TGT​AGA​AGG​CGT​A. Full length gata5 cDNA with a C-terminal HA tag was similarly cloned into pCS2+ EcoRI and XhoI sites using primers CGC​CGA​ATT​CAT​GTA​TTC​GAG​CCT​GGC​TTT and AAT​GCT​CGA​GTC​AAG​CGT​AAT​CTG​GAA​CAT​CGT​ATG​GGT​ACG​CTT​GAG​ACA​GAG​CAC​ACC. Eomes cloning into pCS2+ was performed between EcoRI and XhoI sites. pCS2+eomesaN320K was produced by PCR mutagenesis of the wild type construct using AAA​CTG​AAG​CTA​ACC​AAC​AAG​AAA​GGA​GCA​AAT​AAC​AAC​AAT and TCC​GAA​AGA​TAT​TTC​TTG​TCT followed by recircularization. Eomesa ∆CTD was similarly produced using the following primers TAA​GAA​CTG​CTT​TTC​AAG​ATC​CTT​TAT​CAA​TCC and CGA​ATC​ATA​ATT​GTC​CCT​GAA. The ∆NTD mutation was produced by removing the EcoRI/BstEII fragment from pCS2+eomesa and replacing with the EcoRI/BstEII fragment produced by PCR from pCS2+eomesa with primer pair GCC​CTC​GAA​TTC​ACA​GTT​AAG​AAT​GGC​GCG​GGC​GC and CCC​GCA​GGT​CAC​CCA​CTT​TCC​GCC​CTG​AAA​TCT​CCA.

All capped mRNA were synthesized from plasmids encoding proteins of interest in pCS2+ NotI linearization followed by SP6 transcription as described (Bruce et al., 2003), with the exception tbxta which was produced from pSP64T as described (Marcellini et al., 2003). mRNA quantities for T-box factors were scaled in order to inject equimolar amounts of each mRNA per embryo. One-cell stage embryos were injected with the following quantities: eomesa – 400pg; Eomes∆VR – 410 pg; EomesFL – 420 pg; eomesa∆NTD – 308 pg; eomesa∆CTD – 285 pg; eomesaN320K – 400 pg; eomesb-myc – 286 pg, tbx16-myc – 217 pg; tbxta – 223 pg; gata5-HA – 140 pg; mixl1 – 200 pg. For Tbx16 knockdown one-cell stage embryos were injected with, 0.5 pmol of a previously characterized tbx16 morpholino (GeneTools) (Bisgrove et al., 2005).

Unlabelled in vitro translated protein was produced using rabbit reticulocyte lysates according to manufacturer’s protocol (Promega).

Total RNA was extracted from specified cell types using Rneasy Mini Kits (QIAGEN), and polyA selected using Oligotex mRNA Mini Kits (QIAGEN). 500ng polyA + RNA per lane was size fractionated on a 1.5% agarose/MOPS gel, transferred onto Hybond N membranes (GE Healthcare), and probed with ³²P-random-primed 1 kb XmaI-EcoRV cDNA fragment corresponding to the exon 1–4T-box region.

Cell lysates were prepared using radioimmunoprecipitation assay (RIPA) buffer, subjected to SDS–polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride membranes. Membranes were blocked with 5% milk powder in Tris-buffered saline with Tween 20, incubated in primary antibodies overnight including rabbit anti-Eomes CTD (Abcam, ab23345, 1:2,000), rabbit anti-Eomes NTD (Santa Cruz, sc-98555, 1:1,000) and rat anti-Eomes (eBioscience, 14–4,876, 1:1,000). Secondary antibodies were donkey anti-rabbit horseradish peroxidase (GE Healthcare NA934, 1:2,000) and goat anti-rat horseradish peroxidase (GE Healthcare NA935, 1:2,000). Blots were developed by chemiluminescence using Amersham ECL Prime Detection Reagent (GE Healthcare).

Wild type (+), Eomes ^null/null (null), feeder-depleted ESCs were cultured in DMEM (ThermoFisher) with 15% FCS, 1% non-essential amino acids, 0.1 mM β-mercaptoethanol and 1,000 U/ml recombinant leukaemia inhibitory factor (Millipore). For differentiation ESCs were resuspended at 1×10⁴ cells/ml in DMEM (ThermoFisher) with 15% FCS, 1% non-essential amino acids, 0.1 mM β-mercaptoethanol in hanging drops (10 μL) plated on the inside lids of bacteriological dishes. After 48 h embryoid bodies were transferred in 10 ml medium to 10 cm bacteriological dishes and RNA extracted at the appropriate timepoints.

P19Cl6 embryonal carcinoma cells were cultured in α-MEM (ThermoFisher) supplemented with 10% FCS. To induce differentiation, cells were seeded at 3.7×10⁵ cells/6 cm dish in media containing 1% DMSO (Sigma) and RNA harvested after 72 h.

CTLL cells derived from the ATCC TIB-214 line were maintained at 10⁴–10⁵ cells per ml in complete T cell medium supplemented with IL-2. SNH fibroblasts and HeLa cells were maintained on 0.1% gelatin coated dishes in DMEM supplemented with 10% bovine calf serum (Hyclone).

Cytoplasmic RNA was produced as previously described (Eggermont and Proudfoot, 1993). Total RNA was produced using Rneasy Mini Kits according to manufacturers protocol (QIAGEN). RT-PCR was performed using OneStep RT-PCR Kit (QIAGEN) using the following primers: Total Eomes–TGTTTTCGTGGAAGTGGTTCTGGC and AGG​TCT​GAG​TCT​TGG​AAG​GTT​CAT​TC; Eomes exon 4-6 to distinguish FL and ∆VR isoforms ATC​GTG​GAA​GTG​ACA​GAG​GAC​G and CGG​GAA​GAA​GTT​TTG​AAC​GCC; Gapdh–TGCACCACCAACTGCTTAGC and GGC​ATG​GAC​TGT​GGT​CAT​GAG; Eomes start codon to ∆CTD 3′ UTR–ATATCTCGAGATGCAGTTGGGAGAGCAGCTC and TGG​GCT​CGA​AGA​TGA​AAC​TC; HBB exon 2 to Eomes exon 6 – GCA​CGT​GGA​TCC​TGA​GAA​CT and CGG​GAA​GAA​GTT​TTG​AAC​GCC. For nested PCR to test exon 5-6 splicing association with the long Eomes 3’ UTR the initial primer pair used was ATC​GTG​GAA​GTG​ACA​GAG​GAC​G and CAA​GTA​CGG​AGG​CAG​CTG​AG.

Whole-mount in situ hybridization (WISH) of zebrafish embryos were performed as described (Jowett and Lettice, 1994). Anti-sense riboprobes for noto (Talbot et al., 1995), chrd (Miller-Bertoglio et al., 1997), vgll4l (Nelson et al., 2014), zic3 (Grinblat and Sive, 2001), mixl1 (Alexander et al., 1999) and sox32 (Dickmeis et al., 2001) were produced as described. Blinding and randomisation was performed prior to categorical scoring of WISH embryos to prevent bias.

Clones to test the splicing efficiency at Eomes exon 6 were generated by cloning PCR products using primers ACG​GCA​ATT​GGC​CTC​GAA​CAT​TCT​TGC​TTC and CCA​GCC​ATC​ACT​TTG​GTC​AAA​GGT​GGA​AGG​CAA​AAG into MfeI-BstXI sites of the human HBB gene (GenBank accession no. U01317) within a previously described TAT-inducible expression vector (Ashe et al.,1997). Mutation of splicing sequences for the ∆VR and FL Eomes isoforms were introduced by PCR using primers: ∆VR - CAT​GTA​CAC​GGC​TTC​AGA​AAA​CGA​CAG​GTT​AAC​GCC​AAG​TCC​GAC​GGA​TTC​CCC​TCG​ATC​CCA​TCA​GAT​TGT​CCC​TGG and CTA​CAA​TAT​AAA​GAG​AGA​CAC​TTA​AAA​ATA​AAA​AAC​AAC​CCT​CAC​GTT​GTC​CCC​AAA​CAA​GCT​GCC​TCC​CAG​AAG​C; FL–CATGTACACGGCTTCAGAAAATGACAGGTTAACTCCA TCTCCCACGGATTCCCC and GGA​CAT​TAT​ATA​CAC​CGC​CTC​TTA​TAT​TTT​TAC​ACC​AAC​CCT​CAC​GTT​GTC​CCC​AAA​CAA​GCT​GCC​TCC​CAG​AAG​C followed by recircularization of the resulting PCR products. Deletion of the VR was similarly performed using primers ATC​CCA​TCA​GAT​TGT​CCC​TGG​A and CTA​CAA​TAT​AAA​GAG​AGA​CAC​TTA​AAA​ATA​AAA​AAC​AAC​CCT​CAC​GTT​GTC​CCC​AAA​CAA​GC T​GCC​TCC​CAG​AAG​C. HBB plasmids were co-transfected with a plasmid expressing the HIV transactivator protein TAT (Adams et al., 1988), into HeLa cells using Lipotectamine 2000 according to manufacturers protocol (ThermoFisher).

The Tbr1 subfamily Gene Tree was generated by Ensembl (Yates et al., 2016). Eomes conservation measurements (phyloP) across 60 vertebrate species were visualized in UCSC Genome Browser (http://genome.ucsc.edu/) (Karolchik et al., 2004; Speir et al., 2016). Sequence logos of the Eomes variant region in placental mammals, other tetrapods and teleosts were based on alignment of the same 60 vertebrate species and visualized using WebLogo (Crooks et al., 2004).

Full-length protein alignments were performed using Clustal Omega (Goujon et al., 2010; Sievers et al., 2011; McWilliam et al., 2013) and visualized using JalView (Waterhouse et al., 2009). BLOSUM62 average distance gene tree was also produced using Jalview.

Single-cell (sc) RNA sequencing count data of zebrafish embryonic cells from Wagner et al (2018) was downloaded from GEO (Barrett et al., 2013; Wagner et al., 2018). Raw UMI-filtered count data in CSV format from 6 h post fertilisation (h.p.f.) embryos (GSM3067190) was imported in to R v3.6.2 and analysed using Seurat v3.0.2 (Stuart et al., 2019). Cells with less than 200 features, and features detected in <3 cells were discarded. The remaining count data were then normalised via SCTransform v0.2.1 (Hafemeister and Satija, 2019) with mitochondrial genes passed as a regression variable. Genes were clustered using UMAP utilising the R package uwot v0.1.5 (Melville et al., 2020), with the following parameters: dims = 1:30, n.neighbors = 5, min.dist = 0.001. To assign cell identities to clusters FindAllMarkers was called in Seurat using default parameters. For consistency with the original source publication of the 6 h.p.f. scRNA-seq data we cross-referenced the marker genes for each cluster in the present study with the clusters defined by Wagner et al. (2018). Significant positive markers in each of the 14 clusters defined by our analyses in Seurat were overlapped with the top 20 markers for each identity defined by Wagner et al. (2018). Cell identities were then assigned based on maximum concordance with markers defined by Wagner et al. (2018).

In mouse the three annotated Eomes transcripts give rise to three structurally distinct isoforms including the full length (FL) product, a splice variant having an alternative splice acceptor site within exon 6 leading to loss of a 19 amino acid variant region (∆VR), and a transcript with an alternative mRNA 3’ cleavage site leading to loss of exon 6 and its encoded CTD (∆CTD) (Figure 1A). The highly conserved VR sequence is known to be phosphorylated at three amino acid residues in mouse spleen and kidney (Figure 1A) (Huttlin et al., 2010). The internal exon 6 splicing event emerged in the tetrapod lineage through a synonymous single nucleotide change in an arginine codon (CG>AG) introducing a splice acceptor sequence.

Because the ∆CTD transcript annotation has an incomplete 5′ end, it remains unclear whether it encodes the entire NTD. However, our RT-PCR analysis using primers located in the ∆CTD 3’ UTR and at the FL/∆VR start codon suggests that exon 1 coding information is intact (not shown). The CTD encoded by exon 6 has been shown to function in transcriptional activation (Picozzi et al., 2009), suggesting that the ∆CTD isoform is likely to be functionally compromised in comparison to FL and ∆VR isoforms. Consistent with this, the CTD is more highly conserved than the NTD (Figure 1A). Functional differences between FL and ∆VR isoforms, however, are yet to be examined. Both eomesa transcripts identified in zebrafish encode the same protein (Bruce et al., 2003). Relatively little is known about the single annotated zebrafish eomesb transcript, which appears to be more divergent from the ancestral gene (Figure 1B).

Murine Eomes is expressed in numerous cell types including trophoblast stem cells (TSCs), mesendoderm, and T lymphocytes. To identify Eomes transcripts we initially performed Northern blot analysis (Figure 1C). Transcript sizes corresponding to all three annotated isoforms were detectable but the ∆CTD transcript was underrepresented. The FL and ∆VR annotations display different 3′ UTR lengths. To test whether the two distinct upper bands detected by Northern blot correspond to alternative exon 6 splicing events or merely different UTR lengths we next performed nested PCR (Figure 1D). We found that the long 3′ UTR is associated with both the FL and ∆VR coding isoforms. Strikingly, the ratio of FL/∆VR is similar for both total Eomes and the long 3’ UTR transcripts. The abundance of the different coding transcripts therefore appears to be independent of UTR length. Further analysis through cloning Eomes intron5/exon6 to replace intron2/exon3 of the human HBB gene in an expression construct followed by transfection into HeLa cells revealed that the ratio of FL/∆VR splicing is consistent with the wild type Eomes gene, suggesting that the low levels of the ∆VR isoform are due to weaker splicing consensus sequences, thus favouring the FL isoform (Supplementary Figure S1). However, analysis of Eomes proteins by Western blot indicates that various N-terminal truncations occur which cannot be accounted for by the annotated coding transcripts (Supplementary Figure S2). We conclude that FL is clearly the most abundant of the three annotated coding isoforms. Importantly, this predominant isoform expressed by mouse cells corresponds to the single eomesa isoform in zebrafish.

Zebrafish eomesa loss of function causes less severe phenotypes compared with the dramatic defects observed in Eomes mutant mouse embryos. To further explore mouse and zebrafish Eomes functional capabilities we overexpressed either mouse Eomes FL or ∆VR mRNAs in zebrafish embryos for comparison with those overexpressing zebrafish eomesa.

Zebrafish Eomesa represses ectoderm genes such as vgll4l and zic3 during blastula stages and activates mesendoderm genes including organizer markers noto and chrd at the onset of gastrulation (Bruce et al., 2003; Nelson et al., 2014). Injecting equimolar quantities of zebrafish eomesa mRNA, or mouse FL or ∆VR Eomes isoforms at the one-cell stage, we found that each was equally able to repress vgll4l and zic3 in mid/late blastulas (4 h post-fertilisation - h.p.f.) and induce noto and chrd in early gastrulas (6 h.p.f.; Figures 2A–D).

Zebrafish Eomesa is suggested to induce DFCs, based on observation of ectopic sox17 expression in cells of the outer margin in early gastrulas (7 h.p.f.) on eomesa overexpression (Bjornson et al., 2005). We sought to determine whether these ectopic sox17 + cells express a broader range of DFC markers (sox17, vgll4l and foxj1a), and also whether they can be similarly induced by mouse Eomes. Notably, though Eomesa represses expression of vgll4l during blastula stages, during gastrulation vgll4l is expressed in DFCs, and has recently been identified as a key regulator of DFC proliferation, survival and function (Fillatre et al., 2019). Repression of vgll4l at later stages in DFCs would consequently be inconsistent with promoting DFC formation.

We found that eomesa, and Eomes FL or ∆VR isoforms were all similarly able to upregulate both sox17 and vgll4l in the outer margin of gastrulas. We further note that there was a diversity of phenotypes beyond wild type expression (type I) wherein ectopic dorsolateral expression was markedly observed in individual cells/small clusters outside the primary dorsal DFC cluster (type II), where there was dorsolateral expression with no defined primary cluster (type III), and where ectopic expression was also observed in the ventral margin (type IV) (Figures 2E,F). Conversely, for foxj1a we note that while eomesa, Eomes FL and ∆VR can induce a greater intensity of dorsal staining, and expansion of the dorsal DFC cluster, ectopic expression in the ventrolateral margin is rare compared to other DFC markers (Figures 2G,H).

We conclude that both FL and ∆VR mouse Eomes isoforms are functionally highly similar to zebrafish Eomesa in these contexts. Moreover, Eomesa regulation of vgll4l appears to be context-specific, repressing its expression during blastula stages while inducing its expression in DFCs during gastrulation. We further conclude that additional factors likely to be predominantly dorsally localised are required for robust upregulation of foxj1a compared to sox17 and vgll4l.

Since results above strongly suggest mouse Eomes is functionally similar to zebrafish Eomesa, do eomesa mutants have comparatively mild defects due to functional redundancy with other T-box factors? Eomesb is not appreciably expressed during early zebrafish development (Figure 3A), and is not upregulated in eomesa mutants (Nelson et al., 2014) thus it seems unlikely that it compensates for loss of Eomesa. We recently identified a key role for the non-placental mammal T-box factor, Tbx16 in endoderm formation, with Brachyury homologue Tbxta having a more minor role (Nelson et al., 2017). Both of these factors show zygotic upregulation concomitant with declining eomesa mRNA levels and prior to expression of key markers of presumptive endoderm (e.g., gata5 and mixl1), and endoderm (e.g., gata5, sox32 and sox17; Figure 3A), thus may compensate for the early loss of expression of such markers in MZeomesa mutants (Du et al., 2012).

Moreover, single-cell RNA-seq data (Wagner et al., 2018) demonstrate that tbx16 is robustly co-expressed with the critical endodermal regulator mixl1 in the presumptive endoderm at early gastrulation stages (6 h.p.f.), suggesting the potential for Tbx16 to upregulate mixl1 expression to initiate endoderm specification, as our previous published analyses suggest (Nelson et al., 2017) (Figures 3B,C). Additional specific detail on cluster identities in Figures 3B,C is provided in Supplementary Figure S3.

To test whether Tbx16 functions redundantly with Eomesa during endoderm formation next we performed antisense morpholino knockdown of Tbx16 in wild type and eomesa mutant embryos. We found that while mixl1 expression is reduced on loss of Eomesa or Tbx16 alone, loss of both TFs leads to more striking loss of mixl1 (Figure 3D). Eomesa and Tbx16 therefore collaboratively activate mixl1 expression, strongly suggesting that Tbx16 relieves the requirement for Eomesa in zebrafish endoderm formation.

Eomesa and Mixl1 bind upstream of endoderm master regulator sox32 to positively regulate its expression prior to endoderm specification (Nelson et al., 2014; Nelson et al., 2017). Eomesa, Mixl1 and Gata5 can physically interact and their combined overexpression has been shown to induce ectopic endoderm gene expression in late blastulas and early gastrulas (Bjornson et al., 2005). Combined expression of Tbxta with Mixl1 and Gata5, however, is insufficient to induce sox32 expression (Bjornson et al., 2005). This is consistent with limited co-expression between tbxta and sox32 around the time of endoderm specification (Figure 4A). However, tbx16 and sox32 are substantially co-expressed in the endoderm at the onset of gastrulation, (Figures 3B, 4B). Tbx16 is also critical for endoderm progenitor induction (Nelson et al., 2017). We therefore extended our study to test whether Tbx16 can induce ectopic endoderm marker expression in cells at the animal pole (i.e. in cells where Tbx16, Mixl1 and Gata5 are absent in wild type embryos) through co-overexpression with Mixl1 and Gata5, as Eomesa can.

As expected, combined overexpression of eomesa, mixl1 and gata5 induces ectopic sox32 expression at the animal pole (Figure 4C). However, tbx16 did not synergise with mixl1 and gata5 to upregulate sox32 in the animal pole. This may suggest that Tbx16 and Eomesa are not equally capable of forming a complex with Mixl1 and Gata5 to induce endoderm and/or DFC fate, or alternatively that there are other key components of the complex which are capable of interaction with Eomesa/Mixl1/Gata5, but not Tbx16/Mixl1/Gata5. We conclude that Eomesa and Tbx16 perform similar functions in overlapping processes in the developing zebrafish embryo, but appear to do so via distinct molecular mechanisms.

Tbx16 and Eomesa lack significant sequence homology, especially outside the T-box domain (Supplementary Figure S4). However, Xenopus Eomes and its Tbx16 orthologue VegT have been suggested to display similar specificity in part due to a single shared asparagine residue within the T-box, rendering them functionally distinct from the Tbxta orthologue Xbra, which has a lysine in the equivalent position (Conlon et al., 2001) (Supplementary Figure S4). We therefore sought to address the following questions: 1) Are Tbx16, Tbxta and/or Eomesb capable of inducing Eomesa target genes in early gastrulas; 2) Is the T-box asparagine residue critical for Eomesa function; 3) Are key Eomesa functions dependent on the highly conserved CTD or relatively poorly conserved NTD.

We injected equimolar quantities of mRNA corresponding to each wild type T-box factor, or Eomesa ∆NTD, ∆CTD or N320K mutants and assessed the effect on dorsal mesoderm marker noto and DFC/endoderm marker sox32, and DFC marker vgll4l. Deletion of the NTD or CTD ablated Eomesa ability to induce noto expression, while N320K mutation had no significant effect (Figure 5A). All other T-box factors failed to produce a consistent or compelling induction of ectopic noto (Figure 5A). Consistent with previous results, Eomesa overexpression led to ectopic sox32 expression in the outer marginal cells indicative of increased numbers of DFC-like cells but not endoderm (Bjornson et al., 2005) (Figure 5B). NTD and especially CTD deletions markedly reduced ectopic sox32 induction, however, N320K mutation had little effect (Figure 5B). We further note that the N320K mutation had no discernible effect on the ability of Eomesa to synergise with Mixl1 and Gata5 to induce sox32 expression at the animal pole, suggesting that this mutation within the T-box does not interfere with the known T-box interactions with Mixl1 and Gata5 (Bjornson et al., 2005) (Figure 4C). Overexpression of tbx16, tbxta and eomesb did not lead to ectopic sox32 induction highlighting functional distinctions with eomesa (Figure 5B). Vgll4l expression showed similar induction to sox32 at the margin by both wild type Eomesa and ∆NTD and N320K Eomesa (Figure 5C). We have previously shown by ChIP-seq that at sphere stage Eomesa binds in the first intron of vgll4l (Nelson et al., 2014). Analysis of Tbx16 ChIP-seq data in mid/late gastrulas (8–8.5 h.p.f.) (Nelson et al., 2017) also shows Tbx16 binding at close matches to the known T-box consensus sequence within vgll4l intron 1, suggesting Tbx16 does have the potential to directly participate in regulation of vgll4l during gastrulation (Figure 5D). However, tbx16 overexpression suggests that it is not individually sufficient to strongly drive ectopic vgll4l expression (Figure 5C).

We next tested whether Tbx16 and Tbxta can act combinatorially to induce dorsal marker gene expression. We found that on Tbx16/Tbxta combinatorial overexpression DFC markers vgll4l and sox17 were expressed over a broader region of the dorsolateral margin but showed concomitant reduction in the primary DFC cluster (Supplementary Figure S5A). Conversely, foxj1a expression remained localised to the dorsal margin, again suggesting that foxj1a expression is somewhat dependent on additional dorsally localised regulators (Supplementary Figure S5A). Furthermore, while tbx16 and tbxta are coexpressed with noto in early gastrulas (Supplementary Figures S5B,C), their combinatorial overexpression does not lead to expanded noto expression (Supplementary Figure S5D). We therefore conclude that combined activities of Tbx16 and Tbxta are not sufficient to induce dorsal fates, unlike Eomesa.

Overall this suggests that Eomesb, Tbx16 and Tbxta do not individually have similar abilities to Eomesa in inducing dorsal mesoderm and DFC gene expression, that the previously reported N/K amino acid difference between Eomesa/Tbx16 and Tbxta does not appreciably influence specificity and function in this context, and that deletion of the relatively poorly conserved Eomesa NTD does not result in complete loss of function.

Results above suggest that Eomesa overexpression induces DFC fate based on ectopic expression of markers including sox32 and vgll4l (Figures 2F, 5B,C). To resolve the conflict between the observed Eomesa-mediated repression of vgll4l expression at mid/late blastula stages (4 h.p.f.) but induction by early gastrulation (6 h.p.f.) we explored the role of Eomesa target Sox32 in vgll4l induction. We found that sox32 overexpression through one-cell stage mRNA injection led to a dramatic upregulation of vgll4l expression in early gastrulas (6 h.p.f.; Figure 6A), suggesting localised Eomesa-mediated upregulation of vgll4l at the margin occurs through Sox32 rather than a switch in Eomesa function directly at the vgll4l locus. To test whether Sox32 is required for induction of vgll4l expression we performed knockdown using a previously validated and widely used antisense morpholino (Dickmeis et al., 2001). This knockdown clearly resulted in loss of DFC vgll4l expression (Figure 6B). Furthermore, vgll4l induction in locations outside the dorsal margin, caused by combinatorial eomesa/mixl1/gata5 overexpression was also profoundly abrogated by Sox32 knockdown. Thus, during gastrula stages induction of vgll4l is via Sox32 rather than direct Eomesa activities.

Vgll4l expression was similarly upregulated outside the dorsal margin by combinatorial mixl1/gata5 overexpression, and partially blocked by Sox32 KD (Figure 6B). However, vgll4l expression in mixl1/gata5 overexpressing embryos was not as profoundly reduced on Sox32 KD as was the case for eomesa/mixl1/gata5 overexpression. It is not completely clear whether this is because mixl1/gata5 can activate vgll4l independent of Sox32 function, or due to the absence of Eomesa-mediated repression of vgll4l. Addition of tbx16 to mixl1/gata5 overexpression led to a significant increase in the fraction of embryos exhibiting ectopic vgll4l expression compared to mixl1/gata5 alone (Figure 6B). We note that overexpression of tbx16 alone was insufficient to enhance vgll4l expression in a significant fraction of embryos, and showed no ability to induce sox32 expression (Figures 5B,C). It seems likely that while tbx16 alone does not exert a strong influence at the vgll4l locus, mixl1 and gata5 can provide a context for tbx16 to enhance vgll4l expression. This likely involves Tbx16 binding to vgll4l intron 1 (Figure 5D) rather than occurring via sox32, since Tbx16 cannot induce sox32 expression, either individually or in combination with Mixl1 and Gata5 (Figures 4C, 5B).

Overall the present results combined with previous published reports from ourselves and others suggest a model wherein Eomesa acts within interlocking incoherent type 3 and coherent type 1 feedforward loops (Mangan and Alon, 2003) to repress vgll4l while combining with Nodal downstream effectors Mixl1 and Gata5 to activate sox32, which in turn activates vgll4l around the time of DFC specification. In addition to this, our analyses indicate that both mouse Eomes FL and ∆VR isoforms are functionally equivalent to Eomesa, suggesting phenotypic differences between zebrafish and mouse Eomes loss-of-function mutants are not likely to be driven by functional divergence, but rather redundancy with co-expressed factors in zebrafish such as Tbx16.

T-box transcription factors are an ancient family of genes with many key roles in embryogenesis and disease. Lineage-specific differences that occurred in the family during vertebrate evolution have resulted in altered gene complements and diversity of splice isoforms in distinct evolutionary lineages (DeBenedittis and Jiao, 2011; Papaioannou, 2014). While AS events in specific evolutionary lineages have led to functional diversification of certain T-box factors, we have shown that Eomes loss-of-function phenotypic differences between mouse and zebrafish are unlikely to be due to evolutionary differences in Eomes protein function, but rather a degree of compensation by Tbx16 which is present in zebrafish but not placental mammals.

Though AS is an evolutionary means of increasing functional diversity within the proteome, our data suggests that Eomes exon 6 AS is not functionally important in the context of early development. In the case of the ∆VR splicing event a synonymous mutation created an alternative splice acceptor, however, our data suggests it is hardly used leading to majority production of the FL isoform. That the ∆VR isoform arose and is maintained in the tetrapod lineage may be due to substantial functional similarlity of FL and ∆VR isoforms, leading to a lack of selective pressure.

In overexpression studies the ∆VR isoform has the ability to induce trophoblast markers in embryonic stem cells (Niwa et al., 2005), and cardiac mesoderm markers in embryonal carcinoma cells (Costello et al., 2011). Moreover, both FL and ∆VR isoforms can induce organizer and DFC markers while repressing ectoderm markers on overexpression in zebrafish. These observations provide further evidence of their functional equivalence.

The present data demonstrate that the ∆VR and ∆CTD isoforms are only weakly expressed compared with FL Eomes. While we find no evidence for the functional importance of the VR it is intriguing that it is both highly conserved and known to be phosphorylated (Huttlin et al., 2010). It is possible that these isoforms potentially make substantial contributions in contexts we have not explored. The present evidence, however, suggests that in mice, as in zebrafish the FL isoform is the more important molecule.

Given the complexities of mouse Eomes mutant phenotypes it would be interesting to explore isoform-specific functions in mouse. This could be achieved through either modifying the endogenous Eomes locus in embryonic stem cells (ESCs) such that only specific isoforms could be expressed, or using isoform-specific inducible transgenes in Eomes null mutant ESCs. This could be combined with directed differentiation procedures to determine whether there are detectable isoform-specific functions in relevant cell types. Whether zebrafish Eomesa, or other T-box factors can functionally substitute for mouse Eomes could also be tested in a similar system. Alternatively, the genetically modified ESCs could be used to make mice in an attempt to study isoform and orthologue functions in vivo.

We previously demonstrated that Eomesa and Tbx16 display overlapping genomic binding profiles in early zebrafish embryos (Nelson et al., 2017). Whether they are functionally redundant, however, had not been explored. The present experiments strongly suggest that Eomesa and Tbx16 redundantly regulate the homeodomain transcription factor mixl1, which has key conserved functions in endoderm formation in zebrafish and mouse (Kikuchi et al., 2000; Hart et al., 2002). Mixl1 mutants suffer profound loss of endoderm (Kikuchi et al., 2000). Reduced expression of mixl1 in the margin of late blastulas/early gastrulas is coupled with reduced numbers of endoderm progenitors in late gastrulation in both MZeomesa mutants, and on tbx16 knockdown (Du et al., 2012; Nelson et al., 2017). It therefore seems likely that the enhanced reduction of mixl1 expression on tbx16 knockdown in eomesa mutants in the present study would lead to increased loss of endoderm progenitors. Our previous RNA-seq analyses indicate that expression of tbx16 is not significantly different in MZeomesa mutants (Nelson et al., 2014). It therefore seems likely that Tbx16 partially compensates for loss of Eomesa during zebrafish endoderm formation. Our study therefore highlights a consistent requirement for T-box function in vertebrate endoderm formation. Interestingly, while multiple orthologous T-box factors have similar expression domains in early zebrafish and mouse embryogenesis, those domains are typically expanded in zebrafish (Wardle and Papaioannou, 2008). Coupled with its rapid rate of development and the greater number of T-box factors in zebrafish, there is likely to be a higher degree of T-box factor co-expression, enhancing the probability of redundancy.

While Eomesa and Tbx16 share some redundant functions we also identified key differences. It was previously shown that Eomesa can combine with Mixl1 and Gata5 to drive expression of sox32 at the animal pole (Bjornson et al., 2005). However, Tbx16 does not appear to have the same ability as Eomesa to drive sox32 expression either individually or in combination with Mixl1 and Gata5, even though sox32 expressing cells appear to exhibit tbx16 expression in single-cell RNA-seq data. This is consistent with previous observations that Meomesa and MZeomesa mutants have reduced expression of sox32 during gastrulation without complete loss (Du et al., 2012; Xing et al., 2022). It therefore seems likely that Tbx16 is sufficient to rescue certain Eomesa functions but cannot completely compensate for its loss. It is further notable that Tbx16 does not seem individually able to induce the dorsal mesoderm marker noto as Eomesa can. However, given that Eomesa acts upstream of Nodal (Xu et al., 2014; Xing et al., 2022) it seems likely that major differences in outcome between eomesa and tbx16 overexpression stem from enhanced Nodal signalling on eomesa overexpression. It will be interesting to learn more about the common and unique functional activities of Eomes and Tbx16 that drive target gene expression.

Eomesa and Tbx16 are only distantly related within the T-box family (25.3% of Tbx16 amino acid identity), with the majority of conserved amino acids occurring within the T-box domain. Whether they are likely to act in similar protein complexes to regulate their target genes is therefore unclear. A key study in Xenopus, however, suggested the specificity of target gene induction is primarily mediated by the T-box itself, rather than NTDs and CTDs (Conlon et al., 2001). The same study demonstrated a single asparagine to lysine substitution in Xenopus Eomes and VegT T-box domains, alter their inductive properties to mimic Brachyury (Conlon et al., 2001). Importantly, both Eomesa and Tbx16 (which has been proposed as the zebrafish orthologue of Xenopus VegT (Griffin et al., 1998)) share the same critical asparagine. Our data suggest that the N320K mutation has little effect on induction of Eomesa target genes explored here, is unlikely to prevent T-box interaction with co-factors Mixl1 and Gata5, or substantially account for differences with Tbxta in endoderm and DFC formation. In fact, analysis of single-cell RNA-seq data suggests that the greater importance of Eomesa and Tbx16 in endoderm formation is more likely to be attributable to lesser Tbxta expression in the endoderm. It is therefore possible that Eomesa and Tbx16 also have overlapping roles in endoderm formation downstream of driving mixl1 expression in presumptive endoderm that are yet to be elucidated.

Interestingly, participation of zebrafish Tbx16 in processes controlled by Eomes in mice is not limited to endoderm formation. For example, while Eomes acts upstream of basic helix-loop-helix transcription factor gene Mesp1 to specify cardiac mesoderm in mice (Costello et al., 2011), Tbx16 regulates the orthologous gene mespaa in zebrafish (Garnett et al., 2009). Eomes loss-of-function leads to aberrant mesoderm cell migration during mouse gastrulation, while tbx16 zebrafish mutants also exhibit cell-autonomous defects in mesoderm migration (Ho and Kane, 1990; Arnold et al., 2008a). Remarkably, potentially interesting parallels continue to emerge, such as the requirements for zebrafish Tbx16 and mouse Eomes in blood progenitors (Rohde et al., 2004; Harland et al., 2021). It is therefore possible that the presence of Tbx16 in teleost fish has led to a reduced requirement for Eomes in multiple developmental contexts.

The present study focuses on early embryonic development, however, Eomes is known to have later roles in neurological development, as well as in the immune system. Importantly, Eomes is an key regulator of neurogenesis in the subventricular zone, and loss leads to microcephaly and severe behavioural defects (Arnold et al., 2008b). Though eomesa is equivalently expressed in the telencephalon of developing zebrafish larvae, whether null mutants have an equivalent phenotype is unknown (Mione et al., 2001; Du et al., 2012). If they do not, however, it is unlikely to be due to redundancy with tbx16, which is absent from the developing brain. Similarly, it is unclear whether eomesa mutants exhibit defects in the immune system, such as in T cell differentiation and NK cell development and function as in mammals (Simonetta et al., 2016; D'Cruz et al., 2009). Both eomesa and eomesb are co-expressed in lymphocytes in fish, however, suggesting they may be redundant in the immune system (Takizawa et al., 2007; Takizawa et al., 2014).

While T-box factor redundancy during development is not a novel concept e.g. (Amacher et al., 2002; Garnett et al., 2009; Jahangiri et al., 2012; Gentsch et al., 2013; Nelson et al., 2017), the molecular basis for this redundancy (or indeed T-box factor molecular interactions in general) is not well understood. In future it will be interesting to study whether redundant T-box factors recruit similar co-factors to regulate gene expression, and whether this occurs through conserved or divergent amino acid sequences and structural motifs.

While Eomesa is capable of inducing dorsal mesoderm markers such as noto and chrd in zebrafish embryos, it is notable that their expression is normal in the absence of Eomesa (Bruce et al., 2003; Du et al., 2012). However, it is also notable that tbxta and tbx16 do not show altered expression in eomesa mutants in published WISH and RNA-seq datasets (Du et al., 2012; Nelson et al., 2014). It is therefore possible that Tbx16 and Tbxta are amongst factors compensating for the loss of Eomesa. In support of this, Tbxta has been shown to directly activate noto expression, and tbxta mutants fail to maintain noto expression in mid/late gastrulation stages, leading to loss of notochord (Melby et al., 1997; Morley et al., 2009). Similarly, Tbx16 is required to maintain chrd expression in axial structures at mid/late gastrulation stages (Miller-Bertoglio et al., 1997; Warga et al., 2013). The reduced expression of both noto in tbxta mutants and chrd in tbx16 mutants follows the decline in eomesa mRNA expression levels, suggesting that Tbxta and Tbx16 maintain the expression of dorsal mesoderm markers in the absence of Eomesa. Nevertheless, our results indicate that of these T-box factors, only Eomesa is sufficient to induce ectopic dorsal mesoderm marker expression. This suggests key differences in the molecular functions of these T-box factors. It is possible that co-factors required for dorsal mesoderm induction by Eomesa are localised throughout the margin while those required by Tbxta and Tbx16 are restricted to the dorsal margin. Alternatively, given Eomesa regulates expression of Nodal pathway ligands which are required for dorsal mesoderm fates, it is possible that Eomesa but not Tbxta/Tbx16 is capable of expanding dorsal mesoderm through upregulation of Nodal signalling (Du et al., 2012; Xu et al., 2014; Xing et al., 2022).

Loss of Eomesa leads to upregulation of vgll4l during blastula stages whereas overexpression of eomesa causes repression of vgll4l (Nelson et al., 2014). The present experiments suggests that Eomesa acts within feedforward loops to repress vgll4l expression until activators including Sox32 accumulate to drive vgll4l in DFCs at the onset of gastrulation. Given that Eomesa is maternally contributed and not spatially restricted in early development (Du et al., 2012), while accumulation of vgll4l activators is principally driven by Nodal at the dorsal margin, this suggests a model wherein Eomesa controls the specificity and timing of vgll4l induction. Eomesa mRNA steadily declines during blastula stages as expression of mixl1, gata5, tbx16 and sox32 increase, and is virtually undetectable at the onset of gastrulation, (Figure 3A and Bruce et al., 2003; Figiel et al., 2021). While Eomesa protein does persist through gastrulation (Du et al., 2012) it seems likely that temporal and spatial changes in abundance of vgll4l activators and repressors acting within these feedforward loops cooperatively regulate the specificity of vgll4l expression during DFC specification.

Genetic data, however, suggest that our model is likely to be incomplete. While Sox32 is required for correct DFC formation (Alexander et al., 1999; Essner et al., 2005), upstream regulators mixl1 and gata5 are not required for DFC formation individually or in combination. Rather mixl1 and gata5 seem to be strictly required upstream of sox32 for correct endoderm formation (Reiter et al., 1999; Kikuchi et al., 2000; Reiter et al., 2001). While we cannot discount the possibility of mixl1 and gata5 expression in precursors of DFCs, present evidence suggests that there are either alternative upstream regulators of sox32 in DFCs vs. endoderm, or additional redundant factors in DFCs rescuing the requirement for mixl1 and gata5. However, given the apparent requirement for Nodal signalling in DFC formation (Alexander and Stainier, 1999), it seems likely that whatever the upstream regulators of sox32 expression in DFCs they will be Nodal-dependent. Overall this highlights a lack of understanding of the gene regulatory networks that direct DFC vs. endoderm formation, which will be a key focus of our future work.

Recent evidence suggests Vgll4l is required for tbx16 expression during DFC formation (Fillatre et al., 2019). That Tbx16 binds the vgll4l promoter during gastrulation could suggest that complex regulatory loops control DFC formation and maintenance. The ability of Eomesa to induce ectopic DFCs during early gastrulation combined with expression of mouse Eomes in progenitors of the node and requirement for correct node formation (Arnold et al., 2008a; Costello et al., 2011) suggests the potential for a conserved role in establishment of left-right asymmetry with some degree of redundancy with Tbx16 in zebrafish. However, a role for the vgll4l mammalian homologue Vgll4 in left-right asymmetry has yet to be determined. Further study of the mechanistic parallels in T-box factor mediated formation of zebrafish DFCs and mouse node would be beneficial to gain a more detailed evo-devo understanding of this process.

The diversity of DFC marker gene induction observed in this study was particularly striking, and points to dorsally localised determinants of DFC identity and function that are less readily induced by Eomesa and Eomes. We found that between them Eomesa and mouse Eomes isoforms were able to induce sox32, sox17, vgll4l and foxj1a. Sox32 is required for maintenance of DFC identify and formation of the left-right organiser (Alexander et al., 1999; Essner et al., 2005) while its downstream target sox17 is required for correct left-right organiser function (Aamar and Dawid, 2010). Vgll4l is a key mediator of Hippo signalling and regulates epigenetic programming of DFC by controlling the expression of writers and readers of DNA methylation, influencing DFC proliferation, apoptosis and ciliagenesis (Fillatre et al., 2019). Foxj1a is the master regulator if motile cilia formation (Yu et al., 2008). That the Eomes-mediated induction of foxj1a was more restricted to the dorsal margin than that of other markers suggests that Vgll4l and the Sox and T-box factors known to be involved in DFC formation are not sufficient to fully induce DFC identity. Other localised cues (physical, mechanical, signalling or cell intrinsic factors) are therefore likely to be involved in foxj1a induction.

Overall we conclude that enhanced AS in mammals has not significantly altered Eomes function in early embryogenesis. Rather we conclude that the different degrees of T-box factor co-expression and the presence/absence of additional factors including Tbx16 has modulated the severity of the Eomes null mutant phenotype in the embryo proper between mouse and zebrafish. Furthermore, we conclude that in zebrafish Eomesa participates in DFC formation through directing feedforward loops via sox32 to control vgll4l expression. Our results therefore provide novel insights into evolutionary differences in vertebrate endoderm formation, and the gene regulatory networks involved in controlling the zebrafish left-right organiser formation.