Hox13 genes are required for mesoderm formation and axis elongation during early zebrafish development

A hallmark of early embryonic vertebrate development is the progressive formation of the posterior embryonic body from anterior to posterior (Kimelman and Martin, 2012; Steventon and Martinez Arias, 2017). This is clearly seen with the production of somites, which first form just behind the head and are then added sequentially until the final anterior-posterior (A-P) axis of the embryo is established (Holley, 2007; Pourquié, 2018). A key component of this process is a bipotential neuromesodermal cell progenitor (NMp) population located at the most posterior end of the embryo (the tailbud) during the somite-forming stages, which produces both the spinal cord and musculature (Gouti et al., 2015; Henrique et al., 2015; Kimelman, 2016b; Martin, 2016; Steventon and Martinez Arias, 2017; Wilson et al., 2009). The NMps express both the neural gene sox2 and the mesodermal gene T/Brachyury (known as tbxta or no tail in zebrafish), allowing them to differentiate in either a neural or mesodermal direction. The canonical (β-catenin-dependent) Wnt signaling pathway plays a key role in regulating this switch, such that NMps exposed to high Wnt enter the presomitic mesoderm and differentiate into various mesodermal fates such as muscle, whereas those deprived of Wnt signaling become neural tissue. This mechanism allows the embryo to carefully balance the ratio of spinal cord to somites along the entire A-P axis, although in amniotes it occurs over a prolonged period, whereas in anamniotes the fate decisions largely occur during gastrulation (Martin and Kimelman, 2012; Steventon and Martinez Arias, 2017).

What causes the termination of the embryonic A-P axis is uncertain. The simplest explanation is that the NMps, which progressively differentiate during the somitogenesis stages, are eventually depleted, resulting in completion of the A-P axis. An intriguing alternative idea is that the most 5′ Hox genes (paralog group 13) actively terminate posterior extension along the A-P axis (Aires et al., 2019; Amin et al., 2016; Denans et al., 2015; Mallo, 2018; Steventon and Martinez Arias, 2017; Young et al., 2009). The initial impetus for this view came from the observation that knockout of the mouse Hoxb13 gene caused the addition of two small somites at the tip of the tail, extending the embryo to 67 somites (Economides et al., 2003). Curiously, knockout of the other Hox13 paralogs in mouse did not cause a lengthening of the body axis (Dolle et al., 1993; Fromental-Ramain et al., 1996; Godwin and Capecchi, 1998; Suemori and Noguchi, 2000).

Additional support for the proposed role of Hox13 genes in terminating the A-P axis in mouse came from experiments in which Hoxa13, Hoxb13 and Hoxc13 were overexpressed using a strong transcriptional promoter, resulting in embryos lacking all or most of the tail (Aires et al., 2019; Young et al., 2009). Additionally, chick embryos overexpressing Hoxa13, Hoxb13 and Hoxc13, but not Hoxd13, showed slower ingression of mesodermal cells into the somites from the tailbud (Denans et al., 2015). Moreover, these authors showed that Hoxa13 overexpression caused a reduction in the levels of the endogenous T/Brachyury gene, apparently through downregulation of the Wnt signaling pathway. This led to their proposal that a key job of the Hox13 genes is to reduce the size of the presomitic mesoderm by progressively reducing canonical Wnt signaling.

A major caveat to the proposal that Hox13 genes limit the length of the A-P axis is that, with the one exception of the mouse Hoxb13 mutant that causes a minor increase in the number of somites (as well as overgrowth of posterior neural tissue causing a wider and longer spinal cord), it is based on overexpression studies. We chose instead to take a loss-of-function approach in zebrafish using CRISPR to mutate the two most abundant Hox13 genes in the tailbud. We show here that loss of hoxa13b or a combined loss of hoxa13b and hoxd13a does not cause posterior defects in an otherwise wild-type background. However, we show that hoxa13b and hoxd13a genes genetically interact with tbxta, the zebrafish ortholog of T/Brachyury, such that a combined reduction in Tbxta activity and loss of hoxa13b and hoxd13a leads to severe synergistic posterior embryonic defects. We show that these defects are due to a loss of mesoderm and a concomitant increase in sox2-expressing cells, which parallels a sharp reduction in the expression of the posterior canonical Wnt genes and the retinoic acid (RA)-degrading enzyme Cyp26a1, as is observed in tbxta null mutants. Finally, using a newly created zebrafish line that overexpresses Hoxa13b we show that gain of function of Hox13 protein is not the opposite of loss of function, suggesting that Hox13 gene overexpression studies have the potential to be misleading. We propose that the essential role of Hox13 genes is to provide robustness to the Brachyury-dependent NMp niche, which maintains Wnt signaling as well as preventing RA from accumulating in the tailbud, allowing the NMps to differentiate as mesoderm.

Zebrafish have six paralogs of the Hox13 cluster: hoxa13a, hoxa13b, hoxb13a, hoxc13a, hoxc13b and hoxd13a. Using data from a previous RNA sequencing (RNA-seq) analysis of the tailbuds of mid-somitogenesis zebrafish embryos (Kimelman et al., 2017), we observed that all genes except hoxb13a are expressed at significant levels in the tailbud. We therefore made in situ hybridization probes to all the Hox13 genes except hoxb13a and observed that each of these is expressed in the tailbud at mid-somitogenesis, as expected (Fig. 1A). To determine the relative abundance of each of the Hox13 genes we used quantitative PCR, and found that hoxa13b and hoxd13a were the most abundant and were expressed at similar levels, with the remainder of the Hox13 genes being expressed at lower levels. This result was also observed in tailbud RNA-seq data (Kimelman et al., 2017).

Of the two most abundant Hox13 genes, we initially focused on hoxa13b because a previous overexpression study reported that Hoxa13 inhibited mesodermal cell movements out of the chick tailbud, whereas Hoxd13 had no effect (Denans et al., 2015). Zebrafish hoxa13b is expressed in the future posterior mesoderm as early as the end of gastrulation, and continues to be expressed there until the end of somitogenesis (Fig. 1B). The initial expression of hoxa13b at such an early stage of embryogenesis seemed surprising for a gene proposed to be involved in terminating axis formation. Therefore, we generated a mutant allele of hoxa13b that removed 16 bp in the middle of the coding region, eliminating the essential DNA binding region (Fig. 2A). Embryos homozygous for hoxa13b^Δ16 were phenotypically completely normal, as were adults obtained from these embryos. One possibility for the lack of phenotype is compensation by the other Hox13 genes triggered by nonsense-mediated decay, as observed in zebrafish for other transcripts with a premature termination codon (Ma et al., 2019). However, using qPCR we observed that hoxa13b^Δ16 was expressed at the same level as wild-type hoxa13b, and that the other Hox13 genes were not upregulated in hoxa13b^Δ16 homozygous mutant embryos (not shown).

An alternative possibility for the lack of phenotype in zebrafish hoxa13b^Δ16 embryos is that the other Hox13 genes are redundant with hoxa13b. We therefore mutated the other abundant Hox13 gene, hoxd13a, starting with adults that were homozygous for the hoxa13b^Δ16 mutation. Analysis of hoxd13a expression revealed that it is also activated early during somitogenesis, starting at the eight-somite stage (Fig. S1). We obtained three hoxd13a mutations, which inserted either 4 or 13 base pairs in the middle of the hoxd13a coding region, or deleted 8 base pairs at this site, with all of the mutations eliminating the DNA binding domain (Fig. 2A). In crosses of heterozygous adults we observed embryos with low levels of tail defects from mild to severe (crosses of hoxa13b^Δ16/Δ16;hoxd13a^ins13/+ fish produced 4.2% defective embryos, n=188; Fig. 2B), with a very small number of these defective embryos surviving to adulthood (Fig. S2). We observed a strong selection bias against double mutant females, but were able to obtain one double mutant female (hoxa13b^Δ16/Δ16;hoxd13a^ins4/ins4). Crossing this to a double mutant male of the same genotype produced 12.5% defective embryos (n=40). These results demonstrate that hoxa13b and hoxd13a are redundant because posterior defects appear when both Hox13 genes are mutant.

To analyze the tail defects during the somitogenesis stages, we placed embryos from mutant crosses at 21°C beginning at the start of gastrulation (shield stage). We routinely use a post-shield stage cooling protocol because it allows us to examine embryos at all somitogenesis stages during the day after fertilization, rather than throughout the night as would occur if embryos were kept at the standard temperature of 28.5°C. Provided the embryos are cooled at shield stage or later, wild-type embryos develop completely normally at temperatures as low as 17°C (Kimelman, 2016a). When we raised embryos from crosses of hoxa13b^Δ16/Δ16;hoxd13a^ins13+/− adults at 21°C, we surprisingly observed a high frequency of posterior body defects from mild to severe. To quantify our results, we divided embryos into three classes of defects, with the most severe embryos (class 1) appearing largely identical to embryos with a null mutation in tbxta, originally known as no tail (ntl) mutants (Halpern et al., 1993; Schulte-Merker et al., 1994, Fig. 3A). Interestingly, however, unlike ntl mutants that have both a lack of tail and no notochord, our class 1 mutants all contained a notochord (Fig. 3A). Crosses of hoxa13b^Δ16/16;hoxd13a^ins13+/− fish produced a range of embryos from classes 1 to 3 (Fig. 3B, columns 1 and 2). We were able to identify hoxa13b^Δ16/16;hoxd13a^ins13−/− double homozygous mutant males, and these crossed to hoxa13b^Δ16/Δ16;hoxd13a^ins13+/− females further enhanced the phenotype (Fig. 3B, column 3). We similarly observed strong defects with crosses of hoxa13b^Δ16/Δ16;d13a^ins4+/− and hoxa13b^Δ16/Δ16;d13a^Δ8+/− fish (Fig. 3B, columns 4 and 5). A cross of the one double homozygous female described above to a double homozygous male produced embryos that were virtually all class 1 and 2 (Fig. 3B, column 6).

We next asked whether the posterior defects were only found in hoxa13b;d13a double mutants, or if the same phenotype could be found in hoxa13b single mutants kept at 21°C. Embryos from a cross of hoxa13b^Δ16/Δ16 fish also showed posterior defects at 21°C, although at lower frequencies than with the hoxa13b;d13a mutants, and with half the embryos showing a wild-type phenotype (Fig. 3C, column 2, compared to Fig. 3B, columns 1 and 2). Thus, removal of Hoxa13b function results in some posterior defects, but the effect is enhanced when Hoxd13a function is also reduced.

We previously reported that our wild-type stocks of zebrafish contain a naturally occurring variant of Tbxta containing two amino acid changes that lead to a cold-sensitive phenotype (Kimelman, 2016a). Embryos carrying this mutation, which we called no tail^cs (ntl^cs), have a completely wild-type phenotype when raised at the standard temperature of 28.5°C, but have the ntl null phenotype when raised at 17°C, including a lack of notochord. Because we were seeing cold-sensitive defects in our Hox mutant fish, we wondered whether this mutation might also be present in our Hox mutants, and therefore genotyped the fish described above. All of these fish were found to be homozygous for the ntl^cs mutation, suggesting that this contributes to the observed phenotype. When we raised homozygous ntl^cs/cs embryos at 21°C we observed a minor number of posterior defects, but much fewer than found in embryos that were also homozygous for hoxa13b^Δ16 (Fig. 3C, columns 1 and 2). Similarly, at 18.5°C the absence of Hoxa13b function enhanced the ntl^cs phenotype (Fig. 3C, columns 4 and 5). Stronger results were observed at both temperatures when fish that were homozygous for both hoxa13b^Δ16 and ntl^cs and heterozygous for hoxd13a^ins13 were crossed (Fig. 3C, columns 3 and 6). Our results indicate that at 21°C and 18.5°C Tbxta function is only partially reduced in ntl^cs homozygous embryos, and that removal of Hoxa13b function enhances the strength of the posterior defect, which is further enhanced when Hoxd13a function is also depleted. However, the interaction between tbxta and the Hox13 genes is not due to an overall loss of Tbxta function, as the class 1 mutants appear very similar to ntl null mutants yet have an intact notochord.

Because the above studies were done in a ntl^cs/cs background, we backcrossed our hoxa13b;d13a double mutant fish into a ntl wild-type background. These fish produced normal embryos at both 29°C and 18.5°C, demonstrating that the reduction in Tbxta function produced in ntl^cs embryos, especially at lower temperatures, is necessary in order to cause posterior defects. To confirm that the reduction in Tbxta is the cause of the posterior defects, we injected wild-type and hoxa13b;d13a double mutant embyos in the ntl wild-type background with a very low dose (0.2 ng) of a tbxta morpholino oligonucleotide that we previously showed specifically phenocopies the ntl null mutation at 5 ng (Martin and Kimelman, 2008). We observed a marked hypersensitivity to reduced Tbxta levels when both hoxa13b and hoxd13a were mutated (Fig. S3). These results further confirm the genetic interaction between the two Hox13 genes and tbxta.

To uncover the cause of the posterior phenotype defects, we examined gene expression using in situ hybridization on hoxa13b^Δ16ΔJ16;ntl^cs/cs embryos (hereafter called hoxa13;ntl^cs embryos) placed at the semi-permissive temperature of 18.5°C from shield stage until mid-somitogenesis. We observed a sharp reduction in the expression of tbxta in the NMps in the majority of the hoxa13;ntl^cs embryos, such that the only expression remaining at the most posterior end of the majority of embryos was tbxta expression within the notochord progenitors, which are located just anterior to the NMps (Fig. 4A, gray arrowhead). A strong reduction in tbxta was evident in 53.3% of embryos (n=30); the remaining embryos had less reduced or normal expression of tbxta. Interestingly, the expression of tbxta in the notochord (Fig. 4A, black arrowhead) was normal in all of the hoxa13;ntl^cs embryos, whereas in tbxta/ntl mutants the expression of tbxta in the notochord is absent (Schulte-Merker et al., 1994 and see below). The normal tbxta expression in the notochord supports our observation that even class 1 embryos have an intact notochord (Fig. 3A), and further supports the idea that the deficit in hoxa13;ntl^cs embryos at the semi-permissive temperature is specific to the NMps and is not an overall loss of Tbxta function.

Because tbxta expression was strongly reduced, we examined other mesodermal markers including tbx16 and msgn1, which mark the initial mesodermal progenitors (Fior et al., 2012; Griffin et al., 1998; Yabe and Takada, 2012), and pcdh8, which shows the presomitic mesoderm and nascent somites (Yamamoto et al., 1998). All of these genes were strongly downregulated in the majority of the hoxa13;ntl^cs embryos (tbx16, 65.5%, n=27; msgn1, 63.2%, n=19, and pcdh8, 55.6%, n=27; Fig. 4H-J). Canonical Wnt signaling is essential both for brachyury/tbxta expression and for formation of the mesoderm, and so we examined the two posteriorly expressed canonical Wnt genes, wnt3a and wnt8a. Both were strongly downregulated in hoxa13;ntl^cs embryos (wnt3a, 89.7%, n=29; wnt8a, 90.3%, n=31; Fig. 4B,C). The same mesodermal defects were observed in class 1 ntl^cs homozygous embryos kept at 18.5°C that were wild type for hoxa13 and hoxd13 (Fig. S4), demonstrating that loss of these two genes synergistically enhances the effects of Tbxta reduction. Tailbud Wnt signaling is also regulated by two secreted Wnt inhibitors, Wif1 and Sfrp1a, which are expressed just anterior to the tailbud (Row and Kimelman, 2009). In hoxa13;ntl^cs embryos, expression of both genes was found to be closer to the posterior end of the embryo, further diminishing the levels of active Wnts (Fig. S5). RA is kept out of the posterior end of the embryo by the Tbxta-regulated degrading enzyme, Cyp26a1, which is essential because RA blocks mesoderm formation (Martin and Kimelman, 2010; Olivera-Martinez et al., 2012; Sive et al., 1990). In hoxa13;ntl^cs embryos, cyp26a1 was also strongly downregulated (88.9%, n=27; Fig. 4D). These results indicate that the failure to form mesoderm is a result of loss of canonical Wnt signaling, together with an increase in RA in the posterior of the embryo caused by failure to sustain cyp26a1 expression.

Although the mesodermal genes were downregulated in hoxa13;ntl^cs embryos, not all posteriorly expressed genes were affected. For example, cdx4 and fgf8a were normal or only partially reduced (100%, n=31 for cdx4 and n=32 for fgf8a; Fig. 4E,F). Finally, we examined sox2 as inhibition of Wnt signaling causes sox2 to be expressed in the mesodermal region, which we also observed to varying degrees in hoxa13;ntl^cs embryos (84.6% with expanded sox2 expression, n=26; Fig. 4G). The expansion of sox2 into the mesodermal region is best seen with fluorescent in situ hybridization. In wild-type embryos, sox2 expression is excluded from the mesodermal progenitors (Martin and Kimelman, 2012) (Fig. S6A-C; Movie 1). In hoxa13;ntl^cs embryos with a class 1 phenotype, sox2 expands into this region, filling the entire posterior with sox2-positive cells (Fig. S6D-F; Movie 2). This is very similar to that observed in tbxta/ntl null mutants and morphants injected with the morpholino oligonucleotide that recapitulates the ntl mutant phenotype (Fig. S6G-I; Movie 3), except that in hoxa13;ntl^cs embryos the notochord was still present, whereas in tbxta/ntl null mutants and morphants the notochord was absent (Fig. S6E,H). We also observed the same expansion of neural fate in hoxa13;ntl^cs embryos with two other neural markers, sox19a and pou5f3 (Fig. S7). In summary, our results show a strong reduction in mesodermal gene expression and a concomitant increase in neural fated cells, exactly as shown for tbxta/ntl mutants (Martin and Kimelman, 2012) except that the notochord was unaffected, revealing that the absence of Hoxa13b enhances a reduction in Tbxta activity specifically within the NMps and their mesodermal derivatives.

Despite Tbxta having a large number of direct target genes (Garnett et al., 2009; Morley et al., 2009), we previously showed that individual tbxta/ntl mutant cells are rescued by the surrounding cells when transplanted into a wild-type environment (Martin and Kimelman, 2008). Thus, although tbxta/ntl mutants lack tail mesoderm, transplanted tbxta/ntl mutant cells contribute to tail mesoderm, demonstrating that the defect in tbxta/ntl is non-cell autonomous. This led to our proposal that the essential role of Brachyury/Tbxta is to control a small number of genes necessary to establish an NMp niche that has high Wnt and low RA, which is essential for mesoderm to form. To compare hoxa13;ntl^cs cells with tbxta/ntl mutant cells, we transplanted 30-50 dye-labeled cells at the start of gastrulation into the future posterior mesodermal region of the embryo, and then let the embryos develop for 2 days at the semi-permissive temperature of 18.5°C. For the hoxa13;ntl^cs mutants, we only scored transplants in which the donor embryo, which was left to develop after the small number of cells used in transplantation were removed, developed as a class 1 or class 2 phenotype. Wild-type cells transplanted in this manner ended up in the tail somites as expected (n=18 host embryos scored; Fig. 5A-C;). Similarly, hoxa13;ntl^cs cells also ended up in the tail somites, appearing indistinguishable from neighboring muscle cells (n=96 host embryos scored; Fig. 5D-F). These results demonstrate that, exactly as with tbxta/ntl null cells (Martin and Kimelman, 2008), the defect in hoxa13;ntl^cs mutants is rescued by the surrounding cells. These results indicate that the absence of Hoxa13b enhances the defect caused by decreased Tbxta function in the NMps, and does not produce additional cell autonomous differentiation defects.

Because overexpression analyses suggested a role for the Hox13 genes in cell movement (Denans et al., 2015; Payumo et al., 2016), we compared cell movements in hoxa13;ntl^cs cells at the semi-permissive temperature with movements in morphant cells lacking Tbxta. All embryos were injected with a nuclear EGFP then placed at 18.5°C at shield stage. At the 10-somite stage, hoxa13;ntl^cs embryos that showed a strong posterior phenotype, which would all go on to produce class 1 embryos, were identified and compared with wild-type and tbxta morphants by imaging on a confocal microscope. The speed and directionality of cell movements were analyzed as previously described for wild-type embryos (Das et al., 2019; Lawton et al., 2013; Mongera et al., 2018). Interestingly, the overall shape of the tailbud of hoxa13;ntl^cs embryos was the same as tbxta morphant tailbuds, and quite different from a wild-type tailbud (Fig. 6F,I compared to C). To examine the movements in detail, we examined both the NMps and mesodermal progenitor zone (MPZ), which is where tbx16 and msgn1 are first activated (Griffin et al., 1998; Yoo et al., 2003). In both regions, the speed of movement and the ‘straightness’ of movement were similarly reduced in both the hoxa13;ntl^cs embryos and tbxta morphants relative to wild-type embryos (Fig. 6J-M). These results provide further support for the idea that loss of hoxa13b enhances a partial reduction in Tbxta function within the NMps and mesodermal progenitors that derive from the NMps, producing defects that appear similar to those seen with a complete loss of Tbxta.

Because almost all of the previous studies involved overexpression analysis, we wished to examine the effects of conditionally overexpressing Hoxa13b and Hoxd13a in zebrafish. We produced two transgenic lines that expressed either Hoxa13b or Hoxd13a under the control of a heat-shock (HS) promoter, which allowed us to regulate the timing of transgene expression (HS:hoxa13b and HS:hoxd13a, respectively). mCherry was co-expressed with the Hox13 proteins in order to identify transgenic lines with strong expression (Fig. 7A). Prior to being heat shocked, embryos were allowed to develop until mid-somitogenesis (10-somite stage) to bypass any possible effects caused by prematurely producing the Hox13 proteins before the stages they are normally expressed. Overexpression of Hoxa13b caused a severe posterior truncation, whereas overexpression of Hoxd13a only caused defects in the anterior of the embryo (Fig. 7B). By quantitatively examining mCherry using qPCR, we observed that both transgenic lines produced similar levels of the co-expressed reporter, indicating that the differences were probably not due to differences in expression levels (Fig. 7C). Interestingly, a study in chick that examined cell movement defects caused by Hox13 overexpression found that Hoxa13 caused strong effects, whereas Hoxd13 had no effect (Denans et al., 2015), showing that the effects of overexpression of different Hox13 proteins observed in amniotes are also seen in zebrafish.

Optimally, overexpression causes the opposite effect of loss of function. However, overexpression of transcription factors is potentially problematic because it is possible that, when expressed at higher than endogenous levels, they could bind sites in the DNA that they would not normally bind when expressed at normal levels. Moreover, transcription factors are often in large protein complexes, and overexpression has the potential to create aberrant complexes or disrupt the normal stoichiometry of endogenous complexes. To analyze changes in gene expression caused by overexpression or loss of the Hox13 factors, we isolated tailbud explants from hoxa13;ntl^cs mutant embryos kept at 18.5°C, as well as tailbud explants from HS:hoxa13b embryos, 3 h after a 10 s heat shock, and examined the expression of a selection of genes. In both cases, we cut the explants at the 15-somite stage, selecting for embryos that had a clear morphological defect because these would go on to produce embryos with strong posterior defects, as shown in Fig. 3A and Fig. 7B. To keep the A-P extent of the explants constant, they were all cut to extend from the most posterior end of the embryo to the third most newly formed somite.

In agreement with our in situ hybridization data, we observed decreased expression of tbxta, msgn1, wnt3a, cyp26a1 and tbx16 in the hoxa13;ntl^cs mutant embryos, with little to no effect on cdx4 and fgf8a (Fig. 7D). In embryos overexpressing Hoxa13b, we observed decreased expression in most of these genes, including tbxta, wnt3a and fgf8a (Fig. 7D). Importantly, decreased T/Brachyury expression was also observed with Hoxa13 overexpression in chick (Denans et al., 2015) and Hoxc13 overexpression in mouse causes a reduction in wnt3a levels (Young et al., 2009), demonstrating that our overexpression studies caused similar changes to brachyury and Wnt signaling. Comparing the effects of overexpression with loss of function side by side, overexpression of Hoxa13b also caused downregulation of the genes reduced in the hoxa13;ntl^cs mutant embryos, as well as fgf8a and cdx4. However, Hoxa13b overexpression did not simply cause an overall decrease in gene expression because other genes, such as bmp4, increased under these conditions (Fig. 7D). From these data, we conclude that overexpression of the Hox13 genes is not simply the opposite of loss of function, suggesting that the effects of Hox13 overexpression need to be interpreted cautiously.

The Wnt/Brachyury autoregulatory loop is an essential feature of posterior development in early embryonic vertebrate development (Martin and Kimelman, 2008, 2009), with recent evidence from hemichordates demonstrating that it is an ancestral feature of deuterostomes (Fritzenwanker et al., 2019). In vertebrates, this loop ensures that cells remain in a bipotential progenitor state that can produce either neural or mesodermal fates. When canonical Wnt signaling is lost, embryos fail to produce mesoderm and neural tissue increases (Garriock et al., 2015; Gouti et al., 2017, 2014; Jurberg et al., 2014; Martin and Kimelman, 2009; Nowotschin et al., 2012; Wymeersch et al., 2016). Thus, it is crucial that the loop is maintained until the completion of posterior development. Brachyury has a second crucial role in activating cyp26a1 expression, thus keeping RA levels low at the posterior end of the vertebrate embryo, as high RA inhibits mesoderm formation (Martin and Kimelman, 2010; Olivera-Martinez et al., 2012; Sive et al., 1990). We propose here that the Hox13 genes act to help create a niche at the posterior end of the embryo that has high Wnt and low RA by ensuring that the Wnt/Brachyury loop and Cyp26a1 are robustly expressed throughout much of the somite-forming stages.

Key to our understanding of this pathway is the novel mutation in tbxta (ntl^cs), which allows us to control Tbxta activity depending on temperature (Kimelman, 2016a). Whereas ntl^cs mutants are indistinguishable from wild type at the standard temperature of 28.5°C, when we reduce the temperature to 21°C we begin to see some posterior truncation phenotype, which is further enhanced at 18.5°C. Even at 18.5°C, less than 10% of the embryos show the ntl null phenotype. This sensitized background was very useful for exploring the Hox13 mutants because the percentage of embryos with perturbed phenotypes increased in embryos homozygous for a loss of hoxa13b, and even more so when hoxd13a was also mutant. Possibly, the number of embryos with severe posterior truncations will increase yet further when the other less highly expressed posterior Hox13 genes (hoxa13a, hoxc13a and hoxc13b) are also mutant.

Although the Hox13 genes have been proposed to act to terminate axis extension, we find that they are expressed very early in somitogenesis, with hoxa13b expressed at the start of somitogenesis and hoxd13a expressed at the eight-somite stage, which seems surprising for their proposed role in terminating the axis. Similarly, a quantitative analysis in mouse has shown that Hox13 genes are activated at E9.5 in the neuromesodermal progenitor region, whereas somitogenesis does not end until 4 days later (Wymeersch et al., 2019). In chick embryos, strong expression of the Hox13 genes, as shown by in situ hybridization, was reported to begin around halfway through somitogenesis (Denans et al., 2015). Although the timing of onset does not preclude the possibility that the Hox13 genes are involved in axis termination, their early expression at least raises the possibility of alternative roles.

The evidence for Hox13 genes as axis terminators is based almost entirely on overexpression studies, with the one exception being the mouse Hoxb13 mutant, which adds two somites to the tip of the tail, thus increasing the body from 65 to 67 somites, through enhanced proliferation and decreased cell death in the mesoderm and neural ectoderm (Economides et al., 2003), whereas all other Hox13 mutants show no effect on axis length (Dolle et al., 1993; Fromental-Ramain et al., 1996; Godwin and Capecchi, 1998; Suemori and Noguchi, 2000). Interestingly, the one zebrafish hoxb13 gene, hoxb13a, is not expressed in the tailbud, and so perhaps the regulation of tailbud proliferation and apoptosis by Hoxb13 is an amniote or mammalian adaptation.

Although overexpression is commonly and often successfully used to study the role of different proteins in embryonic development, it can also lead to artifacts, particularly because transcription factors at higher than endogenous levels could bind to aberrant binding sites in the genome. Moreover, as many transcription factors act in protein complexes, overexpression has the potential to disrupt the normal stoichiometry of the complex. In our overexpression studies using temporally controllable heat shock lines, we saw strong effects on embryonic posterior development when we overexpressed hoxa13b. Similarly, a previous study in zebrafish that mosaically overexpressed hoxa13b using a promoter that restricted expression to the mesodermal progenitors and their descendants also observed defects in posterior development (Payumo et al., 2016), in agreement with our results. Why hoxd13a overexpression does not cause strong posterior defects is not yet clear, although it is in line with observations from overexpression studies in chick that also showed strong effects from Hoxa13 overexpression but no effect from Hoxd13 overexpression (Denans et al., 2015). Although the DNA binding regions of the different Hox13 factors (the homeobox) are very highly conserved, the N terminus, which makes up almost all of the rest of the protein, is quite different between Hoxa13 and Hoxd13 proteins.

The most important finding from our overexpression studies is that in most cases the regulation of gene expression by overexpression is not the opposite of loss of function, raising concerns about the interpretation of overexpression studies of the Hox13 proteins. As previously reported in mouse for Hoxc13 overexpression (Young et al., 2009), we see a partial reduction in wnt3a expression when Hoxa13b is overexpressed, and a downregulation of tbxta expression under these conditions, in agreement with the reduction in T/Brachyury observed in chick embryos when Hoxa13 is overexpressed (Denans et al., 2015). Thus, our overexpression results parallel those reported in amniotes. However, we also see a reduction in tbxta and wnt3a in the hoxa13;ntl^cs mutant embryos at the semi-permissive temperature of 18.5°C, demonstrating that the Hox13 genes promote the expression of these genes (and others), rather than inhibiting them. We therefore suggest that the previous overexpression studies should be carefully interpreted.

We propose that one key role of the Hox13 genes is to provide robustness to the Wnt/Brachyury loop because it is crucial that this loop is strongly expressed throughout all of the somitogenesis stages to ensure that mesoderm can be produced from the NMps as the body extends (Fig. 8). If the loop fails at any point during somitogenesis as a result of decreased Wnt or Brachyury, then the NMps fail to produce the somites and neural fates increase because only the sox2-positive derivatives of the NMps are produced (Garriock et al., 2015; Gouti et al., 2014; Martin and Kimelman, 2012; Tsakiridis et al., 2014; Turner et al., 2014). Wnt signaling is crucial for activation of downstream genes, particularly tbx16 in zebrafish and Tbx6 and Msgn1 in mice (Bouldin et al., 2015; Chalamalasetty et al., 2014; Kimelman, 2016b; Nowotschin et al., 2012; Wittler et al., 2007). These downstream genes act to inhibit the NMp state and to activate subsequent genes needed for somite formation (Fig. 8). A second important role of Brachyury is to activate expression of the RA-degrading enzyme Cyp26a1 (Martin and Kimelman, 2010), which keeps the levels of RA low in the NMps as increased RA levels terminate axis extension. Although RA is not necessary in mice or zebrafish for terminating axis extension (Berenguer et al., 2018; Cunningham et al., 2011), it is essential that the levels in the posterior end of the embryo are kept very low (Martin and Kimelman, 2010; Olivera-Martinez et al., 2012; Sive et al., 1990).

The regulation of Wnt signaling in the tailbud is challenging for the embryo. In the NMps, Wnt signaling needs to be regulated at a level high enough for Brachyury expression to be maintained, but not so high as to force the cells into a mesodermal fate. Once the cells exit the most posterior end of the embryo and enter the posterior presomitic mesoderm, they are exposed to higher Wnt levels than found in the NMp zone, which then activates mesodermal gene expression (Bouldin et al., 2015, Fig. 8). However, if Wnt is too high throughout the tailbud it produces deleterious effect such as suppressed neural development (Jurberg et al., 2014; Garriock et al., 2015; Martin and Kimelman, 2012) and aberrant segmentation (Bajard et al., 2014; Garriock et al., 2015; Jurberg et al., 2014). Thus, the role of the Hox13 genes might be to sustain the Wnt/Brachyury loop in the most posterior end of the embryo at a low level that is just sufficient to maintain Brachyury expression, but not so high as to cause deleterious effects.

How then might the Hox13 proteins regulate the Wnt/Brachyury loop? An intriguing possibility is suggested from a study in Caenorhabditis elegans that Hox factors act to ensure robust gene expression by consistently promoting the full activation of target genes by other transcription factors, and as such have been termed ‘guarantors’ of gene expression (Zheng and Chalfie, 2016; Zheng et al., 2015). In the guarantor model, the Hox proteins are not absolutely required for target gene expression and instead are crucial in cases where some cells or groups of cells fall below a critical threshold of expression of an essential gene because of variability in levels of transcription. In this sense, the variable and relatively low level of posterior embryonic defects seen at the fully permissive temperature (29°C) in our cross of double mutant fish in the ntl^cs/cs background can be viewed as embryos that stochastically fall on the lower end of expression of the Wnt/Brachyury loop due to very minor reductions in Tbxta activity without the help of the Hox13 factors to boost the levels. When Tbxta activity is further reduced in these embryos by lowering the temperature, the number of embryos that stochastically fall below the necessary threshold to sustain the loop increases, resulting in a greater number of embryos with a posterior defect. Although a true test of this model requires eliminating all five tailbud-expressed Hox13 genes in zebrafish, the results suggest that even a complete loss of Hox13 function may only be a partially penetrant effect with regards to completion of the A-P axis.

Which genes are regulated by the Hox13 factors in any system remains unknown but, in zebrafish, tbxta/brachyury is an intriguing candidate as it is upstream of both the wnt genes and cyp26a1 (Martin and Kimelman, 2008, 2010), and its expression is strongly downregulated in the hoxa13;ntl^cs embryos at the semi-permissive temperature (Fig. 4). Interestingly, a DNA fragment containing just 2.1 kb upstream of the tbxta start site produced tbxta expression throughout the mesoderm during the gastrula stages, and later expression in the notochord, but it did not activate tailbud tbxta expression, demonstrating that proximal sequences of tbxta do not activate expression in the NMps during the somitogenesis stages (Harvey et al., 2010). Importantly, using a new method for identifying in vivo Hox13 binding sites in tailbud cells has led to identification of a somitogenesis stage tailbud enhancer for tbxta, providing one clear locus of Hox13 regulation of the Wnt/Brachyury loop (Z.Y., C. R. Braden, A. E. Wills and D.K. unpublished).

Wild-type fish were an AB/WIK mixture. Fish were used for crosses at ages between 3 months and 3 years. The ntl^cs mutant fish (ntla w181) has been described previously (Kimelman, 2016a).

In the ntl^cs;hoxa13b^Δ16 mutant (designated as line w243), 20 bp were deleted and 4 bp were added resulting in a 16 bp deletion. In the hoxa13b sequence, CCCAAGTCCTGCACGCAACCCACCACATATGG was changed to CCCAAGagaaATATGG, with the lower case letters indicating novel bases.

In the ntl^cs; hoxa13b^Δ16; hoxd13a^ins13 mutant (designated as line w244), 13 bp were inserted. In the hoxd13a sequence, CCCGTGGACCAC was changed to CCCGTGaagcctcggtgaaGACCAC, with the lower case letters indicating novel bases.

In the ntl^cs; hoxa13b^Δ16; hoxd13a^ins4 mutant (designated as line w245), 2 bp were deleted and 6 bp were added, resulting in a 4 bp insertion. In the hoxd13a sequence, AACCCGTGGACCAC was changed to AACCCGccaataGACCAC, with the lower case letters indicating novel bases.

In the ntl^cs; hoxa13b^Δ16; hoxd13a^Δ8 mutant (designated as line w246), 8 bp were deleted. In the hoxd13a sequence, CAATAAACCCGTGGACCACGG was changed to CAATAAACCACGG.

All CRISPR mutant lines were designed and produced following published methods (Moreno-Mateos et al., 2015; Talbot and Amacher, 2014). The sequence used to make a gRNA for hoxa13b was GGGGGTTGCGTGCAGGACTT, which has a base change at the second base to allow for synthesis by T7 polymerase. The sequence used to make a guide RNA (gRNA) for hoxd13a was GGGGCTTCACCGTGGTCCAC, which has a base change at the second base to allow for synthesis by T7 polymerase.

The hoxa13b and hoxd13a coding regions were amplified from 15-somite stage zebrafish embryo cDNA and inserted into a vector such that the stop codons were removed and a viral 2A peptide (Provost et al., 2007) was placed immediately after the coding region. The mCherry sequence was placed immediately after the 2A sequence. This sequence was placed in a Tol2-hsp70 vector and the resulting plasmid was used together with Tol2 transposase to create stable transgenic lines as previously described (Kawakami, 2004): hsp70l:hoxa13b-2A-mCherry (designated w247) and hsp70l:hoxd13a-2A-mCherry (designated w248). Transgenic lines were heat shocked at 40°C for 30 min.

All animal protocols used here were approved by the University of Washington Institutional Animal Care and Use Committee.

Alkaline phosphate in situ hybridization used standard conditions (https://wiki.zfin.org/display/prot/Thisse+Lab+-+In+Situ+Hybridization+Protocol+-+2010+update). Fluorescent in situ hybridization used a published procedure (Lauter et al., 2011).

The sequence and use of the tbxta/ntl morpholino has been previously described (Martin and Kimelman, 2008). Injected at 5 ng, it completely recapitulates the ntl mutant phenotype.

Embryos were injected at the one-cell stage with 2% fluorescein or rhodamine dextran, and transplanted into the ventral side of a shield stage wild-type embryo using a CellTram (Eppendorf). The hoxa13;ntl^cs donor embryos were also raised after a small number of cells were taken for transplantation and only donors that produced a strong phenotype were scored.

HS:hoxa13b and HS:hoxd13a embryos were heat shocked at the 10-somite stage and kept at 28.5°C until they reached the 15-somite stage (about 3 h post heat shock). The hoxa13;ntl^cs embryos were placed at 18.5°C at shield stage and kept at this temperature until sampling. Embryos were sorted either by mCherry expression for the transgenic embryos or by phenotype (large neural tube and small presomitic mesoderm) for the mutants. Wild-type control embryos were manipulated the same way for each treatment. Tailbuds were isolated at the 15-somite stage using the method we described previously (Manning and Kimelman, 2015) in which the epidermis is removed prior to cutting the explant. The explants began at the boundary of the third most newly formed somite and extended to the most posterior end of the embryo. Thirty tailbuds were sampled and pooled for RNA extraction from each group. RNA was extracted using TriReagent (Invitrogen) and purified with RNA Clean & Concentrator spin columns (Zymo Research). cDNA was produced using the iScript Supermix cDNA kit (BioRad). Quantitative PCR was performed in triplicate using the SsoAdvanced Universal SYBR Green Supermix kit (BioRad) on a CFX Connect Real-Time PCR Detection System (BioRad). The primers used for amplification are listed in Table S1. The level of each specific gene in the explant was normalized to the level of ribosomal 18S mRNA in the explant, and the fold change of gene expression due to treatment (overexpression or mutation of hoxa13b) was calculated according to published methods (Schmittgen and Livak, 2008).

Embryos were injected with 50 ng mRNA encoding H2B-EGFP synthesized in vitro from the plasmid CS2-H2B-EGFP using the mMESSAGE MACHINE SP6 kit (ThermoFisher). Wild-type, hoxa13;ntl^cs and tbxta morphant embryos were placed at 18.5°C at shield stage and kept at this temperature until filming at the 10-somite stage. Embryos were mounted in 1% low-melt agarose and imaged using a spinning disk confocal microscope [inverted Marianas spinning disk system (Intelligent Imaging Innovations, 3i) with an Evolve 10 MHz EMCCD camera (Photometrics) and a Zeiss microscope Plan-NEOFLUAR 25×0.8 immersion objective]. Time-lapse image stacks with a step size of 2 μm were taken every 2 min for a period of 2 h at 25°C. The hoxa13;ntl^cs and tbxta morphant embryos were retrieved from the agarose after imaging and allowed to develop to 2 dpf (days post-fertilization) at 18.5°C to confirm phenotype. Three embryos from each genotype were used for cell movement analysis using Imaris (Oxford Instruments).

The Mann–Whitney test was used for comparison of the qPCR results for mCherry expression in the explants of HS:hoxa13b and HS:hoxd13a embryos after heat shock. Fisher's exact test was used to compare the number of defective embryos (class 1, 2 and 3 phenotypes pooled) after injection of 0.2 ng tbxta morpholino in the wild-type and hoxa13;d13 mutant fish in a ntl^+/+ background.

For the cell movement data analysis, only tracks with a duration longer than 30 min were included. The mean track speed and straightness were calculated using Imaris. Tracks were color-coded by mean track speed or straightness using Imaris to generate Fig. 6A,B,D,E,G,H. From each embryo, 300 to 500 representative tracks were manually and randomly picked from the NMp and MPZ based on the physical position. Tracks from the same region (NMp or MPZ) of three embryos were pooled for each genotype for statistical analysis of cell movement. Dunn's test was used for the post hoc multiple comparisons of cell moving speed and straightness after detection of significant difference using the Kruskal–Wallis test. All statistical analyses were conducted using R 3.6.0. The significance level was set at P<0.05.

We thank Natalie Smith and Adrian Wang for critical comments on the manuscript, Eric Thomas and Andrew Curtwright for advice on CRISPR, and Natalie Smith for invaluable help throughout this project.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.185298.reviewer-comments.pdf