Flower meristem maintenance by TILLERS ABSENT 1 is essential for ovule development in rice

Plants have a unique ability to produce lateral organs, such as leaves and floral organs, continuously throughout their life cycle. This ability depends on the activity of pluripotent stem cells, which are located at the tip of above-ground meristems, such as the shoot apical meristem (SAM) and flower meristem. In these meristems, stem cells self-renew to maintain a constant number, while supplying cells for organ differentiation.

In Arabidopsis (Arabidopsis thaliana), the CLAVATA–WUSCHEL (CLV–WUS) negative feedback loop plays a central role in coordinating stem cell self-renewal with organ differentiation in the meristem. WUS, a homeodomain transcription factor, positively regulates stem cell proliferation (Laux et al., 1996; Mayer et al., 1998). Loss of function of WUS causes a failure in stem cell maintenance, leading to premature termination of the meristem (Mayer et al., 1998). In the vegetative phase, the wus mutant terminates shoot development after forming a few leaves, and wus flowers lack carpels and have reduced numbers of stamens in the reproductive phase (Laux et al., 1996). In contrast to WUS, the CLV signaling pathway plays a major role in repressing the proliferation of stem cells (Clark et al., 1993, 1995; Fletcher et al., 1999; Schoof et al., 2000). The protein encoded by CLV3 is processed into a CLE peptide, which acts as a ligand of CLV1 – a leucine-rich repeat (LRR)-receptor kinase (Clark et al., 1997; Fletcher et al., 1999). The CLV signaling pathway negatively regulates WUS expression, whereas WUS promotes CLV3 transcription by directly binding to its promoter (Brand et al., 2000; Schoof et al., 2000; Yadav et al., 2011).

The genetic mechanisms underlying stem cell maintenance have been also well-studied in grasses such as maize (Zea mays) and rice (Oryza sativa). In particular, several signaling pathways that negatively regulate stem cell proliferation are well documented in both maize and rice (Taguchi-Shiobara et al., 2001; Suzaki et al., 2004, 2006, 2008, 2009; Bommert et al., 2005, 2013; Pautler et al., 2015; Je et al., 2016, 2018; Wu et al., 2018; Suzuki et al., 2019a). In rice, for example, a signaling pathway containing FLORAL ORGAN NUMBER 1 (FON1) and FON2 plays an important role in negative regulation of the flower meristem (Suzaki et al., 2004, 2006). FON1 encodes a CLV1-like LRR-receptor kinase, whereas FON2 encodes a CLE protein closely related to CLV3. Loss of function of either FON1 or FON2 induces stem cell overproliferation, leading to an increase in the number of stamens and carpels. Other CLE genes, such as FON2-LIKE CLE PROTEIN 1 (FCP1) and FON2 SPARE 1 (FOS1), are also involved in the negative regulation of the stem cell maintenance in the SAM and/or flower meristem (Suzaki et al., 2008, 2009). By contrast, our knowledge of positive regulators of meristem maintenance remains limited in rice.

Unlike WUS in Arabidopsis, its rice ortholog, TILLERS ABSENT 1 (TAB1; OsWUS), is not involved in positive regulation of the SAM in the vegetative phase (Tanaka et al., 2015; Suzuki et al., 2019b). TAB1 is not expressed in the SAM and the tab1 mutant produces the same number of leaves as wild type (Tanaka et al., 2015). Instead, TAB1 promotes stem cells during axillary meristem development. In the tab1 mutant, no axillary shoot (tiller) is formed due to a failure in stem cell maintenance at the early stage of axillary meristem development (Tanaka et al., 2015; Hirano and Tanaka, 2020; Tanaka and Hirano, 2020). TAB1 function is partially repressed by FON2 because the TAB1 expression domain is expanded in an enlarged meristem at the early stage of axillary meristem development in tab1 (Tanaka and Hirano, 2020). Therefore, the genetic relationship between TAB1 and FON2 in axillary meristem development in rice is similar to that between WUS and CLV3 in SAM maintenance in Arabidopsis. In the vegetative SAM, WOX4, a close paralog of TAB1, maintains stem cells (Ohmori et al., 2013). In axillary meristem development, WOX4 starts to be expressed at a later stage and is responsible for maintaining the established axillary meristem (Tanaka et al., 2015). In summary, the positive regulator differs depending on the type of meristem in rice, and, as yet, a positive regulator in the flower meristem is not known.

In Arabidopsis, the ovule is initiated from the placenta, which is differentiated from the carpel margin meristem – a meristematic tissue present along the margins of two fused carpel primordia (Skinner et al., 2004; Roeder and Yanofsky, 2006; Reyes-Olalde et al., 2013). Specification of ovule identity is regulated by SEEDSTICK (STK), which acts redundantly with three other AGAMOUS (AG) clade MADS-box genes: AG, SHATTERPROOF 1 (SHP1) and SHP2 (Pinyopich et al., 2003). After specification, WUS plays an important role in ovule development. In transgenic plants in which WUS is knocked out in an ovule-specific manner, integument differentiation is completely inhibited (Groß-Hardt et al., 2002). WUS is specifically expressed in the nucellus but not in the integument, indicating that WUS promotes integument development in a non-cell-autonomous manner. The nucellus-specific expression of WUS is regulated by class III homeodomain leucine zipper (HD-ZIPIII) genes in collaboration with BELL1 (Yamada et al., 2016).

Unlike in Arabidopsis, ovule formation depends on the activity of the flower meristem in rice (Yamaki et al., 2005, 2011). The ovule is differentiated directly from the flower meristem, which remains even after carpel specification (Yamaki et al., 2005). Loss of function of LONELY GUY (LOG), which encodes a cytokinin-activating enzyme, results in a severe defect in the flower meristem, leading to a reduced number of floral organs, such as stamens and carpels (Kurakawa et al., 2007). A weak allele of log displays an ovule-less phenotype, consistent with the observation that the ovule initiates from the flower meristem (Yamaki et al., 2011). Loss of function of rice OsMADS13, an ortholog of STK, results in the conversion of an ovule into a carpelloid structure (Dreni et al., 2007). Thus, the genetic regulators that specify ovule identity seem to be conserved between Arabidopsis and rice.

In our previous study, we showed that the tab1 mutant produces abnormal floral organs (Tanaka et al., 2015); however, the function of TAB1 in flower development remains to be elucidated. In this study, we conducted developmental analysis focusing on maintenance of the flower meristem and ovule formation in tab1. As a result, we found that ovule development is strongly compromised in tab1. Furthermore, expression of meristem markers was found to disappear precociously from the flower meristem in tab1. Collectively, our results suggest that, by maintaining stem cells at a later stage of flower development, TAB1 plays an essential role in ovule development in rice.

We previously reported that tab1-1, a complete loss-of-function mutant, produces the same number of leaves in the vegetative phase as wild type (Tanaka et al., 2015). To examine the function of TAB1 in the reproductive phase, we first examined the number of floral organs in tab1-1. Analysis of 180 flowers indicated that there was no significant difference in the number of lodicules, stamens or pistils between wild type and tab1-1 (Fig. 1A). However, about half of the tab1-1 flowers were abnormal such that the palea/lemma was reduced in size or absent (Fig. S1).

We found that the tab1-1 mutant was completely sterile, as reported previously (Lu et al., 2015). Therefore, to check whether the pistil is functional in tab1-1, we crossed wild-type pollen with tab1-1 pistils. As a result, no seeds were obtained even after performing more than 70 crosses. We carefully observed the morphology of the tab1-1 pistil; however, only minor abnormalities, such as a slightly thinner ovary and an increase in stigma number, were found (Fig. 1B,C,E,F).

To examine the inner features of the pistil, we cleared the tab1-1 pistils and observed them under differential interference contrast optics. In wild type, two long vascular bundles ran from the base of the ovary to the tip of each stigma, and a short vascular bundle reached almost to the central region of the ovary (Fig. 1D). In tab1-1, long vascular bundles similarly stretched towards the stigmas (Fig. 1G,H); however, the short vascular bundle ended in an abnormal position in some cases (Fig. 1G), or branched abnormally and had a mass of proliferated cells in others (Fig. 1H). The internal structure of the ovary also appeared to be affected by tab1 mutation: an elliptical structure, presumed to be an ovule, was clearly observed inside the ovary in wild type (Fig. 1D), but no such structure was detected in tab1-1 (Fig. 1G,H).

Collectively, these results suggest that pistil formation is likely to be partially compromised in the tab1-1 flower.

We prepared longitudinal sections to observe the internal structure of the ovary in more detail. In wild type, an ovule, surrounded by inner and outer integuments, adhered to the inner wall of the ovary and an embryo sac was clearly observed inside the ovule (Fig. 2A). In tab1-1, by contrast, no ovule-like structure was observed. In severe cases, the ovary had a hollow structure, implying that it lacked an ovule (Fig. 2B,D); in others, an unidentified structure with neither embryo sac nor integuments was present inside the ovary (Fig. 2C,D). These observations suggest that ovule formation is severely defective in the tab1-1 mutant.

To examine further the defect in ovule formation in tab1-1, we carried out developmental analysis using marker genes for flower development. First, we analyzed the expression pattern of DROOPING LEAF (DL), which is required for carpel specification (Yamaguchi et al., 2004). We confirmed that DL is expressed in the carpel primordia in wild type (Fig. 3A,C), as reported previously (Yamaguchi et al., 2004). The carpel primordia in tab1-1 had morphology similar to that in wild type and showed a similar expression pattern of DL (Fig. 3B,D), implying that carpel specification is not affected by the tab1 mutation.

Next, we focused on ovule development. At the early stage of development in wild type, a small bulge considered to be an ovule primordium was observed between the carpels (Fig. 3A). By contrast, a flattened structure instead of a bulge was observed between the carpel primordia in tab1-1 (Fig. 3B), suggesting that normal ovule initiation is inhibited by tab1 mutation. At a later stage in wild type, an ovule primordium with initiated integuments was observed inside the ovary (Fig. 3C). In tab1-1, by contrast, the putative ovule primordium lacked the integument-like structure (Fig. 3D).

We also examined the expression of OsMADS13, which has an important role in ovule development in rice (Dreni et al., 2007). As previously reported (Dreni et al., 2007), OsMADS13 was expressed in the ovule primordium and the adaxial side of developing carpels in wild type (Fig. 3E,G). Although a similar expression pattern of OsMADS13 was detected in tab1-1, the morphology of the ovule was aberrant during its development (Fig. 3F,H): in some cases, an abnormal tissue formed instead of an ovule primordium (Fig. 3F); in others, the ovule primordium was completely absent inside the ovary (Fig. 3H).

These morphological observations indicated that ovule development is seriously impaired in tab1-1; however, the expression pattern of OsMADS13 was similar between wild type and tab1-1. Therefore, the defects in ovule initiation and development observed in tab1-1 do not seem to be associated with the function of OsMADS13.

Because the ovule is formed from the flower meristem, we examined the expression pattern of a meristem marker, ORYZA SATIVA HOMEOBOX 1 (OSH1) (Sato et al., 1996; Tsuda et al., 2011), during flower development. In wild type, OSH1 was expressed throughout the flower meristem at an early stage of flower development (Fig. 4A), and a similar pattern of OSH1 expression was detected in most tab1-1 plants (Fig. 4B). At a middle stage, when stamen primordia emerged, OSH1 expression was observed in the flower meristem between stamen primordia in wild type (Fig. 4C). Again, a similar expression pattern of OSH1 was observed in the flower meristem in tab1-1 (Fig. 4D).

At a later stage, after carpel primordia initiation, OSH1 was expressed in a small region adjacent to the carpel primordia in wild type (Fig. 4E); however, such OSH1 expression was not detected in any tab1-1 plant (Fig. 4F). This result suggests that meristematic cells are lost precociously at a late developmental stage in the tab1 flower. Hereafter, we refer to the meristem at this late stage as the ‘final flower meristem’.

Our recent study showed that TAB1 promotes stem cell fate during axillary meristem development (Tanaka and Hirano, 2020). To clarify whether TAB1 is associated with stem cell maintenance in flower development, we analyzed the expression pattern of the stem cell marker FON2 (Suzaki et al., 2006). In wild type, FON2 was expressed in a small central region of the flower meristem at the early stage of flower development (Fig. 5A), as reported previously (Suzaki et al., 2006). In tab1-1, FON2 expression was similarly observed in the early flower meristem (Fig. 5B). At a later stage, when carpel primordia emerged, FON2 expression was restricted to a few outer layers of the final flower meristem in wild type (Fig. 5C). In tab1-1, by contrast, FON2 expression was completely absent (Fig. 5D), suggesting that stem cells had disappeared. This result suggests that TAB1 is required for maintaining stem cells in the final flower meristem.

Next, we analyzed the spatiotemporal expression patterns of TAB1 during flower development. TAB1 transcript was first detected in the apical central region of the flower meristem at the early stage of flower development (Fig. 6A). At the later stage, when carpel primordia emerged, TAB1 expression was observed in the final flower meristem (Fig. 6B). This result supports the idea that TAB1 functions in the flower meristem.

TAB1 expression was also detected during ovule development. Soon after an ovule primordium emerged, TAB1 was expressed in the margin of the ovule primordium (Fig. 6C). Following differentiation of the ovule tissues, TAB1 transcript was observed in the embryo sac and the boundary region between the nucellus and integuments (Fig. 6D).

To examine the genetic relationship between TAB1 and FON2, we analyzed the expression pattern of TAB1 in the fon2-3 mutant, in which the flower meristem is enlarged as a result of overproliferation of stem cells (Suzaki et al., 2006, 2019a). At the early stage, TAB1 was expressed in a wider region in the enlarged flower meristem of fon2-3 compared with wild type (Fig. 6A,E). At a later stage, the expression region of TAB1 was similarly expanded in the final flower meristem in fon2-3 (Fig. 6B,F).

Next, we analyzed the expression pattern of WOX4, which is responsible for maintaining the vegetative SAM in rice (Ohmori et al., 2013) during flower development. The results showed that WOX4 was expressed in the flower meristem at both the early and late stages of flower development (Fig. 6G,H).

To examine the effect of fon2 mutation on the tab1 phenotype, we generated a fon2-3 tab1-1 double mutant and analyzed its phenotype. Consistent with a previous report (Suzaki et al., 2006), the number of pistils was significantly increased in fon2-3 (Fig. S2). There was no significant difference in pistil number among wild type, tab1-1 and fon2-3 tab1-1.

We then focused on ovule morphology. Ovule structure in fon2-3 was similar to that in wild type (Fig. 7A,B,G). As described above, an ovule-like structure was not observed in any tab1-1 plants, and a hollow ovary was detected in the most severe cases (Fig. 7C,G). In fon2-3 tab1-1, by contrast, about 40% of the flowers formed an ovule with an embryo sac and integuments (Fig. 7D,G), although the morphology differed slightly from that of wild type. In addition, about 30% of the fon2-3 tab1-1 flowers formed an ovule-like structure with integuments but without an embryo sac (Fig. 7E,G). Furthermore, the proportion of severe morphological abnormalities, such as the hollow ovary observed in the tab1-1 single mutant, was markedly reduced in the double mutant (Fig. 7F,G). These results suggest that the fon2 mutation partially rescued the ovule defects in the tab1 mutant.

To test whether this rescue was associated with meristem function, we analyzed the expression pattern of OSH1. In situ hybridization analysis showed that OSH1 was clearly expressed in the final flower meristem in fon2-3 tab1-1 (Fig. 4G), similar to its expression in wild type (Fig. 4E). This result indicates that the activity of the final flower meristem that was lost in tab1-1 was restored by the fon2 mutation.

Taken together, these observations suggest that the partial rescue of ovule formation in fon2-3 tab1-1 seems to be associated with restoration of the activity of the final flower meristem.

In rice, TAB1 acts as a stem cell-promoting factor during axillary meristem development (Tanaka and Hirano, 2020). However, TAB1 is not involved in stem cell maintenance in the SAM in the vegetative phase (Tanaka et al., 2015; Suzuki et al., 2019b); instead, WOX4 is required for SAM maintenance (Ohmori et al., 2013). In Arabidopsis, by contrast, WUS promotes stem cell maintenance in both the developing axillary meristem and the established SAM (Laux et al., 1996; Mayer et al., 1998; Wang et al., 2017). Thus, in rice, TAB1 and WOX4 are likely to share, in part, the function of WUS in Arabidopsis.

In the present study, we identified a function of TAB1 in flower development: namely, TAB1 is predominantly required for maintaining stem cells in the final flower meristem at the late stage of flower development (Fig. 8). At this stage, expression of the meristem marker OSH1 and the stem cell marker FON2 disappeared completely in the tab1 mutant. As a result, tab1 plants failed to form the ovule, which should be differentiated from the final flower meristem. Thus, the importance of TAB1 function is evident in the final flower meristem.

In Arabidopsis, loss of function of the stem cell regulator WUS leads to a reduction of the number of floral organs (Laux et al., 1996). However, we observed no significant change in the number of floral organs in tab1 mutants. OSH1 and FON2 were expressed normally in the flower meristem at the early-to-middle stages of flower development in tab1, suggesting that stem cell activity was not affected by the tab1 mutation at these stages. Therefore, there seems to be a putative gene (X) that promotes stem cell fate at the early-to-middle stages (Fig. 8). As deduced from the tab1 phenotype, the action of gene X is expected to be reduced at the late stage. We first considered that WOX4 is a possible candidate for gene X owing to its function in the SAM (Ohmori et al., 2013). Unexpectedly, however, in situ hybridization analysis showed that WOX4 was expressed in the flower meristem at the late stage in addition to the early-to-middle stages. Therefore, it is likely that WOX4 alone is insufficient to maintain the flower meristem at the late stage, and probably also at the early-to-middle stages. Thus, WOX4 does not seem to be a candidate for gene X. Identification of gene X is an important issue to be resolved in a future study in order to understand stem cell regulation in the flower meristem.

Our assumption of the existence of gene X indicates that several positive regulators are involved in stem cell maintenance in rice. Our group have revealed that rice has multiple CLV3-like negative regulators of stem cell maintenance such as FON2, FCP1 and FOS1 (Suzaki et al., 2006, 2008, 2009). The diversification of both positive and negative regulators implies that the genetic network of stem cell regulation is highly complex and varies depending on the type of meristem in rice. In Arabidopsis, a CLV3-related CLE peptide, CLE27, has been reported to act as a negative regulator of meristem maintenance (Je et al., 2016). In addition, the wus mutant forms axillary shoots from adventitious meristems and flowers on the shoots (Laux et al., 1996), suggesting that an unknown positive regulator promotes stem cells in adventitious meristems and flower meristems. Therefore, the regulatory mechanisms underlying stem cell maintenance in Arabidopsis may also not be simple.

In Arabidopsis, the flower meristem is consumed after carpel initiation, and ovules are initiated from the placenta, which develops from the carpel margin meristem (Skinner et al., 2004; Roeder and Yanofsky, 2006; Reyes-Olalde et al., 2013). In rice, by contrast, the flower meristem remains even after carpel initiation, and the ovule is initiated directly from the meristem (Yamaki et al., 2005, 2011).

Various defects regarding ovule formation were observed in tab1. The loss-of-ovule phenotype seems to result from a failure in ovule initiation. The unidentified structure detected inside the ovary instead of the ovule probably results from inhibition of ovule differentiation. Despite these defects, OsMADS13 was expressed in tab1 as it is in wild type; thus, the potential for ovule specification does not seem to be affected by loss of TAB1 activity. So why is ovule development defective in tab1? Considering the role of TAB1 in the final flower meristem discussed above, we suggest that the reduced stem cell numbers and subsequent shortage of ovule initial cells might lead to the ovule defects seen in tab1 mutants. In support of this suggestion, a weak allele of LOG, which is required for maintaining the flower meristem, also results in ovule loss (Yamaki et al., 2011).

When combined with fon2 mutation, tab1 showed partial rescue of the ovule defects. Furthermore, expression analysis of OSH1 revealed that the activity of the final flower meristem in tab1 was recovered by the fon2 mutation. These observations imply that the excess of stem cells in fon2 mutants is probably associated with the recovered meristem activity, eventually leading to the partial rescue of ovule formation in fon2 tab1 plants. Taken together with our suggested role of TAB1, it is likely that TAB1 promotes ovule formation by maintaining stem cells in the final flower meristem (Fig. 8).

In Arabidopsis, the wus mutation is epistatic to clv3; thus, the wus clv3 double mutant has a wus phenotype (Schoof et al., 2000). The reason why the fon2 mutation partially rescues the tab1 phenotype in rice may be related to the unknown gene X that promotes stem cell at the early-to-middle stages of flower development (Fig. 8), as discussed above. The activity of gene X might be elevated as a result of the fon2 mutation, generating an excess amount of stem cells. Previously, we showed that the tab1 defects in axillary meristem development were partially rescued by fon2 mutation (Tanaka and Hirano, 2020). This rescue was found to be due to precocious expression of WOX4. The difference in the phenotypes between tab1 fon2 in rice and wus clv3 in Arabidopsis seems to result from the existence of an additional gene(s) promoting stem cells in rice.

Although the activity of the final flower meristem in tab1 mutants was recovered by fon2 mutation, ovule formation was not fully rescued. In addition, fertility was not restored in fon2 tab1 plants. In situ hybridization analysis indicated that TAB1 is expressed in the developing ovule, in addition to the flower meristem, suggesting that TAB1 plays some role in ovule differentiation. Therefore, the incomplete rescue of the tab1 ovule by fon2 mutation seems to be related to a failure in ovule differentiation, in addition to a deficiency of stem cells. In Arabidopsis, WUS is reported to be involved in ovule development, especially in integument development (Groß-Hardt et al., 2002; Yamada et al., 2016). Indeed, the CLV1::WUS; wus plant, in which loss of WUS activity is specific to ovules, fails to initiate integuments (Groß-Hardt et al., 2002). We observed that integument differentiation was similarly inhibited in the tab1 mutant in rice. Therefore, a role in integument development may be conserved between rice TAB1 and Arabidopsis WUS. However, the integument defect in tab1 was partially rescued by fon2 mutation to some extent, which also suggests that this defect is a secondary effect of insufficient cells for ovule development. The role of TAB1 in ovule differentiation, including integument formation, will be an interesting issue to explore in future studies.

In this study, we have shown that TAB1 is essential for ovule development in rice. TAB1 seems to promote ovule formation by maintaining stem cells in the ‘final flower meristem’ at a late stage of flower development. In Arabidopsis, ovule formation is not associated with the flower meristem, because the ovule is initiated from the placenta. Therefore, the direct necessity of stem cell activity in ovule formation seems to be unique to rice.

Oryza sativa L. ssp. japonica cultivar Taichung 65 (T65) was used as the wild type in phenotype comparisons and in situ hybridization analysis. The tab1-1 and fon2-3 mutants have been reported previously (Suzaki et al., 2006; Tanaka et al., 2015). tab1-1 has a nucleotide substitution in the splicing site of intron 1 that causes premature termination of TAB1 protein translation in the intron (Tanaka et al., 2015). fon2-3 has a point mutation at a conserved amino acid in the CLE domain and shows phenotypes very similar to those of a complete loss-of-function allele (Suzaki et al., 2006). Plants were grown outdoors in pots containing soil.

Pistils were dissected and immersed in a fixing solution (acetic acid and ethanol, in a ratio of 1:3). After 5 days, the samples were transferred to sterile water for 1 h, and then to a clearing solution (chloral hydrate, water, and glycerol in a ratio of 8:3:1) for more than 24 h. Samples were observed under differential interference contrast optics using a BX50 optical microscope (Olympus) coupled with an Axiocam 506 color camera (Carl Zeiss).

Plant tissues, such as pistils and young panicles, were dissected, fixed and dehydrated according to the methods of Toriba and Hirano (2018). The tissues were replaced with xylene, and then embedded in Paraplast Plus (McCormick Scientific LLC). Microtome sections (8 μm) were prepared for histological analysis, mounted on glass slides, and stained with 0.05% Toluidine Blue-O for observation using a BX50 optical microscope (Olympus) coupled with an Axiocam 506 color camera (Carl Zeiss).

To prepare the FON2 probe, RNA was transcribed with T7 RNA polymerase using digoxigenin-labeled UTP (Roche Applied Science), and treated by alkaline hydrolysis, as described previously (Suzaki et al., 2006). The TAB1, OSH1, WOX4, DL and OsMADS13 probes were prepared as described previously (Yamaguchi et al., 2004; Dreni et al., 2007; Ohmori et al., 2013; Tanaka et al., 2015). Microtome sections (8-10 μm) were made from tissues embedded in Paraplast Plus as described above and mounted on glass slides. In situ hybridization and immunological detection were performed according to the methods of Toriba and Hirano (2018). Observation was performed using a BX50 optical microscope (Olympus) coupled with an Axiocam 506 color camera (Carl Zeiss).

We thank A. Takahashi and K. Ichiki for technical assistance.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199932.