Main

The body plan of vertebrates has distinct left–right (LR) asymmetries in the disposition of internal organs. Cells and tissues are instructed as to their left or right identity at very early stages of embryo development. A complex cascade of epigenetic and genetic mechanisms ultimately results in the directional transfer of laterality information to the regions adjacent to the embryo node by the late gastrulation and early somitogenesis stages. The identity of such mechanisms is beginning to be understood and involves signalling by the Wnt, fibroblast growth factor (FGF) and Notch pathways, among others (reviewed in refs 1–5). The LR asymmetric placement of internal organs contrasts with (and is concealed by) the bilateral symmetry of the musculoskeletal and dermal outer layers of the body wall. This bilaterally symmetric arrangement arises from the synchronized segmentation of the paraxial mesoderm into aligned pairs of somites along the anterior–posterior (AP) axis, a process that also takes place in the vicinity of the embryo node and is controlled by the Wnt, FGF and Notch pathways, among others (reviewed in refs 6–9). Thus, during a sizeable developmental window, the same signalling pathways convey both symmetric and asymmetric information for bilaterally synchronized somitogenesis and LR asymmetric patterning, respectively. This seeming paradox poses a challenge that is currently unexplained.

Here we show that the establishment of the LR axis in zebrafish embryos is controlled by two parallel mechanisms, one that depends on H+/K+-ATPase activity and acts very early in development (before the start of zygotic transcription), and a second mechanism akin to the nodal flow of mouse embryos, which takes place in Kupffer's vesicle (KV) during early somitogenesis and depends on the expression of left–right dynein (lrd) and functional cilia. We also identify a mechanism mediated by retinoic acid (RA) signalling that provides a molecular link between the development of AP and LR axes and is essential to buffer the influence of LR asymmetric information on the synchronized and hence bilaterally symmetric elongation of the AP axis. Our findings illustrate the existence of higher levels of coordination between the establishment of embryonic axes and organ patterning10.

Early steps of LR patterning in zebrafish

To address a possible role of H+/K+-ATPase activity in zebrafish LR patterning, we incubated embryos with the specific inhibitor omeprazole from the 1–2-cell stage until bud stage (zebrafish embryo staging is described in ref. 11). LR patterning defects were observed, as demonstrated by reversed heart looping (31% (n = 98) versus 1% (n = 90) in control embryos; Fig. 1a, g). Moreover, expression of the nodal-related gene southpaw (spaw) and its target pitx2, normally restricted to the left lateral plate mesoderm (LPM12,13,14; Fig. 1b, c), seemed randomized in embryos treated with omeprazole (44% (n = 45) and 43% (n = 82), respectively; Fig. 1h, i). Similar results were obtained with SCH 28080, a reversible inhibitor of H+/K+-ATPase activity (Table 1 and data not shown). The fact that these treatments induced high percentages of LR alterations when performed before midblastula transition15 (Table 1) indicates that the H+/K+-ATPase relevant for LR patterning in the zebrafish is maternally transcribed, as it is in Xenopus embryos16. H+/K+-ATPase α immunoreactivity was readily detected in 2-cell-stage embryos, and was subsequently localized to all blastomeres with no apparent asymmetries up to the 1k-cell stage (Supplementary Fig. S1). Thus, in the zebrafish, as in the chick16, the role of H+/K+-ATPase activity during LR patterning seems to depend on post-translational differences, rather than those at the transcriptional level as in Xenopus16. Our findings uncover a very early role of H+/K+-ATPase activity in the control of LR patterning in zebrafish embryos, which, to our knowledge, represents the earliest known step in the cascade of laterality information in this species.

Figure 1: Cascade of LR asymmetric information in the zebrafish.
figure 1

a–f, Control zebrafish embryos have a rightward looping of the heart, revealed by green fluorescence in the mlc2a-GFP transgenic line (a), left-sided expression of spaw (b) and pitx2 (c) in the lateral plate mesoderm (black arrowheads), expression of lrd in KV (d), numerous cilia in KV, as evaluated by acetylated tubulin immunoreactivity (ac-tub; e), and a net leftward flow inside KV, as evaluated by the movement of injected fluorescent beads (f). g–x, Inhibition of H+/K+-ATPase activity by incubation with omeprazole (g–l), inhibition of Notch signalling by incubation with the γ-secretase inhibitor DAPT (m–r) or downregulation of lrd translation by injection of an lrd-MO (s–x) resulted in reverse (leftward) heart looping (g, m and s), ectopic expression of spaw (h and t) in the right LPM, or absence of spaw expression in the left LPM (n and t), and ectopic expression of pitx2 in the right LPM (i, o and u). Expression of lrd in KV was not affected by any of these treatments (j, p and v). Cilia in KV seemed normal in size and distribution in embryos treated with omeprazole (k) or DAPT (q) but were severely distorted in lrd morphants (w). Leftward flow in KV of omeprazole-treated embryos and DAPT-treated embryos was normal (l, r) but was absent or very slow in lrd morphants (x). All embryo views are posterior, dorsal at the top, except a, g, m and s, which are ventral, anterior at the top, and b, c, h, i, n, o, t and u, which are dorsal, anterior at the top. Black and red arrowheads point to normal and ectopic gene expression domains, respectively. In f, l, r and x two superimposed photographs were taken 15 s apart and the second photograph was pseudocoloured in red.

Full size image
Table 1 LR patterning defects induced by manipulations during development
Full size table

In chick embryos, it has been shown that the difference in H+/K+-ATPase activity results in increased levels of extracellular Ca2+ on the left side of the node, which in turn results in a local increase in Notch activity necessary for correct LR patterning17. Downregulation or upregulation of Notch signalling in mouse18,19 and zebrafish19 embryos, respectively, also results in LR patterning defects, indicating a possible evolutionarily conserved role of Notch activity in the LR information cascade. To address whether Notch activity is necessary for LR patterning in the zebrafish, we incubated embryos with the γ-secretase inhibitor DAPT (N-[N-(3,5-difluorophenacetyl-l-alanyl)]-(S)-phenylglycine t-butyl ester)20,21. Incubation with DAPT from bud to early somitogenesis stages resulted in alterations in LR patterning at the level of visceral LR asymmetries (24% (n = 46); Fig. 1m), and in the expression of spaw and pitx2 (38% (n = 50) and 32% (n = 41), respectively; Fig. 1n, o). The fact that these treatments perturbed LR patterning most frequently when performed after 75% epiboly and before the 1-somite stage (Table 1) is in concordance with the proposed role of Notch signalling in the establishment of the LR axis in mouse18,19 and chick17 embryos, and shows that in zebrafish embryos Notch signalling acts downstream of H+/K+-ATPase in the cascade of laterality information.

A role for cilia in zebrafish LR patterning

In the mouse embryo, the initial LR symmetry-breaking event is thought to take place in the node and to be mediated by the directional leftward flow of extracellular fluid generated by the rotation of monocilia in nodal cells22. It has also been proposed that a similar mechanism operates in vertebrate embryos other than the mouse, on the basis of the existence of cilia and the expression of Lrd in the node of Xenopus, chick and zebrafish embryos23, although experimental confirmation of this suggestion is lacking. In contrast, LR asymmetries exist in Xenopus and chick embryos before node formation (reviewed in ref. 24), and our results described above with inhibition of H+/K+-ATPase and Notch activities indicate that this is most probably true of the zebrafish as well. Thus, if Lrd or other motor proteins have a function in LR patterning in these embryos, it is likely to be the amplification of upstream LR information rather than its generation de novo. We addressed this possibility by downregulating lrd translation using morpholino-modified antisense oligonucleotides (MO). Injection of lrd-MO into 1-cell-stage embryos resulted in a high frequency of LR patterning defects, as evaluated by reversed or failed heart looping (23% (n = 86); Fig. 1s) and altered expression of spaw and pitx2 in the LPM (40% (n = 67) and 40% (n = 36), respectively; Fig. 1t, u). As expected, this treatment did not alter lrd expression in KV (Fig. 1d, v).

We next explored the possibility that, as in mouse embryos25, the role of lrd in zebrafish LR patterning was related to cilia function and the generation of a directional fluid flow. For this purpose, we observed the cilia of cells lining KV (the zebrafish equivalent to the mouse node23) in control embryos and lrd morphants. Whereas cilia were readily visible and extended along the entire bottom of KV in control embryos (Fig. 1e; see also refs 23, 26), those of lrd morphants seemed fewer and shorter and were distributed in a patchy fashion in KV, which often seemed smaller and distorted (80% (n = 10); Fig. 1w). We also investigated the existence of a directional flow of extracellular fluid by tracking the movement of fluorescein-labelled latex beads injected inside KV. In control embryos we detected a movement of some of the injected beads towards the left side of KV in most cases (90% (n = 10); Fig. 1f). In contrast, beads injected into lrd morphants failed to move or did so at a much slower pace (90% (n = 10); Fig. 1x). These results indicate the existence of a directional leftward net flow inside KV that depends on lrd function and is associated with a normal distribution and/or length of cilia that line this structure. Our findings therefore provide strong experimental evidence in support of the ‘nodal flow hypothesis’ in vertebrate embryos other than the mouse.

To investigate the placement of lrd function in the cascade of laterality information in the zebrafish embryo, we injected lrd-MO into 1k-cell-stage embryos. This treatment induced LR patterning defects similar to those of embryos injected at the 1-cell stage (Table 1), showing that Lrd functions downstream of H+/K+-ATPase activity during LR patterning. Furthermore, because MO injections at the 1k-cell stage preferentially target cells fated to form KV26, these results argue against possible earlier functions of Lrd during LR patterning24 of zebrafish embryos. Downregulation of H+/K+-ATPase or Notch activities did not result in changes in lrd expression (Fig. 1j, p), cilia distribution or size (Fig. 1k, q) or fluid flow (Fig. 1l, r) in KV. Taken together, our results so far identify a LR information cascade in the zebrafish embryo that involves early epigenetic mechanisms, conserved in Xenopus and chick, that are amplified in KV by a cilia-based mechanism conserved in mouse.

Bilaterally symmetric somites are not a default state

In the course of our characterization of the laterality cascade in zebrafish embryos, we observed that experimental manipulations that blocked the flow of LR information resulted in evident but transient LR asymmetries in somite formation. This phenotype indicated that the normal bilaterally symmetric progression of somitogenesis was somehow under the control of the LR information cascade. Thus, whereas somite formation and segmentation proceeded in a bilaterally symmetric fashion in control embryos27 (Fig. 2a–c), downregulation of H+/K+-ATPase activity from the 1–2-cell stage up to the bud stages (but not from mid-epiboly onwards; not shown) frequently resulted in uneven numbers of somites on the left and right sides when analysed between the 6-somite and 13-somite stages (36% (n = 89); Fig. 2d–f). This asymmetry was not obviously biased in any direction: 41% of the abnormal embryos had one or more somites on the left side and 59% had one or more somites on the right side. Similar results were obtained by blocking the LR information flow at the level of Notch signalling. Consistent with the crucial role of Notch signalling in the molecular clock that regulates somitogenesis28, and with phenotypes of zebrafish mutants in Notch pathway components29,30,31,32, incubation with DAPT from bud to early somitogenesis stages resulted in evident alterations in somite segmentation and maturation (Fig. 2i; see also ref. 21). This treatment also induced LR asymmetries in the number of somites (26% (n = 55); Fig. 2g, h) that were not biased towards either side of the embryo: 46% of the abnormal embryos had one or more somites on the left side and 54% had one or more somites on the right side. These results indicate that the bilaterally symmetric somitogenesis of zebrafish embryos is not a default state but rather that it requires an active mechanism of LR clock coordination, under the control of the LR information cascade.

Figure 2: LR asymmetric somitogenesis in zebrafish embryos.
figure 2

a–c, Control zebrafish embryos have a symmetric number of somites at any given developmental stage, as evaluated by uncx4 (a, b) and myoD (c) expression. d–l, Inhibition of H+/K+-ATPase activity by incubation with omeprazole (d–f), inhibition of Notch signalling by incubation with the γ-secretase inhibitor DAPT (g–i) or downregulation of lrd translation by injection of an lrd-MO (j–l) resulted in an uneven number of somites on the left and right sides, as evaluated by uncx4 expression at the 8-somite stage (d, g and j) and at the 10-somite stage (e, h and k). The expression of myoD revealed asymmetric somitogenesis at the 10-somite stage in embryos incubated with omeprazole (f) and lrd morphants (l), and defects in somite maturation in embryos incubated with DAPT (i). All embryo views are dorsal, anterior at the top. Extra somites are indicated by arrows.

Full size image

The timing of the appearance of somite LR asymmetries after the downregulation of H+/K+-ATPase or Notch activities (between the 6-somite and 13-somite stages) is well correlated with the proposed developmental window for the role of KV in LR patterning26. We therefore next asked whether the disruption of cilia function and leftward fluid flow in KV of lrd morphants were associated with LR asymmetric somitogenesis. In these experiments a small but significant percentage of embryos had LR asymmetries in somite formation (14% (n = 210); Fig. 2j–l); these were unbiased (50% of abnormal embryos had more somites on the left side and 50% had more on the right side) and were limited to embryos between the 6-somite and 13-somite stages. We then analysed zebrafish one eye pinhead (oep) mutant embryos, in which the transfer of LR information to the node is thought to occur normally but in which there is an impaired relay of laterality information from the node to the LPM33. We found that, although oep mutant embryos had alterations of LR visceral asymmetries (not shown; see also ref. 33), somitogenesis proceeded in a bilaterally symmetric fashion indistinguishable from wild-type sibling embryos (not shown). Taken together, our results indicate that the cascade of LR information branches in or near KV to control the LR asymmetric patterning of internal organs and the LR symmetric segmentation of presomitic mesoderm (PSM) independently.

RA signalling coordinates LR and AP axis elongation

We next investigated the mechanism by which the LR information cascade controls the bilateral symmetry of somite formation. Nascent somites are specified in the PSM by the combined action of opposed gradients of Wnt/FGF and RA activities and by the synchronized oscillations of Notch and Wnt activities (reviewed in refs 6, 7, 28, 34). We manipulated the activity of these four pathways to ascertain their involvement in the coordination of LR somitogenesis in the zebrafish. Inhibition of FGF and/or Wnt signalling did not result in asymmetric somitogenesis in zebrafish embryos (Supplementary Fig. S2). In contrast, inhibiting Notch activity with DAPT (Fig. 2g, h) resulted in embryos with an uneven number of somites on each side, indicating that Notch activity is mechanistically involved in the coordination of LR somitogenesis. However, the fact that the observed LR asymmetries in somite formation occurred in an unbiased fashion indicates the possible existence of additional mechanisms that would provide the necessary laterality information to counteract the influence of the LR information flow.

Indeed, blocking endogenous RA production by using an MO against raldh2 (the major source of RA in the context of somite formation35, also known as aldh1a2 or aldehyde dehydrogenase 1 family, member A2) resulted in a strongly biased LR asymmetric somite development. raldh2 morphants initiated somitogenesis in a bilaterally symmetric way that was indistinguishable from that in embryos injected with control MO (Fig. 3a). However, as somite formation proceeded, a significant fraction of raldh2 morphants developed an uneven number of somites on the left and right sides (21%, versus 0% in embryos injected with control MO (n = 137 and 153, respectively); Fig. 3b, c). This asymmetry was evident between the 6-somite and 13-somite stages, after which raldh2 morphants recovered a normal bilaterally symmetric number of somites (Fig. 3d). The LR asymmetry in somitogenesis seemed strongly biased: the left side had more somites than the right one in 76% of the embryos with asymmetric somite formation. The specificity of the phenotypes induced by injecting raldh2-MO was further verified with zebrafish neckless (nls) mutant embryos, which lack RALDH2 activity36, with similar results (Fig. 3e–h and Supplementary Table S1).

Figure 3: RA signalling coordinates LR somitogenesis in zebrafish embryos.
figure 3

a–h, In situ hybridization for uncx4 expression on raldh2 morphants (a–d) or nls mutant embryos (e–h), in which RA signalling is downregulated, reveals bilaterally symmetric somite numbers at the 5-somite (a), 6-somite (e) and 14-somite (d and h) stages, and increased number of somites on the left side at the 7-somite (b), 8-somite (f) and 11-somite (c and g) stages. All embryo views are dorsal, anterior at the top. Extra somites are indicated by arrows.

Full size image

We next investigated the integrity of the LR information cascade in embryos injected with raldh2-MO. No alterations in the left-sided expression patterns of spaw or pitx2 in the LPM were detected in raldh2 morphants (n = 36 and 48, respectively; not shown), which is consistent with previous results from quail37 and mouse38 embryos. In contrast, we found that the left-biased asymmetries in somitogenesis caused by the downregulation of RA signalling required the existence of a normal flow of LR information. The inhibition of H+/K+-ATPase activity or lrd function in raldh2 morphants therefore prevented the left-sided bias of asymmetric somitogenesis that is characteristic of raldh2 morphants (Supplementary Table S1).

Molecular basis of LR asymmetric somitogenesis

The periodicity of somite formation is thought to depend on the action of an oscillator or clock that sets the pace of segmentation in the cells of the PSM. Several components of the Notch signalling pathway have been identified as molecular readouts of this segmentation clock and have been shown to have oscillatory expression patterns (cyclic genes; reviewed in refs 39, 40). We investigated the integrity of the segmentation clock in zebrafish raldh2-morphant or nls mutant embryos by analysing the expression of the cyclic genes deltaC30, her1 (refs 29, 41) and her7 (refs 42, 43). Whereas no differences were observed in the cycle phase44 of either transcript between left and right PMS in control embryos at any developmental stage (Fig. 4a–c, g–i, m–o), clear desynchonization was evident in embryos in which RA signalling was inhibited. Thus, most raldh2-morphant and nls mutant embryos analysed between the 4-somite and 12-somite stages had LR asymmetric expression patterns of deltaC (25% (n = 281); Fig. 4d–f, and not shown), her1 (28% (n = 258); Fig. 4j–l, and not shown) and her7 (30% (n = 147); Fig. 4p–r, and not shown). The degree of asymmetry between the left and right PSM was variable between embryos from the same injection experiment, some having an offset of the cyclic gene expression by one phase (Fig. 4d, f, k, q) and others by two phases (Fig. 4e, j, l, p, r); still others were in the same phase of consecutive cycles (not shown). These results indicate that RA signalling is necessary for the synchronization of the molecular clock between the left and right PSM.

Figure 4: Desynchronization of the molecular clock in raldh2 morphants.
figure 4

a–r, In situ hybridization for deltaC (a–f), her1 (g–l) and her7 (m–r) expression in control zebrafish embryos (a–c, g–i and m–o) showing synchronization between the left and right sides of the embryo, whereas raldh2 morphants (d–f, j–l and p–r) have desynchronized expression patterns. All embryo views are posterior, dorsal at the top. The phase of the cycle (1, 2 or 3) is indicated under each left and right PSM.

Full size image

The observation that extra somites in raldh2 morphants or nls mutant embryos are consistently the last ones formed is difficult to explain by a simple LR clock desynchronization mechanism. Instead, such a phenotype would be compatible with an anterior shift of the right determination front6 or a posterior shift of the left one. To investigate these possibilities we analysed the expression of mespb41 (one of the zebrafish orthologues of the mouse Mesp2 segmentation polarity gene45) after downregulation of RA signalling. In control embryos mespb is expressed in two or three bilaterally symmetric stripes that localize to the anterior half of the forming somite and the most anterior presumptive somite(s) (Fig. 5a; see also ref. 41). In contrast, raldh2 morphants had striking LR asymmetries in the expression pattern of mespb, which most frequently appeared shifted anteriorly on the right side of the embryo (21% (n = 35); Fig. 5b).

Figure 5: RA signalling counteracts the LR information flow during zebrafish somitogenesis.
figure 5

a, b, In situ hybridization for mespb in raldh2 morphants (b) reveals a LR asymmetry in its expression domain compared with control embryos (a). c, d, Expression of fgf8 in control embryos is bilaterally symmetric (c), whereas raldh2 morphants show an anterior extension of the fgf8 expression domain on the right side (d). e–n, In situ hybridization for raldh2 (e–h) and cyp26a1 (i–l) reveals complementary expression domains around the tailbud of control embryos (e, i), which are not altered in embryos treated with omeprazole (f, j) or DAPT (g, k) or in lrd morphants (h, l). m, n, The expression of cyp26b1 and cyp26c1 in early somitogenesis embryos is restricted to the midbrain–hindbrain boundary. All embryo views are posterior, dorsal at the top, except in m and n, which are dorsal, anterior at the top. Arrowheads point to the anterior level of gene expression.

Full size image

Anterior shifts in the expression pattern of Thylacine1 (Thy1), the Xenopus orthologue of mespb, have been reported after downregulation of RA signalling in frog embryos46. Although a possible LR asymmetry in such shifts was not analysed in those experiments (one side of the embryo served as control), the authors showed that the AP level of Thy1 expression was determined by the antagonism of RA and FGF signalling pathways46. To test whether a similar situation occurred in raldh2-morphant zebrafish embryos, we analysed the expression of fgf8, which is normally expressed in a broad domain in the posterior PSM (Fig. 5c; see also ref. 47). We detected an anterior shift in the expression domain of fgf8 on the right side of raldh2 morphants (25% (n = 32); Fig. 5d).

Finally, we investigated the mechanism by which RA signalling becomes lateralized in response to the LR information cascade. One possibility is at the level of expression of enzymes responsible for the synthesis or degradation of RA or its receptors. To address this, we performed in situ hybridization analyses of zebrafish raldh2 (which our results indicate encodes the RA-synthesizing enzyme relevant for preventing LR asymmetric somitogenesis), cyp26a1, cyp26b1 and cyp26c1 (encoding RA-catabolizing enzymes), and rarα2a, rarα2b, rarγ, rxrα, rxrβ, rxrγ and rxrδ (encoding RA receptors) in control embryos of 3-somite and 6–8-somite stages, and after experimental manipulations of the LR information cascade. We could not detect obvious LR asymmetries in the expression of these transcripts at any stage analysed in control or manipulated embryos (Fig. 5e–n, Supplementary Fig. S3, and data not shown), indicating that the control of RA signalling by the LR information cascade is likely to be post-transcriptional. In this respect, a possibility that warrants further investigation is the participation of a mechanism recently uncovered in the mouse embryo node that results in the sided accumulation of RA itself (N. Hirokawa, personal communication).

Taken together, our results demonstrate that the bilateral progression of somitogenesis in zebrafish embryos depends on RA signalling and is tightly linked to the cascade of LR organ asymmetry information. The findings presented in the accompanying paper indicate that this mechanism is conserved in mouse and chick embryos (ref. 48; P. Dollé, personal communication). Specifically, our results are consistent with a model in which the LR information flow causes transient side-restricted changes in the activities of the Notch, FGF and/or Wnt signalling pathways, which are subsequently buffered by the antagonistic action of RA signalling (Fig. 6). In the chick, the normal LR information flow induces transient and subtle lateralization in the expression of Fgf8 (ref. 49), Wnt8C50 and the Notch pathway components Delta-like1 and Lunatic fringe17, in the perinodal region. Even though comparable LR differences in the normal expression of such transcripts are not observed in zebrafish embryos, it is possible that they are too subtle to be detected by in situ hybridization. Alternatively, LR differences in the activity of Notch, FGF and/or Wnt signalling pathways might take place at post-transcriptional levels. In any case, our experiments show that blocking the arrival of LR asymmetric inputs to the node results in an unbiased desynchronization of somite formation between the left and right sides of the embryo. This phenotype is consistent with the existence of LR asymmetric antagonistic influences on somitogenesis on either side of the zebrafish node. In our model, the relief of RA-signalling-mediated buffering of these LR asymmetric influences, induced in raldh2 morphants and nls mutant embryos, would result in their amplification, leading to a desynchronization of the Notch-based molecular clock and to an extension of the fgf8 expression domain on the right side of the embryo. These molecular alterations would in turn result in a delayed maturation of presomitic cells in the right PSM, as compared with the left PSM, and LR asymmetric somite segmentation. Our findings identify a crosstalk between the molecular mechanisms that regulate LR patterning and AP axis extension to ensure that both processes take place in a coordinated manner.

Figure 6: Crosstalk between LR and AP axes.
figure 6

Diagram illustrating the interactions between the mechanisms controlling LR patterning and somite formation during early somitogenesis stages, and the proposed role of RA signalling as a LR buffer of the laterality information flow (blue arrows) in the zebrafish embryo. See the text for details.

Full size image

Methods

Zebrafish strains

Wild-type (AB), mlc2a–eGFP, nlsi26 and oeptz257 zebrafish strains were used in this study. References for these strains are provided in Supplementary Methods.

Pharmacological treatments

Zebrafish embryos were incubated with omeprazole (Sigma), SCH 28080 (Sigma), DAPT (γ-secretase inhibitor IX; Calbiochem) and/or SU5402 (Calbiochem) as detailed in Supplementary Methods.

Morpholino and messenger RNA injections

Capped mRNAs encoding axin or GFP were synthesized and used as detailed in Supplementary Methods. MOs against raldh2, fgf8 or lrd, and a control MO, were designed and used as detailed in Supplementary Methods.

In situ hybridization

Details of the whole-mount in situ hybridization protocol and probes used in this study are given in Supplementary Methods.

Other methods

Details of immunofluorescence analyses and visualization of fluid flow in KV are provided in Supplementary Methods.