Kremen1 restricts Dkk activity during posterior lateral line development in zebrafish

Canonical Wnt signaling regulates many cellular behaviors during embryonic development, such as progenitor cell maintenance, proliferation, migration and differentiation (Grigoryan et al., 2008). The canonical Wnt signaling pathway is activated by a large family of secreted Wnt ligands that bind to Frizzled receptor and LDL receptor-related proteins 5 and 6 (Lrp5/6). The single-pass transmembrane proteins Kremen1 and Kremen2 have been identified as modulators of canonical Wnt signaling (Cselenyi and Lee, 2008; Nakamura et al., 2008). Kremen1/2 are high-affinity co-receptors for Dickkopf (Dkk) proteins, a family of secreted Wnt inhibitors. During development of various organs, including the nervous system, limbs and liver, Kremen1/2 synergize with the Dkk proteins to inhibit Wnt signaling (Davidson et al., 2002; Ellwanger et al., 2008; Lu et al., 2013). Dkk inhibits Wnt activity by competitively binding to Lrp5/6. In cultured cells, Dkk binds to Kremen1/2 and Lrp5/6, thereby forming a protein complex; this ternary complex is endocytosed and probably targeted for degradation (Nakamura et al., 2008). Studies in mice and Xenopus demonstrated that Dkk binding to Kremen1/2 inhibits canonical Wnt signaling, as loss of Kremen1/2 function leads to Wnt overactivation phenotypes, such as defects in anterior CNS development and limb formation (Davidson et al., 2002; Osada et al., 2006; Ellwanger et al., 2008). Furthermore, in zebrafish, Dkk and Kremen1 (a Kremen2 homolog has not been identified in zebrafish) have been shown to regulate mechanosensory organ size by inhibiting Wnt mediated proliferation (Wada et al., 2013). By contrast, in the absence of Dkk, Kremen1/2 were found to bind directly to Lrp5/6 in vitro, and Kremen2 facilitates Wnt activity during neural crest induction in Xenopus (Hassler et al., 2007). Further research is required to more fully elucidate how Kremen1/2 modulate canonical Wnt signaling during development in vivo.

The zebrafish posterior lateral line (pLL) mechanosensory system has proven to be an excellent model for studying how Wnt/β-catenin signaling regulates cellular dynamics during embryonic development. Formation of the pLL occurs through the collective migration of a cohort of ∼100 cells called the pLL primordium (pLLP) (Aman and Piotrowski, 2009; Chitnis et al., 2012). As the pLLP migrates, it deposits patterned clusters of cells that will form mechanosensory organs called neuromasts (NMs). As each maturing NM is deposited from the rostral (trailing) region of the pLLP, proliferating progenitor cells in the caudal (leading) region of the pLLP give rise to new proto-NMs. At the end of pLLP migration, the nascent pLL consists of 5-6 NMs along the trunk of the zebrafish embryo and a terminal cluster (tc) of 2-3 NMs at the end of the tail.

The Wnt/β-catenin and Fgf signaling pathways interact to regulate cellular proliferation, differentiation and migration in the pLLP (Aman and Piotrowski, 2009; Chitnis et al., 2012; Harding and Nechiporuk, 2012). Canonical Wnt signaling is active in the leading region of the pLLP and restricted in the trailing zone, in part by expression of dkk1b (Aman and Piotrowski, 2008). Conditional overexpression of dkk1b results in a dramatic truncation in the pLL due to loss of pLLP organization, decreased cellular proliferation, increased cell death and failed migration (Aman and Piotrowski, 2008; Aman et al., 2011; McGraw et al., 2011). The Fgf signaling pathway is active in the mid-region of the pLLP and is required for the formation of proto-NMs (Aman and Piotrowski, 2009; Chitnis et al., 2012; Harding and Nechiporuk, 2012). Recent work by Matsuda et al. (2013) showed that Wnt signaling through Lymphoid enhancer-binding factor 1 (lef1) regulates NM spacing along the trunk through the expression of the Fgf pathway inhibitor Dual specificity phosphatase 6 (dusp6) in the leading region of the pLLP (Matsuda et al., 2013). In summary, the Wnt pathway must be precisely regulated in the pLLP to ensure proper pLL development.

In this study, we show that Kremen1 acts in the pLLP to modulate Wnt signaling. We have identified a kremen1 mutant strain (krm1^nl10) that displays a premature and progressive loss of Wnt activity in the leading region of the pLLP, as well as decreased cellular proliferation and increased cell death. Our mosaic analysis revealed that mutation of kremen1 leads to non-cell-autonomous defects in the pLLP that are morphologically similar to ectopic activation of Dkk1 (McGraw et al., 2011). Attenuating Dkk levels, either by morpholino knockdown of dkk1b/2 or inhibition of Fgf signaling, results in a partial rescue of the krm1^nl10 phenotype. Analysis of tagged Dkk1b protein in krm1^nl10 and wild-type (WT) embryos revealed a broader spread of Dkk1b in mutant primordia. Based on these data, we propose that an absence of Kremen1 function leads to an expanded range of the Dkk signal and, consequently, premature attenuation of Wnt signaling in the pLLP. Our study provides new insight into Kremen1-mediated modulation of Wnt activity in vivo during organogenesis and suggests a pathway that might be misregulated during disease.

In an ongoing N-ethyl-N-nitrosourea (ENU)-based screen to identify zebrafish mutants with defects in pLL development, we isolated the strain krm1^nl10. By contrast to complete pLLP migration in WT embryos, migration of the pLLP halted midway along the trunk in krm1^nl10 mutants, and the pLLP dispersed into a single line of cells extending from the last deposited NM (Fig. 1A,B). Live imaging demonstrated that new NM deposition failed in the caudal portion of the trunk due to a failure of proto-NM renewal in mutant primordia (Fig. 1E,F; supplementary material Movies 1, 2). As a result, by 2 days post fertilization (dpf), krm1^nl10 mutants showed significantly fewer NMs that were also shifted anteriorly, and an invariable lack of tc NMs (Fig. 1A,B,D,G). Adult krm1^nl10 homozygous mutants were viable, mated naturally and produced offspring, although they lacked NMs on the caudal-most trunk and tail as adults (supplementary material Fig. S1A,B).

Positional cloning of the krm1^nl10 mutation identified a single base pair change from thymine to cytosine in the splice donor site of intron 6 of the kremen1 gene on chromosome 5 (supplementary material Fig. S2A,B). RT-PCR for kremen1 generated a single product in WT embryos and at least two variants in krm1^nl10 mutants (supplementary material Fig. S2C,D). Variant 1, the more abundant product, resulted from improper splicing and insertion of the 100 base pair (bp) intron 6 between exons 6 and 7. Variant 2 resulted from premature splicing of exon 7, which led to the loss of 31 bp in exon 6 (supplementary material Fig. S2D). Both variants contained predicted translational frame shifts, which resulted in premature stop codons and disruption of the CUB domain, an essential domain for Kremen1/Dkk binding (Mao et al., 2002), and in loss of the transmembrane domain (supplementary material Fig. S2E). Injection of 200 pg/nl WT kremen1 mRNA rescued the loss of tc NMs in krm1^nl10 mutants without producing obvious loss of Wnt phenotypes in WT embryos. By contrast, expression of mRNA encoding mutant variants 1 or 2 failed to rescue tc NMs in krm1^nl10 mutant embryos (supplementary material Fig. S2G). Injection of kremen1 morpholino antisense oligonucleotide (krm1-MO) (Gore et al., 2011), along with p53-MO to minimize non-specific cell death (Robu et al., 2007), into WT Tg(cldnB:GFP) zygotes produced embryos missing tc NMs and an anterior shift in the axial level of deposited NMs, phenotypes similar to those observed in krm1^nl10 mutants (Fig. 1B,C,D,G). Together, these data indicate that the krm1^nl10 mutant phenotype results from a loss of Kremen1 function.

At 30 hours post fertilization (hpf), kremen1 was expressed in discrete regions in the developing embryo, including the pLLP, lens, otic vesicle, midbrain-hindbrain boundary and the nascent tail fin fold (supplementary material Fig. S3A). The Kremen1 ligands dkk1b and dkk2 displayed complementary expression patterns to kremen1 in the developing embryo (supplementary material Fig. S3B,C). At 32 hpf, kremen1 was expressed in the leading and mid-zones of WT and krm1^nl10 pLLP (Fig. 2A,B). By 40 hpf, kremen1 expression was restricted to the leading zone (Fig. 2C,D). We found that, in addition to the previously reported expression of dkk1b (Aman and Piotrowski, 2008), dkk2 was also expressed in the migrating pLLP. In WT embryos, both dkk1b and dkk2 were expressed in the mid-region of the pLLP between 32 and 40 hpf (Fig. 2E,G,I,K). dkk1b and dkk2 had similar expression domains in krm1^nl10 primordia (Fig. 2F,H,J,L). By 40 hpf, dkk2 was primarily localized at the last deposited NM, and faint expression was present throughout the mutant pLLP, probably due to a loss of proper patterning in the mutant pLLP during late stages of migration (Fig. 2L). In addition, the Wnt and Dkk receptors lrp5 and lrp6 were expressed throughout the pLLP at 36 hpf (supplementary material Fig. S4A,B). In summary, the expression domains of kremen1, dkk1b, dkk2, lrp5 and lrp6 are consistent with their roles in modulating canonical Wnt activity in the pLLP.

Studies in mice and frogs have demonstrated that Kremen1 functions as an inhibitor of canonical Wnt signaling, and loss of Kremen1 function would therefore be expected to result in a Wnt overactivation phenotype (Nakamura et al., 2008). Zebrafish apc^mcr mutants showed elevated Wnt/β-catenin activity that resulted in increased cellular proliferation and abnormal pLLP migration (Aman and Piotrowski, 2008; Aman et al., 2011). Surprisingly, rather than resembling a Wnt overactivation phenotype, the krm1^nl10 pLL phenotype was strikingly similar to the loss of Wnt signaling found in lef1 mutants (McGraw et al., 2011; Valdivia et al., 2011). lef1 mutant primordia showed premature loss of pLLP organization and truncation of the pLL, due in part to a decrease of proliferative cells in the leading region of the pLLP. Thus, we asked whether cell proliferation is impaired in krm1^nl10 mutant primordia.

Using a bromodeoxyuridine (BrdU) incorporation assay, we found that proliferation was reduced in krm1^nl10 mutant pLLP compared with WT at 34 hpf (Fig. 3A-C). We also observed a significant increase in cell death in krm1^nl10 primordia at 40 hpf using a terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay (17.1% TUNEL-positive primordia in WT embryos versus 40.0% in krm1^nl10; Fig. 3D-F) and caspase immunoreactivity (13% of WT and 27% of mutant primordia showed increased caspase3 immunolabeling; supplementary material Fig. S5A-C). Interestingly, injection of the p53-MO did not rescue NM numbers or spacing in krm1^nl10 mutants compared with uninjected mutant embryos (supplementary material Fig. S11E), suggesting that the loss of proliferation, but not a p53-dependent cell death, is a major cause of the krm1^nl10 phenotype.

To further define the cellular mechanisms underlying the krm1^nl10 phenotype, we used photoconversion of the Kaede fluorophore to mark cells in the leading zone of pLLP at 24 hpf, and then examined their number and location at 48 hpf, when pLLP migration was complete. In WT embryos, labeled cells underwent multiple cell divisions (Fig. 3G,I,K) and populated the terminal NMs. In krm1^nl10 embryos, there was no significant change in the number of labeled cells between 24 and 48 hpf (Fig. 3H,J,K). This result was strikingly different from the behavior of labeled progenitor cells in lef1^nl2 mutants, which remained proliferative but prematurely exited the leading region (McGraw et al., 2011). Thus, krm1^nl10 mutants more closely resemble a phenotype resulting from a global inhibition of Wnt activity by ectopic Dkk1b expression (Aman et al., 2011; McGraw et al., 2011) rather than loss of Wnt transcriptional activity seen in lef1^nl2 mutants.

Because decreased proliferation and increased cell death are consistent with reduced Wnt signaling, we asked whether expression domains of Wnt target genes, lef1, axin2 and sef (Aman and Piotrowski, 2008), were altered in krm1^nl10 mutant pLLPs. At 32 hpf, WT and krm1^nl10 pLLPs showed similar expression patterns for lef1, axin2 and sef (Fig. 4A,B,G,H,M,N; Table 1). By 40 hpf, WT primordia displayed a reduction in expression of Wnt target genes, consistent with previous research showing that Wnt signaling was progressively decreased in the pLLP over the course of its migration (Valdivia et al., 2011; Matsuda et al., 2013). By contrast, krm1^nl10 primordia showed an earlier reduction of target gene expression at 36 hpf and an even more substantial reduction by 40 hpf (Fig. 4D,F,J,L,P,R; Table 1).

To further investigate modulation of Wnt activity in krm1^nl10 mutants, we used a transgenic biosensor line, Tg(7xtcf-siam:eGFP), which responds to Wnt/β-catenin activity (Valdivia et al., 2011; Moro et al., 2012). At 36 hpf, WT embryos showed GFP-positive cells predominantly in the leading region of the pLLP (supplementary material Fig. S6A-A″). By contrast, in krm1-MO-injected embryos, the number of GFP-positive cells was significantly decreased in the leading region of the pLLP (supplementary material Fig. S6B-B″,F), although the overall number of GFP-positive cells was not significantly different from WT (supplementary material Fig. S6E). By 40 hpf, GFP was significantly downregulated throughout the primordia of kremen1 morphants compared with WT embryos (supplementary material Fig. S6C-F). GFP-positive cells in the trailing and mid-region of morphant primordia at 40 hpf are localized at proto-NMs (supplementary material Fig. S6D-D″, yellow arrows). The fluorescent intensity of GFP expression was significantly reduced in morphant primordia compared with WT at both 36 and 40 hpf (supplementary material Fig. S6E). Together with decreased Wnt target expression, these results indicate a significant decline in Wnt activity in the leading region of the pLLP during later stages of mutant pLL formation.

Recent work has shown that expression of the ERK inhibitor dusp6 is regulated by canonical Wnt signaling through lef1 in the leading region of the pLLP and is required to inhibit Fgf activity (Matsuda et al., 2013). Whereas dusp6 was expressed in the leading and mid-region in WT primordia (Fig. 4S,U,W), its expression was absent in the leading region of the krm1^nl10 mutant at 32 hpf (Fig. 4T) and restricted to the mid-zone and last-deposited NMs at 36 and 40 hpf (Fig. 4V,X). However, expression of the Wnt target fgf10a (Aman and Piotrowski, 2008; Matsuda et al., 2013) and the Fgf target pea3 were not grossly altered in krm1^nl10 primordia (supplementary material Fig. S7A-L). Expression of the chemokines cxcr4b and cxcr7b that are required for proper pLLP migration were also not altered in krm1^nl10 mutants compared with WT embryos (supplementary material Fig. S8A-D). Altogether, these data indicate that Wnt-mediated transcription is prematurely and progressively reduced, but not completely abolished in krm1^nl10 primordia.

Because Kremen1 is a transmembrane receptor protein, we reasoned that decreased Wnt activity observed in krm1^nl10 mutants might be rescued by introducing WT cells into krm1^nl10 primordia. Mosaic WT embryos containing WT donor cells showed full extension of the pLL (6/6 embryos; Fig. 5A,B; supplementary material Fig. S9A,A′; Table 2). Surprisingly, presence of WT donor cells in krm1^nl10 mutant pLLPs was unable to rescue pLL extension (0/16 embryos; Fig. 5C; supplementary material Fig. S9B,B′; Table 2). By contrast, similar numbers of WT donor cells were able to rescue pLL formation in lef1^nl2 mutant hosts (5/5 embryos; supplementary material Fig. S9E,E′; Table 2) (McGraw et al., 2011). A similar non-cell-autonomous defect in pLLP migration was observed in mosaic embryos containing Tg(hsp70l:dkk1b-GFP) donor cells that were induced to secrete Dkk1b following heat shock (0/6 embryos showed tc formation; supplementary material Fig. S8D,D′; Table 2) (McGraw et al., 2011). Finally, mosaic embryos containing krm1^nl10 donor cells on a WT background showed full pLL formation (10/10 embryos; supplementary material Fig. S9C,C′; Table 2). These results indicate that loss of Kremen1 function leads to non-cell-autonomous defects in mutant pLLP, similar to those induced by Dkk1b secretion; by contrast, mutant cells behave normally in a WT background.

The failure of WT donor cells to rescue the krm1^nl10 phenotype could be explained by a non-cell-autonomous loss of Wnt signaling in these cells. We tested this by generating mosaic embryos that contained Tg(7xtcf-siam:eGFP) donor cells in WT or krm1^nl10 hosts. At 24 hpf, GFP-positive donor cells were present in both WT and mutant host primordia (Fig. 5E,F,I). By 48 hpf, GFP-positive cells contributed to the deposited NMs, including the tc NMs of WT hosts (Fig. 5G,I; supplementary material Movie 3). By contrast, the majority of donor cells in krm1^nl10 mutant hosts downregulated GFP and possibly underwent cell death, as indicated by the rhodamine-positive cell fragments in the pLLP (blue arrowheads; Fig. 5H,I; supplementary material Movie 4). Together, these data suggest that loss of Kremen1 function results in a non-cell-autonomous decrease in Wnt signaling in krm1^nl10 primordia.

Based on our data, we hypothesized that a loss of Wnt signaling in krm1^nl10 mutant pLLPs resulted from an expansion of Dkk activity. Unfortunately, we were unable to obtain commercial antibodies that recognized zebrafish Dkk1b or Dkk2 proteins to directly test this model. Thus, we used a number of approaches to decrease Dkk activity in WT and krm1^nl10 embryos. We reasoned that if our hypothesis was correct, attenuating Dkk activity should rescue the pLL truncation in krm1^nl10 mutants.

Fgf signaling in the pLLP is known to be required for the proper expression of dkk1b (Aman and Piotrowski, 2008). Thus, we decreased Dkk function in the pLLP by attenuating Fgf signaling with the pharmacological inhibitor SU5402 (Mohammadi et al., 1997). WT and krm1^nl10 embryos were treated with a suboptimal dose of SU5402 (a dose that did not cause pLL patterning defect in WT) or DMSO (control) beginning at 24 hpf, and the extent of pLLP migration and NM spacing were analyzed at 2 dpf. As expected, dkk1b expression was decreased and lef1 expression increased following SU5402 treatment in both WT and krm1^nl10 embryos when compared with DMSO-treated controls (Fig. 6G-N). We found that NM spacing, but not NM number, was rescued in the treated krm1^nl10 mutant embryos, and that WT primordia migrated to the end of the tail to deposit a full complement of NMs (Fig. 6A-F). BrdU incorporation indicated that proliferation was not rescued in SU5402-treated krm1^nl10 mutants compared with DMSO-treated controls and was decreased in SU5402 WT pLLP (supplementary material Fig. S10). These results suggest that reduction of Dkk levels through inhibition of Fgf signaling partially rescues pLLP patterning in krm1^nl10 mutants.

Next, we blocked Dkk function by injecting WT or krm1^nl10 zygotes with dkk1b and/or dkk2 splice-blocking morpholinos (supplementary material Fig. S11A-C). In all conditions, including control embryos, p53-MO was also co-injected to minimize non-specific cell death. Injection of the individual dkk-MOs resulted in a small but significant increase in the number of deposited NMs in krm1^nl10 mutants compared with krm1^nl10 embryos injected with p53-MO (supplementary material Fig. S11D). Co-injection of both dkk1b/2-MOs resulted in a rescue of NM numbers comparable to that of WT (supplementary material Fig. S11D). WT embryos injected with dkk1b, dkk2 or dkk1b/2-MOs did not show a significant change in NM number compared with WT p53 morphants, although their NMs were shifted caudally (supplementary material Fig. S11D; Fig. 7A,C,E). krm1^nl10 embryos injected with dkk1b/2-MOs showed a partial rescue of NM spacing compared with control krm1^nl10 embryos (Fig. 7B,D,E); in a subset of embryos, the pLLP migrated to the end of the tail. In addition, krm1^nl10 embryos injected with dkk1b/2-MOs showed rescue of BrdU incorporation levels (Fig. 7J-N). Both WT and mutant embryos also showed expanded lef1 expression domains in the primordia following dkk1b/2-MO injection (Fig. 7F-I). This is consistent with a previous report, indicating that the size of the Wnt signaling domain in the pLLP regulated NM spacing (Matsuda et al., 2013). Together, these results support the model that expansion of Dkk activity in krm1^nl10 mutants is responsible for the premature loss of Wnt activity in the pLLP. When Dkk function is inhibited by dkk1b/2-MO injection, Wnt activity and cellular proliferation are restored in krm1^nl10 mutants, which in turn results in partial to full rescue of NM number and spacing.

Based on our data, we predicted that Kremen1 acts to restrict the domain of Dkk activity in the pLLP. To examine the behavior of Dkk1b protein during pLLP migration, we generated a tagged version of Dkk1b under the control of the heat shock promoter (hsp70:dkk1b-mTangerine). Mosaic expression of the ectopic Dkk1b-mTangerine protein beginning at 25 hpf resulted in severe pLL truncation and loss of NMs in both WT and krm1^nl10 embryos at 2 dpf (supplementary material Fig. S12). This phenotype is consistent with global induction of Dkk1b expression (McGraw et al., 2011) and indicates that the fusion protein is functional. To examine the behavior of tagged Dkk1b protein during pLLP migration, we heat-shocked injected embryos at 25 hpf and examined the spread of mTangerine fluorescence in WT and krm1^nl10 embryos at 28 hpf, prior to the onset of Dkk1b-induced cell death. We only analyzed primordia containing between one and three Dkk1b-mTangerine-positive cells, so we could unambiguously determine the source of the secreted fusion protein. In mutant primordia, we found a significant increase in the amount of Dkk1b-mTangerine at two cell diameters from the expressing cell compared with WT primordia. Importantly, despite the transient nature of the transgene expression, there was no significant difference in mTangerine levels within the source cells (Fig. 8A-D). These data suggest that Kremen1 is necessary to limit the spread, and possibly the subsequent activity, of Dkk1b in the pLLP.

In this paper, we describe a novel zebrafish mutant strain (krm1^nl10) that carries a genetic lesion in kremen1, a non-obligate co-receptor for the Dkk family of secreted Wnt inhibitors. krm1^nl10 mutants fail to form a complete pLL due to a loss of Wnt signaling in the leading region of the pLLP, which in turn results in a decreased cellular proliferation, increased cell death and loss of Wnt target gene expression. Transplantation of WT donor cells into the krm1^nl10 primordia failed to rescue the mutant phenotype, demonstrating the non-cell-autonomous effects of the mutation. In mosaic embryos, krm1^nl10 donor cells were able to contribute to pLL formation in a WT host, consistent with the role of Kremen1 as a non-obligate receptor for Dkk. Attenuation of Dkk levels either partially or completely rescued pLL formation in the krm1^nl10 mutants. Our overexpression experiments also demonstrated that the Dkk1b fusion protein has a more extensive spread from the source cell in the mutant background. We conclude that Kremen1 restricts the domain of Dkk to the mid-region of the pLLP, probably through mediating the endocytosis of the Dkk-Lrp5/6 complex. In the absence of Kremen1 function, Dkk improperly spreads and inhibits normal Wnt signaling (supplementary material Fig. S13). This represents a previously unidentified mechanism for the precise modulation of canonical Wnt activity in the context of collective cell migration.

Previous in vitro and in vivo studies have demonstrated that Kremen1 acts together with Dkk to inhibit Wnt signaling and that loss of Kremen1 results in overactivation of canonical Wnt signaling (Nakamura et al., 2008). By contrast, we observed phenotypes consistent with Wnt inhibition in the pLLP of the krm1^nl10 mutants. The evidence for this included: truncation of the pLL and an anterior shift in the spacing of the deposited NMs; phenotypes observed during inhibition of Wnt signaling (McGraw et al., 2011; Valdivia et al., 2011; Matsuda et al., 2013); expression of Wnt target genes during pLLP migration were prematurely and progressively decreased in krm1^nl10 mutants; expression of the Wnt sensor line Tg(7xtcf-siam:eGFP) was decreased in kremen1 morphants; and absence of Wnt-dependent dusp6 expression (Matsuda et al., 2013) in the leading region of krm1^nl10 mutant primordia. Additionally, we observed a decrease in BrdU incorporation and an increase in cell death in the pLLPs of krm1^nl10 mutants. Lineage labeling of progenitor cells in the pLLP further revealed that cells in krm1^nl10 mutants failed to divide and remained in the leading region. Altogether, krm1^nl10 mutants displayed a gradual loss of Wnt activity that resembled the phenotype seen after ectopic activation of Dkk1b activity (Aman et al., 2011; McGraw et al., 2011).

Based on the loss of Wnt activity phenotype and additional experiments (mosaic analyses, attenuation of Dkk expression and analysis of tagged-Dkk1b protein), we propose a model in which Kremen1 plays a role in limiting the range of Dkk proteins in the pLLP. We propose that in krm1ⁿ¹⁰ mutants, the absence of Kremen1 results in a failure of Dkk-Lrp5/6 endocytosis, leading to the spread of Dkk throughout the pLLP and to improper diminishment of Wnt activity in the leading region. Generation of mosaic embryos containing WT donor cells in krm1^nl10 mutant hosts revealed that the absence of Kremen1 function resulted in non-cell-autonomous defects, i.e. a failure by donor cells to rescue the krm1^nl10 phenotype. Additional mosaic analyses using Tg(7xtcf-siam:eGFP) donor cells in krm1^nl10 hosts showed a distinct non-cell-autonomous loss of Wnt-mediated GFP expression in transplanted cells. This result is strikingly different from that observed in mosaic lef1^nl2 embryos containing WT donor cells, which showed complete rescue of the pLL (McGraw et al., 2011), again underscoring the non-cell-autonomous nature of the krm1^nl10 mutant phenotype. How could mutation of Kremen1, a membrane-localized receptor, exert non-cell-autonomous defects on cells in the pLLP? As Kremen1 is a receptor for the secreted Wnt inhibitor Dkk and regulates endocytosis of the Kremen1-Dkk-Lrp5/6 complex (Mao et al., 2002), we reasoned that in krm1^nl10 mutants, Dkk was perhaps not properly cleared from the pLLP. We were unable to directly test this model due to a lack of antibodies that recognize zebrafish Dkk proteins. Thus, we used indirect methods, such as attenuation or blocking of Dkk activity and ectopic expression of tagged Dkk1b protein. When we decreased Dkk levels in the mutant pLLPs either by blocking Fgf signaling, which regulates dkk1b expression in the pLLP (Aman and Piotrowski, 2008), or by dkk1b/2 morpholino injections, we found a partial rescue pLL formation in krm1^nl10 mutants. Mosaic expression of tagged Dkk1b protein in WT and krm1^nl10 mutant primordia revealed a significant increase in the spread of Dkk1b-mTangerine in mutant pLLP. Together, these results support our model that Kremen1 functions to restrict the domain of Dkk activity in the pLLP. Future work is required to examine the behavior of native Dkk1b protein in these mutants.

We show that kremen1, but not dkk1b/2, is expressed in the leading region of the pLLP. This raises the question of whether Kremen1 might have another function in addition to acting as a Dkk receptor. Previous work in vitro demonstrated that, in the absence of Dkk1, Kremen1/2 binds to and stabilizes Lrp6 at the cell surface, thus promoting Wnt activity. Additionally, Dkk2 has been shown to stimulate Wnt signaling during neural crest induction (Hassler et al., 2007). The possibility that Kremen1 might also play a role in promoting Wnt activity in the pLLP is compatible with our model, in which Kremen1 is also required to restrict the range of Dkk activity. However, inhibiting Dkk1b/2 function is able to partially or fully rescue pLL formation in krm1^nl10 mutants, which argues against Kremen1 promoting Wnt activity. Future work is needed to determine whether Kremen1 has additional roles in modulating Wnt signaling in the pLLP.

In conclusion, we have shown that Kremen1 modulates Wnt signaling in the pLLP and is required for proper collective cell migration. Our data indicate that in the absence of Kremen1 function, Wnt signaling is prematurely downregulated in the pLLP in a non-cell-autonomous manner through an ectopic expansion of Dkk proteins. As improper Wnt signaling has been implicated in collective cancer invasion (Friedl and Gilmour, 2009), and expression of kremen1 is altered in some cancers (Dun et al., 2010; Murphy et al., 2012), our findings may provide a potential mechanism for Wnt misregulation during disease.

Embryonic and adult zebrafish were staged and maintained according to standard protocols (Kimmel et al., 1995). The transgenic fish strains used were: Tg(-8.0claudinB:lyneGFP)^zf106, referred to as Tg(cldnB:GFP) (Haas and Gilmour, 2006), Tg(neuroD:eGFP)^nl1 (Obholzer et al., 2008), Tg(hsp70l:dkk1b-GFP)^w32 (Stoick-Cooper et al., 2007) and Tg(7xtcf-siam:eGFP)^ia4 (Valdivia et al., 2011; Moro et al., 2012).

The genetic lesion underlying the krm1^nl10 phenotype was identified using standard simple sequence length polymorphism (SSLP) mapping techniques (Gates et al., 1999). To identify polymorphisms, heterozygous carriers of krm1^nl10 on a polymorphic *AB/WIK background were intercrossed to produce homozygous, heterozygous and WT progeny. The krm1^nl10 lesion was located on the distal arm of chromosome 5, in the kremen1 gene (GenBank accession number: BC158176.1). WT and mutant mRNAs were generated using the mMessage kit (Ambion) and were injected at 200 pg/nl into one-cell stage zygotes.

RNA in situ hybridization was performed using standard protocols (Andermann et al., 2002). Antisense RNA probes generated were: lef1 (Dorsky et al., 1999), axin2 (Aman and Piotrowski, 2008), fgf10a (Grandel et al., 2000), pea3 (Raible and Brand, 2001; Roehl and Nüsslein-Volhard, 2001), sef (Aman and Piotrowski, 2008), dusp6 (Lee et al., 2005), dkk1b (Aman and Piotrowski, 2008), cxcr4b and cxcr7b (Dambly-Chaudiere et al., 2007). We cloned full-length cDNA and used it to generated probes for kremen1 (forward RT-PCR primer: AGTGTTATACAGCGAATGG, reverse RT-PCR primer: TTAGTTTCCCACAAGTGGG) and dkk2 (forward RT-PCR primer: ATGCTCACTGTTACGAGGAGT, reverse RT-PCR primer: GTATGGATCGTCCCTTCTTAG). Whole-mount immunolabeling was performed following established protocols (Ungos et al., 2003). The following antibodies were used: rabbit or mouse anti-GFP (Life Technologies, A11122 and A11120, respectively; both 1:1000), rat anti-BrdU (Abcam, ab6326; 1:100), rabbit anti-activated caspase 3 (Cell Signaling, 9664; 1:200), Alexa 488 (Life Technologies, A11001; 1:1000) and Alexa 568 (Life Technologies, A11011; 1:1000). TUNEL labeling was performed using an established protocol modified for fluorescent detection (Nechiporuk et al., 2005). BrdU incorporation was carried out using established protocols (Harris et al., 2003; Laguerre et al., 2005). Adult NMs were labeled with 0.005% 2-[4-(dimethylamino)styryl]-N-ethylpyridinium iodide (DASPEI; Life Technologies) in embryo medium for 20 min at room temperature.

Confocal analyses and time-lapse imaging were performed using an Olympus FV1000 confocal system. Nomarski images were collected using a Zeiss Imager Z1 compound microscope. Individual cells were lineage-labeled using photoconversion of the Kaede fluorophore as previously described (Ando et al., 2002; McGraw et al., 2011). Cell counts were determined using DAPI (30 μM; Life Technologies)-labeled nuclei. For time-lapse imaging, embryos were analyzed between 36 and 48 hpf using established techniques (McGraw et al., 2011). Images were processed using ImageJ software (Abramoff et al., 2004). Brightness and contrast were adjusted using the Adobe Photoshop software package.

All antisense oligonucleotide morpholinos (Genetools) were co-injected with the p53-MO (Robu et al., 2007) at twice the concentration of the experimental MOs. The p53-MO was injected at 10 μg/μl for control experiments. A translation-blocking MO against kremen1 (a gift from the Weinstein laboratory, 5′-AAGCTGCGACTCTCCACGAATCCAT-3′) (Gore et al., 2011) was injected at 0.5 μg/μl. Splice donor site-blocking MOs were designed against dkk1b exon 1/intron1 (dkk1b-MO: GCATATTTCTATGCTTACCTGCGGT) and dkk2 exon2/intron2 (dkk2-MO: AATTGAACAAGCGTACAGTTGCTGC), and injected at 7 μg/μl. RT-PCR was carried out using dkk1b-forward ATGATGCACA-TCGCCATGCTC/dkk1b-reverse GCACACATGCCAGAGACACTAA and the dkk2 primers described above. Fgf was inhibited using 75 μM SU5402 (Calbiochem) in embryo medium, and control embryos were treated with DMSO.

Full-length dkk1b was amplified by PCR from an EST clone (accession#: DV598280) using primers that contained attB1 and attB2 sites. Following amplification, PCR products were recombined into a pDONR221 vector, and the hsp70:dkk1b-mTangerine plasmid was assembled using components of the Tol2kit (Kwan et al., 2007). Ten pg/nl of the plasmid DNA was injected into one-cell stage embryos. At 25 hpf, embryos were heat-shocked at 39°C for 40 min using a thermocycler (Applied Biosystems).

Transplantation experiments were carried out as previously described (Nechiporuk and Raible, 2008). In brief, rhodamine dextran (Life Technologies)-labeled donor cells from blastula stage embryos were transplanted in the left side of gastrula-stage host embryos. Host embryos were imaged at 24 hpf to determine the position of donor cells in the pLLP and at 48 hpf to assess pLL formation.

All data are presented as average±s.e.m. Fluorescence intensity was determined using the ImageJ analysis tool, and values (in arbitrary units, A.U.) were generated by selecting a region of interest (ROI) in the pLLP, collecting mean fluorescence intensity and then subtracting nearby background. For analysis of Dkk1b-mTangerine expression, the ROI was set at two cell diameters (based on DAPI and membrane marker expression) surrounding the Dkk1b-mTangerine-expressing cell, excluding the expressing cell itself. Separate expression levels were collected for the Dkk1b-mTangerine-expressing cell. Calculation of statistical significance was carried out using VassarStats (http://vassarstats.net/index.html).

We thank the Weinstein laboratory at The National Institute of Child Health and Human Development for reagents and David Kimelman and Catherine Drerup for comments on the manuscript.