ABSTRACT
In the avian hindbrain, the loss of premigratory neural crest cells from rhombomeres 3 and 5 (r3, r5) through programmed cell death contributes to the patterning of emigrant crest cells into three discrete streams. Programmed cell death is induced by the upregulation of Bmp4 and Msx2 in r3 and r5. We show that cSFRP2, a WNT antagonist, is expressed in the even-numbered rhombomeres and that over-expression of cSfrp2 inhibits Bmp4 expression in r3 and r5, preventing programmed cell death. By contrast, depleting cSFRP2 function in r4 results in elevated levels of Msx2 expression and ectopic programmed cell death, as does overexpression of Wnt1. We propose that programmed cell death in the rhombencephalic neural crest is modulated by pre-patterned cSfrp2 expression and a WNT-BMP signalling loop.
INTRODUCTION
Genetic studies in C. elegans, Drosophila and mouse have identified genes, such as Ced3 (Yuan and Horvitz, 1990) and Caspases (Tewari et al., 1995), that are required for the conserved intracellular pathway leading to programmed cell death (PCD) (Jacobson et al., 1997). The mechanisms responsible for activating this death programme during development are tissue-specific, and although poorly understood, parts of the signalling mechanism have been identified for some tissues. Thus, Graham et al. (1994), Yokouchi et al. (1996), Zou and Niswander (1996), Ganan et al. (1996) and Merino et al. (1999) have demonstrated that BMP4 is sufficient to induce PCD in the rhombencephalon and interdigital mesenchyme. BMP4 induces PCD via a Msx2-mediated pathway during rhombencephalic and interdigital development, and in P19 cells (Graham et al., 1993; Zou and Niswander, 1996; Ganan et al., 1998; Marazzi et al., 1997). Yet, some studies on interdigital PCD have shown that Msx2 regulates Bmp4 expression (Ferrari et al., 1998). However, all data support the notion that BMP4 protein and Msx2 overexpression lead to an increase in PCD, both in the rhombencephalon and interdigital mesenchyme (Takahashi et al., 1998; Ferrari et al., 1998; Buckland et al., 1998; Yokouchi et al., 1996; Marazzi et al., 1997; Merino et al., 1999), and their attenuation leads to neural tube defects, and abnormal digit patterning or webbing (Foerst-Potts and Sadler, 1997; Winograd et al., 1997; Katagiri et al., 1998; Ganan et al., 1998). The developing hindbrain, which is metamerically subdivided into eight rhombomeres (r; Lumsden, 1990), produces premigratory neural crest (NC) cells along its dorsal margin, the majority of which migrate ventrally to the branchial arches. However, studies on the chick embryo have shown that the odd-numbered rhombomeres, r3 and r5, do not significantly contribute NC cells (Lumsden et al., 1991). Rather, r3 and r5 NC cells undergo PCD in both birds (Lumsden et al., 1991, Jeffs et al., 1992; Graham et al., 1993; Graham et al., 1994; Maden et al., 1997; Takahashi et al., 1998) and mouse (Serbedzija et al., 1996), sculpting discrete streams of r2, r4 and r6 NC cells from the otherwise continuous outflow from the dorsal neural tube. Those r3 and r5 NC cells that escape PCD may exit the neural tube either anteriorly or posteriorly at the adjacent rhombomere boundaries to join the NC streams from even-numbered rhombomeres (Sechrist et al., 1993; Birgbauer et al., 1995; Kulesa and Fraser, 1998). Those that contribute to the r4 NC stream may be incorporated in second branchial arch derivatives (Köntges and Lumsden, 1996). Those that migrate to join the r2 and r6 streams have been shown, by short-term labelling, to migrate ‘towards’ but not into branchial arch 1 nor 3 (Sechrist et al., 1993; Birgbauer et al., 1995; Nieto et al., 1995; Kulesa and Fraser, 2000). However, a study by Couly et al. (1996) based on grafting quail neural folds into chick hosts has shown that r3/r5 NC cells do make a small contribution to first and third branchial arches and later contribute to the intracartilaginous interfaces between r2 and r4 NC-derived cells (r3 derivatives) and between r4 and r6 NC-derived cells (r5 derivatives). However, another long-term r3 and r5 NC fate map has shown that r3 and r5 NC cells contribute only to the neuronal derivatives of r2 and r6 NC streams and not to first or third branchial arch structures (Köntges and Lumsden, 1996). The discrepancies between Couly et al. (1996) and Köntges and Lumsden (1996) could be due to the stage of rhombomere transplantation, as cells within neighbouring rhombomeres have been shown to mix prior to rhombomere boundary formation (Fraser et al., 1990; Lumsden, 1990, 1991). Thus transplantation prior to rhombomere boundary formation, i.e. at 5 somites, leads to donor cells mixing with neighbouring host cells. Normally, the streamed NC cell emigration pattern, created at least in part by PCD, may play a role in keeping the distinct NC cell subpopulations from mixing with each other (Graham et al., 1996) as they enter and populate their respective first, second and third branchial arch territories.
The expression of Msx2 and Bmp4 has been shown to be tightly correlated with PCD in r3 and r5, both spatially and temporally (reviewed by Graham et al., 1996; Takahashi et al., 1998). When these odd-numbered rhombomeres are cultured in isolation, expression of Bmp4 is downregulated and the PCD program is abrogated, thereby allowing the production of viable NC (Graham et al., 1993, 1994). When BMP4 protein is added to odd-numbered rhombomeres cultured in isolation, Msx2 expression is upregulated and the PCD program restored (Graham et al., 1994; reviewed by Graham et al., 1996; Marazzi et al., 1997; Farlie et al., 1999). Yet, when r3 is co-cultured with an even-numbered rhombomere, Bmp4 is expressed in r3 and NC cells undergo PCD (Graham et al., 1994), suggesting that Bmp4 is regulated by an interaction with the even-numbered rhombomeres (reviewed by Graham et al., 1996). Yet, r3 is not freed from its PCD fate when it is isolated prior to rhombomere boundary formation, and even-numbered rhombomeres cannot induce PCD in odd-numbered rhombomeres, in vitro, if the rhombomeres are isolated independently and re-conjoined in vitro (Farlie et al., 1999). Thus, BMP4 (through Msx2) seems to play a significant role in initiating the PCD program, following an inductive signal from the even-numbered rhombomeres.
When even-numbered rhombomeres are treated with recombinant BMP4 protein, the PCD program is not initiated (Graham et al., 1994). By contrast, Msx2 gain-of-function studies report that the NC of even-numbered rhombomeres can undergo PCD (Takahashi et al., 1998). Therefore, BMP4 is not capable of functioning alone and requires a co-factor to induce the Msx2-mediated PCD program within even-numbered rhombomeres.
Factors that could play a role in modulating PCD in the odd-numbered rhombomeres are members of the WNT family, as WNT signalling has been shown to inhibit Bmp4 expression (Baker et al., 1999). The Wnt gene family consists of at least 16 members of secreted glycoproteins. They are thought to act through membrane bound frizzled receptors and have been shown to control cell fate and proliferation during embryogenesis (Parr and McMahon 1994; Hays et al., 1997; Saint-Jeannet et al., 1997; Yoshikawa et al., 1997; Moon et al., 1997; Dorsky et al., 1998). Recently, a group of putative WNT antagonists known as Secreted Frizzled Related Proteins (SFRPs) have been identified (Shirozu et al., 1996; Rattner et al., 1997; Melkonyan et al., 1997; Finch et al., 1997; Leynes et al., 1997; Salic et al., 1997; Xu et al., 1998; Chang et al., 1999). These proteins contain a cysteine rich domain, homologous to that found in the WNT frizzled receptors, and have been shown to bind WNTs (Lescher et al., 1998; Wang et al., 1997, 1997; Ladher et al., 2000). To date, six SFRP family members have been identified in human, mouse and chick. Several Wnt genes, including Wnt1 and Wnt3a are expressed in the hindbrain (Hollyday et al., 1995), and Sfrps have been reported to be expressed in alternating stripes within the mouse rhombencephalon (Leimeister et al., 1998; Hoang et al., 1998). Furthermore, some of SFRP family members (SARP1, SARP2, and SFRP2) have been shown to indirectly modulate cellular responses to pro-apoptotic stimuli (Melkonyan et al., 1997; Wolf et al., 1997; Zhou et al., 1998 ). These data raise the possibility that endogenous WNT antagonists may modulate PCD in the rhombencephalon. To investigate this, we have analysed the function of the chick SFRP2 homologue, cSFRP2. We show the temporal and spatial origins of the rhombencephalic PCD induction signal. In addition, we show that ectopic cSfrp2 expression prevents PCD in the odd-numbered rhombomeres and its absence from the odd-numbered rhombomeres allows PCD to be induced. We therefore propose that rhombencephalic PCD is mediated by WNT signalling.
MATERIALS AND METHODS
Dorsal rhombomere ablations
Hi-Sex Brown eggs (J. K. Needle and Co, Waltham, Herts, UK) were incubated at 38±1°C with 50-60% (v/v) relative humidity to HH stages 8-10 (Hamburger and Hamilton, 1951), as PCD is first detectable with Acridine Orange at stage 9+ (unpublished observation). Using a flame-sharpened 100 μm tungsten wire, transverse cuts (3/4 of the rhombomere depth) were made along the rhombomere encompassing the even-numbered rhombomere to be excised. Embryos were then re-incubated to HH 11-13, whilst PCD is still detectable. The embryos were harvested and stained with Acridine Orange as previously described by Graham et al. (1993).
Nuclepore filter insertion
Hi-Sex Brown eggs were incubated at 38±1°C with 50-60% (v/v) relative humidity to HH stages 8-10. Using a flame-sharpened 100 μm tungsten wire, one transverse cut was made along a rhombomere boundary. A 0.015 μm pore size, or a non-porous Nuclepore filter (Costar), was placed into the cut and embryos were then re-incubated until HH 11-13. The embryos were then harvested and stained with Acridine Orange as previously described by Graham et al. (1993).
Construction of the cSfrp2 retrovirus
The cSfrp2 coding region was amplified from the chick cSfrp2 cDNA (Ladher et al., 2000) by PCR, using oligonucleotide primers to the 5′ ORF sequence CCGCCATGGCGCGCCGCCTC and 3′ ORF sequence CCAAAGCTTCTAACACTGCAGT. The PCR program consisted of 35 cycles of three steps: 95°C for 1 minute, 60°C for 1 minute, and 72°C for 1 minute. The NcoI/HindIII cSfrp2 fragment was then subcloned into the adaptor plasmid pSlax12Nco RCAS subtype B containing cSfrp2 was then created using ClaI/pSLAX12Nco and its orientation was determined by restriction digest analysis and nucleotide sequencing.
Explanted rhombomeres
Hi-Sex Brown eggs were incubated at 38±1°C with 50-60% (v/v) relative humidity to HH 10. The odd-numbered rhombomeres were injected by mouth with DiI and the rhombomeres were isolated as described by Graham et al. (1993). The isolated rhombomeres were then cultured with either cSFRP2- or green fluorescent protein (GFP)-conditioned medium from chick embryo fibroblasts (CEF) infected with either RCAS(B)-EGFP (kindly provided by J. Gilthorpe) or RCAS(B)-cSfrp2. GFP was used as a control to also determine whether supernatant containing the retrovirus could infect the explanted rhombomeres. The CEFs were CaCl2-transfected and grown as described by Morgan and Fekete (1996).
HNK-1 immunochemistry
The explants were fixed and processed as described by Graham et al. (1993), with the following modifications. The entire procedure was performed without detergent as HNK-1 is a cell surface molecule. The HNK-1 antibody (Zymed 07-5703) was added at a dilution of 1:200 and incubated overnight at 4°C. The secondary antibody was a goat anti-mouse FITC (Jackson IRL, 115-096-044) diluted at 1:200 and was incubated for 1 hour at room temperature. The explants were viewed using confocal microscopy.
Whole-mount in situ hybridization
A minimum of five embryos per stage were assayed for all whole-mount in situ hybridization experiments. The in situ protocols used were as described by Grove et al. (1998) with the following modifications. The synthesis of the RNA was extended from 2 hours to 4 hours, and the hybridization step was carried out for 48 hours instead of 18 hours. Explants were processed as whole mounts, however, they were first covered with an 8% gelatin solution, and hybridised for 12 hours only.
cDNA for making riboprobes
An 800 bp mouse Krox20 partial cDNA, containing the zinc finger and 3′UTR, was used as a template (Wilkinson 1992) for antisense transcription. A 700 bp EcoRI-HindIII chick Hoxb2 partial 3′cDNA fragment was used as a template for antisense transcription (kindly provided by A. Kuroiwa, Japan). A 600 bp mouse Fgf8 partial cDNA fragment was used as a template for antisense transcription (Mahmood et al., 1995). A 5′ 350 bp chick Msx2 partial cDNA fragment was used as a template for antisense transcription (Graham et al., 1993). A 1 kb chick Bmp4 partial cDNA fragment was used as a template for antisense transcription (Francis et al., 1994). A 1.68 kb chick cSfrp2 cDNA fragment was used as a template for antisense transcription. A 500 bp chick TCF4 partial cDNA encoding a fragment between the conserved HMG box and β-catenin binding domain was used as a template for antisense transcription (kindly provided by L. Paganess and C. Tabin, Harvard University, USA). The Wnt1 cDNA was used for antisense transcription as described by Hollyday et al. (1995).
The cSfrp2 whole-mount in situ hybridized embryos were vibratome sectioned at 60 μm according to standard procedures.
Electroporation
Electroporation of retroviral constructs was used instead of viral infection methods because we required expression of the constructs by HH 9-10, and traditional infection methods would have yielded low infection and high mortality rates. The RCAS(B)-EGFP construct (kindly provided by J. Gilthorpe) was either electroporated alone or co-electroporated with either RCAS(B)-Wnt3a (kindly provided by C. Tabin; Kengaku et al., 1998) or RCAS(A)-Wnt1 (kindly provided by A. Brown (Rudnicki and Brown, 1997) or RCAS(B)-antisense cSfrp2. A dual pulse isolated stimulator (Intercept TSS10) and silver electrodes were used to electroporate (20 volts) the constructs into HH 6-7 chick hindbrains in ovo. Each electroporation contained a solution of 0.5 μg/ μl of each DNA construct and 0.06% Fast Green dye dissolved in 1×PBS.
RESULTS
Cell death is induced in an anterior-to-posterior wave
In vitro studies have shown that BMP4/Msx2 are implicated in the PCD of odd-numbered rhombomere NC cells that is induced by the even-numbered rhombomeres (Graham et al., 1993, 1994). To further examine this phenomenon, we performed in vivo ablations of the dorsal part of even-numbered rhombomeres to determine the precise timing of PCD induction from the even-numbered rhombomeres.
Even-numbered rhombomeres were ablated at HH 8-10 (Fig. 1A,B) and the embryos were allowed to develop until HH 11-13. They were analysed for cell death by Acridine Orange (AO) staining and simultaneously for the expression of gene markers by whole-mount in situ hybridization. Following dorsal ablations, the expression of either Hoxb2/Krox20 or Fgf8/ Krox20 was determined to ensure that the correct even-numbered rhombomere had been removed. In all cases, Hoxb2/Krox20 and Fgf8/Krox20 expression was as anticipated (compare Fig. 1E,F), except that Krox20 is down-regulated in r3 after r2 ablation at HH 8. Hoxb2 expression, however, was normal (compare Fig. 1C,D), indicating that the down regulation of Krox20 expression in r3 was not due to its removal. A gap between the Krox20 expression in r3 and r5 is seen in the basal plate of r4 that was not removed.
An analysis of TUNEL and AO staining has revealed that they both identify the same spatial pattern of PCD in the avian hindbrain (data not shown). However, AO stains slightly younger staged embryos (Fig. 2A). This is not surprising because AO is specific for condensing chromatin, a characteristic that appears earlier than the 3′ DNA nicks detected by TUNEL. In addition, it appears that TUNEL is not specific for cells undergoing PCD, as cells undergoing necrosis also stained positive for TUNEL (Grasl-Kraupp et al., 1995). The spatial pattern of PCD seen with AO in the avian hindbrain starts in r3 at the 8-somite stage, and at the 10-somite stage AO+ cells are found in r2 and at the boundary between r1/r2 (Fig. 2A). This anterior pattern of cell death then persists until HH stage 13 and the posterior pattern of cell death, specifically in r5, begins at 11 somites according to AO staining (Fig. 2A). By HH stage 12, the anterior and posterior patterns are both present and additional AO+ cells can be seen in the r4 and r6 NC migratory routes. AO+ cells are not found after HH stage 13 in the rhombencephalon. In vitro analysis and homotopic grafting has revealed that AO+ cells in adjoining even-numbered rhombomeres originate from the odd-numbered rhombomeres (unpublished observation; Graham et al., 1993).
Ablation of r2 at HH 8 resulted in a 50% reduction in r3 PCD, while it had no effect on r5 (Table 1). Removal of the other even-numbered rhombomeres (r4 or r6) did not effect PCD in either r3 or r5.
By contrast, ablation of any even-numbered rhombomere at HH 9 had no effect on the levels of PCD in r3. However, removal of r2 at the 6-somite stage (HH 9–) or r4 at the 8-somite stage (HH 9+) resulted in a 70% reduction in the level of PCD in r5 (Table 1). Ablation of r6 had no significant effect on either r3 or r5 PCD.
The most robust effect on the levels of PCD in either r3 or r5 was seen after ablating r4 at HH 10. After r4 ablation, r3 PCD was reduced by 78% (Fig. 2C; Table 1), and r5 PCD was reduced by 90% (Fig. 2C; Table 1). At this stage the level of PCD within r3 or r5 was not affected by removal of r2. While loss of r6 had no effect on r3 PCD, PCD in r5 was reduced by 76% (Table 1).
Msx2 and Bmp4 expression was analysed to verify that this reduction of PCD in r3 and r5 is accompanied by a decrease in PCD-mediating molecules (Graham et al., 1993, 1994). In all embryos in which r4 had been ablated at HH 10, Bmp4 (n=17) and Msx2 (n=29) expression was lost in r3 and r5 (Fig. 1G-J), as expected (Graham et al., 1993, 1994).
Therefore, the first detectable induction of PCD in r3 arises at HH 8 from r2 and switches to r4 at HH 10 (Fig. 2E). In the case of r5, the first detectable induction of PCD arises at HH 9 from r2 at the 6-somite stage and from r4 at HH 9+ (Fig. 2E). Subsequently, this signal shifts to r4 and r6 at HH 10 (Fig. 2E). Thus, the PCD inducing signals seem to follow an anteroposterior wave of maturation similar to that of neural crest emigration.
Neural crest cell death is mediated by a diffusible molecule
To further characterise the PCD signal arising from the even-numbered rhombomeres, barriers were inserted at rhombomere boundaries. Either non-porous or porous (0.015 μm pore diameter) Nuclepore filters were used to test the effect of blocking signal transmission and to distinguish between diffusible and cell contact-mediated signals.
Filters were inserted when the even-numbered rhombomeres begin to transmit their PCD inducing signals. Insertion of non-porous filters in the r2/3 boundary (n=11) at HH 8, in r4/5 (n=19) at HH 9-10, and in r3/4 (n=13) at HH 10, resulted in an 80% reduction of PCD within either r3 or r5 (n=31/35; Fig. 2D). This confirmed results from the ablation of even-numbered rhombomeres. By contrast, the insertion of porous filters at equivalent positions and stages did not change the level of PCD in either r3 or r5 (n=27/32; data not shown). Therefore, the even-numbered rhombomeres produce a diffusible signal between HH 8-10 (Fig. 2E), which induces PCD in the NC of r3 and r5. Although the even-numbered rhombomeres produce an apoptotic-inducing signal, they do not undergo PCD themselves. Secreted proteins that are candidates for the role of apoptotic inducers include members of the SARP family (Melkonyan et al., 1997), which includes several secreted frizzled-related proteins (SFRPs) (Leimeister et al., 1998). We therefore analysed the expression of chick Sfrp homologues in the hindbrain.
PCD is controlled by a WNT-BMP loop
cSfrp2 is initially expressed throughout the hindbrain (HH 7-9–; Fig. 3A,B) but is subsequently down-regulated in the odd-numbered rhombomeres at HH 9 (Fig. 3C,D). Expression in the even-numbered rhombomeres at HH 9+-12 is graded, being highest at the ventricular side and lowest at the pial side of the dorsolateral neural tube (Fig. 3D). The timing of this down-regulation correlates closely with the onset of PCD in the odd-numbered rhombomeres, thus suggesting that cSFRP2 may diffuse to the dorsal neural tube.
To investigate whether the down-regulation of cSfrp2 in the odd-numbered rhombomeres is sufficient to induce PCD, we cultured DiI-labelled HH 10 r3 explants in medium conditioned either by cells expressing cSFRP2 from a retrovirus or by control cells expressing GFP from a retrovirus. r3 explants cultured in medium conditioned by control cells produced many DiI-labelled NC cells (53% of total NC cells were labelled with DiI; n=5; Fig. 4E). The addition of medium conditioned by cSFRP2-expressing cells altered neither the total number of migratory NC cells nor DiI-labelled r3 neural crest cells (n=6; Fig. 4E). Thus, exogenous cSFRP2 does not affect the number of r3 neural crest cells produced when r3 is rescued from PCD.
However, a dramatic effect was seen with exogenous cSFRP2 when DiI-labelled r3 explants were cultured in conjunction with r4, which normally induces cell death in r3 (Graham et al., 1994). cSFRP2-conditioned medium prevented PCD of r3 NC when co-cultured together with r4 (40% of the total neural crest was labelled with DiI, n=5; Fig. 4C,E) whereas control conditioned medium did not alter r3 neural crest PCD (4% of the total neural crest was labelled with DiI, n=6; Fig. 4A,E). Thus, cSFRP2 seems to function by inhibiting the effect of the PCD inducer arising from the even-numbered rhombomere.
To verify that cSFRP2 was acting by inhibiting PCD in the explants, we analysed the expression of Bmp4. In the control cultures, Bmp4 expression was found in 83% of explants (n=5/6, Fig. 4B), whereas in the explants cultured in cSFRP2-conditioned medium, Bmp4 expression was extinguished in 63% of explants (n=5/8, Fig. 4D). Thus, cSFRP2 causes a loss of Bmp4 expression, which is known to be sufficient to induce rhombencephalic cell death (Graham et al., 1994).
To ascertain whether cSFRP2 is required to prevent PCD in the even-numbered rhombomeres, we sought to knockdown cSFRP2 by electroporation of retroviral antisense cSfrp2. Each set of constructs was electroporated in conjunction with a GFP-expressing retrovirus, to monitor electroporation efficiency. The electroporated embryos were then assayed for PCD and subsequently for expression of genes known to be involved in rhombencephalic PCD, and controlled using the level of Sfrp2 mRNA. In control HH 6-7 GFP electroporated embryos, the levels of PCD and Msx2 expression were unchanged from that seen in uninjected embryos (n=10/14; Fig. 5A,B), and GFP was found in bilateral patches of the hindbrain (insert, Fig. 5A). Therefore, the electroporation procedure did not lead to any changes in the levels of PCD within the rhombencephalon. HH 6-7 embryos were used for electroporation of antisense cSfrp2. We found that the majority of embryos were normal in appearance and the hindbrains that showed high levels of GFP were assayed for PCD using Acridine Orange, Sfrp2 and Msx2 expression. An increase in PCD was seen in 96% of embryos (P<0.005, n=21/22) and elevated Msx2 expression (n=17/17) within r4 at HH 11 (Fig. 5E,F), and in the lateral neural tube. The level of Sfrp2 mRNA was drastically reduced in embryos that were injected with antisense Sfrp2 (Fig. 5F), thus the antisense Sfrp2 was binding sufficiently to the sense Sfrp2 mRNA and causing its degradation by RNase-H. Intriguingly, we observed a marked increase in the levels of PCD and Msx2 expression in the dorsolateral neural tube, where cSfrp2 is normally expressed (Fig. 5E,F), indicating that the antisense cSfrp2 was also functioning at the exact site of cSfrp2 expression.
cSFRP2 is known to act as an antagonist of WNT8 (Ladher et al., 2000). Therefore, another approach that should result in a loss of cSFRP2 function would be to overexpress a Wnt gene that is normally found in the hindbrain at the relevant stages; candidates include Wnt3a and Wnt1. When Wnt3a was electroporated into the hindbrain at HH 6-7 (n=43), embryos developed with abnormally flattened heads in 54% of cases (n=23/43). Of the remaining embryos (n=20), 4 developed with an open neural tube. In the embryos with high levels of co-electroporated GFP expression, levels of PCD were analysed, and 75% (P<0.005) of these had elevated levels of PCD (n=15/20) and Msx2 expression (n=8/10) in r4 at HH 9-13. These effects were much more marked when Wnt1 was electroporated, where all of the 9/11 embryos with strong GFP and Wnt1 expression failed to close their neural tubes, and both Msx2 expression and PCD were strongly upregulated in the open neural folds at all levels of the hindbrain, including r4 (Fig. 5C,D). Consistent with this, cSfrp2 mRNA injected into Xenopus embryos antagonises the dorsalizing effects of WNT1 but not those of WNT3a (L. Dale and D. E., unpublished data). Thus, both methods of interfering with cSFRP2 function, antisense knockdown and increasing the levels of ligand, resulted in increased PCD and elevated Msx2 expression. In addition, cSFRP2 did not alter the total number of rhombomeric NC cells produced (see Fig. 4E).
Following reports that WNTs can signal either through a β-catenin-dependent or -independent pathway (Boutros et al., 1998; Li et al., 1999), and that the downstream WNT effector Tcf4 is expressed in r3 and r5 in mouse embryos (Cho and Dressler, 1998), we set out to examine the possible role of TCF4 in NC PCD in chick. We first analysed the expression of Tcf4 in normal embryos and then after altering the pattern of PCD. We found that, unlike in the mouse, Tcf4 is expressed all along the cranial neural folds at HH 9 (6-somite stage), and throughout the HH 10+ hindbrain (data not shown). Following gain or reduction of cSFRP2 function, Tcf4 expression was Tcf4 is downregulated in the even-numbered rhombomeres (N=9/17), suggesting that Tcf4 is transcriptionally repressed by WNT3a (data not shown). Thus, our preliminary results suggest that SFRP2 acts via a β-catenin-independent pathway.
DISCUSSION
The neural crest primordia of rhombomeres 3 and 5 undergo extensive PCD. The induction of PCD depends on the presence of even-numbered rhombomeres (Graham et al., 1993), and is influenced by BMP4 and Msx2 (Graham et al., 1993, 1994). However, the signalling cascade leading to PCD in the hindbrain is largely unknown. This study sheds further light on the induction of PCD, in particular the role of WNT signalling via the antagonistic effects of cSFRP2.
We have found that the even-numbered rhombomeres produce a secreted molecule that diffuses into the odd-numbered rhombomeres to induce PCD. We have identified r2 as the source of the early PCD-inducing signal for r3 at HH 8. Likewise, r4 is required to induce PCD in r5 at HH 9. In both cases, signalling occurs prior to the detection of PCD within the particular rhombomere, indicating that this early inducing signal is not the PCD execution signal. Rather, it is probably the initial signal necessary for the later execution signal at HH 10. The production of these early and late PCD signals by the even-numbered rhombomeres follows an anterior-posterior wave similar to that of subsequent neural crest emigration.
The secreted PCD-inducing molecule, while produced by the even-numbered rhombomeres, can apparently induce PCD only in rhombomeres 3 and 5. The even-numbered rhombomeres must therefore either contain an inhibitor against the induction of PCD, or they lack a co-factor required for the PCD inducer function. We propose that PCD in the even-numbered rhombomeres is actively inhibited, and that cSFRP2 is the molecule responsible for the inhibition. cSFRP2 has appropriate characteristics as a candidate effector of inhibition: it is expressed specifically in the even-numbered rhombomeres at the critical stages, and gain or reduction of function experiments indicate that cSFRP2 interferes with the PCD program. Gain of cSFRP2 function leads to loss of Bmp4 expression and arrest of PCD in r3 and r5, whereas interference with cSFRP2 function leads to increase of Msx2 expression in the even-numbered rhombomeres and the induction of PCD.
It is likely that cSFRP2 functions in the control of PCD via the suppression of WNT signalling (Shirozu et al., 1996; Rattner et al., 1997; Melkonyan et al., 1997; Finch et al., 1997; Leynes et al., 1997; Salic et al., 1997; Xu et al., 1998; Chang et al., 1999; Ladher et al., 2000; Lee et al., 2000). Involvement of SFRPs in PCD has been shown for mammary gland, ovarian corpus luteum and prostate (Wolf et al., 1997), breast adenocarcinoma (Melkonyan et al., 1997), and immortalized breast epithelial cells (Zhou et al., 1998).
Several Wnt genes are expressed in the developing hindbrain and are candidates for mediating PCD: Wnt1 and Wnt3a are expressed along the dorsal midline of the hindbrain (Hollyday et al., 1995), whereas Wnt8c is restricted to the dorsal midline of r4 (Hume and Dodd 1993). Together, Wnt1 and Wnt3a have been shown by overexpression or null mutation to function during NC induction; their involvement with NC PCD is therefore difficult to assess (Augustine et al., 1993; Ikeya et al., 1997; Saint-Jeannet et al., 1997; LaBonne and Bronner-Fraser, 1998). An analysis of PCD in the Wnt1 and Wnt1;Wnt3a null mutants reported no change in hindbrain PCD (Serbedzija et al., 1996; Ikeya et al., 1997). However, contrary to the previous report of PCD in mouse r3 and r5 NC (Serbedzija et al., 1996), a recent examination has failed to detect this PCD (Trainor and Krumlauf, personal communication).
To determine whether or not Wnt3a and/or Wnt1 function to modulate PCD in the chick rhombencephalon, we have analysed the effect of their overexpression. The embryos showed upregulation of PCD in the even-numbered rhombomeres, similar to the antisense reduction of cSFRP2 function. During normal hindbrain development, Wnt1 and Wnt3a expression starts at HH 7, a stage when cSfrp2 expression is ubiquitous. PCD would thus be prevented by cSFRP2 until, at HH 9, cSfrp2 expression is downregulated in r3 and r5. At this stage and rhombomere level, WNT signalling is no longer antagonised and PCD is initiated.
We have found that WNT1 is more effective than WNT3a at increasing levels of Msx2 expression and PCD in the neural folds. This is consistent with the findings of Lee et al. (2000), who showed that mSFRP2-expressing cells antagonise the dermomyotome-inducing activity of WNT1 and WNT4, but not that of WNT3a. Furthermore, cSFRP2 antagonises the dorsalizing activity of WNT1 but not that of WNT3a in Xenopus assays (L. Dale and D. E., unpublished data).
Previous studies on the function of Wnt genes have not provided evidence for the involvement of a specific Wnt in hindbrain PCD, whereas their involvement has been shown in determining NC cell fate (Dorsky et al., 1998), expansion of NC populations (Saint-Jeannet et al., 1997; Ikeya et al., 1997), and neural tube closure (Augustine et al., 1993). These various effects on development could be attributed to the specific dose dependence effects of the WNT proteins (Dorsky et al., 1998; Wilder and Perrimon, 1995). Here, we have shown that both WNT3a and WNT1 could be involved in regulating PCD in r3 and r5. This is consistent with the phenotypes of Wnt mutant mice, which exhibit similar developmental defects to those seen when PCD is arrested, as in Apaf1 mutant mice or caspase inhibition, which develop with incomplete fusion of the neural folds (Augustine et al., 1993; Yoshida et al., 1998; Weil et al., 1997). The effects may often be concealed, since Wnt genes act not only during neurulation and neural crest induction, but also during gastrulation; thus, Wnt mutant mice exhibit severe defects which are associated with this earlier event. Wnt1 mutants show a loss of midbrain and cerebellar structures only (McMahon and Bradley, 1990).
WNT signalling activates dishevelled (DSH), which can then act along two distinct pathways: one of these involves antagonising ZestWhite-3 kinase/GSK3β to stabilise β-catenin and enable its nuclear translocation (reviewed by Shulman et al., 1998; Wodarz and Nusse 1998); the other pathway, independent of β-catenin, involves the kinases JNK/SAPK (Boutros et al., 1998; Li et al., 1999). JNK signalling is involved in inducing cell death in many cell types (reviewed by Basu and Kolesnick 1998). Furthermore, Jnk is expressed in the hindbrain of mice, and double Jnk1;Jnk2 mutant mice have reduced PCD in the hindbrain and their neural folds fail to close (Kuan et al., 1999).
Further evidence for β-catenin-independent WNT signalling in hindbrain NC PCD comes from analysing other components of the β-catenin pathway, such as TCF4, which is required to bind β-catenin for nuclear translocation and transcription of target genes (Brannon et al., 1997; Brunner et al., 1997; reviewed by Eastman and Grosschedl 1999; Korinek et al., 1998). Our results indicate that Tcf4 is expressed throughout the hindbrain at equivalent stages in chick, suggesting that during PCD, WNTs may signal through a β-catenin-independent pathway.
Our study indicates an important role for WNT signalling in the induction of PCD in the developing rhombencephalon. We have shown that Bmp4 and Msx2, both closely associated with PCD, are downstream of WNT signalling. However, WNT signalling alone appears not to be sufficient to cause PCD, since odd-numbered, Wnt-expressing rhombomeres cultured in isolation do not undergo PCD (Graham et al., 1993). Culturing r3 in conjunction with r4 in cSFRP2-conditioned medium, culturing r3 in isolation (Graham et al., 1994), or removing the PCD inducer originating from the even-numbered rhombomeres, all result in the down-regulation of Bmp4 expression and arrest of r3 PCD. This indicates that the induction of Bmp4 expression (and death) is regulated by the combination of WNT signalling and a PCD-inducing signal diffusing from the even-numbered rhombomeres.
Once Bmp4 expression is initiated, it acts on the odd-numbered rhombomeres to induce Msx2 and PCD (Graham et al., 1994). In contrast, when BMP4 protein is added to even-numbered rhombomeres, they do not undergo PCD (Graham et al., 1994). Consequently, BMP4 alone is unable to cause PCD in the presence of cSFRP2. Furthermore, Msx2 gain-of-function studies have shown that Msx2 is capable of inducing PCD within the even-numbered rhombomeres (Takahashi et al., 1998). This suggests that another, WNT-dependent factor is required to work in conjunction with BMP4 to initiate the Msx2 and PCD.
We propose that the apoptotic program in the rhombencephalon is initiated by WNT signalling (Fig. 6). The ubiquitous expression of the WNT antagonist cSfrp2 in the early hindbrain initially prevents PCD. Later in rhombencephalic development, cSfrp2 expression becomes downregulated in the odd-numbered rhombomeres. In precise correlation with cSfrp2 down-regulation, PCD becomes detectable in the odd-numbered rhombomeres (Lumsden et al., 1991). Presumably, the lack of cSFRP2 in the odd-numbered rhombomeres allows WNT to act in combination with the diffusible signal from the even-numbered rhombomeres to induce Bmp4 expression. BMP4, in turn, acts with a WNT-dependent molecule (possibly TCF4 or JNK) to activate Msx2 expression, and thus an autocrine loop induces PCD in the odd-numbered rhombomeres.
Acknowledgments
We are grateful to Frank Schubert for critical reading of this manuscript. We wish to thank Tony Brown for his generous gift of the retroviral Wnt1 construct, Cliff Tabin for Wnt3a virus and for the Tcf4 cDNA, David Wilkinson for the Krox20 cDNA, Atsushi Kuroiwa for the Hoxb2 cDNA, Ivor Mason for the Fgf8 cDNA, and Anthony Graham for the Msx2 cDNA. We also thank Jon Gilthorpe for his retroviral EGFP construct, and for his technical help with the electroporator, also Iano Campbell and Amata Hornbruch for their excellent technical help. This work was supported by the European Union, the Wellcome Trust, and the Medical Research Council.