Expression of a novel FGF in the Xenopus embryo. A new candidate inducing factor for mesoderm formation and anteroposterior specification

In 1987 two groups found that basic fibroblast growth factor (bFGF/FGF-2) showed mesoderm-inducing activity when applied to isolated animal caps from Xenopus blastulae (Slack et al., 1987; Kimelman and Kirschener, 1987). The characteristics of the induction resemble the ventral-type inductions provoked when the laterovegetal and ventrovegetal blastomeres are combined and cultured with animal blastomeres (Dale and Slack, 1987), and it has become generally accepted that this interaction is probably mediated by a FGF-like molecule in vivo (Smith, 1989). There are, however, reasons for doubting that bFGF itself is the factor responsible. We have recently shown that neutralizing antibodies to Xenopus bFGF fail to inhibit inductions in transfilter experiments, in which a tissue explant, known to be a source of endogenous mesoderminducing activity, is separated from the competent tissue by membranes enclosing a fluid-filled space (Slack, 1991). A further problem arises because bFGF does not have a signal sequence of the type normally required for secretion from cells. We have shown that bFGF overexpressed from mRNA injected into early embryos shows rather little mesoderm-inducing activity unless a secretory signal sequence is added to it (Thompson and Slack, unpublished observations). This is in line with several transfection experiments on mammalian cells that have shown no secretion of the molecule without the addition of a heterologous signal sequence (reviewed Rifkin and Moscatelli, 1989), although there is one recent report that bFGF can be secreted from fibrosarcoma cell lines by an as yet uncharacterized pathway (Kandel et al., 1991).

Recently Amaya et al. (1991) succeeded in inhibiting the function of FGF receptors in the early Xenopus embryo by overexpression of an injected mRNA encoding a mutant form of a Xenopus receptor lacking the cytoplasmic domain. The mutant receptor produces non-functional dimers with at least some types of the endogenous FGF receptors. The resulting embryos have reduced quantities of mesodermal derivatives and defects in the posterior, suggesting an interference with one or more of the processes of mesoderm induction, dorsalization and anteroposterior specification. This is the best direct evidence that a FGF-like molecule is an inducing factor in the early amphibian embryo. But if bFGF itself is not the ligand then what might be?

bFGF is a member of a multigene family with at least 7 members (Robinson, 1991) among which there is considerable cross reaction with the known FGF receptors and hence overlap of their biological activity (Johnson et al., 1990; Bottaro et al., 1990; Keegan et al., 1991). Five members of this family, namely int-2 (FGF-3), kFGF (FGF-4), FGF-5, FGF-6 and KGF (FGF-7), have recognized secretory signal sequences and are therefore better candidates for secreted inducing factors than bFGF (FGF-2) or aFGF (FGF-1). We have previously investigated the biological activities of int-2 and kFGF using in vitro translation products of synthetic mRNAs. kFGF showed about the same specific activity as bFGF in a mesoderm-induction assay, while int-2 had considerably lower activity (Patemo et al., 1989). In this present study, we have cloned a novel FGF from Xenopus embryo cDNA. This molecule, which we have provisionally called embryonic FGF (eFGF), has a secretory signal sequence and the protein has mesoderm-inducing activity. It is expressed maternally with zygotic transcription reaching maximal levels during the gastrula stage. In the neurula, expression is concentrated at the posterior end of the axis. We believe that eFGF is an important new candidate for an inducing agent that could function as an endogenous mesoderm-inducing factor (MIF) and may also have a role in the formation of the anteroposterior axis.

XeFGF was cloned from Xenopus embryo cDNA using the polymerase chain reaction. Total RNA was prepared from mixed stage Xenopus embryos using a LiCl precipitation method (Mohun et al., 1984). First strand cDNA was synthesized using random hexamer primers and was used as a substrate for amplification by the polymerase chain reaction.

Several pairs of degenerate deoxy-oligonucleotide primers were designed from the two mammalian kFGF sequences available (Yoshida et al., 1987; Brooks et al., 1989) and a product of the predicted size was obtained with this primer pair.

XK(10D): 5TA(T/C)TG(T/C)AA(T/C)GT(I/C)GG(I/C)AT(I/C)GG1 3′

XK(14D): 5′(T/C)TC(A/G)TA(I/C)GC(A/G)TT(A/G)TA(A/G)TT(A/G)TT3′

Taq polymerase was used with an amplification regime of 35 cycles of 30 seconds at 55°C, 30 seconds at 72°C and 30 seconds at 94°C. The product was cloned into the EcoRV site of the Bluescript n KS+ vector (Stratagene). The initial PCR clone, XeFGF(0.25), was 252 bp long and dideoxy sequencing of both strands, using Sequenase (USB), revealed a predicted 70% identity with human kFGF at the amino acid level.

The two ends of the cDNA were obtained by a modified RACE protocol (Frohman, 1990). Primers XK(1), XK(2), XK(3) and XK(6) were designed from the sequence of XeFGF(0.25). For the 3′ end first strand cDNA was synthesized using a poly (T) primer which incorporated a Not I site at its 5′ end. The NotI site is underlined:

poly (T) (Not): 5′ TGCATGCGGCCGCT(15) 3′.

This cDNA was used as a substrate for the polymerase chain reaction with the primers poly (T) (Not) and

XK(3): 5′ TTTTACCGGACGGAAGGATAAAT3′.

Products from this reaction were subjected to a subsequent nested reamplification using the primers poly (T) (Not) and

XK(1): 5′ AGTTTACTGGAGCTTTCCCCAGTGGAA 3′.

The products from this reaction were then cloned into the EcoRN/NotI site in Bluescript II KS+. Several clones were sequenced and found to encode the complete carboxy terminus of a putative kFGF homologue. For the 5′ end, cDNA was synthesized off the gene-specific primer

XK(2): 5′ CATGCCATTTATCCTTCCGTC 3′.

A poly (A) tail was then added at the 3′ end of the first strand cDNA using terminal transferase. The tailed cDNA was used in a PCR reaction using the poly (T) (Not) primer and the nested primer

XK(6): 5′ CATGCCATTTATCCTTCCGTC 3′.

Products were cloned into the EcoRV/NotI site of Bluescript IIKS+. Sequencing revealed two types of 5′ clones quite divergent at the nucleotide level. All clones sequenced showed a potential ATG intiation codon with upstream stop codons within 50 bp of this site.

The whole coding region of both genes was then amplified from random primed cDNA using primers XeFGF(5′) and XeFGF(3′), which were designed using sequence data obtained from both the 5′ and 3′ RACE products such that they were perfect matches for both cDNAs. Products of the predicted size were cloned into the EcoRV site of Bluescript Il KS+. Recombinant clones were identified by colony lifts and filter hybridization (Sambrook et al., 1989) with XeFGF(0.25) oligolabelled to a specific activity of 10⁹ cts/min/μg (Pharmacia). The hybridization and washing conditions were as described for the Southern blot analysis. The cDNAs now designated XeFGF(i) and XeFGF(ii) were then sequenced on both strands. Two clones were sequenced for XeFGF(i), which were identical. Three clones were sequenced for XeFGF(ii), two being identical and one showing a T to C change at position 429. This change does not affect the amino acid sequence and may represent a genetic polymorphism. The sequences were analysed and homology searches carried out using the Intelligenetics and University of Wisconsin Genetics Computer Group sequence analysis packages. The nucleotide sequences have been submitted to EMBL and the accession numbers are X62593 and X62594.

Xenopus genomic DNA was prepared from adult female blood (Blin and Stafford, 1976). Southern blot analysis was carried out as described in Sambrook et al. (1989) with the following modifications. 15 μg of genomic DNA was digested with 100 Units of the appropriate restriction endonuclease overnight at 37°C. Digested DNA was separated on a 0.8% agarose gel and after denaturation, separated fragments were transferred to Hybond-N Nylon membrane (Amersham) in a vacuum blotting apparatus (Hoefer). The blot was hybridized with XeFGF(0.25) oligolabelled to a specific activity of 10⁹ cts/min/μg (Pharmacia). The final stringency of the wash was 30 minutes with 0.1× SSPE, 0.1% SDS at 68°C. Washed filters were exposed to XAR-5 X-Omat film (Kodak) for 7 days.

XeFGF(i) was subcloned into the transcription vector pSP64T (Krieg and Melton, 1984). Mutagenesis around the proposed initiating ATG was accomplished by PCR, using primers Xe(64A) and Xe(64B). The former primer provides a sequence around the initiating coding which has been proposed by Kozak (1986) to be optimal for the translation of eukaryotic mRNAs. This sequence is shown underlined.

Xe(64A): 5′ GCCGCCACCATGGCTGTTCCATCGGCCCTGGTG 3′

Xe(64B): 5′ TTATCATATCCGTGGCAAGAAATGGG 3′

In vitro transcriptions were carried out with SP6 RNA polymerase (Kreig and Melton, 1987). RNA was translated in nuclease-treated rabbit reticulocyte lysate with or without canine pancreatic microsomes (Promega). The lysates were analysed by SDS-PAGE and assayed for mesoderm-inducing activity following the methods of Godsave et al. (1988).

The pET T7 expression system allows the production of both fusion and non-fusion protein from cloned DNA (Rosenberg et al., 1987). XeFGF(i) was excised from the Bluescript vector with an internal Ncol site and the BzwiHI site in the vector polylinker. Cutting with Ncol has the effect of removing 44 amino acids from the N terminus of the recombinant protein including the potential hydrophobic signal peptide. This fragment was then cloned into the NcoI/BamIII site in the translation vector pET 8c. This cloning strategy does not produce a fusion protein.

BL21(DE3) pLys S bacteria were transformed with the XeFGF(i)-pET construct, grown in liquid culture and induced to produce protein following the methods of Rosenberg et al. (1987). The supernatant from lysed bacteria was made 0.5 M with respect to NaCl, and loaded onto a 3 ml column of heparin Sepharose (Pharmacia). The column was washed with 0.5 M NaQ, 10 mM Tris-HCl pH 7.0 and then eluted with 2.5 M NaCl, 10 mM Tris-HCl pH 7.0,1 mM EDTA. This fraction was then desalted and further purified on an HPLC heparin affinity column using a gradient of NaCl from 0.5 M to 2.5 M.

Gene-specific ³²P-UTP-labelled antisense RNA probes, XeFGF(i)GS and XeFGF(ii)GS were generated from the more divergent 5′ ends of the coding region of the cDNAs. Probe protections were carried out according to the methods of Kreig and Melton (1984). Hybridization was carried out overnight at 45°C. RNAase digestion was with 40 μg/ml RNAase A and 700 unityrnl RNAase T1 for 1 hour at 37°C. The probes give protected bands of 295 and 307 nucleotides for XeFGF(i) and XeFGF(ii), respectively. Exposures were of 7-14 days using XAR X-Omat film (Kodak). Some RNAase protections were also carried out using XeFGF(0.25) (protected length 200).

The most commonly used internal standard for probe protections is the expression of EF1-α gene (Krieg et al., 1989) but this increases as development proceeds and hence makes comparison between different developmental stages difficult. We have adopted expression of the house-keeping gene ornithine decarboxylase (ODC) as a gene that is uniformly expressed (Bassez et al., 1990). It is also relatively abundant, thus enabling the specific activity of the probe to be reduced to levels appropriate to a particular experiment.

First strand cDNA was prepared by random primed reverse transcription of RNA from mixed stage Xenopus embryos. Using this as a template, a 381 bp fragment of the Xenopus ODC gene, corresponding to nucleotides 1310-1691, was amplified by PCR using the primers

ODC(5′): 5′ CACATGTCAAGCCAGTTC 3′ and

ODC(3′): 5′ GCCTATACATTGATGCTG 3′.

The amplification product was cloned, in both orientations, into HindII cut Bluescript II KS+ to generate two plasmids ODC(1) and (2). Both ODC plasmids can be linearized by cutting with BglII at a position which represents position 1599 in the original cDNA. The antisense probes synthesized give a protected band of 91 nucleotides.

In situ hybridization on Xenopus embryo sections was carried out using ³⁵S-radiolabelled antisense RNA probes. Embryos were fixed overnight in 4% paraformaldehyde at 4°C, washed in PBSA and stored in 70% ethanol at –20°C until required. They were then dehydrated in ethanol, passed through butanol before orienting and embedding in Fibrowax (BDH) at 60°C. Sections were cut at 6-8 μm, collected onto H₂O on 3-aminopropyltriethoxysilane-coated slides and allowed to dry overnight at 37°C. Sections were dewaxed through Histoclear (National Diagnostics) and rehydrated through an ethanol series. Prehybridization steps were as described by Perry-O’Keefe et al. (1990) except that, immediately after the first fixation step following rehydration, the slides were incubated in 2% H₂O₂ in PBSA for 4-5 hours. This step was suggested by David Wilkinson and bleaches pigment from embryos allowing silver grains to be distinguished from pigment under dark-field illumination. Following washing in PBS, protease K was used at 5 μg/ml at 37°C for 30 minutes before a second post-fixation step in 4% paraformaldehyde in PBSA for 20 minutes at room temperature. After treatment with HC1 and acetic anhydride, slides were dehydrated and air dried before adding the hybridization solution containing probe. Hybridization solution was 50% formamide, 10% dextran sulphate, 5× Denhardts, 2×SSPE and 200 μg/ml tRNA. Antisense RNA probes were synthesized from XeFGF(0.25) using 60 μCi of ³⁵S-UTP (1000 Ci/mmol) in the transcription reaction. Probes were hydrolysed to a size of 100-150 bases with alkaline bicarbonate and used at a final concentration of about 40 pg/μl/kb probe complexity. Hybridization was carried out under siliconized coverslips at 55°C for 16 hours. Posthybridization steps were as described by Wilkinson and Green (1990) with high-stringency washes in 50% Formamide, 2× SSPE, 20 mM DTT at 65°C for 30 minutes and treatment with RNAase A at 20 μg/ml at 37°C for 30 minutes. Slides were dehydrated and air dried before dipping in NTB-2 emulsion (Kodak). They were then exposed for 10-14 days at 4°C before developing with D19 developer (Kodak). After fixing, slides were washed extensively in water before counterstaining with Giemsa as described by Perry-O’Keefe et al. (1990). Under dark-field illumination silver grains appear white whereas pigment grains appear red.

Degenerate primers were designed and synthesised in accordance with the known sequences for mammalian kFGF and used to prime a polymerase chain reactions on mixed stage Xenopus cDNA. A band of the predicted size was obtained with primers XK(10D) and XK(14D) which, on sequencing, had a 70% identity of derived amino acid sequence to part of the first and second exons of mammalian kFGF. The 3′ and 5′ parts of the cDNA were then cloned by a RACE protocol. The RACE procedures yielded sequence data that were suggestive of two copies of the gene in the Xenopus genome. The complete coding region was then amplified from cDNA using primers designed from the now accurately known sequences.

This confirmed that two types of closely related cDNAs were present (Fig. 1). This result is not unexpected as Xenopus laevis is believed to be pseudotetrapioid as a result of a genome duplication event occurring approximately 30 million years ago (Bisbee et al., 1977; Kobel and du Pasquier, 1986). The two cDNAs isolated differ by 8% in nucleotide and 11% in derived amino acid sequence. It would seem likely that these two genes represent the divergent pseudoalleles that arose from the genome duplication event. Similar pairs of sequences have been reported on a number of previous occasions (eg Shuldiner et al., 1991). Sequence analysis of the full-length coding regions revealed that the original PCR product, XeFGF(0.25), was derived from XeFGF(i). Evidence that XeFGF(0.25) spans part of exon 1 and exon 2 of the Xenopus gene comes from another clone isolated from a gastrula cDNA library that showed a match to XeFGF(0.25) up to Ser 94 followed by a non-matching pyrimidine-rich stretch (data not shown). This is exactly the position of the first exon-intron junction in human kFGF (Yoshida et al., 1987) and suggests that the clone came from an unspliced mRNA or genomic DNA fragment. It also suggests that the boundary between the first exon and the first intron has been conserved between XeFGF and those mammalian FGFs for which the genomic organization is known.

A genomic Southern blot was prepared and hybridized with XeFGF(0.25) (Fig. 2). None of the enzymes used cut within the probe. Therefore the presence of multiple bands in the EcoRI and PstI digestions would again suggest the presence of two pseudoalleles with restriction site polymorphisms in the sequence of the first intron, although a polymorphism between true alleles of either pseudoallele cannot be excluded.

When our sequences are aligned with those of mammalian FGFs the closest matches are obtained to FGF-6, and to kFGF from which the degenerate PCR primers were originally designed. In Fig. 3 the two sequences are shown aligned with human FGF-6 and kFGF. From this it may be seen that the similarity is much weaker for the 70-80 N-terminal amino acids than for the remainder of the sequence. Indeed there is more evolutionary divergence in this region for the human and mouse sequences (Brooks et al., 1989) and this would seem to suggest a reduced requirement for absolute sequence conservation in this part of the protein. The figures for percentage derived amino acid identity over the full-length proteins are XeFGF(i): 61% to HFGF-6, 57% to HkFGF; XeFGF(ii): 61% to HFGF-6, 58% to HkFGF. If the more divergent 70-80 N-terminal amino acids are left out of the comparisons, the identity approaches 70% for both kFGF and FGF-6.

Because it is possible that neither of these genes is the true homologue, or that eFGF represents a common ancestor that has undergone a gene duplication in the mammalian line, we are provisionally calling our molecule eFGF (embryonic FGF) and deferring identifying it as the amphibian homologue of any of the piammalian FGFs reported so far. This is prefixed with an “X” for Xenopus and the presumed pseudoalleles resulting from tetraploidization are designated (i) and (ii). So we shall call our molecules XeFGF(i) and XeFGF(ii) in this paper.

In order to show the evolutionary relatedness of XeFGFs to the existing members of the FGF family, the match between our sequences and all human FGFs is displayed as a dendrogram in Fig. 4. This was constructed from all the pair-wise comparisons using the PILEUP program of the University of Wisconsin analysis package. For reference, the amino acid identity between both eFGFs and Xenopus bFGF is 34%.

Various structural features can be deduced from the derived amino acid sequences. The predicted relative molecular mass is 21000 for XeFGF(i) and 22000 for XeFGF(ii). The isoelectric point for both is very high at 11 which is very similar to those of the FGF-6 and kFGF proteins. In Fig. 5 are shown hydrophobicity plots for the N-terminal sections of both proteins. Both sequences have a hydrophobic stretch of amino acids at their N terminus characteristic of a secretory signal sequence. The predicted signal peptide cleavage sites calculated from Von Heijne’s rules (Von Heijne, 1986) are shown with arrows. The methionine indicated as number 1 is probably the initiating residue because sequence analysis of the 5′ RACE products showed stop codons in all three reading frames within 50 bp upstream of this ATG.

At position 30 (31 for ii) is a sequence Asn.Asp.Thr which could serve as an acceptor site for N-finked glycosylation. Despite the N-terminal divergence between the XeFGFs, kFGF and FGF-6, they all have such a site in this region. This would argue for some evolutionary pressure to retain this potential glycosylation site. It is however, not a prime requisite for biological activity as our results and others (Miyagawa et al., 1991) show that N-terminally truncated forms can act both as MIFs and transforming growth factors. The potential heparin-binding domains of bFGF as reported by Yoshida et al. (1989) are quite well conserved in XeFGF.

The experiments described here were performed in order to verify the predictions made from the sequence. Synthetic mRNA coding for XeFGF(i) was prepared and translated in vitro with rabbit reticulocyte lysate. When microsomes were included, two extra bands appeared about 2000 smaller and 2000 larger than the primary translation product of about 21000 (Fig. 6). These probably represent processing products with first the removal of the signal peptide to give the smaller band and then the addition of sugars to the NDT site to give the larger band. The lysates were found to be positive when tested for MIF activity (Godsave et al., 1988). It is not possible to prepare large amounts of the protein by this technique, therefore it was decided to express XeFGF(i) in the bacterial expression system previously used by Kimelman et al. (1988) for bFGF.

XeFGF(i) was subcloned into the vector pET 8c. The construct used produces a protein truncated by 44 amino acids at the N terminus, which includes the potential hydrophobic signal sequence. Human kFGF similarly truncated is active in MIF assays (unpublished results). On induction, the bacteria produced a protein of relative molecular mass 17000 as predicted from the removal of 44 N-terminal residues. This was purified from the bacterial lysate by conventional and HPLC heparin chromatography (Fig. 7). The purification showed that the induced protein does bind tightly to heparin, eluting from the heparin HPLC column at about 1.1 M NaCl. Using the serial dilution MIF assay of Godsave et al. (1988), it was found to have a specific activity of about 10⁶ uniting, very similar to the value for bacterially produced Xenopus bFGF (Green et al., 1990). Most of the inductions were the translucent vesicles or “ventral type” inductions typically produced by members of the FGF family (Fig. 8), although we have noticed a something which is very rarely seen following treatment with bFGF.

The potential significance of XeFGF as an inducing factor obviously depends on its expression, both temporal and spatial, in early embryonic development. We have studied expession of the mRNA using RNAase protection and in situ hybridization.

Fig. 9A and B shows RNAase protections for whole embryo RNA for ten stages of development from egg to stage 41, Due to the presence of two presumed pseudoalleles, protections were carried out using probes, XeFGF(i)GS and XeFGF(ii)GS, which are specific for the divergent 5′ ends of each cDNA. It is conventional to use the translation factor EFl-a as an internal control in the analysis of the developmental expression of mRNA in Xenopus. It does, however, suffer from a major problem in that its expression continues to rise throughout the period of development in which we are interested, making the quantitative comparison between different stages difficult. In this study,we have used the housekeeping gene ornithine decarboxylase (ODC) as the internal reference. Independent experiments have shown that the ODC message represents a fairly constant proportion of total RNA regardless of developmental stage or region of the embryo from which the RNA is prepared (data not shown).

From Fig. 9A it can be seen that, using XeFGF(i)GS as a probe, maternal expression is clearly detectable but that the level rises sharply by the onset of gastrulation. It then falls again, after stage 12 and the closure of the blastopore, to a low and fairly constant level. The temporal expression that is seen using the XeFGF(ii)GS probe is somewhat different. There is no detectable maternal component, zygotic transcription however, follows that of XeFGF(i) but is at a level 5- to 10-fold lower. Differential transcription of pseudoalleles in Xenopus has recently been reported for two insufin genes (Shuldiner, 1991).

So it can be seen that XeFGF mRNA is present throughout the period of early development, which includes the time when mesoderm induction is thought to be occurring, and peaks in gastrulation suggesting an additional role during this active phase of cell movements and regional specification. The peak level of expression for XeFGF(i) and XeFGF(ii) respectively is about 1-2% and 0.1-0.2% that of ODC, putting them into the intermediate to scarce category of messenger RNA abundance. A rough estimate based on RNA loading and exposure time suggests that the expression level is somewhat higher than the maternal level of bFGF mRNA (unpublished data) or the zygotic level of activin B (Thomsen et al., 1990).

A number of RNAase protections were carried out on dissected pieces of embryos using the XeFGF(i)GS probe. There was little or no difference between the maternal mRNA content of animal, marginal and vegetal pieces from blastulae (Fig. 10A). In stage 10 gastrulae there is an excess in dorsal quarters over ventral three-quarters (Fig. 10B), and a considerable excess in vegetal over animal halves (Fig. 10C). In neurulae (stage 13-19) and lozenge stages (stage 20-24), expression appears exclusively in the posterior third, and is not seen in the anterior or the middle thirds (Fig. 10D). Although the resolution of these studies is necessarily not very high, they do complement the in situs and confirm that the zygotic expression is concentrated near the blastopore lip and, as the axis is formed, in the posterior of the forming body.

In situ hybridizations were carried out on sections from a variety of stages from oocytes to the prelarval stage 40, and some results are shown in Fig. 11. The probes XeFGF(0.25) and XeFGF(i)GS gave similar results. A signal is visible in early (stage 1-2) but not later oocytes, showing that the maternal expression commences early in oogenesis. No signal is visible at early embryonic stages, the maternal mRNA now presumably being below the detection limit of in situ hybridization. At stage 10 a distinct signal appears in the mesoderm near the dorsal lip. As the lip develops, this extends to a complete circle around the embryo. As the blastopore closes, the expression domain becomes concentrated into the posterior of the forming axis, ending up in the early neurula once more dorsal to the blastopore and present in both the ectodermal and mesodermal layers of the forming axis (Fig. 11 A). At stage 20 (lozenge) a similar signal is apparent, and at stage 34 (tailbud) this posterior domain narrows down into the tailbud blastema. The surrounding epidermis shows no signal, but the undifferentiated region that produces the neural tube, notochord and somites of the tail shows a strong hybridization signal (Fig. 11C). By this stage there also appears a stripe of expression straddling the midbrainhindbrain border and occupying the ventral half of the neural tube at this level (Fig. 11B). A third region of expression visible in the tailbud stage is the myotomes. This seems fairly uniform with no concentration in the anterior or posterior of the body, and the level of the signal is somewhat lower than in the brain or tailbud (Fig. 11D).

XeFGF is most closely related to the mammalian genes FGF-6 and kFGF. From sequence similarity it has not been possible to decide if it represents the amphibian homologue of either of these genes. Indeed, the sequence similarity of XeFGF with both putative homologues is rather lower than that for bFGF, where the Xenoptzs-human difference is 84% (Kimelman et al., 1989). One possibility is that XeFGF represents a gene ancestral to both FGF-6 and kFGF, which has undergone duplication and subsequent divergence in the mammalian line. Another possibility is that it represents a new member of the FGF family with an as yet undiscovered direct homologue in mammals, and this is currently under investigation. The true relation of XeFGF to other members of the gene family must await a more comprehensive set of sequences and more information about gene organization and structure.

We have isolated two versions of XeFGF which are probably pseudoalleles arising from tetraploidization. The expression patterns of the two forms differ in that XeFGF(i) is expressed maternally while XeFGF(ii) is not, and the zygotic expression of XeFGF(i) is at a level 5-10× higher than XeFGF(ii). We might expect evolutionary divergence of pseudoalleles to lead more easily to loss of function in one copy than to gain of function, so it may be that the expression pattern of XeFGF(i) represents the ancestral and more general condition.

The N-terminal amino acid sequence has the characteristics of a cleavable signal peptide and the results of the in vitro translation experiments argue that XeFGF is a secreted molecule. Like other members of the FGF family, XeFGF protein will bind with high affinity to heparin and has mesoderm-inducing activity. In general, the character of the inductions resembles the ventral type inductions produced by the ventrolateral part of the vegetal hemisphere (Dale and Slack, 1987) rather than the axial inductions provoked by activins (Smith et al., 1990; Thomsen et al., 1990). We have, however, also obtained a number of axial inductions which are not necessarily found at the higher concentrations. This effect is currently being studied further.

Work in our laboratory shows that bFGF is probably not secreted in the embryo (Thompson and Slack, unpublished observations) making it an unlikely inducing molecule. There is, however, good evidence that a FGF-like molecule has a role in mesoderm induction. Firstly, we know that FGF receptors are expressed maternally and are present on the surface of all cells in the blastula (Gillespie et al., 1989; Musci et al., 1990). The level of receptor accessible at the cell surface rises and falls over the blastula period, during which mesoderm induction is generally thought to take place (Jones and Woodland, 1987). Secondly, Amaya et al. (1991) have shown that destroying the function of a population of FGF receptors by overexpression of a dominant negative mutant leads to embryos lacking derivatives of the lateral and ventral mesoderm. Since mesoderm induction probably commences well before the mid-blastula transition (MBT) and the onset of zygotic transcription, all components of the process should be maternally transcribed. We have shown that mRNA encoding XeFGF is present at these early stages. Since it has the appropriate biological activity, the mRNA is present at the right stage, and the protein is a secreted molecule, XeFGF must be considered as a candidate for a mesoderm-inducing molecule in vivo. We are presently attempting to prepare neutralizing antibodies that could be used in conjunction with the transfilter apparatus of Slack (1991) in order to test this possibility directly. It is true that the maternal transcripts are fairly uniformly distributed, but it is possible that there are different controls on translation or secretion in animal and vegetal hemispheres.

The expression pattern also suggests that XeFGF may function during interactions which occur after mesoderm induction is over. The temporal peak of expression coincides with the period when we believe that the lateroventral mesoderm is being dorsalized by the organizer (Dale and Slack, 1987; Lettice and Slack, unpublished) so it might be involved in this process. A concentration of expression in the posterior becomes marked during gastrulation and is maintained in the tailbud at later stages. This domain is not tissue specific, since it is seen both in the posterior mesoderm and the posterior neural plate in the neurula. We have recently argued from a survey of the classical literature that the posterior is the dominant end for interactions controlling the specification of the anteroposterior level (Slack and Tannahill, in press). Again, the work of Amaya et al. (1991) shows posterior axis deficiencies in embryos overexpressing a dominant negative form of a FGF receptor. We are thus tempted to speculate that XeFGF may also have a role as an inducing factor released from the posterior of the axis, which is involved in anteroposterior specification. The tailbud, which is perhaps comparable to the proliferating mesenchyme cells that underlie the apical epidermal ridge of the limb bud (Cooke, 1975), continues to proliferate and to export cohorts of cells of successively more posterior character during the tailbud stages. At least two differentiated tissues, the neural tube and somites, are produced from the tailbud. It is not known whether there are distinct precursor cells for each tissue but the XeFGF localization does not show any regionalization within the tailbud and it is possible that it is here playing some role in maintaining the tailbud in an undifferentiated state. The later expression domains, in the hindbrain and the myotomes, also appear at the tailbud stage and are presumably associated with different and independent functions.

Recently reported evidence strongly suggests an important role for a FGF-like factor in the events of regional specification in the Xenopus embryo, particularly mesoderm induction. The biological activity, the presence of maternal mRNA, and its probable secretion makes XeFGF, at present, the best candidate for such a molecule. In addition the domains of later zygotic expression are suggestive of a role in the formation of the body axis. In order more clearly to define the role of XeFGF in the embryo, we are currently preparing antibodies in order to look at the protein, studying the effects of various developmental perturbations on XeFGF expression, and studying the morphological effects of ectopic expression and functional ablation of XeFGF.

We should like to thank Gabriele Johnson and Emma Bums for technical assistance and Paul Martin for a critical reading of the manuscript.