The nature of the mesoderm-inducing signal in Xenopus: a transfilter induction study

The last three years have seen an explosion of work describing the mesoderm-inducing activity of a number of cytokines belonging to the FGF and the TGF/3 superfamilies (reviewed Smith, 1989; Slack, 1990). Several of these factors have been found in early Xenopus embryos either as mRNA or as protein or both, and are therefore candidates for a role as real endogenous mesoderm-inducing factors. At present there is feverish work going on devoted to the establishment of the exact expression pattern of each factor (eg Thomson et al. 1990; Koster et al. 1991), but these results, although important and necessary, are not going to tell us whether a particular factor is actually involved in mesoderm induction.

In a previous paper, a set of nine criteria was suggested for establishing whether or not a substance could qualify as a morphogen (Slack and Isaacs, 1989). Six of these deal with in vitro effects and expression data that have been or are being addressed by other studies. The remaining three deal with properties that cannot be known from expression data alone: (1) The factor should be exported from the signalling region. (2) It should be transmitted to the responding tissue. (3) Inhibition of synthesis, transport or action by mutation or by biochemical antagonists should inhibit the response.

All three of these can be studied using an old embryological technique known as the transfilter experiment. In this, the signalling and the responding tissues are separated by a membrane of known permeability properties and the characteristics of the signal are deduced from the types of membrane that it is able to cross (Grobstein and Dalton, 1957). In this paper, the term ‘signal’ will denote an embryological process that may depend on one or more than one substance. It has already been shown that mesoderm induction can take place across Nucleopore filters with a pore size too small to admit cytoplasmic protrusions (Grunz and Tacke, 1986; Gurdon, 1989). The factor or factors responsible must therefore be able to be secreted from vegetal cells and to diffuse across a short liquid gap to the responding tissue.

In a preliminary set of experiments, the transfilter setup was modified to enable the effects of neutralizing reagents to be tested and it was shown that low concentrations of heparin, but not of other sulphated polysaccharides, could inhibit the induction (Slack et al. 1987). Heparin also binds tightly to cytokines of the FGF family and, in the case of the MIF activity of bFGF but not its mitogenic activity, it is inhibitory in vitro. Because of this, the fact that heparin inhibited the transfilter induction was at the time thought to suggest that bFGF was an essential component of the signal.

In the present paper, the experiments are extended to test the effects of follistatin, a potent inhibitor of activin, and of neutralizing antibodies to bFGF. These inhibitors inhibit their respective cytokines in vitro but have only limited effects on the transfilter induction. This suggests that, contrary to prevailing opinion in the field, neither activin nor bFGF are major components of the mesoderm-inducing signal emitted from the vegetal cells, although either or both might still have a role to play within the responding tissue or in inductive processes at later embryonic stages.

Xenopus bFGF was prepared by H.Isaacs using the bacterial expression plasmid XF140 as described in Kimelman et al. (1988). It was purified by both conventional heparin sepharose chromatography and by a subsequent HPLC heparin sepharose column with gradient elution (see Slack and Isaacs, 1989). Bovine activin A was provided by D.Huylebroek (Innogenetics, Ghent). It was a conditioned medium from HeLa cells transfected with an vaccinia virus-derived expression vector (see Smith et al. 1990). Xenopus activin A was partially purified from XTC serum-free conditioned medium by phenyl sepharose chromatography following the method of Smith et al. (1988). Xenopus activin B was prepared using the mammalian expression plasmid pcDNAl-Xact/SBc provided by G.Thomsen (Harvard University; see Thomsen et al. 1990). This was used to transfect COS cells by the DEAE dextran/chloroquine method, and these cells were used to condition serum-free medium. Control medium conditioned by COS cells transfected with vector alone was without activity.

Follistatin purified from porcine ovarian follicular fluid was a gift from Dr H. Sugino (Institute for Physical and Chemical Research, Wako, Saitama, Japan). This protein was discovered as a factor inhibiting secretion of FSH by pituitary cells, and its activin-binding activity is described by Nakamura et al. (1990). It has not previously been shown that it can neutralize activin in vitro. Anti-XbFGF (CH) was prepared in the laboratory by immunization of a rabbit with purified Xenopus bFGF. Anti-bovine bFGF (DOG) was a gift from D.Gospodarowicz (UC Medical Center) and was previously used in the study of Slack and Isaacs (1989). The antibodies were protein A purified to reduce toxicity.

The heparin used in the inhibition experiments was from Sigma, grade I from porcine intestinal mucosa. Heparin is a glycosaminoglycan, which is initially synthesized as a repeating polymer of D-glucuronic acid and N-acetyl glucosamine. The N-acetyl glucosamine then becomes progressively sulphated on the 2(N) and 6 (O) positions, and the glucuronic acid becomes progressively epimerised to L-iduronic acid and sulphated on the 2 position (Comper, 1981). The result is a very heterogeneous polymer with a large number of possible internal sequences. The best known biological activity of heparin is its anticoagulant activity, which depends on binding to antithrombin HI although this is due to a different carbohydrate sequence from binding to FGF (Choay et al. 1983; Barzu et al. 1989).

The suramin was a gift from L. Madhaven (Dept. Biochemistry, Oxford). It is a synthetic polyaromatic compound containing six sulphonic acid groups, and has been used as an anti-trypanosomal agent. It is known to antagonise the effects of several cytokines (Coffey et al. 1987).

Xenopus embryos were obtained by artificial fertilization using methods described previously (eg Godsave et al. 1988). Salines are based on normal amphibian medium (NAM: Slack, 1984). Disaggregation of vegetal explants was carried out in ‘PhoNaK’, which is 50 nw sodium phosphate pH 7.0, 35mM NaCl, ImM KC1. For experiments in which dorsal and ventral tissue explants were used, the dorsal side of embryos with a clear dorsoventral pigmentation difference was labelled with Nile Blue at the 4- to 8-cell stage.

The apparatus used is shown in Fig. 1. It consists of a stainless steel cylinder of internal diameter 7 mm within which can be placed various membranes and spacing rings of diameter 6 mm. The use of the correct spacing rings is essential since the tissues must be slightly compressed to immobilize them, but not compressed so much as to cause damage. Spacing rings were punched out of tin foil of known thickness (Goodfellow Metals Ltd., Cambridge) using a leather punch.

Thicknesses of 0.5 mm, 0.25 mm and 0.15 mm were found to be optimal for whole endoderm (VP), half endoderm (DVP or VVP) and ectoderm (AP) explants, respectively (for sizes of explants see Fig. 4). Two types of Nucleopore membranes were used in the experiments, with pore sizes of 0.4 and 3.0 μm. Nucleopore membranes are about 10μm thick and are made of polycarbonate coated with polyvinylpyrrolidone. The pores are created by neutron bombardment with subsequent etching so that they have circular cross section and run straight through the membrane. Although membranes with small pores have more pores per unit area than those with large pores, this does not make up for the reduction of crosssectional area of each pore, and in our case the 0.4μm membrane has 20× less pore area overall than the 3.0μm membrane. For the transfilter neutralization experiments, the membrane assembly between the tissues was a sandwich consisting of a sheet of nylon mesh of thickness about 0.1 mm (Gallenkamp) and a 3 μm pore Nucleopore membrane on either side. In this arrangement, the purpose of the nylon is to create a wide enough liquid gap between the tissues for the neutralizing reagent to trap the inducing factor. The purpose of the Nucleopores is to prevent contact of the tissues, which can occur across the nylon mesh alone.

For transfilter experiments 0.1ml of medium, containing the inhibitor where appropriate, was placed in the well. Then the bottom spacing ring was put in, then the inducing tissue with blastocoelic surface up, then the membranes and second spacing ring, then the test tissue with blastocoelic surface down, and finally a sheet of dialysis membrane and the weight to immobilize the ectoderm. The combinations were set up in the afternoon and incubated at 25°C overnight. The next morning, when control embryos had reached about stage 20, both top and bottom tissues appeared in good condition (Fig. 2A). Cell lysis did not exceed a few cells, probably those damaged during insertion of the explants into the apparatus. The test tissues, which in this paper were always explants of stage 8 animal pole ectoderm, were removed for culture in NAM/2 in 24-well plates for 2 further days. By this time control embryos had reached stage 41 and the induced or uninduced state of the explants was obvious down the dissecting microscope (Fig. 2B). Most of the explants were examined histologically using methods described in Godsave et al. (1988).

The following grading system was used to score the test tissues:

0: Jumbled epidermis only, may contain cavities but no mesodermal tissues and no stratification of epidermis (eg Fig. 5B,D,G,H).

1: Concentric arrangement of stratified epidermis, loose mesenchyme and mesothelium. May contain blood-like cells and small wisps of muscle.

2: Still a concentric arrangement but with a substantial clump of muscle (eg Fig. 5C,E,F).

3: Patches of muscle and mesenchyme with no concentric pattern within stratified epidermal vesicle (eg Fig. 5A).

4: As 3. with clumps of neuroepithelium.

5: Contains notochord with associated neuroepithelial and muscle masses. Often with otocysts and dense (neural crest type) mesenchyme. Stratified epidermis still present but not necessarily continuous around the outside.

These are entirely independent of the transfilter neutralizations and are carried out in Terasaki plates by a simple modification of the standard serial dilution assay for mesoderm-inducing factors. One row of wells contains a set of dilutions of an active cytokine and another row contains the same dilutions of the cytokine together with a fixed concentration of an inhibitor (Fig. 3). Usually the concentration of inhibitors used were also those used in the transfilter experiments. At least one hour is allowed for reaction before the animal caps are added to the wells. After 2-3 days culture each row should contain induced explants on the high cytokine side and uninduced explants on the low cytokine side of an end point (for photo see Fig. 4 of Slack, 1990). For serial 2-fold dilutions, the number of wells that the end point is shifted by the inhibitor towards the high cytokine side is taken as log₂ (units ml⁻¹ cytokine inhibited by this concentration of inhibitor). For purposes of comparison the inhibitions are normalised by dividing the inhibitor concentration by this activity and assuming that this represents approximately the concentration of inhibitor required to neutralize one unit ml⁻¹ of the cytokine.

Using the methods described above, the control positive induction rate approached 100 % either for a single 0.4μm membrane or for the sandwich arrangement. As in the work of Grunz and Tacke (1986), SEM study of the 0.4 μm membranes after use showed no evidence of cytoplasmic contacts. In the case of the sandwich arrangement, it is not possible to fix all the components together for EM examination, but cellular contact seems exceptionally unlikely since the gap is several cells wide, there is a total lack of adhesion of either tissue component to the Nucleopores and no material was ever seen in the nylon grid at the light microscope level.

Although the frequency of inductions was very high, the histological grade was usually low, most cases being grade 1 or 2. The mean grade of the controls, at 2.2, includes two grade 5 and one grade 4 cases, but these were exceptional and if they are left out the figure would be 1.7. So, as in the previous transfilter studies referred to above, the inductions are predominantly of a ventral character and it is probable that not all components of the normal signal are being transmitted effectively.

Since it was already known that heparin could block the transfilter mesoderm-inducing signal, the conditions for effective neutralization could be established using heparin. Neutralization was not found using nucleopore membranes alone, even when a stack of as many as five 3 μm membranes was used. It is probable that this is because there is no real liquid gap between the membranes and the amount of heparin in the pores themselves is inadequate to prevent passage of the signal. On the other hand, neutralization is effective and reliable with the sandwich arrangement in which there is a liquid gap about the thickness of the nylon (100μm). With a near 100% positive rate for animal-vegetal pairs and a near 100 % negative rate for similar pairs cultured in the presence of heparin (see Table 3), it was felt that this system does enable secreted factors to be neutralized by suitable inhibitors and hence potentially provides a discriminating method for the identification of at least those endogenous mesoderminducing factors that are transmitted in this apparatus.

For experiments on the regional specificity of transfilter induction, stage 8 blastulae were dissected into five pieces as shown in Fig. 4. A single 0.4μm membrane was used to separate inducing and responding tissues. The visual results (Table 1) showed that both dorsal and ventral vegetal tissue gave 100 % inductions while the marginal zone pieces sometimes did so and the animal pole pieces never did. Histological examination of the test tissues showed that there was a slight difference in dorsoventral grade between inductions provoked by the dorsal and ventral vegetal pieces but this was much less pronounced than that found in animal-vegetal combinations (Dale et al. 1985). Animal pole pieces did not produce any inductions although there were quite commonly found to be cavities in the explants. These were larger than the small extracellular spaces usually seen in isolated animal caps, but there was no stratification of the epidermis and no mesothelium was seen.

The transmission time was examined with the nylon sandwich in order to be comparable with the neutralization experiments. The test tissue explants were removed at 1h, 3h and overnight (about 16h). As shown in Table 1, many of the 3h cases were positive, indicating that transmission of the signal and an irreversible response to it can be completed within this time, although the maximum effect takes longer. The overnight incubation is experimentally convenient and so was used for the other experiments reported here, but it can be assumed that the results represent processes occurring before the loss of ectodermal competence to the mesoderm-inducing signal (Jones and Woodland, 1987).

In vitro neutralization tests were carried out in order to establish the effectiveness of each of the inhibitors used. These experiments are carried out using the serial dilution method shown in Fig. 3 and are completely independent of the transfilter experiments. The results (Table 2) show that follistatin is a very potent inhibitor of activin. Bovine activin A was inhibited at about 1ng ml⁻¹ per MIF unit and Xenopus activin A and activin B at concentrations about 10× higher. Although the effect of follistatin on pituitary cells has been assumed to be due to inhibition of activin, this is the first time that the inhibition has been demonstrated in vitro. Both the antibodies to bFGF neutralized activity, the DOG antibody (prepared against bovine bFGF) at about 20 μg ml⁻¹ per MIF unit, and the CH antibody (prepared against Xenopus bFGF) at about 0.5 pg ml⁻¹ per MIF unit. Heparin is, as previously shown, a potent inhibitor of bFGF, but showed only slight activity against the activins. Suramin inhibited bFGF and activin A to a limited degree but the concentration of 50 pM used in the transfilter neutralizations would only be expected to neutralize 1–2 MIF units ml⁻¹.

The results are presented in Table 3 and some typical histological sections are shown in Fig. 5. Some experiments were carried out with whole vegetal cores (VP) and others with ventral vegetal halves (WP). Both types of inducing tissue gave a high percentage of positive inductions with the sandwich assemblies just as they had done with the single 0.4 μm membranes (Table 1; Fig. 2B). The average histological grade of the VP inductions was 2.2 and of the WP inductions was 1.7, which is similar to what was found with the single 0.4 μm membranes above. So it seems that the 20 × greater pore area of the 3 pm membranes approximately compensates for the extra 105 pm of distance that the signal has to travel. Inductions of this histological grade are what would be produced in vitro by about 10 MIF units ml⁻¹ of bFGF and perhaps rather fewer units ml⁻¹ of activin A (Green et al. 1990). This means that an inhibitor concentration that inhibits more than 10 MIF units ml⁻¹in vitro should inhibit the transfilter induction. If it does not do so, the implication is that another factor is responsible in vivo.

Follistatin failed to inhibit the transfilter inductions even at lug ml⁻¹, which should inhibit about 100 MIF units ml⁻¹ of Xenopus activin A or B (Fig. 5E). The histology of the inductions showed a slight reduction of average grade compared with the controls although this was not statistically significant (P>0.05 in t-test). It therefore seems unlikely that activin is a major component of the signal. Neither of the antibodies against bFGF were able to inhibit the transfilter inductions even at concentrations capable of inhibiting several hundred MIF units ml⁻¹ of bFGF. Again there was a slight, but not statistically significant, reduction of grade. Many of the anti-FGF experiments were performed with ventral vegetal tissue (WP) as the inducer since it was felt that the properties of bFGF corresponded most closely to the ventral vegetal component of the three signal model (Slack et al. 1989). But the results make it very unlikely that bFGF is a major component of the signal even on the ventral side of the embryo. When high concentrations of follistatin and anti-FGF were used together there was a reduction of inducing activity and average grade that just reached statistical significance (0.05>P>0.02 in t-test). This may mean that these substances or cross-reacting relatives are responsible for a proportion of the signal, but other substances would certainly have to be involved as well.

In view of the predominantly ‘negative’ results using specific inhibitors, it is important to show that inhibition is possible even if the inhibitor used is not very specific. So a number of further experiments were performed with heparin and an almost complete inhibition was achieved at 200μgml⁻¹ (Table 3 and Fig. 5). Suramin was also tested because it has been used as an inhibitor of various cytokines by other workers in various types of experiment. The results show that it is also an inhibitor of the mesoderm-inducing signal, giving almost complete inhibition at 50 μM.

However, there is an important difference between the effect of suramin and that of heparin. This was shown when they were tested on conventional animal-vegetal (Nieuwkoop) combinations assembled with no intervening membrane. Heparin had no effect on these while suramin produced a dramatic inhibition (Fig. 6). The mean gradings were control 4.0 (4 cases), heparin 4.1 (200μg ml,8 cases), suramin 0.5 (50gM, 14 cases). The only difference between a Nieuwkoop combination and a transfilter combination is the presence of the liquid gap in the latter. This suggests that heparin must be exerting its effect in the transfilter experiments by intercepting the signal in the liquid gap. Suramin on the other hand could be inhibiting by an action at any stage, for example an inhibition of signal secretion, signal transfer, or some later event within the responding tissue.

The problem of the identification of the endogenous mesoderm-inducing factor(s) in Xenopus has turned out to be harder than originally expected. This is partly because there are now quite a few substances showing mesoderm-inducing activity at picomolar or low nanomolar concentrations. At the last count there were 11: aFGF, bFGF, kFGF, FGF-5, ECDGF, int-2, activin A, activin B, TGFj3-2, TGF/J-3, BMP-4. It is probable that some others such as the vg-1 related (vgr) proteins will soon join the list. Although at the time of writing only bFGF had definitely to be shown to be present as active protein during the stages when mesoderm induction is occurring (Kimelman et al. 1988; Slack and Isaacs, 1989), it is probably only a matter of time before the presence of several other active proteins is confirmed (J. Smith, personal communication; H. Isaacs, personal communication). So we have several candidates, but to prove that a given substance is a mesoderm-inducing factor in vivo, we need to show not only that it is present at the right time and place, but also that it is secreted from the vegetal cells.

The transfilter neutralization technique offers a new way of investigating this and overcomes several of the potential problems of other methods, such as the injection of inhibitors into whole embryos. First, the large and well-controlled gap between the tissues enables the concentration of inhibitor in the relevant position to be accurately known. Second, the use of a small test tissue explant and the separation of the tissues means that morphogenetic movements have little bearing on the outcome. Finally, inhibitors have been used whose effects can be independently assessed at least semi-quantitatively by in vitro neutralization experiments.

The present experiments confirm the previous ones of Grunz and Tacke (1986) and Gurdon (1989) in showing that at least some of the normal components of the mesoderm-inducing signal emitted by the vegetal cells are capable of crossing a liquid gap. However, it is clear that the histological grade of transfilter inductions is substantially lower, or more ventral, than that found in animal-vegetal combinations (see Fig. 6 and also Dale et al. 1985). This may simply be a quantitative effect arising from the considerable reduction in crosssectional area available for diffusion across a Nucleopore membrane. Alternatively, if there are indeed several qualitatively distinct components to the mesoderm-inducing signal, it is possible that those required for the axial induction are less efficiently transmitted than those needed for the ventral induction. This would also explain why the difference between the inductive capacity of dorsal and ventral vegetal pieces seems much less pronounced in these experiments than in animal-vegetal combinations (Dale et al. 1985; Dale and Slack, 1987). So we cannot prove that all components of the natural signal are transmitted across the liquid gap, but it is clear that some active material is and so it seems a legitimate exercise to try to identify the molecule(s) responsible.

The results using follistatin and anti-bFGF antibodies show clearly that the signal from the vegetal cells does not consist of activin A or activin B or bFGF alone. Even when both inhibitors are used together at high concentrations, there is only a slight reduction of the effect. At best the signal may consist of several substances of which these are minor components, or with which the inhibitors show some slight cross neutralization. The fact that the signal can be totally inhibited in this experimental setup is shown by the results using heparin. Because heparin does neutralize the MIF activity of bFGF efficiently in vitro, its neutralization of the transfilter induction was originally taken as evidence for an essential role for bFGF, but it now seems that this association is fortuitous and that the heparin must be exerting its effect in some other way. The present study shows that heparin, in contrast to suramin, only shows its inhibitory effect when there is a liquid gap between the animal and vegetal explants and so is presumably not able to penetrate the explants themselves.

For those who wish to save a role for the activins, it may be argued that they are only responsible for the dorsal, organizer-inducing, signal. This seems to be poorly transmitted in the transfilter apparatus where ventral type inductions predominate. It is also possible to argue that either activins or bFGF or both are indeed secreted but in some form or as part of some complex that is not recognised by the inhibitors used. This cannot be excluded by the present data but would again indicate the existence of other molecules essential to the process.

Even if activins and bFGF are not secreted from vegetal cells at all, the results do not exclude a role for them at some stage of mesoderm induction after the receipt of the initial signal by the responding cells. For example, one or more of these factors might be synthesized and released in the responding tissue but still be an obligatory link in the causal chain. It is also quite possible that members of the activin and FGF families are involved in events that occur at later developmental stages than mesoderm induction; for example, the specification of different anteroposterior body levels (Ruiz i Altaba and Melton, 1989; Cho and de Robertis, 1990).

It is generally thought that mesoderm induction commences in the early blastula, some time before the beginning of zygotic transcription at the mid-blastula transition (Jones and Woodland, 1987). This suggests that we should consider candidates that are expressed maternally. There are several of these, including bone morphogenetic protein-4 (Koster et al. 1991), vg-1 related proteins (G. Thomsen and D. Melton, 3rd International Xenopus Meeting, Les Diablerets, Switzerland, 1990) and other members of the FGF family (H. Isaacs, personal communication). When specific inhibitors become available each of these candidates can be examined using the transfilter neutralization technique. In the meantime, a simple test for a factor might be sensitivity of its MIF activity to heparin in vitro. Unless it is sensitive at least to a level of about 10μgml⁻¹ per MIF unit, it is unlikely to be a major component of the mesoderm-inducing signal.

I am grateful to Emma Burns for assistance with histology, and to Gabriele Johnson for assistance with neutralization tests.