Control of digit formation by activin signalling

The developing limb constitutes a simple and useful model for the analysis of morphogenesis in vertebrates. The early limb bud consists of a core of undifferentiated mesodermal cells covered by an ectodermal jacket. At the distal margin of the bud the ectoderm exhibits a specialized thickening termed the apical ectodermal ridge (AER) which plays a pivotal role in limb development. Limb outgrowth is accomplished by proliferation of the mesoderm subjacent to the AER. This proliferating region is termed the progress zone. Chondrogenic differentiation of the limb mesoderm starts in the central core of the bud and progresses distally by the successive incorporation of cells leaving the progress zone (Summerbell et al., 1973). The primordium of the humerus/femur (stylopodium) is formed first, followed by the ulna-radius/fibula-tibia (zeugopod) and the skeletal pieces of the hand/foot (autopod) are the last to be formed. Concomitantly with the formation of the skeletal rudiments a number of mesodermal cells are eliminated by apoptotic cell death contributing to sculpt the final shape of the limb (Hurle et al., 1996). In the last few years, considerable evidence has been accumulated indicating that the coordination between growth, cell death and differentiation of the limb mesoderm is controlled by the interaction of signalling molecules acting at the tip of the developing limb (Niswander and Martin 1993; Macias et al., 1997; Merino et al., 1998). The maintenance of the progress zone mesoderm in an undifferentiated and proliferating state is carried out by FGFs (FGF-2, -4, -8 and -9) produced by the AER (reviewed by Martin, 1998). The control of chondrogenic differentiation and cell death involves a signalling pathway mediated by BMPs (BMP-2, -4 and -7; Zou and Niswander, 1996; Kawakami et al., 1996; Yokouchi et al., 1996; Macias et al., 1997; Zou et al., 1997) and TGFβs (Gañan et al. 1996; Merino et al., 1998). These findings point to the occurrence of a triple interactive molecular mechanism controlling the growth and differentiation of the limb mesoderm, including distal signals for proliferation (FGFs), proximal signals to establish the zones of chondrogenesis (TGFβs), and intermediate signals (BMPs) promoting growth and differentiation in the zones committed to chondrogenesis and apoptosis in the undifferentiated mesoderm. Up to now, TGFβ2 is the only identified signal with a potential to determine the zones of chondrogenesis (Gañan et al., 1996;

Merino et al., 1998). Exogenous application of this growth factor into the interdigital regions inhibits interdigital cell death and induces the formation of an ectopic digit through BMP signalling (Gañan et al., 1996; Merino et al., 1998). However, expression of the tgfβ2 gene at the tip of the chondrogenic digital rays occurs relatively late in development (Merino et al., 1998), and transgenic mice deficient in tgfβ2 have normal digits (Sanford et al., 1997). These findings indicate that other chondrogenic signals could be involved in limb skeletogenesis. To further understand this question we have tested the possible role of activins in the onset of chondrogenesis in the developing limb.

Activins are members of the TGFβ superfamily which play a fundamental role during embryonic development (see Stern et al., 1995). The activins are dimeric proteins, composed of disulphide-linked βA or βB subunits. Three main forms of activins, activin A (βAβA), activin B (βBβB) and activin AB (βAβB) are distinguished by their dimeric composition, although other members of the family containing different subunits have been also identified (Hotten et al., 1995; Oda et al., 1995). Activins exert their effects by forming heterodimeric complexes with type I and type II serine/threonine kinase receptors which act as a signal transducer and ligand-binding component respectively (see Baker and Harland, 1997). Since there are different forms of activin type I and type II receptors, the biological effects of activins may be related to the receptor expressed by the responding target cells. An additional complexity of activin signalling is that other members of the TGFβ superfamily can also bind to the activin receptors (Ebner et al., 1993; Matzuk et al., 1995a), and that activin signalling may be modulated by follistatins. Follistatins, constitute a group of proteins encoded by the same gene which are generated by alternative splicing or by proteolytic cleavage. Follistatins neutralize activin bioactivity by abolishing its binding to the activin type II receptors (see, De Winter et al., 1996). Activin signalling has been implicated in early limb development (Stern et al., 1995) and in skeletal muscle differentiation (Link and Nishi, 1997), and it may be also possible that it has a role in limb skeletogenesis. Exogenous administration of activin A promotes chondrogenesis in limb bud micromass cultures (Jiang et al., 1993, see also Chen et al., 1993). In addition, although there are no detailed studies of the pattern of expression of activin genes in the course of limb skeletogenesis, activin and activin receptor transcripts have been identified in limb prechondrogenic aggregates (Roberts et al., 1991; Nohno et al., 1993; Feijen et al., 1994; Verschueren et al., 1995). While Inhibins are known to have chondrogenic activity in limb mesodermal cultures (Chen et al., 1993), the expression of the alpha subunit in the embryo is restricted to the developing gonads at the stages of organogenesis (Feijen et al., 1994).

In this study we show that βA and βB subunits of activin, activin type II receptors (ActRII) and follistatin genes all exhibit a regulated pattern of expression in the developing chick limb autopod, strongly suggestive for a role of activin signalling in the establishment of the skeleton, joints and tendons of the digits. This potential function of activin signalling was explored by different experimental approaches. These include ectopic application of human recombinant activins, blocking activin signalling by follistatin treatment, and by analyzing changes in the pattern of expression of activin-related genes following experimental manipulations of the limb leading to alterations in the skeletal pattern of the digits. Our findings indicate that activins play a role in the developing autopod as proximal signals that appear to specify the position of the digital rays. This effect is contrary to the role of FGFs, which maintain the mesoderm undifferentiated at the progress zone. The changes in the pattern of expression of the BMP receptor gene bmpR-1b following treatments with activin and the potential of the BMP antagonist, Noggin, to downregulate activin βA gene expression indicate that activin signalling acts in cooperation with BMP-signalling.

We used Rhode Island chick embryos ranging from 3 to 9 days of incubation (stages 20-35, Hamburger and Hamilton, 1951). Eggs were windowed at the desired stages and experimental manipulations of the limb were performed in the right wing or leg bud.

Local application of the different growth factor and proteins were performed using heparin or Affi-Gel blue beads. The beads were incubated in PBS or in the selected recombinant human protein solutions (see below) and implanted into the limb mesenchyme. Most experiments were performed at stages 28-29 and the beads were implanted at the tip of digit III or in the third interdigital space. In other cases, the beads were implanted at different locations and stages as indicated in Results.

Surgical removal of the AER was performed in the leg bud at stages 27-28 using fine tungsten needles. After the operation the eggs were returned to the incubator and the embryos killed at different time intervals to study changes in gene expression by in situ hybridization.

For ZPA grafting experiments, the posterior margin mesoderm was excised from stage 22 wing buds and grafted into the anterior wing margin of host embryos at stage 20. After the operation the embryos were returned to the incubator and used for gene expression studies.

The morphology of the limbs was studied after cartilage staining with Alcian green as described previously (Gañan et al., 1996). The pattern of cell death was analyzed by vital staining with neutral red and by Tdt-mediated dUTP nick end labeling (TUNEL) in tissue sections as described previously (Macias et al., 1997). In some experiments the morphology of the limb was studied by scanning electron microscopy. In other experiments, the experimental and control limbs were examined histologically following fixation in 3% glutaraldehyde and Araldite embedding.

Affi-Gel blue (Bio-Rad) or heparin acrylic beads (Sigma) were employed as carriers for administration of the selected proteins. Beads ranging between 80 and 150 μm in diameter were selected, washed in PBS and incubated for 1 hour at room temperature in the selected protein solution. Recombinant human TGFβ1 (R&D Systems) was used at 2 μg/ml; recombinant human FGF-2 (R&D Systems) at 1 mg/ml; recombinant human Noggin (Regeneron Pharm Inc. Tarrytown, N.Y.) at 1 mg/ml; recombinant human activin A (obtained through the National Hormone and Pituitary Program, NHPP, of the National Institute of Diabetes and Digestive and Kidney Diseases, NIDDK) at 0.73 mg/ml. Activins B and AB (obtained from Dr S. Choe, Salk Institute) were used at 0.1 mg/ml and 0.7 mg/ml respectively. Follistatin (rhFS-288, obtained through NHPP, NIDDK) was used at 0.1 mg/ml.

Specific probes for bmpR-1b (BRK-2), actRIIa and actRIIb were provided by T. Nohno. Chicken tgfβ2 and gdf-5 probes have been described previously (Merino et al., 1998 and 1999). Fragments of chicken activin βA (733bp), activin βB (502bp) and follistatin (638bp) genes were obtained by RT-PCR. First strand cDNA was synthesized with a mixture of random hexamers (Promega) and 1 μg of total RNA from day 7.5 autopod. The following primers (5′ to 3′) were used: for chicken activin βA gene 5′ primer 5′-CCTGAGATGGTGGAAGCA-GTA-3′ and 3′ primer 5′-GCAAGCATCATCAGGAAG-3′; for chicken activin βB gene, 5′ primer 5′-CGGACACTAGCAAC-AAGTGG-3′ and 3′ primer 5′-ATGCAACAGGAGTTCACGG-3′; for chicken follistatin gene, 5′ primer 5′-ATGGAAGATCACACAG-CGC-3′ and 3′ primer 5′-GCTAATCCAAAGGATCTGCC-3′. PCR reactions were performed in a total volume of 100 μl using Taq DNA polymerase (Gibco BRL). The cycling conditions were 1 minute at 94°C for denaturation, 2 minutes at 55°C for annealing, 3 minutes at 72°C for elongation, and then 10 minutes at 72°C after the last cycle (35 cycles). The PCR products were subsequently cloned into pGEM-T (Promega) and the authenticity of the fragments was confirmed by dideoxy sequencing.

In situ hybridization of control and treated limbs was performed in whole-mount specimens and in tissue sections. For whole-mount in situ hybridization, samples were treated according to their size and stage of development with 10 to 30 μg/ml of proteinase K for 30 minutes at 20°C. Hybridization with digoxigenin-labeled antisense RNA probes was performed at 68°C. Reactions were developed with BCIP/NBT substrate or with purple AP substrate (Boehringer-Mannheim). In situ hybridization in tissue sections was performed using digoxigenin-labeled antisense RNA probes as described by Zou et al. (1997). Specificity of labeling was controlled using sense RNA probes.

A mouse Wnt3a cDNA encoding the entire open reading frame was cloned into the replication-competent retroviral vector RCAS (Kengaku et al., 1998). Primary chick embryonic fibroblasts were then transfected and the supernatant was collected, concentrated by ultracentrifugation and microinjected in stages 9-11 wing or leg bud primordia.

All the studied members of the activin signalling pathway show precise temporo-spatial patterns of expression in the developing autopod suggesting a role of this signalling pathway in the morphogenesis and differentiation of the skeletal elements of the autopod. Expression of these components is also observed in the differentiating muscles.

Expression of activin βA subunit is a precocious marker of the digital rays and interphalangeal joints. βA transcripts start to appear at stages 25-26. As shown in Fig. 1A-D, βA transcripts mark the zone of formation of the prechondrogenic digital aggregates first in digits IV and III, and then in digits II and I (Fig. 1E). In the course of development, and as digits incorporate successive phalangeal elements, βA expression is maintained at the tip of the digits. However, it is absent at more proximal levels except in the zones of joint formation where βA transcripts also form prominent domains of expression (Fig. 1E-F). The appearance of βA transcripts in the developing joints is coincident with the expression of gdf-5 (Fig. 1F-G) which is an early marker of the metatarso-phalangeal and interphalangeal joint regions (Storm and Kingsley, 1996; Merino et al., 1999). As limb development proceeds and digits complete the incorporation of phalangeal elements, βA transcripts disappear from their distal tip. Thus, digits I and II lose their distal labeling at stages 32-33 (Fig. 1H) and digit III and IV at stages 33-34. During this period βA transcripts remain associated with the interphalangeal joints (Fig. 1H).

Expression of activin βB subunit showed a spatial pattern of expression similar to that of βA, but its temporal appearance was different (Fig. 1I-J). Up to stage 30, βB transcripts were not detected in the autopod. From stage 30, βB transcripts appeared associated with the zones of joint formation (Fig 1I). The appearance of these joint domains followed the same sequence than that observed for βA. However, they were always detected some hours after the appearance of βA transcripts. Expression of βB transcripts at the condensing tip of the digits was nearly correspondent with the period in which the digits stop distal outgrowth, thus showing a sequence of appearance opposite to that of βA. The first digital tip showing βB transcripts was digit I at stage 31 (Fig. 1I), followed by digit II and the tips of digits III and IV lacked βB transcripts up to stage 32 (Fig. 1J).

While transcripts of both actRIIa and actRIIb are associated with the zones of expression of activin β subunits, they also show additional domains of expression in the differentiating cartilage of the phalanges. ActRIIa is mainly expressed in the developing muscles. However, at stage 28, low expression levels are also detected in the undifferentiated mesenchyme surrounding the digital rays (Fig. 1K-M). Well defined domains of expression of this receptor in the skeletal elements of the autopod are only detected in the diaphysis of the skeletal pieces as they initiate differentiation into hypertrophic cartilage (Fig. 1M). At these relatively advanced stages, actRIIa transcripts are also intense in the peripheral mesenchyme of the joints when they initiate the process of cavitation (Fig. 1M). An additional region of expression of actRIIa is the myotendinous junction of the autopodial tendons (not shown).

ActRIIb, as observed for actRIIa, is expressed in the developing muscles (not shown) and in later stages in the hypertrophic cartilage and in the developing joints in the course of cavitation. However, unlike actRIIa, actRIIb exhibits prominent domains of expression in the stages of digit formation in the mesenchyme surrounding the distal tip of the digit and the zone of joint formation (Fig. 1N-Q). At stage 28, actRIIb expression is low and appears concentrated around the condensing tip of the developing digits (Fig. 1N). From this stage, the expression domains associated with the distal tip of the digits are more intense and surround the precartilaginous condensation (Fig. 1O-Q). Well defined domains of actRIIb expression are also detected in these stages in the mesenchyme around the zones of joint formation (Fig. 1P-Q). From stages 31-32, expression is intense in the hypertrophic cartilage (Fig. 1R) and in the mesenchyme of the joint interzone (not shown). At these stages, expression of actRIIb is also detectable at low levels in the boundary of the autopodial tendons (not shown).

To analyze the function of activins in the formation of the limb skeleton, beads incubated in recombinant activins were implanted into the limb bud at different stages and positions. Implantation of activin A-beads at stages 22 to 24 was not followed by skeletal alterations (n=22). These experiments were performed both in wing and leg buds, and the beads were implanted in the anterior margin, posterior margin and progress zone mesoderm. From stages 25-26, the beads were implanted in the progress zone mesoderm corresponding to the interdigital regions. In all cases the treated limb exhibited a stage dependent pattern of skeletal alterations. Implantation of activin A-beads at stages 25 or 26 was followed by excrescences in the metatarsal cartilages which, in the most severe cases, caused the fusion of two adjacent skeletal pieces (Fig. 2A; n=6). When the beads were implanted at stage 27, cartilage excrescences were also formed but they were associated with the proximal phalanges (n=7). When the beads were implanted at stages 28 or 29, an extra digit consisting of two or three phalangeal elements was formed in 23 out of 29 treated embryos (Fig. 2B) and ectopic chondrogenesis lacking an identifiable morphology was observed in 5 of the remaining 6 embryos. At stage 30, in the majority of the cases, activin A-beads induced a small ectopic cartilage lacking digit morphology (n=8). By stage 31, when the interdigital mesoderm begins to be removed by programmed cell death, ectopic chondrogenesis was only found in 2 out of 5 experimental embryos.

Since the digital expression of the activin βA and βB subunits shows differences indicative of the presence of activin A at the beginning of digit formation and/or activins AB or B at the end of digit formation, we analyzed the effect of local administration of these activins in the interdigital region. No significant differences in the ability to induce ectopic chondrogenesis were detected between activin AB or B and activin A. Ectopic chondrogenesis or extra digit formation was observed in 12 out of 16 embryos following implantation of activin B-beads in the interdigital region at stages 28-29, and in 6 out of 8 embryos after implantation of activin AB-beads.

In view of the similarities in the chondrogenic inducing effects among the different activins and between activins and TGFβs (Gañan et al., 1996) we analyzed possible molecular differences preceding the appearance of cartilages following the application of these growth factors (activins or TGFβs). In these experiments, we analyzed the induction of bmpR-1b and activin βA and tgfβ2 genes themselves. BMPR-1b is a type I receptor for BMPs. Previous retrovirus-mediated overexpression experiments using either dominant negative (Zou and Niswander, 1996) or constitutive active constructs (Zou et al., 1997) indicated that this receptor was the transducer of the chondrogenic signals in the limb bud. Our findings showed that the implantation in the interdigital space of beads incubated with either TGFβ1 (n=30), activin A (n=34), activin B (n=28), or activin AB (n=20) was followed by the ectopic expression of the above mentioned genes in a similar temporal pattern (Fig. 3A-C). Ectopic expression of bmpR-1b was detected between 10 and 15 hours after implantation of beads incubated with activins (Fig. 3A) or TGFβ1. Ectopic expression of activin βA (Fig. 3B) and tgfβ2 (Fig. 3C) occurred between 15 and 20 hours following the implantation of the various beads. This suggests that activation of these genes occupied a subsequent step in the molecular cascade triggered by the application of TGFβ1-or activin-beads. In accordance with this interpretation, implantation of beads incubated with the BMP antagonist Noggin at the tip of the digits downregulated the expression of activin βA (Fig. 3D; n=8). A similar finding was previously observed for the tgfβ2 gene (Merino et al., 1998).

Our present findings, along with previous studies on the role of TGFβ2 in digit formation (Gañan et al., 1996; Merino et al., 1998), showed that βA, tgfβ2 and βB transcripts were progressively and additionally expressed at the tip of the digits and, further, that the proteins encoded by those genes were able to induce extra digits when they were exogenously administered into the interdigital mesoderm. To analyze possible functional implications of these redundant chondrogenic signals, beads bearing activins or TGFβ at the concentration required to induce extra digits were implanted at the tip of the developing digits between stages 27-29.

Implantation of TGFβ-beads was followed by truncation of the digits at proximodistal levels dependent on the stage of treatment. However, regardless of the level of truncation, in 11 out of 21 experimental limbs the distal phalanx of the truncated digits exhibited the characteristic claw morphology of the distal phalanges (Fig. 2D). The remaining truncated digits were characterized by the absence of a number of phalanges but the tip of the digit was rounded. Histological examination of the treated limbs showed that truncation was due neither to abnormal cell death nor to the disintegration of the AER (not shown).

Treatments with activins caused minor phalangeal alterations, including joint fusion and shortening or thickening of the phalanx that the bead was associated with (not shown). However, at difference of TGFβ-treatments, truncations were unusual (2 out of 6 experimental embryos treated with activin A; 3 out of 9 embryos with activin B; and 0 out of 10 embryos with activin AB) and when present the distal phalangeal element showed a rounded tip rather than the claw morphology observed after TGFβ treatment.

Since the expression of activin βA (which is the first molecular marker of the digits) appeared first in the most posterior digits, we explored whether ZPA influences its expression. To test this, a small fragment of stage 22 limb mesoderm containing the ZPA was grafted in the anterior margin of the wing bud of stage 20 host embryos. This operation resulted in the induction of extra anterior domains of the βA gene, which were detectable 48 hours after the ZPA graft, showing a clear association with the mirror digit duplication induced by the graft (Fig. 4A).

In previous studies we observed that extra digits can be induced by surgical removal of the AER from the interdigital regions (Hurle and Gañan, 1987). This finding suggested a role of AER in inhibiting the expression of chondrogenic signals in the subjacent mesoderm. To test this hypothesis, AER was removed from the interdigital regions at stage 28 (n=14), which induced ectopic tgfβ2 (Fig. 4B), that was followed by a small domain of activin βA (not shown) that preceded the appearance of the ectopic cartilage. In addition, surgical removal of the AER from the tip of digit III at stage 28 (n=12) caused an initial upregulation of βA activin expression at the tip of the treated digit (Fig. 1C) followed by intense downregulation detectable 24-30 hours thereafter (Fig. 4D). However, since these findings could be related to a local production of TGFβs in the course of healing of the ectodermal wound (Martin et al., 1993), complementary experiments to increase AER function were also peformed.

The effects of increasing AER function were initially analyzed by implanting beads bearing FGF-2 at the tip of the growing digits (n=6). Under these conditions, undifferentiated cells lacking activin βA expression accumulated at the tip of the digit and its physiological expression remained restricted proximally in association with the previously formed digit condensation (Fig. 4E). To further analyze this antagonistic effect of FGFs, activin βA expression and digit formation were studied in limbs in which the function of the AER was experimentally increased by retrovirus-mediated misexpression of the Wnt3a gene (n=16). Under these conditions non-ridge limb ectoderm is induced to express functionally active AER genes leading to the formation of limbs characterized by a hypertrophied AER function (Kengaku et al., 1998). In the most severe phenotypes, limbs misexpressing Wnt3a did not form digital cartilages and were characterized by a narrow autopod lacking activin βA expression (Fig. 4F-I).

The possible function of follistatin in modulating activin signalling moved us to analyze whether it was expressed during the development of the autopod. Expression of follistatin in early stages of limb development has been studied by Amthor et al., (1996). At these early stages (prior to stage 23), follistatin transcripts are associated with the developing somites and with the myogenic cells invading the limb bud. From stages 25-26, in addition to the myogenic masses of the zeugopodial region, follistatin transcripts are also observed in the autopod associated with the zones of formation of the digital rays (Fig. 5A-D). These digital domains of follistatin expression are arranged dorsally and ventrally to the differentiating cartilage marking the future regions of tendon formation (Fig. 5B-D), and are particularly abundant at the distal tip of the digit close to the zone of expression of activin βA transcripts (Fig. 5D). The appearance of the different domains of follistatin expression follows the same temporal sequence as digit formation, appearing first associated with digits IV and III (stages 25-26) and then with digits II and I (stages 28-29). Up to stages 32-33, expression of follistatin is most intense in the distal region of the digit dorsally and ventrally to the zone of expression of activin βA, but considerable labeling is also present along the entire digital trajet of the tendons (Fig. 5D). From these stages, follistatin transcripts are concentrated in the mesenchyme surrounding the tendinous blastema, in the outer layer of the perichondrium and less intensively in the mesenchyme of the joint interzone (Fig. 5E-F).

In view of the temporal and spatial coincidence in the expression of follistatin and activin βA subunit at the distal growing tip of the digits, we explored a possible causal relationship between both genes by analyzing the effect of exogenous administration of activin A on the expression of follistatin (n=7). In these experiments ectopic expression of follistatin was detected following implantation of activin A-beads in the interdigital regions. This domain of follistatin expression was first detectable 20 or 24 hours after the treatment when a mesodermal condensation was apparent around the bead (not shown), and expression was mantained in association with the extra digits (Fig. 5G).

Follistatin treatments were designed to block the function of activin A. Implantation of follistatin-beads into the limb bud (anterior margin, posterior margin or progress zone mesoderm) prior to stage 25 was not followed by alterations in limb skeletogenesis (n=29). From stage 25 up to stage 30, which corresponds to the period in which βA transcripts are expressed in the digital rays, implantation of follistatin-beads at the tip of the growing digits was followed by intense inhibition of chondrogenesis. Treatments at stages 25-26 inhibited, in all cases (n=13), the formation of the metatarsal cartilage but digit formation was often restored at the level of the phalangeal elements (Fig. 2E-G). An interesting feature of these autopods lacking the metatarsal cartilage of digit III, was that, in most cases (9 out of 13), the phalangeal pieces of digit III became aligned with the metatarsal of digit II (Fig. 2F-G) leading to partial or total truncation of digit II (Fig. 2G). From stage 27 implantation of follistatin-beads at the tip of the growing digits resulted in the formation of truncated digits lacking all the phalangeal pieces that should be formed distally to the bead (Fig. 2H).

Chondrogenesis induced by activin A was inhibited in all cases when the activin A-beads were postincubated in follistatin prior to their implantation in the interdigits (n=11). Similarly, activin-A-induced chondrogenesis was inhibited, although to a lesser extent (5 out of 8 cases), when a bead of follistatin was co-implanted with a bead of activin A (not shown). To exclude the possibility that the inhibition of chondrogenesis by follistatin was not specifically due to its antagonistic effect on activins, we tested the effect of follistatin on TGFβ-induced chondrogenesis. For this purpose two beads, one incubated in TGFβ1 and one incubated in follistatin, were implanted in the same interdigit (n=14). In these experiments, ectopic chondrogenesis was never abolished, although the extra digits observed following this double treatment were in general thinner and smaller than those induced by TGFβs alone. Interestingly, joints were in most cases absent from these extra digits (Fig. 2C).

Implantation of follistatin-beads in the interdigital regions (n=20) was followed by normal development and interdigital cell death was not modified even when two beads were implanted in the same interdigit within a 12-hour interval (not shown).

Previous studies have reported the presence of activin βA transcripts in mouse limb prechondrogenic condensations (Feijen et al., 1994). Here, we show that the expression of βA in the limb chondrogenic condensations is restricted to the digits while it is absent from the condensations of more proximal skeletal elements. We further show that this gene constitutes the earliest marker for the digital condensations (metatarsal and phalanges). Its expression starts at stage 25 coincidentally with the formation of the metatarsal pieces and is maintained at the growing tip of the digits until the last phalanx is formed (stages 32-34). Expression of actRIIb is intense at these stages in the mesodermal cells surrounding the distal tip of the digit indicative of a role of this receptor in the transduction of the activin signals. The lack of βB transcripts in the digits until the formation of the last phalanx indicates that only activin A is present in the initial stages of digit formation. This pattern of gene expression, along with the chondrogenic promoting effect of activin A in limb mesoderm micromass cultures (Jiang et al., 1993), is strongly indicative of a role of activin A in the formation of the digits. The induction of ectopic digits observed here following implantation of beads bearing activin A into the interdigital limb mesoderm and the inhibition of digit formation after treatments with follistatin, which is a natural antagonist of activins (de Winter et al., 1996), are in full agreement with this interpretation. In addition, and most importantly, our findings provide evidence for a stage-dependent specificity of the chondrogenic function of activin A in the course of limb development. Prior to stage 25, treatments with activin A-beads are ineffective in promoting ectopic chondrogenesis. From this stage, and coincidentally with the presence of βA transcripts in the digital prechondrogenic aggregates and actRIIb in the mesenchyme surrounding the digital condensations, activin A-beads implanted in the interdigital regions induce the formation of extra digits or ectopic chondrogenesis. Conversely, implantation of follistatin-beads blocks digit growth but has no effect on the formation and growth of the proximal skeletal elements of the limb (zeugopod and stylopod). As for activin-mediated chondrogenesis, inhibition of chondrogenesis by follistatin begins at stage 25. In addition, the chondrogenic effect of activin A-beads is fully blocked when the beads are postincubated in follistatin prior to their implantation. Although follistatin, in addition to antagonizing activins (de Winter et al., 1996), is also able to neutralize some BMPs (Fainsod et al., 1997; Iemura et al., 1998), it is unlikely that the inhibitory effect of follistatin in the formation of normal and activin-induced digits is only explained by its potential to antagonize BMPs. follistatin does not block chondrogenesis induced by treatment with TGFβs, which also involves BMP-signalling (Merino et al., 1998), and in contrast to other specific BMP-antagonists such as Noggin (Merino et al., 1998) or Gremlin (unpublished observations), does not inhibit interdigital cell death, a biological process present in the autopod at these stages which is also mediated by BMPs (Zou and Niswander, 1996; Yokouchi et al., 1996; Macias et al., 1997).

Previous in vitro studies indicate that the function of activin A in chondrogenesis is to recruit cells to form the chondrogenic condensation by increasing cell adhesion (Jiang et al., 1993). As will be discussed below, our study shows that the role of activin A in digit chondrogenesis is also to induce the expression of bmpR-1b, thus making the cells sensitive to the chondrogenic influence of BMPs (Zou et al., 1997).

An additional finding of this study is that the progressive incorporation of phalangeal elements into the digits is accompanied by the progressive expression of genes encoding factors with chondrogenic potential. In a previous study, we observed that tgfβ2 is expressed at stage 29 when the distal part of the digits are being formed (Merino et al., 1998), and TGFβs administered exogenously are able to induce extra digits (Gañan et al., 1996). Here, we show that coincidentally with the appearance of the condensation of the last phalanx of each digital ray, activin βB is also expressed at the tip of the digits, suggesting that activins AB and/or B are also involved in this advanced period of digit formation. Again, our functional approach indicates that both activin B and activin AB are able to induce extra digits. Thus, the progressive establishment of the digit skeleton involves the successive addition of activin A, TGFβ2 and activin B and/or AB all of them with strong chondrogenic potential. This redundancy in chondrogenic signals may explain why the experimental disruption of tgfβ2 (Sanford et al., 1997) or activin genes (Matzuk et al., 1995b) in the mouse does not result in a mutant digit phenotype. A comparable degree of redundancy is also observed in the control of the proximodistal outgrowth of the limb which is mediated by at least 4 different FGF isoforms (FGF-2, FGF-4, FGF-8 and FGF-9) produced by the AER (see review by Martin, 1998) and in BMP signalling associated with the control of proliferation, differentiation and cell death of the progress zone mesoderm (Niswander and Martin, 1993; Macias et al., 1997) which includes at least three different BMPs (BMP-2, BMP-4 and BMP-7; Francis-West et al., 1995; Laufer et al., 1997).

The occurrence of truncated digits in which the last phalanx, in spite of its proximal location, displays a distal claw morphology, which is observed after application of TGFβ at the tip of the digit, suggests that the progressive addition of chondrogenic signals might contribute to the distal determination of the phalanges. Also remarkable in this study is the specificity of activin/TGFβ signalling in the formation of digits. This indicates that other signals may account for the onset of chondrogenesis at the zeugopod and the stylopod segments of the limb. From this observation, it is tempting to propose that changes in the nature of the chondrogenic signals may lead to the formation of morphologically different prechondrogenic aggregates, thus contributing to the establishment of differences in the skeletal morphology of the different segments of the limb. This may establish a link between the early patterning of the limb skeleton by homeobox genes and the morphology of the skeleton. The mesoderm cells responding to activin/TGFβ signalling corresponding to the domains of expression of hoxA13 and 5′hoxD genes (Nelson et al., 1996). In addition, digit duplication following the grafting of ZPA in the anterior margin of the limb (which is accompanied by rearrangement in the domain of expression of hoxd genes; Izpisua-Belmonte and Duboule, 1992) is followed by the anterior duplication of the digital domains of expression of activin βA subunit. Also, in support of that interpretation are recent studies reporting differences in the chondrogenic behaviour of the limb mesoderm in relation with the pattern of hoxd gene expression observed in vitro (Jung and Tsonis, 1998) and in retrovirus-induced misexpression experiments (Goff and Tabin, 1997).

Previous work has established that BMPs (BMP-2, BMP-4 and BMP-7) play a double role in the developing autopod promoting cartilage growth and interdigital cell death (Zou and Niswander, 1996; Macias et al., 1997; Capdevila and Johnson, 1998; Merino et al., 1998). Signalling through the type I receptor BMPR-1B appears to account for the chondrogenic effects of BMPs (Kawakami et al., 1996; Zou et al., 1997; Merino et al., 1998). Here we show that, as observed previously for TGFβs (Merino et al., 1998), the formation of extra digits following treatment with activins is preceded by the induction of an ectopic domain of bmpR-1b. This expression is followed by the appearance of activin βA and tgfβ2 transcripts either in treatments with TGFβ or with activins. In addition, treatment with the BMP antagonist Noggin downregulates the physiological expression of βA at the tip of the digits, a feature also observed previously for tgfβ2 (Merino et al., 1998). These findings indicate the occurrence of a complex cascade of molecular events accounting for the formation of digits which is triggered by the activation of bmpR-1b. All the tested growth factors (activins A, B, and AB, and TGFβs) are able to induce bmpR-1b, but the pattern of expression of these factors in vivo points to activin A as the physiological signal for this function. The activation of bmpR-1b makes the cell sensitive to the chondrogenic effect of local BMPs, thus blocking its alternative function promoting cell death. In addition, signalling through BMPR-1b positively regulates the expression of tgfβ2 and activin βA establishing a loop which ensures the progressive outgrowth of the digits. The appearance of βB transcripts in the induced cartilages requires longer periods, suggesting that other mechanisms account for the regulation of activins B and/or AB.

Limb outgrowth is sustained by proliferation of the mesodermal cells in the progress zone. In this region, the mesodermal cells are maintained in an undifferentiated and proliferating state due to the influence of FGFs produced by the AER (see review by Martin, 1998). The above results indicate that activin A signalling triggers differentiation of the progress zone cells when they leave the influence of the AER. Our experimental designs to analyze possible interactions between activins, AER and FGFs clearly shows that FGFs play an opposite but complementary function to activins. Application of FGFs at the tip of the digits downregulates βA expression and delays chondrogenesis. Most remarkably, expression of activin βA subunit and digit chondrogenesis are both fully inhibited in limbs overexpressing Wnt3a. Previously, it has been demonstrated that Wnt3a is an inductor of the AER and its misexpression causes a hyperactive ridge with fgf-8 being expressed at the limb bud margin and in a considerable number of cells of the dorsal and ventral ectoderm (Kengaku et al., 1998). Changes of βA expression following local removal of the AER from the tip of the digits illustrates the complementary roles of FGFs and activins. A few hours after AER removal, βA is upregulated, thus showing the antagonistic effect of FGFs and activin. However, by 24 hours after AER removal, βA expression undergoes progressive downregulation which can be explained by the arrest in the incorporation of cells coming from the progress zone caused by AER removal.

The formation of joints is a complex process which involves at least three successive steps: first, chondrogenic differentiation is arrested in the zone of joint formation; second, cell death removes mesenchymal cells from the joint interzone; and third, differentiation of the capsulo-synovial apparatus including vascular invasion (Mitrovic, 1977; Mori et al., 1995). Several signalling molecules, including GDF-5 (Storm and Kinsley, 1996; Brunet et al., 1998, Merino et al., 1999), BMP-2 (Macias et al., 1997), TGFβ2 (Merino et al., 1998), and BMP-6 (unpublished observations) have been identified in the developing joints of the digits, but their precise function in the different steps of joint formation awaits clarification. Our present observations show that activin βA and βB subunits and actRIIa and actRIIb exhibit a regulated pattern of expression in the different steps of joint formation indicative of a function of activin signalling in digit arthrogenesis. In this study, the absence of joints in TGFβ-induced extra digits after treatment with follistatin is the only functional evidence for a role of activins in joint formation.

We have observed that follistatin gene expression constitutes a precocious marker of tendon formation. In addition, although activins are not detected in the early developing tendons, maximal expression of follistatin is always observed at the distal tip of the digits close to the domain of activin expression, and ectopic administration of activins induces follistatin expression followed by tendon formation. Thus, our finding may be indicative of a molecular coordination between the formation of the skeletal primordia and their corresponding tendons. However, at the present we can only speculate about the possible role of follistatin in tendon formation. Since the tendinous blastemas are composed of cells able to chondrify, (Milz et al., 1998) it can be proposed that follistatin avoids the excessive extension of the prechondrogenic aggregates thus establishing the limit between tendon and cartilage. Alternatively, the presence of follistatin in the tendinous primordia may be related to the BMP-antagonic effect of follistatin (Fainsod et al., 1997; Iemura et al., 1998).

Sonia Pérez-Mantecón is acknowledged for technical assistance. Our thanks to L. Niswander for bmpR-1b probe, to T. Nohno for actRIIa, actRIIb and brk-2 probes and to S. Choe for activins B and AB. Activin A and follistatin were obtained through NHPP, NIDDK, NICHHD, USDA. This work was supported by grants from the DGICYT to J. M. H. (PM95-0090) and Y. G. (PM96-0020) and from Fundación Marqués de Valdecilla to J. M. H. R. M. is the recipient of a grant from the Spanish ‘Ministerio de Educación y Cultura’ (Reincorporación de doctores y tecnólogos). J. C. I.-B. is the recipient of a grant from the Fundación BBV.