Molecular approaches to vertebrate limb morphogenesis

A dominant force in embryonic development is the coordinate expression of the genetic program (Davidson, 1986). For example, early insect pattern formation is orchestrated at least in part by a network of transcriptional regulators, i.e. nuclear proteins that act in a cell-autonomous fashion (Akam et al. 1988; Ingham, 1988). However, cells in an embryo do not necessarily behave autonomously but are influenced by a variety of extracellular signals, some of which act over short distances (cell-matrix interactions) while others are long-range signals acting over many cell diameters. One can guess that long-range signals are especially helpful if embryos consist of a large number of cells that have to develop in a coordinate fashion (Wolpert, 1969; Crick, 1970; Meinhardt, 1982). Well-known examples of extracellular signals that can influence cell fate and cell differentiation are growth factors that interact with cell surface receptors (e.g. Smith, 1989), and small molecule hormones (e.g. steroids, thyroid hormones) that bind to specific nuclear receptor proteins (e.g. Evans, 1988; Beato, 1989).

Is there any evidence that small molecule substances take part in pattern formation? Saunders and Gasseling (1968) discovered that transplanting posterior chick wing bud mesenchyme to the anterior margin of a host wing bud results in a mirror-image symmetrical duplication of the host’s hand plate. Instead of the normal pattern with digits 2, 3, and 4, a 432234 pattern will frequently develop (Fig. 1). The tissue capable of inducing duplications is known as zone of polarizing activity (ZPA) or polarizing region. ZPA activity is not species specific as it is found in all amniotes so far examined (reviewed in Tickle, 1980). Moreover, selected other embryonic tissues such as ventral (but not dorsal) tail bud mesenchyme (Saunders and Gasseling, 1983) and Hensen’s node (Hornbruch and Wolpert, 1986) induce duplications when grafted into the chick wing or leg bud. One interpretation of the fact that other tissues are biologically active in the duplication assay is that the ZPA is a non-specific inducer (see e.g. Saunders and Gasseling, 1983). However, an equally valid argument is that at a molecular and cellular level, the mechanism of pattern formation in different parts of the embryo are related.

Wolpert and colleagues (e.g. Tickle et al. 1975) have suggested that the ZPA releases a diffusible, labile morphogen that spreads across the limb bud and thereby forms a concentration gradient. The character of each digit would be specified by the concentration of the morphogen. The model of a diffusible small molecule signalling compound has been contested. Iten and colleagues, for example, have proposed that the limb pattern is formed as a consequence of local interactions between cells (see e.g. Javois, 1984; Bryant and Muneoka, 1986). Oster et al. (1985) put forward a model of limb morphogenesis that is based on a combination of matrix deswelling and mechanical forces between cells. However, when these models were proposed little was known about their cellular and molecular basis. Therefore, it is essential to find the ZPA morphogen or to identify the specific molecules that can account for the postulated pattern generating cell-cell interactions.

Is there a rationale to identify a morphogen molecule? The ZPA has a volume ⩽0.1 μl. Assuming a morphogen concentration of 20 nM and a molecular weight of the morphogen of 300, then 1000 ZPAs would contain less than one nanogram of morphogen. Without prior knowledge of its chemical properties the morphogen would be extremely difficult to identify. Not discouraged by these considerations, Alberts and Tickle as early as 1976 began to search for the putative ZPA morphogen (Bruce M. Alberts, personal communication). They believed it to be a hydrophobic substance because a hydrophobic molecule could readily diffuse laterally in the plasma membrane and thus move from one cell to the next without having to transverse the voluminous, hydrophilic extracellular matrix and also to risk entering a capillary and be removed from the limb bud. Alberts and Tickle prepared organic solvent extracts of ZPA tissue. The extracts were absorbed onto small inert beads that were then implanted like ZPA to the anterior wing bud margin (Bruce M. Alberts, personal communication). In addition, Alberts and Tickle impregnated beads with various ‘off the shelf compounds such as dibutyryl cyclic AMP or thalidomide and implanted the beads into buds. None of these efforts led to altered limb patterns, yet as we will see below the reasoning and strategy were basically correct.

In the early eighties. Tickle and colleagues (1982) and also Summerbell (1983) discovered that retinoic-acid-impregnated pieces of DEAE paper, newsprint or ion exchange beads induce pattern duplications when implanted at the anterior wing bud margin (Fig. 1). Subsequent work showed that in addition to inducing morphologically identical duplications, ZPA grafts and retinoic acid treatment display a very similar dose-, time- and position-dependence (Table 1). This similarity led to the obvious question of whether retinoic acid is the morphogen proposed by the above gradient model. Alternatively, might retinoic acid simply convert cells surrounding the implant into ZPA, which in turn provides the actual signal? There is in fact some support for this suggestion, because tissue next to the retinoic-acid-releasing implant can give rise to duplications when grafted into a host bud (Summerbell and Harvey, 1983). As pointed out by Tickle et al. (1985), the two possibilities are not necessarily mutually exclusive. It could be that retinoic acid released from the implant indeed generates ZPA but that at the same time this induced ZPA also synthesizes and releases retinoic acid. To resolve these issues it is important to determine whether ZPA cells can synthesize retinoic acid. A first step in this direction has been the demonstration that limb buds can synthesize retinoic acid in situ from its biosynthetic precursor retinol (Thaller and Eichele, 1988). To demonstrate in situ metabolism, beads impregnated with [³H]retinol were implanted at the posterior wing bud margin next to the ZPA. A number of metabolites were generated in this assay, including retinal (the intermediate between retinol and retinoic acid) and retinoic acid. It will now be important to compare the rate of conversion of retinol to retinoic acid in ZPA and non-ZPA tissue to find out whether the ZPA is special with regard to retinoic acid production.

A question of considerable interest is whether applied retinoic acid forms a concentration gradient across the limb bud or whether the blood circulating in the vascularized limb bud would prevent the establishment of a stable gradient. Initial studies performed with retinoic acid showed that retinoids can generate a concentration gradient in the limb bud (Tickle et al. 1985). Detailed analyses by Eichele and Thaller (1987) using the morphogenetically active synthetic retinoid TTNPB (E-4-[2-(5,6,7, 8-tetrahydro-5,5,8,8-tetramethyl-2-naphthalenyl)-l-propenyl]benzoic acid) showed that the distribution of applied retinoids along the anteroposterior axis was exponential, and that stable gradients can be set up (Fig. 2). However, if the bead was removed, the gradient collapsed within a few hours (Eichele et al. 1985). This observation implies that the gradient of applied retinoid (TTNPB or retinoic acid) is a steady-state gradient, i.e. its maintenance requires the continuous presence of a source that is balanced by a ‘sink’ in the form of clearance by blood circulation and by enzymatic degradation of the retinoid.

The diffusion coefficient (D) of 1TNPB in the limb bud is about 10⁻⁷cm²s^-1 (Eichele and Thaller, 1987). This value for D suggests that retinoids are not freely diffusible, but interact with cellular retinoic-acid-binding protein that is found in the limb bud (Kwarta et al. 1985; Maden and Summerbell, 1986). Knowing D affords an estimate of the time required to establish a diffusion gradient as 3 to 4 hours. This time span is in a range compatible with the time scale of pattern specification in developing vertebrate limbs. The main conclusion is that retinoids provided from a local source such as a bead or the ZPA can readily set up a diffusion gradient in the limb bud, but that it is necessary to maintain the source, otherwise the gradient will dissipate. In a broader sense, these studies demonstrate that diffusion gradients of hydrophobic substances are feasible in tissues.

The seminal discovery of the effect of retinoic acid on limb morphogenesis prompted the obvious question of whether limb buds contain endogenous retinoic acid. To find out, Thaller and Eichele (1987) extracted homogenates of large numbers of limb buds with organic solvent mixtures. The extracts were analyzed by high-performance liquid chromatography (HPLC). These analyses clearly demonstrated the presence of all-trans-retinol, all-trans-retinoic acid, and all-trans-retinal, as well as approximately 6 additional retinoids whose identities are currently being determined (Thaller and Eichele, unpublished observations). A limb bud at Hamburger-Hamilton stage 21, a stage when applied retinoic acid induces extra digits, contains about 6.5 pg of endogenous retinoic acid, corresponding to a mean tissue concentration of 25 nM. This is close to the concentration needed in the bud tissue to induce a full set of additional digits (20-30 nM). Hence physiological doses of applied retinoic acid induce duplications. The tissue level of all-trans-retinol at stage 21 is approximately 600 nM and that of all-trans-retinal about 10 nM. It is important to realize that the concentrations given here are total retinoid concentrations. A substantial fraction of each retinoid is probably specifically bound to protein (see below), but extraction with organic solvents will denature these proteins and release the bound ligand.

To examine whether retinoic acid is enriched in the posterior region, as one would expect if retinoic acid is the morphogen released by the ZPA, limb buds were dissected into a smaller posterior portion containing the ZPA and a larger ZPA-free anterior piece (see Fig. 3, insert). The concentration of retinoic acid in each of the two pieces was determined by HPLC and is shown in Fig. 3. It amounts to 50 nM posteriorly and 20 nM anteriorly. Hence there is 2.5 times more RA in the ZPA than in non-ZPA tissue. By contrast, retinol is almost uniformly partitioned at a concentration of approximately 600 nM. It goes without saying that retinoic acid will not be distributed in a step-wise fashion (Fig. 3), but since retinoic acid is a small molecule, it will spread across the limb bud in the form of a smooth gradient. We assume that this gradient resembles that of 1T NPB shown in Fig. 2. If we accept this line of reasoning, it becomes clear that the gradient of endogenous retinoic acid would span a concentration range of about half an order of magnitude. This implies that cells in the limb bud would have to sense relatively small concentration differences. That they are capable of doing this can be deduced from the dose-response analyses: it was found that to generate a pattern with an additional digit 4 requires about 5 to 10 times more retinoic acid or 1TNPB than is required to form a pattern with an extra digit 2 (Tickle et al. 1985; Eichele and Thaller, 1987). Possibly, sensing a shallow gradient requires some form of an amplification mechanism (de The et al. 1989). However, while important for the detailed mechanism of action, such a mechanistic issue should not detract from the main point that cells are able to interpret small changes in retinoic acid concentration.

How do cells ‘measure’ retinoic acid? This question touches upon a central yet poorly understood aspect of gradient models, that of thresholds (Slack, 1987). The idea is that a particular threshold concentration of a morphogen specifies a certain structure such as a digit 4 (high concentration) or a digit 2 (low concentration). If the concentration difference between the high and the low end of a gradient were 10000 fold, it would be easy to establish distinct threshold levels. However, the laws of diffusion preclude gradients of small molecules to be so steep, at least over the dimension of a limb bud. One is faced then with the problem of how a shallow gradient can be read and interpreted in such a way as to specify a sequence of digits. The burden of this task is most likely put on the ‘recording mechanism’, much in the same way as the measurement of weak signals, e.g. light from a distant star, depends on a sophisticated recording device.

A critical element of this recording mechanism almost certainly is the retinoic acid receptor which has molecular properties suitable for the measurement and interpretation of a gradient. So far, three distinct receptors for retinoic acid (RAR) have been reported. They are known as a retinoic acid receptor (α-RAR), β retinoic acid receptor (βRAR) (Guiguère et al. 1987; Petkovich et al. 1987; Benbrook et al. 1988; Brand et al. 1988) and y retinoic acid receptor (γRAR) (Krust et al. 1989; Zelent et al. 1989). The three RARs are encoded on separate genes, but their sequences reveal that they are highly homologous and that they belong to a multigene family that includes the receptors for steroid hormones, thyroid hormones and vitamin D₃. It is well established that these receptors are transcription factors that upon ligand binding are targeted to specific regulatory sequences in the 5′ region of ligand-controlled genes (e.g. Evans, 1988; Beato, 1989). As a consequence of receptor binding, the target gene is activated. Hence αRAR, βRAR and γRAR provide a direct link between the small molecule ligand and the expression of yet to be identified target genes (see also below). It is worth noting that hormone receptors can be very diverse, as best exemplified by the thyroid hormone receptor family (Sap et al. 1986; Weinberger et al. 1986), which displays alternative splicing (Izumo and Mahdavi, 1988; Hodin et al. 1989), opposite-strand transcription (Lazar et al. 1989; Miyajiama et al. 1989) and negative transcriptional regulation (Damm et al. 1989). It is possible that the receptors for retinoic acid exhibit a similar diversity. This would be valuable for a fine tuning of transcriptional regulation.

A major unresolved question is how the shallow gradient of retinoic acid (see Figs 2 and 3) can account for the formation of several distinct digits. The problem is particularly intriguing because the experimentally determined mean concentration of endogenous retinoic acid in the limb bud is about 20–30 nM, while the K_d for the steroid hormone receptor family (to which the receptor for retinoic acid belongs) typically ranges between 0.05 and 0.6 nM (Green et al. 1986; Weinberger et al. 1986; Sap et al. 1986; Benbrook and Pfahl, 1987; Dobson et al. 1989). The dilemma is that at such a high ligand concentration, all RAR molecules would be saturated with ligand, which precludes any form of regulation. If retinoic acid and its gradient have any role in pattern specification, then the concentration of free retinoic acid in the limb bud cells must somehow be lowered to a value in the range of the dissociation constant of RAR. The way to resolve this issue is to realize that limb buds contain much more cellular retinoic-acid-binding protein (CRABP) than RAR. Hence, most retinoic acid is actually bound to CRABP, resulting in a free retinoic acid concentration within the range of a K_d typical for a nuclear receptor. In what follows we will calculate the concentration of free retinoic acid in the cell under the assumption that the system is in equilibrium.

As can be seen in Fig. 3, the total retinoic acid concentration in the posterior quarter of the limb bud is 50 nM, and 20nM in the anterior three quarters. Fig. 2 shows that a retinoid locally applied from a bead will spread across the limb bud in the form of an exponential (see also Eichele and Thaller, 1987). It is reasonable to assume that endogenous retinoic acid distributes in a similar fashion and not stepwise as the experimentally determined distribution of Fig. 3 seems to suggest. The question is what concentration range the endogenous retinoic acid gradient spans. For the following order of magnitude calculations, we will use a gradient of 70 nM (posterior) to 15 nM (anterior). Comparison of Figs 2 and 3 suggests that these values are reasonable, although they are of course not experimentally determined.

How much of the total retinoic acid is inside cells? It is not known how retinoic acid is partitioned between the cell and the extracellular matrix. However, CRABP is abundant in cells, while no retinoid-binding proteins have been detected in the extracellular matrix. Hence, we will assume that most retinoic acid is cellular. CRABP and RAR are the two known cellular proteins that tightly bind retinoic acid (there are certainly metabolic enzymes that interact with retinoic acid, but they remain unidentified and probably are not abundant). Limb bud CRABP has been characterized, and has an apparent K_d of 2.0–2.2nM (Kwarta et al. 1985), similar to the K_d of CRABP that was isolated from testis (4nM; Ong and Chytil, 1978). An unusually high apparent K_d of 140–280 nM was reported for chick limb bud CRABP by Maden and Summerbell (1986); the reason for this discrepancy is unclear. We will use a K_d of 2nM. Both papers are in agreement on the CRABP content in limb bud: 25 pmol mg^-1 cytosolic protein (Kwarta et al. 1985) and 14–28 pmol mg^-1 cytosolic protein (Maden and Summberbell, 1986). Given that a stage 21 limb bud has a volume of 0.75 μl (Eichele and Thaller, 1987), a cytosolic protein content of about 15 μg per bud (C. Thaller, unpublished observation), and an average CRABP content of 20 pmol mg^-1, the estimated concentration of CRABP amounts to about 400nM. CRABP is not uniformly distributed across the chick limb bud (Maden et al. 1988). False color-image analysis of immunohistochemically stained CRABP in limb bud sections reveals that the CRABP gradient is about 3-fold, and in opposite direction to the gradient of ligand. These authors make the point that a CRABP gradient opposing that of retinoic acid steepens the gradient of free retinoic acid.

We can estimate the amount of bound and free retinoic acid, in the presence of CRABP and RAR as follows:

It seems reasonable to assume that the cellular concentration of RAR is similar to that of other nuclear receptors (3 to 7 nM; Koblinsky et al. 1972; Katzenellen-bogen et al. 1983). Because [CRABP_total]⪢ [RAR_tot] equation (la) can be simplified to

where [RA_total] is the experimentally determined concentration of retinoic acid depicted in Fig. 3. Moreover

Substituting equations (lb) and (2) in terms of [RA_frec] yields

Since

substituting (3) and (4) into (5) yields

Equation (6) can be rearranged to calculate [RAf,._ec], the unbound retinoic acid available for binding to RAR. Table 2 shows the results of such calculations for four scenarios: (1) no gradients (lines 1 and 2), (2) a 4.6-fold retinoic acid gradient (lines 3 and 4), (3) a 4.6-fold retinoic acid gradient and an opposite 3-fold CRABP gradient (lines 5 and 6), (4) a 4.6-fold retinoic acid gradient and a very steep CRABP gradient (lines 7 and 8). Two interesting conclusions can be drawn from Table 2:

1. A comparison of [RA_total] with [RA_free] clearly shows that, except for the scenario of line 7, most retinoic acid will be bound to CRABP. Hence, as qualitatively predicted above, the concentration of free retinoic acid is indeed in the range of the K_d of a nuclear receptor.

2. If CRABP is uniformly distributed, the gradient of free and total retinoic acid are similar (5.4- vs. 4.6-fold) and shallow. However, a merely 3-fold CRABP gradient in the opposite direction will give rise to a 20fold gradient of free retinoic acid. The striking conclusion from these first approximations is that the combination of ligand and binding protein puts the free retinoic acid concentration into such a range where moderate changes in ligand concentration that occur in space (Fig. 3) and time (Thaller and Eichele, in preparation) could result in substantial changes of receptor occupancy and a concomitant change of receptor-mediated transactivation.

These approximate calculations invite speculation on the role of CRABP in development. For example, a cell could regulate its retinoic acid sensitivity simply by altering CRABP expression, in addition to directly regulating retinoic acid concentrations. Thus, local domains of varying retinoid sensitivity can be created without having to alter the entire organism’s circulating retinoid levels. To think along these lines is especially attractive in view of the observations that retinoic acid seems to regulate cellular retinol-binding protein and perhaps CRABP expression as well (Kato et al. 1985). Thus a modest retinoic acid gradient can be converted to a potent signal, by interactions with CRABP and RAR.

An important advance in understanding the mechanisms of pattern formation was made possible by the discovery of the homeobox, a 180 bp motif that encodes the DNA-binding domain of a multigene family of transcriptional regulators (see Gehring, 1987; Herr et al. 1989 and references therein). Genes containing a homeobox were first discovered in the genome of Drosophila melanogaster (McGinnis et al. 1984a; Scott and Weiner, 1984). A combination of genetic studies with analyses of the spatiotemporal expression pattern in wild-type and mutant embryos has demonstrated that the orchestrated expression of this class of genes contributes in an important way to the specification of the body pattern in insects (reviewed e.g. in Akam et al. 1988; Ingham, 1988). McGinnis et al. (1984b) were the first to show that the homeobox is also present in the genomes of vertebrates. This raised the question of whether the high degree of conservation of gene structure implies conservation of corresponding functions in such disparate organisms as insects and mammals. The answer to this question is not easily forthcoming. In Drosophila, the spatial exp.ession pattern of homeobox genes can be correlated with subsequently formed structures, e.g. Ubx and Antp expression specifies metameric identities (e.g. Scott et al. 1983; Peifer et al. 1987). In vertebrates such correlations are not as straightforward to establish. Moreover, the most valuable insights in the case of Drosophila have come from comparisons of the expression patterns in wild-type embryos and in loss-of-function or null mutants, a route that is presently not accessible for vertebrate systems.

In the past four years, a number of laboratories have embarked on the strategy of spatially and temporally mapping the expression of various homeobox genes during embryonic development of several vertebrate organisms, most notably the mouse (reviewed e.g. by Dressier and Gruss, 1988 and Wright et al. 1989). These studies demonstrate that most homeobox genes are clustered in the genome, that they are regionally expressed and that their spatiotemporal expression pattern is developmentally regulated. Similar to the situation in Drosophila, the expression pattern along the anteroposterior axis reflects the position of the genes in the cluster (e.g. Duboule and Dollé, 1989; Graham et al. 1989). While it is attractive to think that homeobox genes define the organization of the vertebrate embryo in a similar way as they do in the fly, that is, by combinatorial patterns of expression, such a view needs much more experimental substantiation than is currently available. A potentially fruitful avenue towards understanding the role of homeobox genes in vertebrate pattern formation are interspecies comparisons of homolog expression patterns. Classical embryology has gained much from comparative studies and such an approach might also be fruitful at a molecular level. While this strategy is still relatively new, a few interesting observations and insights have emerged. In keeping with the general theme of this review, the data discussed below pertain mainly to the developing limb.

Development of the limb can be divided into two phases. During the first phase the limb bud forms by a bulging out of the embryonic flank. At this stage the bud consists of apparently uniform mesenchyme encased in a jacket of ectoderm. During the second phase the mesenchyme terminally differentiates into an intricate pattern of tissues such as cartilate, bone, muscle, dermis etc. In the chick, the first phase encompasses Hamburger-Hamilton stages 16 to 22/23, while the second phase goes from stage 23 onward. The corresponding stages in mouse are day 8.75 to day 10.75, and day 10.75 onward. Experimental manipulations in the chick that result in a change of pattern must be performed during the first phase to be effective. For example, the local application of retinoic acid yields digit duplications only if performed prior to Hamburger-Hamilton stage 22 (Summerbell, 1983). Thus it makes sense to assume that the global pattern of the limb is specified primarily during the first phase of development; hence gene expression studies aimed at pattern formation ought to focus on phase 1. In what follows we will briefly discuss recent studies that have revealed regionalized (i.e. non-uniform) expression of homeobox genes in limb buds of mouse, Xenopus and chicken.

Among the first studies to demonstrate regionalized expression of homeobox genes was that of Oliver et al. (1988; for a schematic illustration of the spatial expression pattern of this and the other homeoboxes, see Fig. 4). Using an antibody, they observed that in Xenopus embryos, XIHbox 1 protein is expressed in both ectoderm and mesenchyme in the forelimb, but in the hind limb is expressed only in the ectodermal layer. They further observed that XIHbox 1 protein forms a gradient that runs from anterior to posterior, and proximal to distal. Staining of forelimbs of mouse and chick with the same antibody reiterates this pattern of expression (Oliver et al. 1988, 1989). Examination of older mouse embryos shows that expression is transient, because by day 13 (approximately stage 27 in chick), staining is extremely weak and limited to the distal ectoderm (Oliver et al. 1988). The authors suggest that XIHbox 1 may play a morphogenetic role in the limb, and is perhaps repressed by retinoic acid, whose concentration gradient runs counter to that of XIHbox 1 protein. This hypothesis is provocative in light of recent studies which have examined the newt homolog of XIHbox 1 (NvHbox 1 or FH-2) as a gene possibly playing a role in limb regeneration (Savard et al. 1988; Tabin, 1989). Expression of NvHbox 1 is clearly higher in proximal than distal blastema. It is well established that application of retinoic acid to regenerating urodele limbs proximalizes the regenerate (Maden, 1982; Thoms and Stocum, 1984). If distal blastema are first generated by amputation, and the animals are then treated with retinoic acid, one might expect that the proximalizing effect of retinoic acid increases the level of NvHbox 1 transcripts. Neither study found such an increase. Savard et al. (1988) as well as Tabin (1989) suggests that NvHbox 1 may either not be involved in the specification of the proximodistal axis or that retinoic acid interferes at a level that does not involve the regulation of NvHbox 1 expression.

Using another antibody probe, Oliver et al. (1989) have recently characterized the expression pattern of Hox 5.2 in early mouse, frog and chick limb buds. They found that Hox 5.2 expression is complementary to that seen with XIHbox 1. Moreover, Hox 5.2 is predominantly expressed in the distally located progress zone, a region harboring the pool of undifferentiated cells that are responsible for limb outgrowth (Summerbell et al. 1973). Hence, Hox 5.2 could be involved in regulation of limb outgrowth. Dollé and Duboule (1989) have independently isolated Hox 5.2 and their in situ hybridization studies reveal a pattern identical to that reported by Oliver and colleagues. Oliver et al. suggest that one of the reasons for XIHbox 1 and Hox 5.2 being expressed in two non-overlapping sets of cells could be mutual repression. Homeobox cross-regulation is not unprecedented, e.g. in Drosophila Ubx represses Antp transcription by direct binding of Ubx proteins to DNA sequences near the Antp Pl promoter (Hafen et al. 1984; Carroll et al. 1986; Beachy et al. 1988).

A third example of a homeobox gene that is non-uniformly expressed in the limb bud is Hox 7.1. In situ hybridization showed that in early mouse limb buds of 9.5 days (forelimb is equivalent to stage 20/21 chick) Hox 7.1 is expressed distally, in the progress zone (Robert et al. 1989; Hill et al. 1989). The authors raise the possibility that Hox 7.1 is associated with mesenchymal-ectodermal interactions since it is expressed in the progress zone that lies directly underneath the apical ectodermal ridge (AER). In the chick limb bud, the AER is necessary for normal development of the underlying mesenchyme and vice versa (Kieny, 1960, 1968). For example, if the ridge is removed, the resulting limb is truncated (Saunders, 1948; Summerbell, 1974). Moreover, grafting of a piece of flank ectoderm onto the dorsal face of the limb bud leads to a second AER and subsequently to a supernumerary limb (Carrington and Fallon, 1986). Hill et al. (1989), who have also studied Hox 7.1 expression in mouse embryos, make the additional point that in early limb buds (9.5 days), Hox 7.1 is expressed at high levels along the posterior margin (see their Fig. 7F). This region coincides with the ZPA. In later limb buds (13.5 days), maximal expression of Hox 7.1 is found in the tissue of the interdigital spaces; that is, in those cells that are destined to die (Saunders and Gasseling, 1962). Thus the initial pattern of expression seen may not only pertain to mesenchymal-ectodermal interactions, but may also be a preparatory stage for later, more focused expression in the interdigital spaces.

An in situ hybridization study showed that the chicken homeobox gene Ghox 2.1 is expressed in the proximoanterior portion of the bud (Fig. 5). The domain of expression does not correlate with subsequent cytodifferentiation patterns. Wedden et al. (1989) have pointed out that part of the region of Ghox 2.1 expression later undergoes programmed cell death (Fig. 4A). However, they emphasize that Ghox 2.1 is unlikely to act as a global signal for inducing cell death in the limb, because no similar zone of expression is seen along the posterior margin that also undergoes programmed cell death (see Fig. 4A).

Northern analysis of RNA shows that the following homeobox genes are also expressed in the developing or regenerating limb: in chicken homeobox genes, Ghox 1.6, 2.2, 2.3 (Sundin, Pang and Eichele, unpublished data; Wedden et al. 1989), in newt, FH-1 (Tabin, 1989), in humans homeobox gene cl (Simeone et al. 1987) and, in the mouse, the En-1 (Joyner and Martin, 1987; Davis and Joyner, 1988) genes. It will be important to see whether some of these transcripts are non-uniformly expressed in the limb bud. What is also needed is a systematic examination of the expression pattern of all known homeobox genes at the critical stages of limb formation. The description of the normal expression pattern will have to be complemented by analyses of limb buds of mutant embryos or of embryos whose limb buds are experimentally manipulated.

In sum: it is clear from this synopsis that the early limb rudiment, despite being a rather simple tissue, undergoes complex changes in gene expression that precede terminal differentiation. The homeobox genes mentioned here are expressed in a very distinct spatiotemporal pattern in the limb bud. Their pattern of expression does not simply anticipate or reflect the arrangement of subsequently generated terminally differentiated tissues, but forms a prepattern that perhaps subdivides the limb into regions of specific morphogenetic fate.

Since the receptors of RA are transcriptional regulators, it is likely that retinoic acid affects pattern formation through the regulation of one or more key genes early during development of the limb bud. At present it is not known which genes are directly regulated by retinoic acid, but there are some promising candidates. For example, it has been found that retinoic acid treatment will induce the expression of several homeobox genes, either in differentiating teratocarcinoma cells (e.g. Colberg-Poley et al. 1985: Deschamps et al. 1987; Schulze et al. 1987; Mavilio et al. 1988; Dony and Gruss, 1988; La Rosa and Gudas, 1988b) or in primary cultures of embryonic brain (Deschamps et al. 1987). The most comprehensive analysis of a homeobox gene induced by RA treatment of teratocarcinoma cells has been performed by La Rosa and Gudas (1988a,b). They carried out a differential screen to identify any mRNA species that selectively increase early upon retinoic-acid-induced differentiation of mouse F9 teratocarcinoma cells and obtained a cDNA clone of homeobox gene Hox 1.6. Both induction and maintenance of Hox 1.6 expression required the continuous presence of retinoic acid. Experiments with inhibitors of RNA and protein synthesis suggested but did not prove transcriptional regulation, as data from in vitro nuclear run-off transcription were not reported. Hence, a second possibility is that retinoic acid stabilizes Hox 1.6 mRNA. Finally, induction of Hox 1.6 might also be a secondary effect, due to other changes in the F9 cells that accompany the early commitment to differentiation. Strong evidence for the regulation of transcription could be obtained by fusing the upstream region of a retinoic-acid-induced gene to a reporter gene, and then showing that this reporter gene is also induced by retinoic acid. This approach allows one to dissect the regulatory region into DNA sequence elements responsible for regulation by retinoic acid. Should these sequence elements also be sites that bind RAR, this would provide strong evidence that RAR directly regulates the gene.

A particularly illuminating example of the gene fusion approach is provided by a recent study of growth hormone (GH) transcription. Bedo et al. (1989) have found that retinoic acid treatment of human GH1 pituitary cells leads to a dramatic increase in GH expression especially when retinoic acid is applied in conjunction with either glucocorticoids or thyroid hormone. This same synergism is observed when one measures the expression of a CAT reporter gene fused to the growth hormone 5′ regulatory region, supporting the view that, in this case, regulation by retinoic acid is at the level of transcription. Umesono el al. (1988) have also shown that a short DNA sequence, the thyroid hormone response element, can confer retinoic acid inducibility upon a neutral reporter gene. This is an intriguing observation in view of the 62% protein sequence homology between the DNA recognition domains of human RAR and human β thyroid hormone receptor, and suggests the possibility of competition or synergism between the two receptors for cognate regulatory sites. Finally, there might exist morphogenetically active compounds other than retinoic acid. If other morphogens are present in the limb bud, it is possible that their receptors interact with RAR to specify cell fate and pattern formation. This mode of action is reminiscent of the intricate regulatory network through which the bithorax complex, for example, assigns segment identities to the Drosophila embryo (Peifer et al. 1987).

In conclusion, the establishment of the anteroposterior limb pattern can now be rephrased in terms of a signal transduction mechanism consisting of (1) the enzyme(s) that synthesize retinoic acid, (2) the signal in the form of retinoic acid, (3) receptors that function as retinoic-acid-dependent transcription factors, and (4) target genes that are responsible for generating the actual pattern. The question posed at the beginning of this review was whether there is any evidence for the old idea that small molecules are involved in pattern formation. It seems to us that a broad variety of experiments, in part reviewed here, qualify retinoic acid as such a molecule.

We wish to thank Drs George Flentke and Jack Kirsch for helpful discussion. Work from the authors’ laboratory is supported by grants HD 20209 from the National Institutes of Health and NP 630 from the American Cancer Society. S.M.S. and S.E.W. were supported by a fellowship from MDA and NATO, respectively.