Epithelial mesenchymal interactions, the ECM and limb development


Professor Peter Lonai, Department of Molecular Genetics, The Weizmann Institute of Science, Rehovot, 76100 Israel


It has been long since recognized that cellular interactions are not always direct, i.e. they do not always take place between cells contacting each other, or between cells that emit soluble factors and other cells, which respond to it. In contrast, cross-talk between cells is frequently based on signals attached to the extracellular matrix (ECM). Thus besides proximate cell-to-cell contact, certain interactions are mediated by the ECM in a sequence: cell-to-matrix, matrix-to-cell. ECM-mediated interactions may take place within a group or sheet of cells or across adjacent cell sheets. A modified mat-like ECM, the basement membrane, separates adjacent cell sheets and mediates their interactions. Since cell sheets separated by basement membranes are an elementary feature of metazoan histology, interactions with the basement membrane have considerable importance. Recently accumulated evidence emphasizes the importance of ECM-mediated interactions. It is becoming increasingly evident that the ECM functions not only as an architectural component, but it is involved also in signal transduction. This evidence derives from four main sources: from the structure of receptor-ligand complexes, from Drosophila and C. elegans genetics, from cell biological observations and from the analysis of mammalian development. In this review, I will touch upon recent evidence, illustrated by examples of FGF signalling in vertebrate limb development. Although the involvement of matrix components is not yet proven for all cases directly, the strength of multiple indications suggests that a better understanding of ECM-mediated interactions will shed new light on cell differentiation.

Signalling molecules and the ECM

FGF signalling is involved in multiple stages of embryonic development, including vertebrate limb development (for reviews see Yamaguchi & Rossant, 1995; Martin, 1998; Naski & Ornitz, 1998). An important common property of FGF and FGFR is the affinity to heparan sulphate proteoglycans (HSPGs) and their high affinity for the glycosaminoglycan heparin (Burgess & Maciag, 1989). Heparin was recognized to be required for efficient signal transduction by FGFs. Cells deficient in HSPG will respond to FGF only after the addition of heparin (Yayon et al. 1991). Heparan sulphates are long polymers of repeating disaccharides, which can create a multiplicity of variants on a par with amino acids (Turnbull et al. 2001). Recent evidence demonstrates that FGF isotypes differ in their affinity to heparan sulphates in solution (Powell et al. 2002), as well as in tissues (Chang et al. 2000), suggesting that they contribute to the specificity of receptor–ligand interactions.

The molecular relationship between heparan sulphates, FGF and FGFR has been determined by crystallography. FGF-FGFR and heparin form a ternary complex containing two FGF molecules interacting with two receptor monomers. The four protein chains form a grove, which binds heparin chains (Schlessinger et al. 2000) and this binding is a prerequisite for receptor activation. Structural relationships suggest that different FGF:FGFR complexes accommodate different proteoglycans.

The affinity of FGFs to heparin was discovered more than 10 years ago. Since then heparin binding by additional growth factors, such as members of the hedgehog family, EGF, PDGF and TGFβ variants, has also been recognized. Specific heparan sulphates are synthesized from sulphated sugars by complex enzymatic reactions. A number of these enzymes were investigated by mutant analysis in Drosophila. The Sugarless and Sulphateless genes encode enzymes required for the biosynthesis of heparan sulphate glycosaminoglycans. Sugarless and Sulphateless mutant embryos have phenotypes similar to those lacking the functions of two Drosophila fibroblast growth factor receptors, Heartless and Breathless (Lin et al. 1999). The Drosophila gene Sulphateless, which encodes a homologue of a vertebrate heparan sulphate sulphotransferase, is essential for Wg (Wnt in vertebrates) signalling (Lin & Perrimon, 1999), whereas tout velu, encoding another enzyme of heparan sulphate biosynthesis, is required for the movement of modified hedgehog derivatives on the cell membrane. Interestingly, tout velu does not influence FGF and Wnt signalling, demonstrating a high level of specificity in the regulation of extracellular signals by ECM-bound heparan sulphates (The et al. 1999). This biochemical and molecular evidence demonstrates the direct involvement of ECM components in morphogenic signalling.

Basement membranes and signalling across cell sheets

A large body of cell biological research suggests connections between the ECM-bound heparan sulphates and various signalling molecules. Classical studies on the mechanism of epithelial mesenchymal interactions have suggested that the ECM separating the interacting cell sheets are important for salivary gland (Kallman & Grobstein, 1966) and lung (Wessels, 1970) organogenesis. Involvement of ECM components in epithelial differentiation has been shown by antibody inhibition in organ culture. Antibodies to laminin, one of the matrix-forming proteins of the basement membrane, inhibit epithelial morphogenesis in kidney (Klein et al. 1988) and lung (Schuger et al. 1995) development. The importance of laminin receptors, such as dystroglycan (Durbeej et al. 1995; Williamson et al. 1997) and integrin β1 (Fassler & Mayer, 1995), supports this notion. Requirement for the early expressed laminin-1 isoform at the primitive streak stage has been demonstrated by targeted disruption of Lamc1, the gene encoding laminin γ1 (Smyth et al. 1999).

The basement membrane contains a number of glycoproteins. Flat intercalating polymers of laminin and type IV collagen heterotrimers establish its mat-like structure. They bind perlecan and other proteoglycans (Timpl, 1996), which makes the basement membrane a scaffold for signalling molecules in the vicinity of interacting cell layers. Basement membranes have considerable importance in pathology. Established basement membranes form barriers between cellular compartments, which are dynamically broken down and rebuilt during branching morphogenesis and angiogenesis. Thus, they are barriers, which only certain cells such as white blood cells or metastatic tumour cells can penetrate. Their re-formation during branching morphogenesis and tumour angiogenesis is an important facet of malignant differentiation. Epithelial sheets attached to a basement membrane cover the internal and external surfaces of most organs. Anchorage of epithelial cells to the basement membrane saves them from programmed cell death, a threat, which malignant cell can avoid successfully (Frisch & Ruoslahti, 1997).

Three phases of basement membrane formation, synthesis of its network-forming elements, their assembly into a complex tertiary structure and its remodelling, can be distinguished. Intensive research clarified that under physiological temperature, salinity and concentration individual laminin α, β and γ chains form heterotrimers, bind to their integrin or dystroglycan receptors and assemble into large flat multimers, which associate with the characteristic glycoproteins and HSPGs of the basement membrane (for review see Colognato & Yurchenco, 2000). The second network-forming component of the basement membrane is heterotrimeric collagen IV, which also forms multimeric complexes (for review see Brown & Timpl, 1995). Recent results demonstrate that key elements of basement membrane assembly are their receptors, integrins and dystroglycan (Li et al. 2002). The breakdown of basement membranes is an important step in their remodelling. Its major factors are metalloproteases (Vu & Werb, 2000) and heparanase (Vlodavsky et al. 1999). Modification of these enzymes may constitute a lead to important targets for cancer therapy.

Laminin and collagen IV synthesis and the FGF system

Compared to our rather detailed knowledge of basement membrane assembly and remodelling, the synthesis of basement membrane components and their positive regulation is less well understood. Early research centred on retinoic-acid-induced differentiation of F9 embryonic carcinoma cells. These stem cells form polar epithelia, similar to the parietal or visceral endoderm of the early mammalian embryo. Differentiation of F9 cells results in increased laminin and collagen IV transcription and it was recognized that the three loci encoding the laminin hetero-trimer and the two encoding type IV collagen are activated together during retinoic-acid-induced differentiation (Kleinman et al. 1987).

Recent studies in a similar system, embryonic stem (ES) cell differentiation into embryoid bodies, brought further insight. ES cells aggregate into embryoid bodies when they cannot adhere to the culture matrix. They contain two cell layers, the polar visceral endoderm and the internal columnar epithelium that is similar to the epiblast of the early pregastrulation embryo. The two cell layers are separated by a basement membrane and the centre of the embryoid body is cavitated by apoptotic death of cells that could not adhere to the basement membrane (Coucouvanis & Martin, 1995).

Stable transfection of truncated Fgfr2 cDNA into ES cells revealed that FGF signalling through the PI3K-PKB/Akt pathway is required for their differentiation (Chen et al. 2000). Moreover, the dominant negative FGFR mutation abrogated laminin α1, β1 and γ1 as well as collagen IV α1 and α2 synthesis in the mutant (Fig. 1). Interestingly, externally added laminin-1 partially rescued embryoid body differentiation, in as much as it induced differentiation of the columnar ectoderm of the epiblast layer. This suggested that FGF signalling is required for the synthesis of the network forming elements of the basement membrane by the visceral endoderm and that they, through basement membrane assembly, activate epithelial differentiation (Li et al. 2001a).

Figure 1.

Expression of truncated, dominant negative, Fgfr2 cDNA abrogates embryoid body differentiation and the expression of laminin α1 and collagen IV α1. Confocal microscopy. A & B laminin α1; C & D Collagen Iva. A & C wild type, B & D mutant embryoid bodies. Five days of culture. False colours, red: specific antibody; green: phalloidin staining of actin filaments to show morphology.

An important mechanistic aspect of receptor tyrosine kinase-induced laminin and collagen IV synthesis was revealed when the involvement of PKB/Akt was investigated. PKB/Akt is an anti-apoptotic cytoplasmic serine threonine kinase, which together with its homologues mediates numerous pathways connected to programmed cell death, insulin synthesis and cell proliferation (for review, Brazil & Hemmings, 2001). Stable transfection of ES cells with constitutively active PKB greatly enhanced laminin and collagen synthesis. In addition, transactivation demonstrated that this regulation is manifest at the level of transcription, where yet unknown PKB/Akt-activated transcription factors activate the promoter/enhancer regions of laminin β1 and collagen IV α1 (Li et al. 2001b).

According to these findings, signalling through FGFR, and probably other receptor tyrosine kinases as well, exert positive control on laminin and collagen IV synthesis and by it promote basement membrane assembly and epithelial differentiation. Hence ECM-mediated receptor ligand interactions, besides the control of signalling growth and differentiation, feed back to ECM formation. This interaction may function as an amplification device for receptor-mediated signalling.

Localization of receptor tyrosine kinases and their ligands in adjacent cell sheets

Early in situ hybridization experiments revealed that Fgfr1 and Fgfr2 are predominantly expressed in mesenchymal or epithelial tissues (Orr-Urtreger et al. 1991). Comparing the expression pattern of the kit receptor with that of its ligand (Keshet et al. 1991) or PDGFA and its receptor Pdgfra (Orr-Urtreger & Lonai, 1992) demonstrated that polypeptide growth factors and their tyrosine kinase receptors are frequently localized on adjacent cell layers.

As the mutually excusive splice variation of FGFRs has been discovered (Miki et al. 1992), additional examples for reciprocal localization of receptor–ligand pairs was also discovered. Expression of the b alternative of Fgfr2 (or Fgfr2b) was found to be restricted to external and internal epithelia, whereas its second variant, Fgfr2c, was mostly found in mesenchymal tissues (Fig. 2; Orr-Urtreger et al. 1993). Moreover Fgfr1 and 3 also have mutually exclusive splice variants, which differ both in their binding specificity and localization (for review, Johnson & Williams, 1993). Numerous observations suggest that Fgfr variants using the IIIb immunoglobulin-like domain are distinguished by epithelial expression, whereas those using the IIIc domain are preferentially expressed in mesenchymal tissues. Significantly, the ligand binding specificity of these receptor variants seems also specific. Accordingly, b-type epithelial receptor variants bind FGFs distinguished by mesenchymal expression, while c-type Ffgfr in the mesenchyme recognize FGFs expressed in epithelia (Ornitz et al. 1996). It follows that in many cases FGF signalling follows epithelium-to-mesenchyme or mesenchyme-to-epithelium routes. These results suggest that a significant proportion of receptor–ligand interactions take place across adjacent cell sheets, and that these interactions most likely include the basement membrane separating them.

Figure 2.

Fgfr2c is expressed in bones and other mesenchyme, whereas Fgfr2b is transcribed mostly in epithelia of the surface ectoderm and alimentary canal as well as in the endothelia of blood vessels. Radioactive in situ hybridization 15.5 d.p.c. mouse embryo forelimb. A. bright-field, B & C. dark-field illumination. B. Fgfr2c, C. Ffr2b hybridization.

FGF interaction loops in limb development

The role of FGF signalling in limb development was first indicated by replacing the function of the AER in proximal distal limb outgrowth with externally added FGF to cultured limb buds, from which the AER has been removed (Niswander & Martin, 1992). The AER expresses Fgf4, 8, 9, 19 and Fgfr2b (for review Martin, 1998). Further understanding was obtained from over-expressing Fgfs in the flank of chick embryos, where it led to de novo outgrowth of complete ectopic limbs (Cohn et al. 1995). This suggested that limb outgrowth is controlled by epithelial FGF signals. Discovery of Fgf10 in the progress zone mesenchyme and its effect on Fgf8 expression in the ectoderm leading to de novo limb outgrowth suggested that a regulatory loop exists between the Fgf10 in the progress zone and Fgf4 or Fgf8 in the AER (Ohuchi et al. 1997). Further evidence for such a loop was provided by the targeted disruption of Fgfr2 locus, which caused midgestation lethality due to defective placentation and complete absence of limb buds (Xu et al. 1998). Removal of the placental defect by fusing homozygous Fgfr2–/– ES cells with tetraploid wild-type embryos allowed longer survival of the mutant, which displayed complete loss of limb development, with widespread defects of branching morphogenesis (Arman et al. 1999). This phenotype was very similar to the targeted disruption of Fgf10 (Min et al. 1998). Evidence for interaction between Fgfr2b in the AER with Fgf10 in the progress zone derives from the targeted disruption of the exon encoding the IIIb domain. This mutation represents a phenocopy of the Fgf10 mutation (De Moerlooze et al. 2000; Revest et al. 2001). It follows that an important interaction of limb development takes place between Fgf10 in the progress zone mesenchyme and Fgfr2b in the AER.

Whether Fgfr2b is necessary and sufficient for AER formation is, however, less clear. The complete abrogation of limb development in Fgfr2b mutations (Arman et al. 1999; De Moerlooze et al. 2000) would argue for this; nevertheless, observations by Revest et al. raise a note of caution. They detected transient FGF8 expression and epithelial metaplasia in the limb field of the Fgfr2b mutant. This observation may be explained by the involvement of an additional FGFR, whereas Fgfr2b would be required for the maintenance and protracted function of the AER.

The evidence for interaction between Fgf10 in the mesenchyme with Fgfr2b in the epithelium describes one of the two FGF ligand–receptor interactions in limb development. The bulk of evidence, however, concerns the multiple Fgf expressed in the AER. Without discussing the function of the FGF expressed in the AER in detail here (Sun et al. 2002; for review see Tickle & Munsterberg, 2001), the existence of an epithelial mesenchymal interaction loop, consisting of Fgfs of the AER and an FGFR in the mesenchyme, seems to be highly probable. Fgfr2c is expressed in the limb bud mesenchyme (Orr-Urtreger et al. 1993) and has affinity for FGFs expressed in the AER; nevertheless, a role for this receptor as a mediator of AER-derived FGF signals has been ruled out by gene targeting, which caused bone development defects with normal limb outgrowth and morphogenesis (Eswarakumar et al. 2002). A better candidate for the receptor of this interaction, Fgfr1, is expressed in the progress zone mesenchyme and chimeras created by aggregating Fgfr1–/– ES cells with normal embryos displayed limb defects (Saxton et al. 2000). Hence, the FGF control of limb outgrowth is based on receptor–ligand interactions, which mediate bidirectional cross-talk between epithelia, the AER and the limb bud surface ectoderm, and the mesenchyme of the progress zone.

FGF signalling and morphogenic migration

Cell motility along morphogen gradients or due to expansion of developing organs is an important aspect of development. Cell migration takes place along ECM substrates. Involvement of FGF signalling in morphogenic migration was first indicated in the tracheal system of the Drosophila embryo (Klambt et al. 1992). As an extension it has been also shown that branches of the tracheal system expressing the Drosophila FGFR homologue migrate towards their ligand in the surrounding mesenchyme (Sutherland et al. 1996), a feature that resembles FGF-mediated outgrowth of vertebrate limbs. Similar involvement of a C. elegans homologue was observed in the migration of sex myoblasts (DeVore et al. 1995).

In the mouse, targeted disruption of Fgfr1 affects late gastrulation causing defective development of the posterior mesenchyme (Deng et al. 1994; Yamaguchi et al. 1994). Chimeras constructed from homozygous Fgfr1–/– ES cells carrying a β-galactosidase reporter and normal pre-implantation embryos revealed that in the centre of this defect is the retarded migration of mesoderm precursors across the primitive streak (Ciruna et al. 1997). The mechanism of action of Fgfr1 was revealed by the chimeric analysis of gastrulation as well as limb development. Localization and behaviour of cells carrying homozygous mutations of an Fgfr1-associated phosphatase, shp2, and Fgfr1 itself, showed that homozygous mutant cells do not populate the progress zone mesenchyme and that the mutation, rather than conveying proliferation signals, influences cell shape, cell movement and adhesion (Saxton et al. 2000). Returning to the role of Fgfr1 in gastrulation Ciruna and Rossant more recently demonstrated that Fgfr1 controls Snail and E-cadherin expression as well as Wnt3a signalling and through this orchestrates the epithelial to mesenchymal transition of mesoderm in the primitive streak. Thus a molecular link is established between FGF and Wnt signalling in the control of cell-to-cell and cell-to-matrix adhesion (Ciruna & Rossant, 2001).

Recent chimera experiments suggest that Fgfr2 complements the role of Fgfr1 in limb development. While homozygous Fgfr1 mutant cells do not colonize the progress zone mesenchyme, Fgfr2 mutants fail to colonize the AER and owing to their presence in the chimera their normal counterpart fails to establish the normal dorsal ventral compartment boundary (Gorivodsky and Lonai, in preparation). In Fgfr2 chimeras the AER was disrupted and, as shown by its Fgf8, BMP4, Dlx2 and En-1 markers, AER fragments were displaced relative to the dorsal–ventral compartment boundary. Since Fgfr2b is required for AER maintenance and protracted Fgf8 expression, the AER fragments had to be mainly populated by normal cells. Interestingly, En-1, which is a marker of the ventral ectoderm and is required for the dorsal–ventral patterning of the limb, was expressed also dorsally in the chimeras and, and shown in Fgfr2–/– embryos, its expression was controlled by Fgfr2b. It follows that displacement of the normal limb pattern had to be the effect of mutant cells on their wild-type neighbours in the chimeric embryo.

Taken together, loss of Fgfr2 caused two types of defects. Lack of sustained AER differentiation was cell-autonomous, probably caused by an innate defect of the mutant ectoderm to differentiate into the columnar epithelium of the AER, which may be connected to its lack of En-1 expression. The second defect, the dorsal–ventral displacement of the wild-type component, was non-cell-autonomous, indicating that the mutant defect was extracellular. Certain differentiation events, including morphogenetic migration, require cell-to-matrix interactions (Fig. 3). Altabef et al. have shown that the AER-fated ectoderm is derived from a broad area of the ectoderm, suggesting that cells have to migrate to occupy the site of AER differentiation (Altabef et al. 1997, 2000). It follows that Fgfr2, or more accurately its b splice variant expressed in the surface ectoderm, is required for the migration of the AER-fated ectoderm and for its differentiation in connection with signals mediated by Fgf10 in the progress zone.

Figure 3.

Fgfr2 mutant cells interfere with the morphogenic migration of wild type cells in Fgfr2b↔ Wt chimeras. Schematic figure, showing the trunk before limb outgrowth. Blue area, dorsal; pink area, ventral compartment. Segments A to D represent areas with different ratios of wild-type (white) and mutant (red) cells. Right side of the figure shows how the ratio of wild-type and mutant cells influences the position of the AER. The figure does not emphasize whether the displacement is due to steric hindrance by undifferentiated mutant cells, or to lack of routing signals in areas occupied by the mutant cells (see also text).

These results suggest that in common with the control of limb outgrowth by FGF signalling, complement-ary loops of epithelial mesenchymal interactions may also characterize other developmental pathways, and that they affect extracellular properties such as cell adhesion and cell migration. The picture drawn by the results discussed here enhances the importance of the ECM in gene regulation and differentiation and may lead to fruitful experimental hypotheses.