Embryonic gene expression patterns of TGF β1, β2 and β3 suggest different developmental functions in vivo

Transforming growth factor (TGF) β was originally identified on the basis of its ability to induce a transformed phenotype in normal rat kidney fibroblast cells when grown in soft agar (Moses et al. 1981; Roberts et al. 1981). Since the cloning and molecular characterisation of TGF β1 (Derynck et al. 1985), it has become clear that this molecule is just one member of a large multigene family (ten Dijke et al. 1988; Jakowlew et al. 1988; Kondaiah et al. 1990; Melton, 1990). Members of the extended superfamily are found in a wide variety of species as diverse as insects, amphibia, birds and man (Padgett et al. 1987; Weeks and Melton, 1987; Jakowlew et al. 1988; Derynck et al. 1985). The more divergent members of this family include substances that were identified as being of central importance in developmental processes, such as the Decapentaplegic gene (Padgett et al. 1987), which is involved in dorsoventral patterning in Drosophila, and the potential amphibian mesoderm-inducing factors (Weeks and Melton, 1987; Tannahill and Melton, 1989; Smith, 1989).

The true TGF βs are eight in number (Melton, 1990) and share much greater amino acid sequence homology than that between members of the extended TGF β related super-family. These genes have, however, only been isolated from vertebrates so far and, as yet, no mammalian homologues of TGFs β4 to β8 have been identified.

Each TGF β has diverse biological activities in vitro. Each can influence cell growth and/or differentiation in a positive or negative manner. The specific action of the growth factor is dependent on the cell type and culture conditions (Roberts and Sporn, 1990b). However, in many in vitro bioassays, the different isoforms are functionally interchangable (Grayar et al. 1989; Roberts et al. 1990a), though some marked specificities of action have been noted in certain biological systems (Rosa et al. 1988; Jennings et al. 1988; Ohta et al. 1987).

It is debatable whether the different members of the TGF β gene family serve different biological functions in vivo or whether they might indeed be functionally interchangable in vivo. Although much can be learnt from in vitro studies on TGF β biological activity, it is essential to know the sites of synthesis of each of the gene products in order to gain an understanding of the in vivo biological function of these molecules.

In previous studies, the localisations of TGF pi polypeptide and RNA were examined during murine embryogenesis (Heine et al. 1987; Wilcox and Derynck, 1988; Lehnert and Akhurst, 1988; Akhurst et al. 1990a,b). These studies confirmed the in vitro findings of the importance of TGF β1 in haematopoiesis, angiogenesis and osteogenesis, and also indicated that TGF β1 plays a role in epithelial-mesenchymal interactions during morphogenesis. A study of TGF β2 gene expression has also been made by Pelton et al. (1989). These types of global expression study are essential before going into any detailed analyses of the in vivo role of these isoforms in a particular developmental process.

Here we report the use of murine gene-specific probes to compare the global expression patterns of TGF β1, β2 and β3 during murine embryogenesis from 9.5 to 16.5 days post-coitum (p.c.) and the results are discussed in terms of the putative functions of these proteins in mammalian embryonic processes in vivo.

Mouse embryos were obtained from Parkes females mated with NIH males. The day on which the copulation plug was found was considered as day 0.5. The embryos were fixed in 4 % paraformaldehyde in phosphate-buffered saline and embedded in paraffin wax.

Riboprobes were alpha ³⁵S-UTP labelled to a specific activity of 10⁹ disintsmin using the appropriate T3 or T7 transcription system. The probes were digested to an average length of 100 nt by a controlled alkaline hydrolysis (Cox et al. 1984), and used at a final concentration of 40pg μl^-1 (4×10⁴ disintsmin^-1μl^-1).

The TGF β1-specific antisense probe was a 600 nucleotide Kpnl-Apal fragment subcloned into Bluescribe (Stratagene) in antisense orientation with respect to the T3 promoter. The subclone was derived from the full-length murine TGF β1 cDNA (Derynck et al. 1986), which was kindly provided by Dr R. Derynck (Genentech). The subclone corresponds to the precursor region of the TGF β1 polypeptide (amino acids 68–268).

The control probe used was a full-length TGF β2 human cDNA, kindly supplied by Dr G. Bell (unpublished). It was also subcloned into the Bluescribe vector in sense orientation with respect to the T7 promoter.

The TGF β2 DNA probe was obtained by amplification of the reverse transcriptase product of total mouse embryo RNA using the polymerase chain reaction (Saiki et al. 1988). The oligonucleotide primers spanned the initiation and termination codons. This probe was identical in nucleotide sequence to that reported by Miller et al. (1989a). For the in situ hybridisations a 501 nucleotide Pstl-Sacl fragment, spanning amino acid residues 81 to 249, was subcloned into Bluescribe in anti-sense orientation with respect to the T7 promotor.

The TGF β3-specific probe was a 732 nucleotide fragment, spanning amino acid residues 8 to 251 of the precursor polypeptide, in pBluescript KSII (Stratagene) orientated such that antisense probe was generated using the T3 promoter (Denhez et al. 1990).

All three gene-specific probes were sequenced to check the fidelity of the sub-cloning. Sequence homologies between the three gene-specific probes were between 35 and 47 % (Pelton et al. 1989; Denhez et al. 1990).

RNA was isolated from C2C12 (mouse myoblasts) by homogenisation of cells in guanidine isothiocyanate buffer and ultracentrifugation through a caesium chloride cushion as described (Davis et al. 1986). After phenol extraction and ethanol precipitation, 10 μg of each total RNA was separated by electrophoresis through a 1 M formaldehyde-agarose gel (Davis et al. 1986) and blotted onto Nytran membrane (Schleicher and Schuell, Kleene, NH). ³²P-labelled TGF P cDNA was prepared by random primed labelling (Boehringer-Mannheim, Indianapolis, IN) of insert DNA (isolated from one of the cDNA clones containing nucleotides 24–750 for mouse TGF μ3, 1-940 for TGF μ2 and the coding region for TGF μ1) which had been previously purified by gel electrophoresis and Gene Clean (La Jolla, CA). Prehybridisation, hybridisation and washing of the filters was done as described (Church and Gilbert, 1984).

In situ hybridisations were performed at very high stringency, as previously described (Akhurst et al. 1990a). The sections were washed as follows: 5×SSC, 0.1% β-mercaptoethanol, 50°C, 20 min; 2xSSC, 50% formamide, 1% /J-mercaptoethanol, 65°C, 20 min; five washes in 0.5 M NaCl, 10 mM Tris pH7.5, 5 mM EDTA, 37°C, 10min each; one wash in previous buffer plus 10 mg ml^-L RNAase A (Sigma) 37°C, 30 min, 0.5 M NaCl, 10mM Tris pH7.5, 5 mM EDTA, 37°C, 15min; 2×SSC, 50°C, 15 min; and finally 0.1×SSC, 50°C, 15 min. The slides were dipped in Ilford K5 emulsion and autoradiographically exposed for 3 days or up to two weeks. After development the slides were haemotoxylin- and eosin-stained and examined using an Olympus BK2 microscope using bright-field and dark-ground optics. The slides were photographed using Kodak Panatomic X film.

Each piece of data was reproducibly obtained from a minimum of three independent experiments using at least three embryos from each developmental stage. It should be noted that, under the very high-stringency hybridisation and washing conditions used in this study, each gene-specific probe gave the same hybridisation pattern as the corresponding full-length probe.

Previous studies in one of our laboratories (Heine et al. 1987; Denhez et al. 1990) and by other workers (Miller et al. 1989a,b) have shown, by RNA gel blot analysis, that TGF β1, β2 and β3 RNAs show different developmental profiles of expression from 8 days p.c. to birth. In order to compare the spatial and temporal expression patterns of TGF β1, β2 and β3 transcripts during murine embryogenesis, gene-specific probes were generated from the divergent precursor coding region of each cDNA (see Materials and methods). To check that the probes showed no cross-reactivity, they were used on Northern blots of mouse myoblasts expressing each of the three isoforms. Indeed, each probe detected only the predicted transcript sizes for each respective gene. In the cells examined, the TGF β2-specific probe detected two of the four transcript sizes previously observed for this gene (Miller et al. 1989a), (Fig. 1).

In situ hybridisation was performed on sections of mouse embryos from 9.5 to 16.5 dayspost-coitum (p.c.) using the ³⁵S-labelled gene-specific antisense riboprobes, and as a negative control, a sense probe for human TGF pi, which gave no specific hybridisation signal. Each gene shows a unique transcript distribution during development (Table 1), thus supporting the case that there is no cross-reactivity between the probes under the hybridisation and wash conditions employed.

The transcript distribution of TGF β1 is more restricted than that of either TGF β2 or β3, since the latter two show widespread hybridisation to mesenchyme that could be correlated with areas of cell migration and mesenchymal condensation (see below). The transcript distribution of TGF pl RNA is not detailed here since this has been reported extensively elsewhere (Wilcox and Derynck, 1988; Lehnert and Akhurst, 1988; Akhurst et al. 1990b).

Septation of the primitive cardiac tube commences at around 9 days p.c., as mesenchymal cells, which contribute to the valves and septae, appear within the cardiac jelly between the endocardial and myocardial layers of the early heart. By 12.5 daysp.c. this process is virtually complete, though morphogenesis continues until cardiac valve maturation, following birth (Fananapazir and Kaufman, 1988; Akhurst et al. 1990a). We have previously demonstrated that TGFβ1 RNA is expressed in very early endocardial cells, and that, at 9.0 days p.c., this expression becomes limited to the endothelial cells that contribute to mesenchymal cushion tissue. This restricted endocardial TGF β1 RNA expression persists to one week post-partum (Akhurst et al. 1990a).

Here we show that, as early as 9.5 daysp.c., before overt appearance of mesenchymal cushion tissue, there is a regionally restricted distribution of TGF β2 transcripts within the myocardium. Hybridisation of this probe is seen in the region of the atrio-ventricular (av) junction and in the myocardium of the outflow tract, underlying the cardiac jelly, but not in the ventricular or atrial myocardium (Fig. 1B). Myocardial TGF β2 RNA expression continues to at least 10.5 days p.c., with a clear spatial restriction to myocardium underlying mesenchymal cushion tissue both in the av region and underlying the bulbar cushion tissue of the outflow tract (Fig. 1E). There is also some TGF β2 expression in the cushion tissue per se at this stage (Fig. 1E).

By 12.5 days p.c. myocardial expression of TGF β2 no longer persists, but transcripts are still detectable in the mesenchymal component of the maturing heart valves at later stages of development (Fig. 1H). Cardiac expression of TGF β3 RNA is limited to mesenchymal condensations at the base of the heart valves at 14.5–16.5 days p.c. (Fig. 1I).

Within certain major blood vessels, TGF β2 expression is restricted to the tunica media and TGF β3 expression is seen both in the tunica media and tunica intima (Fig. 2C).

In previous studies, TGF β1 RNA expression was observed in areas of ossification, including osteoblasts, osteocytes and osteoclasts (Sandberg et al. 1988a,b; Lehnert and Akhurst, 1988). In this study, TGF β2 and β3 RNA were not seen at high levels within these cel) types, although Pelton et al. (1989) showed hybridisation of a human TGF β2 probe to murine osteoblasts.

The earliest abundant embryonic expression of TGF β3 RNA observed in this study was seen at 10.5 days p.c. and was restricted to the intervertebral disc anlagen (Figs 3A,C, 4A,B). This expression pattern persists in the caudal region of the embryo up to 16.5 days p.c. This RNA is also abundant in mesenchyme within, and adjacent to, the perichondria of non-ossifying cartilage, including tracheal cartilage (Fig. 3G), the nasal septum and otic capsule (Fig. 5C,G,H). Transcripts are also observed in the perichondria of cartilage models of the vertebrae and ribs (Fig. 31,J), though we have not observed it at any stage in chondrification or ossification of the major long bones. We therefore conclude that the expression of TGF β3 appears to be associated with areas of precartilage condensation and active growth and differentiation of chondroblasts. Hypertrophic chondrocytes no longer express this gene.

Contrary to previous reports (Lyons et al. 1989; Pelton et al. 1989), TGF β2 RNA is seen at high levels in the precartilaginous blastemae of the limb buds at 12.5 days p.c., before there is any morphological distinction between these cells and the adjacent mesenchyme (Fig. 3K). In this respect, it is interesting that TGF β2 RNA is observed at a number of mesenchymal sites that might be involved in cell proliferation, migration and/or condensation, for example in the fronto-nasal mesenchyme (Fig. 6K) and secondary palate (Fitzpatrick et al. 1990). As development of the long bones proceeds, TGF β2 RNA expression becomes restricted to the proliferating chondroblast zone of the growth plate at the termini of the long bones and digits (Fig. 3M,N). No expression of TGF β2 RNA is seen in mature cartilage or in areas of intramembranous or endochondral ossification.

The lungs are formed as two endodermal outgrowths, which invade the splanchnic mesoderm, thus generating the lung buds. These simple endodermal tubes are then induced to branch distally, and the lung expands as rapid growth occurs at the terminal end buds and in the surrounding mesenchyme. Within the bronchioles, there is a transition in pulmonary epithelial cell morphology from proximal, columnar to distal, cuboidal epithelium. The fully differentiated distal alveolar epithelial cells are further flattened in morphology. These morphological changes correspond chronologically to events in pulmonary epithelial cell differentiation (Joyce-Brady and Brody, 1990).

TGF β3 RNA is seen submucosally in the trachea and proximal bronchi at 12.5 days p.c. (Fig. 4A,B). Simultaneously, intense expression is seen in the immature columnar epithelial cells of the growing bronchioles. As lung outgrowth proceeds, and the distal branches of the respiratory tract differentiate to a cuboidal epithelium, TGF β3 RNA expression remains restricted to the columnar epithelial cells of the proximal respiratory tree (Fig. 4H). No TGF β3 expression is seen in the epithelia of growing terminal end buds (Fig. 4F), or in differentiated epithelium of the alveoli; thus by 16.5 days p.c., when alveolar epithelium is widespread, epithelial expression of TGF β3 is negligible within the lung (Fig. 4K).

TGF β2 RNA, in contrast, is limited to the cuboidal epithelium of the growing terminal end buds at 12.5 to 14.5 days p.c. and to the differentiating alveolar epithelium which persists at least to 16.5 days p.c. (Fig. 4E,I,J).

A striking observation is the presence of TGF β3 RNA in the pulmonary mesothelium at all stages examined (Fig. 4B,K). TGF β3 RNA is, in fact, expressed in all mesothelia including that of the pericardium, diaphragm, liver and lung. In the diaphragm, both the muscular and mesothelial components express this gene.

We previously demonstrated that TGF β1 RNA is abundant in the epithelial component of structures that are actively involved in the process of morphogenesis. These include the whisker follicle, salivary gland, tooth bud (Lehnert and Akhurst, 1988) and secondary palate (Akhurst et al. 1990b; Fitzpatrick et al. 1990). Each of these epithelia also expresses high levels of either TGF β2 and/or TGF β3 RNA (Table 1). The invaginating whisker follicles at 14.5 daysp.c. express all three TGF β3 transcripts. At 12.5 daysp.c., prior to branching, the epithelium of the salivary gland contains TGF β2RNA, though by 14.5 daysp.c., when branching is in progress, this is no longer discernible. The TGF pl probe also hybridises with invaginating dental epithelium (data not shown) and the medial edge epithelium of the secondary palate co-expresses TGF/J1 and pi (Fitzpatrick et al. 1990).

In general, TGF β3RNA does not show widespread epithelial expression. It is limited only to the early columnar, bronchial epithelium, the whisker follicles and the medial edge epithelium of the secondary palate (Fitzpatrick et al. 1990).

No expression of TGF β1 or β2 RNA was observed in the developing olfactory, visual or auditory apparatus apart from expression of TGF β3 in areas of chondrification within the otic capsule and nasal septum (see above). However, striking hybridisation with the TGF β2 probe was seen in specific epithelia of these organs.

At 10.5 days p.c., the otic vesicle shows very low levels of hybridisation to the TGF β2 probe (data not shown). As the ear develops, the otic vesicle gives rise to the complex labyrinth of the inner ear. By 14.5 days p.c., localised epithelial thickenings have appeared within the cochlea, saccule and utricle in relation to the ingrowth of nerve fibres (Swanson et al. 1990). These sensory epithelial cells show intense hybridisation with the TGF β2 probe (Fig. 5A,B,D), which continues to at least 16.5 days p.c. At this stage, the simple epithelium lining the inner ear does not express TGF β2 RNA.

Likewise expression of TGF β2 is seen in nasal epithelium. At 14.5–16.5 daysp.c. when nasal morphogenesis is virtually complete, hybridisation is seen in the olfactory epithelium lining the nasal septum and conchae (Fig. 5E,F). At this time, there is a gradient of expression of TGF β2 RNA, with most intense hybridisation on the distal, more differentiated, side of this pseudostratified epithelium (Fig. 51). The epithelial component of Jacobson’s organ also shows TGF β2 RNA expression (Fig. 5E,F).

In the eye, TGF β2 is expressed at low levels within the outer neuroblastic layer of the retina at 14.5–16.5 days p.c. (Fig. 5J,K) and more intensely in the anterior dividing epithelium of the lens (Fig. 5J,K). Expression is intense in the region of the ciliary body and the ora serata (Fig. 5J,K,L).

In general, we have not observed widespread expression of TGF β1 or β3 in nervous tissues. However, TGF β2 RNA is clearly transiently expressed at low levels in the ventral spinal cord. TGF β2 transcripts are not detectable at 9.5 or at 12.5 days p.c., but appear transiently between 10.5 and 11 days p.c. (Fig. 6). The time of appearance of TGF β2 transcripts would correlate with the period of motor neuron differentiation.

At 10.5 days the TGF β2 probe also shows hybridisation to the forebrain (Fig. 6). Patchy expression of TGF β2 and β3 RNA is seen in various parts of the brain at later stages, but as yet we have not examined this in detail.

The accumulated data from this study on TGF β2 and β3 and those that are published on TGFβ1 are presented in Table 1.

A major assumption in the interpretation of RNA localisation studies is that the detected RNA transcripts are translated to a biologically active polypeptide. This may be a major over-simplification for molecules such as TGF βs. Post-transcriptional mechanisms are known to control production of biologically active TGF βs in certain biological situations, not only in secretion of the protein product but in activation of the latent forms (Roberts and Spom, 1990b).

In general, it has been shown that the embryological localisations of RNA transcripts and protein product for TGF β1 show a strong correlation indicative of both autocrine and paracrine mechanisms (Heine et al. 1987; Lehnert and Akhurst, 1988; Akhurst et al. 1990a). However, there are no data available on the fraction of this protein that is biologically active.

As yet, there is no published information on the embryological protein distributions of TGF β2 and β3 with which to compare our data. The distributions of TGF β2 RNA observed by us in the ventral areas of the developing spinal cord and prosencephalon are in general agreement with the localisation of TGF β2 protein observed by K. C. Flanders, D. S. Cissel, A. B. Roberts, S. Watanabe, R. LaFayatis, M. B. Sporn, and K. Unsicker (unpublished).

There appear to be some discrepancies between our data on TGF β2 RNA expression and that of earlier studies (Pelton et al. 1989; Lyons et al. 1989). Pelton et al. (1989) concluded that the only epithelial expression of TGF β2 is observed in suprabasal kératinocytes in the 18.5 day p.c. skin and in developing hair follicles. They observed no TGF β2 expression in pulmonary epithelia, where we clearly see this RNA, but report high levels of TGF β2 RNA in regions of ossification and in the submucosa of the gut, sites that were negative for TGF β2 RNA in this study.

Explanations for these discrepancies include strainspecific differences in TGF β expression patterns between mouse stocks, examination of embryos of slightly differing stages or that the choice of riboprobe and hybridisation conditions used in the earlier studies resulted in some cross-reactivity between members of the TGF β family. In this respect, it is notable that the expression of TGF β2 in the ventro-lateral nervous system seen in this study had previously been reported by us when using a full-length TGF β1 probe at reduced hydridisation stringency (50 % formamide, 50°C; Lehnert and Akhurst, 1988), and that the hybridisation and wash conditions used by Pelton et al. (1989) were considerably less stringent than these. We believe that mouse strain-specific variation would be minimal since the distinctive differential expression patterns observed in this and other studies is also observed during human embryogensis (Gatherer et al. 1990).

An intriguing observation in the present study is the localised expression of TGF β2 in the early heart. Transcripts of this gene are found in the myocardium underlying regions of septation and valve formation, but not in atrial or ventricular myocardium. At this stage of cardiogenesis, the TGF β1 gene is expressed specifically in the overlying endothelium of the presumptive heart valves (Akhurst et al. 1990a).

The mechanism of av cushion tissue formation involves induction of an endothelial-mesenchymal transition by the underlying myocardium. Tissue interactions controlling this morphological transformation have been the subject of intensive investigation (Krug et al. 1985; Mjaatvedt et al. 1987; Mjaatvedt and Markwald, 1989; Potts and Runyan, 1989). It has been found that a regionally restricted signal, emanating from the myocardium, acts to induce morphological transformation of the overlying endocardium. The latter tissue also shows regional competence to respond to this signal (Mjaatvedt et al. 1987; Mjaatvedt and Markwald, 1989).

Potts and Runyan (1989) proposed that, in the chick, TGF βs may contribute to this regional induction signal, since antibodies against TGF β inhibit the endothelial-mesenchymal transition in vitro. Furthermore, in combination with TGF β, ventricular myocardium, which is not normally competent as an inducing tissue, gains the ability to promote the endothelial-mesenchymal transition of av endóthelium in vitro. The regional localisation of both TGF β1 and β2 in the endothelium and myocardium which interact to form cushion tissue is entirely consistent with these growth factors playing such a role in vivo.

TGF β2 may be more fundamentally involved than TGF β1 in the induction process since the induction signal is known to arise from the myocardium and is only transiently present. It is striking that myocardial expression of TGF β2 RNA is restricted to the early period of cushion tissue formation, whereas endothelial TGF β1 expression does not show this temporal restriction (Akhurst et al. 1990a). Recently it has been shown that a protein related to TGF βs, namely BMP-2A, shows a similar expression pattern to TGF β2 in the myocardium underlying the av region (Lyons et al. 1990), introducing the possiblity that these two gene products interact in the induction of mesenchymal cushion tissue. The function of endocardial TGF β1, in contrast, may be in modulation of endothelial cell growth and of later morphogenetic events (Akhurst et al. 1990a).

All three TGF βs are expressed in epithelial cells at some stage during development, though TGF β2 shows most widespread epithelial expression. All three TGF βs are potent growth inhibitors of epithelial cells in vitro (Grayar et al. 1989). Furthermore, some epithelial cells are induced to differentiate in the presence of TGF β (Jetten et al. 1986; Masui et al. 1986; Reiss and Sartorelli, 1987). It would be informative to know whether epithelially derived TGF βs are mainly involved in autocrine regulation of growth and differentiation and/or in paracrine interactions between the epithelium and the underlying mesenchyme.

In the case of TGF β1, it has been clearly demonstrated that epithelially synthesised TGF β1 is predominantly localised in the adjacent mesenchyme (Heine et al. 1987; Lehnert and Akhurst, 1988), which led us to propose that the main function of the majority of epithelially derived TGF β1 is in paracrine regulation of morphogenetic interactions (Lehnert and Akhurst, 1988; Akhurst et al. 1990b).

In each of the cases where epithelial TGF β1 RNA expression is seen, this RNA is co-expressed with either TGF β2 and/or TGF β3 RNA (see Table 1). In these examples, therefore, TGF β2 and β3 might similarly be involved in control of morphogenesis. Nevertheless, in the absence of protein localisation data for TGF β2 and TGF β3, it is difficult to make postulates about the different modes of action of these epithelially derived TGF βs. TGF β is known to induce differentiation of squamous bronchial epithelial cells in vitro’, thus, in pulmonary epithelia, expression of TGF β2 and β3 could be correlated with cell growth, differentiation and/or morphogenesis.

Unlike TGF β1 and β3, epithelial expression of the TGF β2 gene is not only limited to morphogenetically active tissue, but is also seen in epithelial cells of established structures that are in the process of differentiation. These include epithelia of the sense organs, alveolar epithelium, hyperplastic nodules of palatal epithelium (Fitzpatrick et al. 1990) and the supra-basal kératinocytes of 18.5 day p.c. skin (Pelton et al. 1989). Glick et al. (1989) have shown that TGF β2 levels are elevated in vitro and in vivo in kératinocytes growth-inhibited or induced to differentiate by retinoic acid or by calcium ions, respectively. They also showed that the growth-inhibitory effects of retinoic acid in vitro can be blocked by antibodies to TGF β2. On the basis of this information, it is tempting to speculate that the endogenous in vivo function of TGF β2 in some epithelia is in autocrine modulation of growth and/or differentiation, as previously suggested by Pelton et al. (1989).

No data exist on the protein localisation of epithelially derived TGF β2 in embryos. However, in retinoic acid-treated adult epidermis the TGF β2 protein is found exclusively within the epithelium (Glick et al. 1989), giving support to the hypothesis of a predominantly autocrine role for epithelially synthesised TGF β2. In contrast, TGF β1 protein induced in the epidermis in response to phorbol ester (Akhurst et al. 1988), is localised predominantly in the dermis (unpublished observations from this laboratory). In this context it may be pertinent to note that TGF β2, unlike the other TGF βs, lacks an integrin-binding (RGD) sequence, which may be involved in tissue-targeting of the growth factor (Roberts and Sporn, 1990b).

In situations where it is possible to distinguish the differentiated cell compartment from the stem cell compartment, as in the skin (Pelton et al. 1989; Glick et al. 1989), hyperplastic nodules (Fitzpatrick et al. 1990) and the olfactory epithelium (this study), it is clearly the differentiating cells that possess abundant quantities of TGF β2 RNA, implying a role in maintenance of homeostasis of the epithelium (c.f. Akhurst et al. 1988). A possible additional interpretation is that epithelial expression of TGF β2 is associated with innervation of epithelia. It is striking that all of the sensory epithelia, including optic, otic and olfactory, express TGF RNA.

In conclusion, we have shown that each TGF β isoform is expressed with a distinct pattern, and that this distribution can be correlated with specific embryological processes. This implies that, in vivo, these TGFβ isoforms serve different biological functions, despite their similar molecular structures and in vitro biological activities. It remains to be seen whether these isoforms are functionally interchangeable in vivo.

We would like to thank Derek Gatherer and Sigrid Lehnert for assistance with in situ hybridisation, Liz Duffie for assistance with photography, Drs S. Mackay and R. Smith for confirmation of anatomical details and Drs A. Balmain and D. Fowlis for critical reading of the manuscript. Research in this laboratory is funded by the Medical Research Council and the Cancer Research Campaign and was assisted by a gift from CARE (The Scottish Association for Care and Support after Diagnosis of Fetal Abnormality).