Epidermal differentiation and keratin gene expression

The epidermis is a stratified squamous epithelium, which forms the protective covering of skin. Only the innermost, basal layer of the epidermis has the capacity for DNA synthesis and mitosis. Under an as yet unidentified trigger of terminal differentiation, a basal cell will begin its journey to the skin surface. In transit, it undergoes a series of morphological and biochemical changes culminating in the production of dead, flattened, enucleated squames, which are sloughed from the surface, and continually replaced by inner cells differentiating outward. The epidermis manifests its protective role by building a three-dimensional network of interconnected keratinocytes, each containing an extensive cytoskeleton of specialized 10 nm keratin filaments, encased by a membranous envelope of highly cross-linked proteins. Mitotically active basal cells adhere to the basement membrane and underlying dermis via specialized, calcium-activated adhesion plaques, called hemidesmosomes (Jones and Green, 1991). Hemidesmosomes contain unique anchoring proteins, including the α₆β₄ integrin heterodimer and several proteins identified by autoimmune antibodies from sera of patients with bullous pemphigoid. Basal cells interact laterally and suprabasally with their neighbors through calcium-activated membranous plaques, called desmosomes (Jones and Green, 1991). Despite their ultrastructural similarity to hemidesmosomes, desmosomes are composed of distinct proteins, including desmoplakins, desmogleins and desmocollins.

Intracellularly, hemidesmosomes and desmosomes connect with the cytoskeletal network of keratin filaments. The pattern of keratins in the skin is complex. As in basal cells of all stratified squamous epithelia, basal epidermal cells display a keratin network composed of the type II keratin K5 (58 kDa) and the type I keratin K14 (50 kDa) (Fuchs and Green, 1980; Moll et al., 1982; Nelson and Sun, 1983; Roop et al., 1987). These keratins constitute -15-25% of basal cell protein. As basal epidermal cells differentiate, they downregulate expression of K5/K14 and induce expression of a new set of differentiation-specific keratins (Fuchs and Green, 1980; Moll et al., 1982; Roop et al., 1987). For most body regions, the type II keratin, KI (67 kDa) and the type I keratin, K10 (56.5 kDa), is expressed (Fuchs and Green, 1980; Roop et al., 1987). As epidermal cells proceed through differentiation, they switch on the expression of an additional type II keratin, K2 (65 kDa) and type I keratin (56 kDa) (Collin et al., 1992). In palmar and plantar skin, an additional type I keratin, K9 (63 kDa) is expressed suprabasally (Fuchs and Green, 1980; Knapp et al., 1986). During wound-healing, the type II keratin, K6 (56 kDa) and type I keratin KI 6 (48 kDa) are induced suprabasally (Mansbridge and Knapp, 1987). These keratins are also typically expressed in the outer root sheath (ORS) of the hair follicles (Stark et al., 1987; Kopan and Fuchs, 1989a) and in the posterior segments of tongue epithelium (Moll et al., 1982). Other suprabasal keratin pairs, such as K4 and K13, are not expressed in the skin, but have been found in other stratified squamous epithelia (for a review, see Sun et al., 1984).

Even though terminally differentiating cells in the epidermis are post-mitotic, they nevertheless are metabolically active, and the differentiation-associated changes in keratin expression are regulated transcriptionally (Stellmach et al., 1991). In the inner spinous layers, keratins are a mixture of residual K5/K14 and newly synthesized differentiationspecific keratins. As spinous cells continue their path to the skin surface, they devote most of their protein synthesizing machinery to manufacturing differentiation-specific keratins. Given that keratins are also highly stable in filament form, suprabasal keratins eventually constitute >85% of the total protein of a differentiated squame. Although the functional significance of the keratin switch has not yet been unequivocally resolved, K1/K10 filaments aggregate to form tonofibrillar bundles, which are thicker than tonofila-ment bundles in basal cells (for a review, see Montagna and Lobitz, 1964). This process also takes place in vitro, and appears to be dependent upon the pair(s) of keratins (Eichner et al., 1986). The increase in filament bundling may enhance the ability of keratins to be among the sole survivors of the terminal differentiation process.

As spinous cells reach the granular layer, they undergo a final tailoring in protein synthesis, producing filaggrin, a histidine-rich, basic protein, which may be involved in bundling tonofibrils into large macrofibrillar cables (Dale et al., 1978). This process may impart to keratin filaments their final protection against the destructive phase, which soon ensues. Several other changes occur in the later stages of differentiation. Membrane-coating granules, made earlier, fuse with the plasma membrane and release lipids into intercellular spaces of granular and stratum comeum cells (for a review, see Schurer et al., 1991). In addition, glutamine- and lysine-rich proteins are deposited on the inner surface of the plasma membrane (Rice and Green, 1979). Some of these proteins, such as involucrin, are made early during differentiation (Simon and Green, 1985 and references therein). Others, such as loricrin, are synthesized later (Mehrel et al., 1990). As each differentiating cell becomes permeable during the destructive phase, a calcium influx activates epidermal transglutaminase, which then catalyzes formation of ε-(γ-glutamyl) lysine isopeptide bonds. The envelope proteins are thereby cross-linked into a cage to contain the keratin macrofibrils (for a review, see Greenberg et al., 1991). As lytic enzymes are released, all vestiges of metabolic activity terminate, and the resulting flattened squames are merely cellular skeletons, chock full of keratin macrofibrils. The stratum comeum, composed of squames sealed together by lipids, is an impermeable fortress, keeping microorganisms out and essential bodily fluids in.

In development, progenitor cells for the epidermis are embryonic basal cells (for a review, see Montagna and Lobitz, 1964). They are the first cells to express epidermal markers, synthesizing very low levels of K14 and K5 (Dale et al., 1985; Kopan and Fuchs, 1989a and references therein). If embryonic basal cells come into contact with specialized mesenchymal cells, called dermal papilla cells, they downregulate K5/K14 expression, and begin to grow downward, forming what will ultimately become a hair follicle. In the absence of stimulus from dermal papilla cells, an embryonic basal cell upregulates K5/K14 expression, concomitant with a commitment to an epidermal cell fate. Shortly thereafter, concomitant with stratification, KI and K10 are made in the suprabasal layers of the developing epidermis. In contrast, six different programs of differentiation take place in the hair follicle, giving rise to terminally differentiated hairs. The program of differentiation in the epidermis is much simpler than that of its appendages, although based on body location, regional differences can give rise to differences in the pat-tem of gene expression associated with the epidermal program of differentiation (for a review, see Fuchs, 1990).

Terminal differentiation in the epidermis continues throughout life. It takes approximately 2-4 weeks for an epidermal cell to leave the basal layer and reach and be sloughed from the skin surface. Thus, the adult epidermis is rejuvenated every few weeks.

Among the most significant contributions to epidermal cell growth in culture are the pioneering studies of Rheinwald and Green (for a review, see Rheinwald, 1980), who realized that epidermal cells are not only dependent upon epidermal growth factor (EGF), but also certain fibroblast factors, one of which is likely to be keratinocyte growth factor (KGF). Thus, when seeded on a lawn of X-irradiated or mitomycin C-treated fibroblasts, and in medium containing the appropriate serum lot, cAMP-inducing agent, growth factors and various additional nutrients, epidermal cells grow for several hundred generations in culture. This system is optimal for generating rapidly growing ker-atinocytes, although many of the biochemical changes characteristic of terminal differentiation, such KI, K10 and filaggrin expression, do not occur appreciably (Fuchs and Green, 1980; Asselineau et al., 1986).

A number of laboratories have contributed to methods aimed at optimizing terminal differentiation in vitro, and from these studies, clues to extracellular controls of kera-tinization have been obtained. One of these factors is calcium, a necessary prerequisite for a variety of differentia-tive processes, including stratification, assembly of desmosomes and activation of epidermal transglutaminase (Rice and Green, 1979; Hennings et al., 1980; Watt et al., 1984; Duden and Franke, 1988). Another important class of regulators are retinoids, which diminish features of terminal differentiation in stratified squamous epithelia, both in vivo (Fell and Mellanby, 1953) and in vitro (Yuspa and Harris, 1974; Fuchs and Green, 1981). When retinol is removed from serum, cultured epidermal cells display many of the morphological and biochemical characteristics of terminal differentiation (Fuchs and Green, 1981). These features include stratification, cell adhesiveness, reduced cell motility, and expression of KI and K10.

While calcium and retinoids are clearly important, cultures grown on feeder layers in the presence of low concentrations of retinol and high concentrations of calcium produce basal keratinocytes and 1-2 layers of squames, but generate very few spinous and granular cells. However, when epidermal cells are cultured on floating lattices of collagen and fibroblasts at the air-liquid interface, their differentiation is significantly enhanced (see Asselineau et al., 1986 and references therein). Subsequent studies showed that, when coupled with consideration of optimal serum lots, calcium, retinoids and other nutrients, this culture method results in a near optimal balance between growth and differentiation, which is both morphologically and biochemically a faithful representation of epidermis in vivo (Kopan et al., 1987; Choi and Fuchs, 1990). Fig. 1 illustrates the four major programs of differentiation that take place in the epidermal raft cultures grown under optimal conditions.

The search for an optimal culture system uncovered a number of extracellular regulators controlling the balance between growth and terminal differentiation. Stimulators of keratinocyte growth include EGF (Rheinwald, 1980), TGF-α (Coffey et al., 1987; Barrandon and Green, 1987), 10⁻⁶ to 10⁻¹⁰ M retinoic acid (Fuchs and Green, 1981; Kopan and Fuchs, 1989b), KGF (Finch et al., 1989), and the cytokines, IL-6, IL-l_α and IL-8 (for a review, see Luger and Schwarz, 1990). Of these, EGF and TGF-a have been the most extensively studied. EGF and TGF-α are both ligands for the tyrosine kinase-activatable EGF receptors, located primarily on the surface of basal cells (Green et al., 1987). However, these factors do not act simply as mitogens (Barrandon and Green, 1987; Vassar and Fuchs, 1991). In fact, when keratinocyte colonies are small, cell growth is exponential and usually not markedly influenced by additional growth supplements. It is only when colonies become larger that supplemental factors play a significant role in colony expansion. This acquired factor dependency appears to arise from the fact that colony growth occurs predominantly at the periphery, and hence larger colonies are dependent upon factors that increase cell migration. TGF-α (Barrandon and Green, 1987) and retinoids (Fuchs and Green, 1981) are especially potent in enhancing migration of ker-atinocytes, and it may be that the growth stimulatory effects of these factors are at least in part related to this ability.

TGF-α differs from EGF in that it is synthesized by ker-atinocytes both in vitro and in vivo (see Elder et al., 1989). Since epidermal cells autoregulate their own growth via TGF-α production, it is perhaps not surprising to find that such control can go awry, leading to uncontrolled growth. Several reports have implicated an increase in EGF receptors and/or TGF-α levels with epidermal tumorigenesis (Yamamoto et al., 1986; Ozanne et al., 1986). Psoriatic epidermis also contains higher levels of TGF-α than normal skin (Elder et al., 1989). Interestingly, when transgenic mice were engineered to overexpress TGF-α in their basal epidermal cells, newborn mice exhibited some, but not all, of the features typical of psoriatic skin (Vassar and Fuchs, 1991). This included a physical appearance of scaliness at the skin surface. Histologically, the scaliness was accompanied by an increase in epidermal proliferation, leading to a proportional increase in spinous, granular and stratum corneum layers (Fig. 2A, K14-TGF-α mouse skin, 9 days; Fig. 2B, normal mouse skin). While these mice did not develop squamous cell carcinomas, they did produce benign skin papillomas upon wound healing (Fig. 2C-D; Vassar and Fuchs, 1991; Vassar et al., 1992). Unlike chemically induced papillomas, these papillomas did not have mutations in the Ha-ras gene (Vassar et al., 1992). Thus, while there are additional steps such as Ha-ras in the keratinocyte growth cycle that can be altered, thereby leading to epidermal hyperproliferation and malignant transformation (for examples, see Bailleul et al., 1990 and Greenhalgh et al., 1990), TGF-α seems to be centrally involved in controlling the proper balance between growth and differentiation. Finally, TGF-α overexpression in mice elicited some effects, such as epidermal hypertrophy, which were seemingly unrelated to hyperproliferation (Vassar and Fuchs, 1991).

Keratinocyte growth factor (KGF) is made by stromal fibroblasts and belongs to the seven-member fibroblast growth factor family (for a review see Aaronson et al., 1991). It differs from other FGFs in that its mitogenic activity appears to be specific for keratinocytes, and not fibroblasts or endothelial cells. Similar to those of TGF-a and EGF, the KGF pathway for keratinocyte growth stimulation is activated by a tyrosine kinase surface receptor (Rubin et al., 1989). In vitro studies have shown that the stimulating effects of KGF on DNA synthesis in keratinocytes is 2-10-fold stronger than that of autocrine keratinocyte growth factors such as aFGF, TGF-α and EGF (Rubin et al., 1989). Taken together with the finding that KGF mRNA expression is dramatically elevated after skin injury (Werner et al. 1992), KGF appears to be a major factor in promoting skin cell proliferation.

Given its paracrine effects on the epidermis, KGF has also been implicated as a major player orchestrating the epithelial-mesenchymal interactions that are necessary for epidermal cells to grow, develop and differentiate (for a review, see Aronson et al., 1991 and references therein). Other mesenchymal factors that are likely to play a role in this process include extracellular matrix components (Bissell et al., 1982) and cell surface-associated proteins (Sar-iola et al., 1988; Hirai et al., 1992).

Given that the basal layer of the epidermis has tyrosine kinase receptors for both autocrine (TGF-α) and paracrine (KGF) factors, the question arises as to whether the effects elicited by activation of these receptors are similar. To answer this question, we recently used the same keratin promoter to drive expression of KGF that we used previously to drive expression of TGF-α (Guo et al., 1993). K14-KGF expression consistently exhibited a more potent effect than K14-TGF-α on generating an increase in the number of cells within the basal layer of the epidermis (Fig. 2E, K14-KGF mouse skin, newborn; Fig 2F, control mouse skin. newborn). In part, this overcrowding may be due to a more potent effect of KGF than TGF-α on epidermal proliferation in vivo. However, a number of major differences existed between the KGF animals and the TGF-α animals that were difficult to explain on the basis of relative degree of potency of these two factors.

One difference was that KGF-expressing epidermal cells were far more crowded and compact than the more loosely interconnected TGF-α-overexpressing epidermal cells. Another difference was in terminal differentiation. While TGF-α-overexpression led to a proportional increase in the number of spinous, granular and stratum corneum layers, KGF expression led to epidermal cells, which prematurely reached the skin surface before completing their program of terminal differentiation (Vassar and Fuchs, 1991; Guo et al., 1993). In addition, while TGF-α overexpression caused epidermal hypertrophy and a larger cytoplasm-to-nuclear ratio (Vassar and Fuchs, 1991), KGF expression produced keratinocytes with a smaller cytoplasm-to-nuclear ratio (Guo et al., 1993). Finally, while TGF-a overexpression in mice was not sufficient to elicit epithelial transformation without TPA treatment or wounding (Vassar and Fuchs, 1991), KGF overexpression led to gross transformations of tongue and skin in adult mice (Fig. 2G and H; Guo et al., 1993). Collectively, these studies suggest that the second messenger signalling that takes place after activation of the KGF receptor may be different from that elicited by activation of the EGF receptor.

One of the most striking phenotypic aberrancies of newborn, high-expressing KGF transgenic mice was the near complete suppression of hair follicle formation concomitant with thickened and grossly wrinkled skin. While other explanations are possible, it seems most likely that a critical, threshold level of K14-KGF expression at a critical time during embryonic development may have prevented some stem cells from choosing a hair follicle cell fate and fun-neled them into an epidermal cell fate instead. This notion is especially intriguing given that (1) hair follicles and epidermis are derived from the same embryonic stem cells, and (2) K14 expression occurs at a low level in embryonic basal cells at a time when the choice between the hair follicle and epidermal pathways of development is made (Kopan and Fuchs, 1989a).

It is well-established that the developmental choice to hair follicle morphogenesis is based upon the interaction between an embryonic basal cell and a specialized mesenchymal component, the dermal papilla cell (Wessels and Roessner, 1965). In the natural situation, hair follicle morphogenesis seems to be favored, rather than inhibited, when this mesenchymal-epidermal interaction takes place (Kollar, 1970; Dhouailly et al., 1978; Hirai et al., 1992). The inverse correlation between keratinocyte-derived KGF expression and hair follicle number suggests the possibility that KGF expression by an embryonic basal cell may be able to interfere with signals normally transmitted by dermal papilla cells. Further studies will be necessary to assess the extent to which this intriguing possibility may be correct, and if so, precisely how epidermal KGF is able to interfere with this commitment decision.

A major question in epidermal biology is when does an epidermal cell undergo a commitment to terminally differentiate? If commitment takes place in the basal layer, then the population of basal keratinocytes should be heterogeneous. Certainly, as judged by their cell-cycling properties, this appears to be the case (Barrandon and Green, 1987; Potten and Morris, 1988). By this criterion, there seem to be at least two populations of basal cells. One population, stem cells, have a long lifespan, a long cycling phase and short S period (for a review, see Potten and Morris, 1988). The second population, transit amplifying cells or paraclones, undergo only a limited number of divisions prior to terminally differentiating (Barrandon and Green, 1987). Moreover, paraclones seem to arise from stem cells, suggesting further that a sequence of events leading to terminal differentiation may take place prior to movement of a cell out of the basal layer (Barrandon and Green, 1987).

A priori, it might seem attractive to visualize terminal differentiation as a passive event, where an occupied basal layer provides the force to push a cell to the first suprabasal layer, and thereafter, its absence of contact with the basement membrane might trigger cell cycle withdrawal and the cascade of events leading to keratinization. One component of the basement membrane, fibronectin, seems to play a pivotal role in inhibiting the differentiation process, although it does not appear to interfere with cell cycle withdrawal (Adams and Watt, 1990). During suspension-induced differentiation of human keratinocytes, transcription is inhibited for the genes encoding the fibronectin receptor, α5β1 integrin, and concomitantly, the receptors lose their ability to bind fibronectin, and subsequently are lost from the cell surface (Hotchin and Watt, 1992). While loss of fibronectin binding seems to enhance the differentiative process, it is not likely to be the initial trigger of differentiation, since integrin distribution in the basal layer seems to be uniform, whereas cell cycle withdrawal and other features of epidermal differentiation can occur within the basal layer (Watt, 1984; Potten and Morris, 1988; Choi and Fuchs, 1990; Adams and Watt, 1990).

Withdrawal from the cell cycle seems to be a prerequisite for irreversible commitment of a keratinocyte to terminally differentiate. The most extensively studied negative regulators of epidermal cell growth are the TGF-βs, which act at picomolar concentrations to inhibit DNA synthesis and cell division (Shipley et al., 1986; Kopan et al., 1987; Bascom et al., 1989a). The effect of TGF-βs on growth is reversible, at least within a 48 hour frame, and is not accompanied by gross changes in protein or RNA synthesis (Bascom et al., 1989a). While TGF-βs inhibit growth of basal cells, their natural expression in epidermis seems to be predominantly in suprabasal, differentiating layers. Indeed during mammalian development, expression of TGF-P2 mRNA (Pelton et al., 1989) and a related Vgr-1 mRNA (Lyons et al., 1989) coincide with epidermal stratification and keratinization. An interesting parallel occurs in Drosophila, where a TGF-β-like gene called decapenta-plegic (dpp) is expressed dorsally in early embryos, and mutations in the dpp gene cause a homeotic transformation of dorsal into ventral epidermis in the fly (Padgett et al., 1987; Arora and Nusslein-Volhard, 1992). In adult mammalian epidermis both in vivo and in vitro, TGF-β1 and TGF-β2 mRNAs are low, but detectable in the differentiating layers (Glick et al., 1989; Bascom et al., 1989b; Pelton et aL, 1989, and references therein). In adult epidermis, TGF-P mRNAs can be upregulated (1) in an autoregulatory fashion (Bascom et aL, 1989b), and (2) when mouse skin is treated with tumor promoting agent (TPA) (Fowlis et al.,1992), retinoic acid (Glick et al., 1990, 1991; see below) or calcium (Glick et al., 1990), agents known to influence epidermal keratinization. Curiously, TPA also induces TGF-α expression, presumably in basal epidermal cells (Pittlekow et al., 1989), a phenomenon which may explain the seemingly antagonistic effects of TPA, enhancing growth and inducing differentiation in epidermis. Collectively, the timing and location of TGF-β mRNA expression suggests that members of the TGF-β family may be involved in maintaining the cessation of growth in the differentiating cells of epidermis.

Elucidating the functional significance of TGF-p expression has been hampered by the fact that TGF-β are produced and secreted by cells in a latent form, which must then be activated prior to interaction with TGF-β receptors on the cell surface (Lyons et al., 1988; Pandiella and Massague, 1991). Mere elevation of TGF-p mRNA expression is therefore not automatically an indication that active TGF-β are being produced. An example of this is the production of latent TGF-β 1 by human keratinocytes cultured under serum-free conditions at neutral pH (Bascom et al., 1989a). Thus, even though the correlation between suprabasal TGF-β mRNA expression and cessation of growth in keratinizing cells is compelling, a causal relationship has not yet been unequivocally demonstrated, nor is it clear to what extent the expression of TGF-β mRNAs in vivo is an accurate reflection of active TGF-β secretion.

Because the effects of TGF-βs on basal cell growth appear to be largely reversible, it has been assumed that TGF-βs alone are not sufficient to induce terminal differentiation, a process thought to be irreversible. In support of this notion were early in vitro studies, showing that the biochemical indicators of terminal differentiation were not induced upon treatment of keratinocytes cultured on plastic with TGF-βs (Kopan et al., 1987; Bascom et al., 1989a). More recent studies with differentiating culture systems have revealed that at greatly elevated levels, TGF-βs can influence biochemical markers of keratinization, but at these high levels, they inhibit rather than promote, KI, K10 and filaggrin expression (Mansbridge and Hanawalt, 1988; Choi and Fuchs, 1990). At high concentration, TGF-βs also enhance expression of K6 (56 kDa) and K16 (48 kDa), keratins more commonly associated with suprabasal layers of epidermis undergoing (1) wound-healing (Mansbridge and Knapp, 1987) and (2) hyperproliferation (Weiss et al., 1984; Stoler et aL, 1988). Studies using floating epidermal cultures have shown that induction of K6 and K16 is accompanied by morphological changes typical of squamous cell carcinomas, including increased stratification, vacuolization and coilocyte formation (Choi and Fuchs, 1990). Interestingly, TGF-βs seem to elicit these changes in keratinizing epidermal cells at least in part independently of their action on basal cells. Collectively, these studies have led to the notion that the epidermal phenotype associated with woundhealing and many hyperproliferative diseases may not relate to hyperproliferation, per se, but rather is a reflection of environmental changes, which in some circumstances, may include active TGF-βs (Mansbridge and Hanawalt, 1988; Kopan and Fuchs, 1989b; Schermer et aL, 1989; Choi and Fuchs, 1990).

A priori, the apparent reversibility of TGF-β-mediated growth inhibition may seem ironic in light of more recent findings that (1) TGF-β mRNA expression is largely confined to suprabasal, terminally differentiating cells, and (2) TGF-βs can upregulate their own expression (Bascom et al., 1989b and references therein). However, the effects of TGF-βs at various stages of epidermal differentiation seem to be quite different (Choi and Fuchs, 1990), and hence it may be relevant that the reversibility studies were conducted on dividing cells of the population, while TGF-β mRNA synthesis and autoregulation seem to take place in non-dividing cells. In addition, it seems increasingly apparent that while cessation of cell growth by TGF-βs may be necessary, it is not sufficient for commitment to terminal differentiation. In a model where a cascade of biochemical changes is necessary to enter and maintain the differentiation state, TGF-β expression may be among the early steps in the pathway.

Recently, the TGF-β receptors have been cloned (Lopez-Casillas et al., 1991; Wang et al., 1991; Lin et al., 1992; Ebner et al., 1993). While the type III receptor is a surface proteoglycan and may regulate the ligand-binding ability of cells, the type I and II receptor encode a transmembrane serine-threonine kinase, indicating an important role for intracellular signalling. While the sequence of events that follow TGF-β activation of its receptors remains to be elucidated, the retinoblastoma gene product, pRb, seems to be involved. Paradoxically, however, it is the underphosphos-phorylated, rather than the phosphorylated, form of pRb that has a growth-suppressive effect, and TGF-β1 seems to maintain pRb in the underphosphorylated state (Laiho et al., 1990). In turn, pRb inhibits transcription of genes involved in growth control, such as c-myc, and curiously activates transcription of several growth inhibitory genes, including those encoding TGF-βs (Pietenpol et al., 1990; Kim et al., 1992 and references therein). As further studies are conducted, the molecular mechanisms underlying the TGF-β-mediated influence on epidermal growth and differentiation should become more apparent.

While TGF-βs seem to accentuate abnormal and inhibit normal differentiation, retinoids at high concentrations have long been known to have inhibitory effects on both forms of epidermal differentiation. Thus, for example, in organ culture of chick epidermis, vitamin A can induce its transition from a keratinizing to a secretory epithelium (mucous metaplasia) (Fell and Mellanby, 1953). Retinoic acid at 10⁻⁶ M in mammalian epidermal cultures can suppress dif-ferentiative features, including K1/K10 expression (Fuchs and Green, 1981), K6/K16 expression (Kopan and Fuchs, 1989b), cornified envelope production (Yuspa and Harris, 1974) and filaggrin expression (Fleckman et al., 1985). Many of these effects are at the level of mRNA expression (Fuchs and Green, 1981; Kopan et al., 1987; Kopan and Fuchs, 1989b) and transcription (Tomic et al., 1990; Stell-mach et al., 1991).

10⁻⁶ M retinoic acid inhibits proliferation and concomitantly induces secretion of active TGF-p2 in mouse ker-atinocyte cultures and in mouse skin (Glick et al., 1989; 1990). Conversely, when exposed to 10⁻⁷ to 10⁻⁶ M retinoic acid, human keratinocytes show an increase in proliferation and cell migration, with no detectable TGF-β2 induction (Choi and Fuchs, 1990). Whether these differences are a reflection of species-specific variations in dose-response to retinoids or variations in culture conditions awaits further investigation. Nevertheless, the discovery of an autocrine regulatory loop between retinoids and TGF-βs in some keratinocytes under some conditions is exciting, and suggests that both factors may be involved in some common regulatory pathways.

Determining the mechanism of action of retinoids on epidermal differentiation has been hampered by the fact that there are a number of intracellular regulators of retinoids. Among the first to be identified were cellular retinol binding protein (CRBP) and cellular retinoic acid binding protein (CRABP), fatty acid binding-like proteins initially thought to mediate the steroid hormone-like effects of retinoids on epidermal gene expression (Chytil and Ong, 1983), but now thought to play a role in the storage or transport of retinoids. More recently, a family of retinoid receptors has been identified with sequence homologies to classical steroid receptors (see Mangelsdorf et al., 1992 and references therein). Retinoic acid receptors (RAR) are transcription factors activated upon binding of RA, whereas RXR receptors become activated upon binding of 9-cis RA, (Zhang et al., 1992b). No analogous retinol receptors have been detected, and it therefore seems likely that the effects of retinol are mediated via intracellular conversion to RA or 9-cis RA and subsequent interaction with RARs or RXRs, respectively.

The control of gene transcription by RARs and RXRs is extraordinarily complex, involving a multitude of both indirect and direct mechanisms. RARs can heterodimerize with each other, with thyroid hormone receptors, and with RXRs (Forman et al., 1989; Kliewer et al., 1992; Yu et al., 1991; Zhang et al., 1992a). At least some of these interactions appear to change the DNA affinity and activity of RARs (Husmann et al., 1991; Zhang et al., 1992a). RARs can bind to thyroid response elements, retinoic acid response elements (RAREs) and retinoid X response elements (RXREs) (Mangelsdorf et al., 1991 and references therein). The repertoire of complex DNA interactions exhibited by RARs and RXRs is further expanded by their capacity to interact with API proteins, thereby conferring indirect transcriptional control of genes (Nicholson et al., 1990).

RAR-α, RAR-γ and RXRα are expressed in human epidermis (Elder et al., 1992 and references therein). RAR-γ and RXRα are expressed in skin and only a few other organs, whereas RARa is more broadly expressed (Krust et al., 1989; Mangelsdorf et al., 1992). Although in situ hybridization studies have not yet localized RXR mRNAs in skin, they have shown that RAR mRNAs are most abundant in the keratinizing layers of epidermis (Noji et al., 1989). Although the direct binding of RA-activated, suprabasal RARs to epidermal genes remains to be unequivocally demonstrated, RA-mediated biochemical changes in epidermal differentiation do appear to be mediated by retinoid response elements (Tomic-Canic et al., 1992). The suprabasal location of retinoid receptors might explain why RA inhibits suprabasal functions, and why culturing cells at an air-liquid interface enables suprabasal cells to differentiate via movement away from the retinol-containing serum.

While it is generally accepted that retinoids inhibit terminal differentiation in the epidermis, it has only recently been suggested that retinoids also act to enhance certain features of the differentiated process (Asselineau et al., 1989). This notion has been supported by examining the behavior of keratinocyte lines expressing a truncated RARγ capable of suppressing action at retinoid response elements (Aneskievich and Fuchs, 1992). In contrast to control lines, the tRARγ lines are basal-like and fail to differentiate. The mechanism by which this tRARγ functions is complex, and it does not seem to act merely as a dominant negative inhibitor of RARs. This said, if RARs/RXRs are able to regulate terminal differentiation in positive as well as negative fashions, this could explain the wide-reaching and often disparate effects of retinoids on epidermal cells in vivo and in vitro. As future studies are conducted, the pathways by which specific RARs or RXRs operate to control the expression of critical genes involved in the differentia-tive process should become more apparent.

When cultured in medium containing 0.05 mM calcium, murine keratinocytes grow as a monolayer, because desmosome assembly is inhibited (Hennings et al., 1980). Upon a switch to high (1.2 mM) calcium, desmosomes form and cells stratify. After 3-7 additional days in culture, epidermal transglutaminase is induced and squames containing cornified envelopes appear in the medium. These data indicate that many features of terminal differentiation can be induced by calcium in vitro. Calcium also seems to play a role in mediating keratinization in vivo: certain calcium ionophores have been shown to enhance action of TPA in promoting epidermal differentiation (Jaken and Yuspa, 1988).

Recent studies have indicated that calcium can act to control the transcription of terminal differentiation-specific genes, including keratins 1 and 10 (Yuspa et al., 1989; Rosenthal et al., 1991) and loricrin (Hohl et al., 1991). In addition, Glick et al. (1990) showed that calcium induces a marked increase in TGF-β2 mRNA expression in murine keratinocytes, thereby implicating calcium in both early and late stages of terminal differentiation.

In some cases, calcium can have a more pronounced effect on regulating differentiation-specific changes in cellular architecture than it has on transcription. Thus, for example, when human cells are cultured in low calcium medium, withdrawal from the cell cycle and involucrin synthesis still occur (Watt, 1984). Similarly, while desmosomes cannot assemble in low calcium, desmosomal proteins are nevertheless synthesized in both low and high calcium medium (Watt et al., 1984; Duden and Franke, 1988). Finally, while changes in calcium concentrations do not seem to have a major effect on filaggrin expression in human keratinocytes (Fleckman et al., 1985), the leader peptide of profilaggrin has a calcium binding domain, suggesting that calcium plays an important role in the formation of keratohyalin and/or the subsequent processing of profilaggrin to filaggrin (Presland et al., 1992). Hence, like other known regulators of epidermal differentiation, calcium seems to have pleiotropic effects.

In summary, we are left with a picture whereby a keratinocyte becomes a terminally differentiating cell as a consequence of a series of biochemical checks and balances. There seem to be several early changes necessary for entry into the differentiation pathway, and among these are likely to be TGF-Ps and other agents that slow or inhibit DNA synthesis and cell division. In normal epidermis, movement of a cell from its basement membrane may contribute to, but does not seem sufficient for, instigating a cascade of biochemical changes that culminate in irreversibly sealing the differentiative fate of a keratinocyte. Once a basal cell has left the innermost layer and its biochemical program has begun to change, it seems to be further influenced by retinoids, TGF-βs and calcium primarily in a fashion that appears to allow retailoring of the architecture of the keratinizing layers above. Further investigation will be necessary to determine whether this capacity to redesign the suprabasal program of differentiation affects basal as well as suprabasal cells.

A knowledge of the major transcription factors controlling keratinocyte-specific gene expression will be of central importance in the quest to elucidate the molecular mechanisms underlying epidermal differentiation. A number of keratinocyte genes have been isolated and characterized, and these serve as a foundation for pursuing factors controlling keratinocyte specificity. They include genes encoding (a) basal keratins K5 (Lersch et al., 1989) and K14 (Marchuk et al., 1984), (b) suprabasal keratins KI (Johnson et al., 1985), K10 (Rieger and Franke, 1988), K6 (Tyner et al., 1985) and K16 (Rosenberg et al., 1988), (c) a cornified envelope protein, involucrin (Eckert and Green, 1986) and (d) filaggrin (Rothnagel and Steinert, 1990; Presland et al., 1992). Sequence comparisons provide some clues to possible common regulatory elements. One is a putative retinoid response element, found upstream of a number of epidermal genes, and recently suggested to be functional for the human keratin 14 gene (Tomic-Canic et al., 1992). Another is the CK 8-mer sequence 5′AANCCAAA 3′, found upstream from a number of epidermal genes (Blessing et al., 1987; Cripe et al., 1987). Blessing et al. (1989) showed that a ∼90 bp fragment containing a TATA box and a CK 8-mer sequence provided a 4-fold enhancement of expression of a CAT reporter gene, but Chin and Chow (1989) was unable to demonstrate function for the CK 8-mer sequence in the E6/E7 promoter of a human papillomavirus genome. Hence, if a CK 8-mer-like sequence is involved in keratinocyte gene expression, it may act in concert either in multiple copies or with other regulatory elements in addition to a TATA box.

Several years ago, we identified a different sequence, 5′GC CTG C AGG C3′, located 5′ from the TATA box of the human K14 gene, that appears to act in conjunction with a distal element to control its transcription in ker-atinocytes (Leask et aL, 1990). This site also bound a transcription factor, AP2, which was significantly more abundant in epidermal keratinocytes than in other cell types in culture (Leask et al., 1990, 1991). The cDNA for AP2 has been cloned, and in situ hybridization studies have revealed that in vivo, AP2 is expressed in a highly restricted tissuespecific fashion, with cells of neural and epidermal lineage being the predominant AP2 expressers (Mitchell et al., 1991). In addition, the Xenopus equivalent of AP2 has been shown to play a role in embryonic development of the frog epidermis (Snape et al., 1990, 1991). AP2 has been implicated broadly in the regulation of other endogenous and viral genes that are typically expressed in keratinocytes (see also Royer et al., 1991), and AP2 sites seem to play some role in restricting keratinocyte-specific expression in vivo (Byrne and Fuchs, 1993). This said, AP2 does not appear to be sufficient for keratinocyte specificity on its own (Leask et al., 1990).

Recently, we found that 90 bp of K5 promoter, missing RARE, AP2 and CK8-mer binding sites, was sufficient to direct expression of a reporter β-galactosidase gene predominantly to the epidermis and tongue of transgenic mice, albeit in the suprabasal, rather than basal, compartment, and with some promiscuous expression in a few non-keratinocyte-containing tissues (Fig. 3A, K5βgal90 transgenic foot skin; compare with Fig. 3B, illustrating foot skin from transgenic mouse expressing K5βgal6000, containing 6000 bp of 5′ K5 sequence; see also Byrne and Fuchs, 1993). Replacement of the AP2 site did not restore basal expression although it may have restricted some promiscuity in expression (Byrne and Fuchs, 1993). The severely truncated K5 promoter segment also exhibited cell type specificity in culture (Byrne and Fuchs, 1993). These findings enabled us to focus on a relatively small segment of DNA, which we showed bound a number of proteins, some of which are common to both the K5 and K14 promoters (Fig. 3C). These include Spl, which has a ubiquitous but not necessarily homogenous distribution in vitro (Robidoux et al., 1992), and a protein complex, which we have tentatively termed 1-2, which binds in the vicinity of the transcription and translation initiation regions of these two promoters. The protein complexes that bind to the 90 bp promoter segment are present in a variety of cultured cell types, and thus, at least in vitro, the restricted expression conferred by the truncated K5 promoter does not appear to be achieved by an epidermal-specific, DNA-binding transcription factor. Whether this is also the case in vivo remains to be shown. In this regard, it is notable that in vivo, Spl seems to be expressed in a tissue-specific fashion, while in vitro, it is not (Saffer et al., 1991; Robidoux et al., 1992).

Our in vitro studies suggest that the specificity of the truncated K5 promoter may be governed by cell-type specific variations, either in the modifications of the factors that bind to the truncated K5 promoter, or in the non-DNA binding proteins that might associate with these DNA-binding proteins. Whatever the mechanism, protein complex 12 seems to be involved since mutations that interfere with protein 1-2 binding also obliterate cell-type specific expression of the K5 promoter in vitro (Byrne and Fuchs,1993). The protein 1-2 binding sequence (GTTC-CTGGGTAAC) is similar to the consensus binding sequence (GTTAATNATTAAC) for the POU-home-odomain transcription factors HNF-la and HNF-10, involved in liver-specific gene expression (reviewed by Johnson, 1990). While proteins 1-2 neither share the celltype specificity nor the precise binding specificity of HNF-lp and HNF-ip, further studies will be necessary to ascertain whether they might nevertheless be related members of the same family.

This said, it is also possible that this complex may be involved in initiating transcription (Smale and Baltimore, 1989). Such transcription initiating factors have been variously termed 8 (Hariharan et al., 1991), YYI (Shi et al., 1991), TFII-I (Roy et al., 1991), NF-E1 (Park and Atchison, 1991), UCRBP (Flanagan et al., 1991), and are either identical binding factors or belong to a similar classes of factors (Roy et al., 1991; Seto et al., 1991). These factors interact with TFIID, and they can bind DNA and initiate transcription even in the absence of a TATA box. These putative roles for protein 1-2 are intriguing, especially since these types of elements are not necessarily fixed and may be found distal to the initiation site (Nakatani et al., 1990). Additional studies will be needed to fully elucidate the role of proteins 1-2 in keratinocyte specificity.

While some headway has been made in the quest to uncover the factors and sequences that orchestrate tissuespecific gene expression in the epidermis and other stratified squamous epithelia, underlying questions remain concerning how transcriptional factors act at a molecular level to control the balance between gene expression in the basal and suprabasal layers. In this regard, it is clear that epithelial-specific and basal-specific gene expression can be uncoupled, since (a) 90 bp of the basal K5 promoter can target expression in transgenic mice to the suprabasal epidermal layers (Byrne and Fuchs, 1993) and (b) 1500 bp of the suprabasal KI promoter can target expression in transgenic mice to both the suprabasal and the basal layers (Rosenthal et al., 1991).

Finally, while calcium and retinoids are likely to play a role in the molecular switch of basal and suprabasal epidermal genes, there are probably additional environmental cues, based on our recent studies with the truncated 90 bp K5 promoter. The expression pattern of this gene was unusual in epidermis and tongue in that it was not analogous to any known keratin gene, but rather appeared to be a regional-dependent subset of the combined activities of a number of suprabasal genes, including those encoding K1/K10, K6/K16, K4/K13 and K9 (Byrne and Fuchs, 1993). Thus, the truncated promoter appeared capable of responding to regional variations, and displayed an environmental sensitivity that is absent in the intact promoter from which it was derived. The peculiar behavior of the transgene in skin suggests that the regulatory elements controlling expression of a keratin in one location of the body can differ from those governing expression even in the same tissue, but in a different region of the body. If true, the regulation of keratin gene expression is significantly more complex than previously realized. In the coming years, a major focus in the field of epidermal differentiation will be on keratinocyte-specific and differentiation-specific transcription factors as possible mediators of extracellular regulators and keratinocyte fate.

I am grateful to all of the past and present members of my laboratory for their numerous contributions to the research efforts that were described in this review. In particular, I would like to thank the following scientists for their substantial contributions to the research reviewed here: Dr Robert Vassar, Dr Carolyn Byrne, Dr Brian Aneskievich, Dr Andrew Leask, Dr Raphael Kopan and Lifei Guo. I would also like to thank Lifei Guo, Dr Carolyn Byrne and Dr Brian Aneskievich for providing the photographs used in the Figures in this paper. Work in the author’s laboratory is funded by a grant from the National Institutes of Health and from the Howard Hughes Medical Institute.