Suprabasal change and subsequent formation of disulfide-stabilized homo- and hetero-dimers of keratins during esophageal epithelial differentiation

Keratin is a group of about 30 cytoskeletal proteins that form intermediate filaments in almost all epithelial cells (Moll et al., 1982; Lynch et al., 1986; Heid et al., 1986). Detailed studies of keratin expression in epidermis both in vivo and in cultured cells showed previously that the basal cells of this stratified squamous epithelium synthesize a basic keratin K5 and an acidic keratin K14 (forming a keratin “pair”; Eichner et al., 1984; Sun et al., 1984), while the suprabasal cells synthesize mainly K1 and K10 keratins (Woodcock-Mitchell et al., 1982; Skerrow and Skerrow, 1983; Schweizer et al., 1984; Stoler et al., 1988). During wound healing, cell culture and other conditions which somehow suppress the expression of K1 and K10 keratins, the suprabasal cells synthesize instead K6 and K16 (and/or K17) keratins (Sun and Green, 1978; Fuchs and Green, 1978; Stoler et al., 1988). Similar studies on corneal epithelium showed that, as in the epidermis, its basal cells make K5/K14 keratins, and that under nonpermissive conditions its suprabasal cells synthesize K6/K16. However, under permissive conditions the more differentiated corneal epithelial cells make K3/K12 (rather than the epidermal K1/K10; Schermer et al., 1986, 1989). Fewer studies were conducted, however, on the in vivo and in vitro differentiation-related changes in keratin expression of esophageal epithelium - a third important keratinocyte prototype - which is “parakeratinized” in rabbit, human and bovine (Doran et al., 1980; Tseng et al., 1984). This epithelium expresses as its main differentiation products another keratin pair K4/K13 (Moll et al., 1982; Sun et al., 1984). Immunohistochemical staining data generated using monoclonal antibodies specific for these keratins have previously shown that these two keratins are mainly detected in the suprabasal cell layers of esophageal epithelium (Van Muijen et al., 1986). A similar suprabasal expression pattern was observed from in situ hybridization experiments (Knapp et al., 1986; Rentrop et al., 1986; cf., Leub et al., 1988). Although these data imply that K4/K13 keratin proteins are localized mainly suprabasally, immunohistochemical staining results can be complicated by false negatives due to antigenic masking (Woodcock-Mitchell et al., 1982; Franke et al., 1983). In situ hybridization generates more reliable RNA localization data, which, however, do not always correspond to the protein data. Therefore, it remains to be established on a protein level that K4 and K13 keratins are indeed localized specifically in the suprabasal compartment.

An important feature of keratinocytes is that they undergo terminal differentiation, forming superficial squamous cells containing disulfide-crosslinked keratins. Although it is well known that such crosslinking serves to improve the physical stability of the keratin network, virtually nothing is known about the biochemical details of this process. For example, it is unclear whether the crosslinking occurs between the acidic and basic keratin members of a pair, thus forming a stabilized heterodimer (or oligomers), or between keratins of the same kind forming homodimers (or oligomers). It is difficult to address this kind of question by studying the skin differentiation-related keratin pair K1/K10 or the hair differentiation-related keratin pairs, because our preliminary experiments indicate that under conditions favoring disulfide bond formation these keratins rapidly form crosslinked products far beyond the dimer stage, making then extremely difficult to analyze (Sun and Green, 1978; Pang and Sun, unpublished observation). In this regard, esophageal epithelial keratins provide an optimal model system for tackling this particular problem.

In this paper, we describe: (i) the characterization of keratins synthesized by rabbit esophageal epithelium both in vivo and in culture by one- and two-dimensional immunoblotting; (ii) a direct, biochemical analysis of keratins associated with the basal and suprabasal compartments of rabbit esophageal epithelial culture; and finally (iii) the characterization of two heterodimer (K4/K13) and one homodimer (K13-K13) that become disulfide-crosslinked during an advanced stage of in vivo esophageal epithelial differentiation. These results provide strong evidence that K4 and K13 keratins are associated mainly with the suprabasal compartment of esophageal epithelium, and clearly indicate that the formation of disulfide-crosslinked homo- and heterodimers of K4 and K13 keratins is an important step during an advanced stage of in vivo esophageal epithelial differentiation.

Water-insoluble extracts from a variety of human, bovine and rabbit epithelial tissues and cultures were prepared. Human tissues, except corneas, were acquired from autopsy; bovine tissues from freshly slaughtered cows; and rabbit tissues from freshly killed adult New Zealand White (NZW) rabbits. The human corneas were excised from eyes kindly provided by New York Eye Bank. In vivo epithelial tissues were harvested by (i) scraping either at room temperature (cornea, esophagus, intestine, bladder) or over solid CO₂ (epidermis) or (ii) incubating the tissues in 2.5 mg/ml dispase (Boehringer Mannheim) in Dulbecco’s Modified Eagle’s Medium (DMEM) for 16 h at 4°C to separate the epithelium from its underlying mesenchymal tissue. Confluent cell cultures of rabbit corneal epithelium, rabbit epidermis, rabbit esophageal epithelium, human epidermis, human mesothelium, cow snout epithelium and MDBK cells were rinsed with phosphate-buffered saline (PBS) and harvested with a plastic scraper (Costar).

All tissues were immediately transferred to cold 25 mM Tris-HCl (pH 7.4), 0.6 M KC1, 1% Triton X-100, plus a mixture of five protease inhibitors to remove the water-soluble proteins (Woodcock-Mitchell et al., 1982; Eichner et al., 1984, 1986). The remaining cytoskeletal preparation containing mainly keratins (“native” intermediate filament) was extracted with 9.5 M urea, 10 mM Tris-HC1 (pH 7.4) and stored frozen at −70°C.

In studies of invivo disulfide-crosslinking, a mid-esophagus was isolated from a freshly killed NZW rabbit and its epithelium was scraped off and extracted as above in the presence or absence of 10 mM N-ethylmaleimide (NEM). The remaining cytoskeletal preparation was then solubilized directly with either 9.5 M urea or 1% SDS buffer.

AE8 is a mouse monoclonal antibody raised against SDS-denatured rabbit esophageal keratin extract, and was selected for its specificity to acidic K13 keratin (see Results). AE1, AE3, AE14 and aIF monoclonal antibodies were prepared as described previously. AE1 recognizes acidic K10, K11, K14, K15, K16 and K19 keratins (Sun et al., 1984; Cooper et al., 1985); AE3 recognizes all known basic keratins (Sun et al., 1984; Cooper et al., 1985); AE14 recognizes the basic K5 keratin characteristic of basal cells of almost all stratified squamous epithelia (Lynch et al., 1986); and a-IFA recognizes most intermediate filament proteins (Pruss et al., 1981; Cooper et al., 1985).

Cultures of human epidermal cells, rabbit epidermal cells and rabbit esophageal epithelial cells were grown in the presence of lethally irradiated 3T3 feeder cells in DMEM containing 17% fetal bovine serum, 0.5 μg/ml hydrocortisone and 15 ng/ml epidermal growth factor. Cell cultures were maintained at 37°C in a humidified atmosphere of 5% CO₂ in air (Rheinwald and Green, 1975, 1977). MDBK cells (a bovine kidney-derived simple epithelial cell line) were obtained from the American Type Culture Collection (ATCC CCL22) and were grown in DMEM containing 10% calf serum.

Tissues were embedded in OCT medium, and snap frozen in isopentane near its freezing point. Frozen sections (6-8 μm) of rabbit esophageal epithelium were stained by indirect immunofluorescence as described (O’Guin et al., 1985).

Postconfluent, stratified cultures of rabbit esophageal and corneal epithelial cells were treated with 2 mM EGTA in calcium-free DMEM (MA Bioproducts) containing 17% fetal calf serum for 5 hours at 37°C. This was followed by vigorous pipetting of the supernatant medium over the colonies. Cells that dislodged into the supernatant were washed with PBS, collected by centrifugation and processed for SDS-PAGE (Schermer et al., 1986).

Frozen sections (7 μm thick) of rabbit esophageal epithelium from a freshly killed animal were stained with DACM (N-[7-dimethylamino-4-methyl-3-coumaring]-maleimide), which is specific for SH groups (Ogawa et al., 1979). A Zeiss fluorescence microscope equipped with an ultraviolet filter was used to detect the fluorescence of DACM.

Rabbit esophageal epithelium from a freshly sacrificed animal was frozen and horizontal sections of 12 μm were cut (Fuchs and Green, 1978). Total proteins were extracted immediately from the sections with sample buffer containing 1% SDS with or without 1% beta-mercaptoethanol. Every third section was loaded on three separate gels and proteins were separated by SDS-PAGE.

Keratin extracts were incubated in 50 mM Tris-HCl (pH 8.5), 5 mM dithiothreitol, and 9.5 M urea for 30 min at 37°C, centrifuged for 10 min at 50,000 g, and the supernatants dialyzed against 10 mM Tris-HCl (pH 7.5) for 16 h at 4°C (Steinert et al., 1976; Sun and Green, 1978; Aebi et al., 1983; Hatzfeld and Franke, 1985; Eichner et al., 1986).

Air-oxidized, reconstituted keratin filaments were digested with 0.001 to 0.01 mg/ml of trypsin (Sigma) in 10 mM Tris-HCl (pH 8.7) at room temperature for 30 min. The reaction was stopped by adding excess amount of trypsin inhibitor (Sigma).

To facilitate the analysis of esophageal keratins, we immunized Balb/c mice with rabbit esophageal keratins, generated hybridomas, and screened by Elisa for antibodies that reacted with esophageal keratins without cross-reacting with human epidermal keratins. A panel of such esophageal-specific antibodies were cloned and characterized, and the properties of one of them, AE8, is described here because of its superior stability, broad species reactivity and immunostaining properties.

To establish the keratin chain-specificity of this antibody, we used it to perform one- and two-dimensional immunoblots against the keratins of a selected panel of human, cow and rabbit epithelia. These studies indicated that AE8 recognized the major acidic keratin of esophageal epithelia of all three animal species (Fig. 1). The tissue-distribution and two-dimensional gel electrophoretic positions of this keratin, which had an apparent molecular mass of 51 kDa in man, 41 kDa in cow, and 46 kDa in rabbit, clearly established that it was keratin 13 (Fig. 2).

Although it is well known that keratins of cultured epidermal keratinocyte differ from their in vivo counterparts (Fuchs and Green, 1978; Sun and Green, 1978; Eichner et al., 1984), the difference between the keratin pattern of cultured rabbit esophageal epithelial cells and that of in vivo rabbit esophageal epithelium is even more striking in that the former contained large amounts of K5/K14 “basal” keratins, K6/K16 “hyperproliferation” keratins, plus the K4/K13 differentiation type keratins, while the latter contained almost exclusively K4/K13 (Figs 3 and 4). Such a difference could be explained by several considerations: (i) the in vivo esophageal epithelium (with >30 cell layers) is much thicker than the cultured epithelial colonies (only <5-6 layers); hence the basal K5/K14 keratins contribute quantitatively much less to the overall in vivo keratin pattern. (ii) Even though under normal conditions K5/K14 synthesis is turned off in the suprabasal layers, under hyperproliferative conditions this synthesis may continue even in the suprabasal compartment (Stoler et al., 1988; Leube et al., 1988). (iii) Keratinocytes cultured under standard, submerged conditions tend to synthesize relatively little differ-entiation-related keratins (in this case the K4/K13 keratins), and to synthesize an elevated level of the so-called hyperproliferation-related K6/K16 keratins (Weiss et al., 1984; Cooper et al., 1985; Sun et al., 1985).

Direct analysis of the keratin proteins in different compartments of a stratified epithelium requires cell fractionation. Although we have not been able to adequately separate the basal vs suprabasal cells of normal rabbit esophageal epithelium, we succeeded in separating these two cell populations from cultured rabbit esophageal epithelial cells using an EGTA-protocol. In this protocol (which we originally developed to separate the basal and suprabasal cells of rabbit corneal epithelial culture; Schermer et al., 1986), confluent cultures of heavily stratified esophageal epithelial cells were treated with a solution of EGTA, which resulted in the selective detachment of all suprabasal cells (Schermer et al., 1986). We then analyzed the keratins of the basal (lanes labeled b in Fig. 5) and suprabasal (s) cells, compared with those of an unfractionated cell population (u), by one- and two-dimensional immunoblotting (Figs 5 and 6). The results indicated that the isolated basal cells contained mainly K5/K14 (Figs 5 and 6); there was also a small amount of a 55 kDa band (Fig. 5a, lane b), which was AE3-negative and most likely represented vimentin of the contaminating feeder/fibroblasts. The suprabasal cells contained the same basal K5/K14 keratins, but in addition they contained K6/K16 and K4/K13 keratins (Figs 5 and 6).

Consistent with its K13 specificity, AE8 stained positively the frozen sections of all human, bovine and rabbit epithelial tissues known to contain K13 (Fig. 7 and data not shown). In esophageal epithelium it stained strongly all the upper cell layers (Fig. 7d). The basal cells showed only weak but discernibly filamentous staining (Fig. 7d). The staining pattern of rabbit dorsal tongue epithelium is interesting in that, although the bottom of deep rete ridges showed “suprabasal” staining, the shoulders of these rete ridges exhibited almost “uniform” staining (basal cell also stained positively; Fig. 7e). Taken together, these in vivo and cultured cell results provided strong support to the notion that in esophageal epithelium K13 was present in relatively small amounts in basal cells, but its level increased significantly in suprabasal cell layers (Van Muijen et al., 1986; Knapp et al., 1986; see Discussion).

As mentioned earlier, esophageal epithelial keratins are uniquely suitable for studying disulfide bond-mediated crosslinking of keratins. Using a histochemical staining agent, DACM (Ogawa et al., 1979), we showed that although the entire thickness of esophageal epithelium contained free sulfhydryl group (Fig. 8a), only the superficially located, terminally differentiated squamous cells were disulfide-rich (Fig. 8b). To determine whether this increase in disulfide content parallels the covalent crosslinking of keratins, we horizontally sectioned rabbit esophageal epithelium, extracted their total proteins with 1% SDS, and analyzed these proteins by SDS-PAGE (with or without mercaptoethanol; Fig. 9). The keratins in these fractions were then visualized using a mixture of several mouse monoclonal anti-keratin antibodies (Fig. 9b). Most of the viable esophageal epithelial cell layers contained, as expected, the K4 and K13 monomers. However, the superficial cell layers contained three additional high molecular mass (110 kDa, 100 kDa and a weak 90 kDa) keratin bands which were sensitive to reduction (Fig. 9c). Although in these fractions the crosslinked bands accounted for only 20-30% of the total keratin, the true value must be significantly higher, since these superficial cell fractions were heavily contaminated by lower viable cell layers (see Discussion).

We determined the composition of these crosslinked ker-atins by immunoblotting using two monoclonal antibodies - the AE3 antibody, which has been shown previously to react with all known basic keratins including the major esophageal basic keratin K4 (Sun et al., 1984; Lynch et al., 1986), and AE8 antibody, which (as shown in Figs 1 and 2) recognizes K13. The results indicated that the 110 kDa and 100 kDa bands (labelled I and II, respectively, in Figs 9 and 10) were heterodimers of the 59 kDa K4 and the 41 kDa K13, while the 90 kDa band (III) was a homodimer of K13. Additional analyses of these samples by twodimensional “diagonal” gel electrophoresis (Hatzfeld and Weber, 1990), in which the sample was reduced after the first-dimensional separation, established clearly that bands I and II were indeed heterodimers made up of about equimolar amounts of K4 and K13, whereas the 90 kDa band III was a homodimer of K13 (Fig. 11).

To rule out the possibility that these crosslinked bands were formed artifactually during the extraction procedure, we solubilized total proteins from intact rabbit esophageal epithelium under a number of conditions including 9.5 M urea or by heating the tissue directly in 1% SDS (Fig. 9b, lane 4). We also extracted the tissue in the presence of 10 mM N-ethylmaleimide in order to block free sulfhydryl groups that are known to be required for disulfide exchange to occur (Fig 9b, lane 4). Similar results were obtained under all these conditions, suggesting that the three high molecular mass bands as shown in Figs 10 and 11 must represent genuine crosslinked keratin species existing in vivo.

To determine whether one can reproduce the formation of these disulfide-crosslinked dimers under well-defined in vitro conditions, we reconstituted esophageal keratin filaments by dialyzing urea-solubilized rabbit esophageal keratins into 10 mM Tris-HCl (pH 7.8) in the absence of a reducing agent (Lee and Baden, 1976; Steinert et al., 1976). Analysis of such air-oxidized filaments by SDS-PAGE showed the formation of bands I and II, with a yield of about 20-30% (Fig. 12). Diagonal gel analyses showed that, like the in vivo bands I and II, these two bands represented heterodimers of K4 and K13 (Fig. 13). Small amounts of tetramers consisting of K4 and K13 in a 1:1 molar ratio were also detected (Fig. 13).

Although under mild oxidative conditions esophageal keratins can be crosslinked in vitro to form heterotypic dimers and tetramers, no homodimer of K13 (band III) was formed. To determine whether reconstitution in the absence of a reducing agent might have resulted in precocious disulfide crosslinking (giving rise to bands I and II), which somehow hindered band III formation, we first reconstituted filaments in the presence of a reducing agent, which was then removed during a second step of dialysis. Although the two heterodimers were again generated, K13 homodimer still did not form, indicating that under in vitro conditions we can reproduce only partially the disulfide crosslinking reactions occurring in vivo.

To determine whether the disulfide crosslinking of heterodimers involved cysteine residues located in the alphahelical rod domains or the non-helical end domains, we digested air-oxidized, reconstituted filaments with trypsin (Fig. 14). As the trypsin concentrations increased from 0.001 mg/ml to 0.01 mg/ml, the non-helical regions of the keratins were progressively digested, leaving behind a trypsin-resistant helical core 40-45 kDa in size (see the reduced samples in Fig. 14a, lane 4; Steinert et al., 1983). Analyses of samples in the absence of a reducing agent showed that the crosslinked heterodimers (bands I and II) were gradually degraded to a 80-90 kDa trypsin-resistant species (Fig. 14a, lane 4) containing both basic K4 (Fig. 14b) and acidic K13 keratins (Fig. 14c). These results suggest that at least some of the disulfide crosslinks must involve cysteine residues residing in the trypsin-resistant rod domains of K4 and K13 keratins.

By analyzing the keratins of the basal and suprabasal compartments of cultured rabbit esophageal epithelial cells, we have been able to provide for the first time direct evidence that: (i) esophageal basal cells contain primarily K5/K14 keratins (Figs 5 and 6). We and others have previously shown by immunohistochemical and in situ hybridization techniques that basal cells of several stratified squamous epithelia contain K5 and K14 (Woodcock-Mitchell et al., 1982; Stoler et al., 1988; Leube et al., 1988). Biochemical analysis of the keratins of highly purified basal cells of the epidermis and corneal epithelium provided direct proof that these two keratins indeed represent the major keratins of the basal layer (Skerrow and Skerrow, 1983; Bowden et al., 1984; Schermer et al., 1986, 1989). We have now shown that this is also the case for esophageal epithelium, a third important keratinocyte prototype, thus extending the generality of this concept. (ii) K4 and K13 keratins are associated exclusively with the suprabasal compartment of cultured rabbit esophageal epithelial cells (Figs 5 and 6). This result confirms and extends the findings of Van Muijen et al. (1986), who showed that their IC7 and 2D7 monoclonal antibodies selectively stained the suprabasal cells of human esophageal epithelium, and those of Rentrop et al. (1986), who showed by in situ hybridization that K13 mRNA is localized mainly in suprabasal cells of mouse tongue (for conditions allowing the expression of K13 in basal cells, see below). Our results are also fully compatible with those of Schweizer et al. (1988), who showed that mouse K4/K13 and K1/K10 are associated with the isolated suprabasal cells of mouse forestomach epithelium. (iii) The K6/K16 keratins are located in the suprabasal compartment (Figs 5 and 6). By analyzing keratins of basal and suprabasal cells isolated from cultured rabbit corneal epithelium, we have previously provided the first direct evidence that K6/K16 keratins are associated with the suprabasal compartment (Schermer et al., 1986, 1989). A similar conclusion was reached by Stoler et al. (1988), who demonstrated immunohistochemically that an antibody to K6 keratin stains the suprabasal cells of human psoriatic epidermis. We have now proved that, in cultured rabbit esophageal epithelial colonies, K6/K16 are also associated exclusively with the suprabasal compartment (Figs 5 and 6). Taken together, these data provide strong evidence that K6/K16 keratins are markers for the suprabasal compartment of all three major prototypes of stratified squamous epithelium, i.e. epidermis, corneal epithelium and esophageal epithelium.

We described in 1984 that K6 and K16 keratins may be regarded as markers for “hyperproliferation” (Weiss et al., 1984). This was based on our finding that these keratins are present in all hyperproliferative human epidermal diseases that we have analyzed so far including psoriasis, actinic keratosis and squamous cell carcinoma, and that these two keratins are synthesized in large quantities in all kinds of cultured keratinocytes (McGuire et al., 1984; Weiss et al., 1984). However, it is important to note that although all hyperproliferative keratinocytes seem to express K6/K16, the reverse is not true, i.e. not all K6/K16-positive stratified epithelia are hyperproliferative. For example, outer root sheath cells (of hair follicle), which are by no means “hyperproliferative”, express K6/K16 perhaps as their normal differentiation products (Stark et al., 1987). Cultured human foreskin epidermal cells continue to synthesize K6/K16 even when the cells become senescent and apparently stop growing (Schermer et al., 1989). Cultured rabbit corneal epithelial cells continue to synthesize K6/K16 even when cell proliferation is completely inhibited by DNA synthesis inhibitors (Schermer et al., 1989). On the basis of these data and the suprabasal location of K6/K16, we think that their expression is perhaps more related to “non-permissive” conditions under which the expression of the differentiation-related suprabasal keratins (i.e. K1/K10 of the epidermis, K3/K12 of cornea and K4/K13 of esophagus) is suppressed. Under such conditions, K6/K16 appears to be turned on by a default mechanism. If so, then these two keratins may be regarded more appropriately as markers for an advanced stage of an “alternative pathway” of keratinocyte differentiation (Galvin et al., 1989; Schermer et al., 1989).

A corollary of the above idea is that the synthesis of K6/K16 should be inversely related to that of the differentiation-related keratins, which is precisely what we have observed in both human epidermis (Weiss et al., 1984) and cultured rabbit corneal epithelial cells (Schermer et al., 1986, 1989). The biochemical basis for this “mutually exclusive” expression of these two major classes of suprabasal keratins (the differentiation type versus the hyperproliferation type) is not understood. However, these two types of keratins most likely can be co-expressed in the same suprabasal cells, since almost 100% of the suprabasal cells in cultures of rabbit esophageal and corneal epithelium are K13- and K3-positive, respectively (Schermer et al., 1986, and data not shown).

Although in confluent cultures of rabbit esophageal cells the K13 keratin is found exclusively in the suprabasal cell layer and thus may be regarded as a marker for an advanced stage of “esophageal-type” differentiation, under several in vivo and cell culture conditions this keratin can be seen to be associated with the basal cells: (i) although there is no doubt that the suprabasal cells of normal rabbit esophageal epithelia are stained much more intensely by AE8 than the basal cells, these in vivo basal cells consistently show weak but discernibly filamentous AE8 staining (Fig. 7d), suggesting that they do contain small amounts of K13. (ii) In rete ridges of rabbit dorsal tongue epithelium K13 is clearly present in the basal cells that line the “shoulder” of the rete ridges, suggesting that, as far as K13 marker is concerned, these basal cells may have attained a more differentiated state than the basal cells located at the bottom of the rete ridges (Fig. 7e). This expression pattern is reminiscent of our previous finding on corneal/limbal epithelium where we found that central corneal epithelial basal cells, unlike limbal basal cells, are K3-positive; this and several other observations have led us to develop the concept that corneal epithelial stem cells reside in the basal layer of limbal zone (Schermer et al., 1986; Cotsarelis et al., 1989; Tsai et al., 1990). The K13 expression pattern may suggest a similar situation; i.e. the K13-negative basal cells located at the bottom of the rete ridges may represent the stem cells of dorsal tongue epithelium (Hume and Potten, 1976). (iii) K13 keratin can be readily detected in the basal-like A431 cells (data not shown). These results indicate that K13 expression is not incompatible with cell proliferation (for related data on other differentiation-related keratins, K1, K3 and K4, see Blessing et al., 1989; Schermer et al., 1986; and Leube et al., 1988; respectively). They also raise the interesting possibility that the molecular mechanisms underlying the suprabasal expression of K4/K13 and the corneal K3/K12 keratins (all of which have been seen to express in specific populations of basal cells) may be more “flexible” than that of epidermal K1/K10, which are rarely seen to be expressed in the basal layer.

Our data indicate that the formation of cystine-stabilized keratin dimers is a major event that occurs during the terminal stage of esophageal epithelial differentiation (Figs 8 and 9). It is interesting to note that, unlike epidermal and hair keratins, which form high molecular mass disulfidecrosslinked polymers (Sun and Green, 1978; Woodcock-Mitchell et al., 1982; Heid et al., 1986; Lynch et al., 1986; Pang and Sun, unpublished observation), the crosslinking of esophageal epithelial keratins is largely limited to the formation of dimers (Figs 10-13). This particular feature of esophageal keratin has important functional implications (see below).

The three crosslinked dimers that we identified in esophageal epithelium are unlikely to be artifacts produced during sample preparation. First, the formation of these dimers parallels closely the disulfide abundance in vivo as assayed with DACM (Figs 8 and 9). Second, the same three crosslinked dimers can be extracted not only with 9.5 M urea but also by solubilizing the tissue in 1% SDS (Fig. 10b, lane 4). Inclusion of EDTA, or 10 mM NEM, which blocks sulfhydryl groups needed for disulfide exchange, also did not alter the pattern of these crosslinked products (Fig. 10b and data not shown). Finally, at least the two heterodimers can be faithfully reproduced on reconstituted keratin filaments (Figs 12 and 13).

The two heterodimers have several implications. First, they have the same protein composition (Figs 10 and 11) and are therefore isoforms possibly generated through the crosslinking of different sulfhydryl groups. Second, as mentioned above, they can be reproduced on reconstituted filaments, suggesting that soluble proteins or enzymes removed during the preparation of cytoskeleton are not necessary for their formation (Figs 12 and 13). Third, since such heterodimers can survive trypsinization, which removes the head and tail domains of the keratin molecule (Fig. 14), the crosslinking must involve at least some of the cysteine residues of the helical cores. Fourth, the yield of the heterodimer crosslinking is relatively high in vivo (certainly greater than 50%; Fig. 9), but is relatively low on reconstituted fila ments (<20%; Fig. 12). This may reflect an imperfection or polymorphism in certain structural aspects of the reassembled filaments (see below). Alternatively, this may be due to subtle differences between keratins of the lower, viable cell layers and those of the superficial cell layers in terms of secondary modifications such as phosphorylation and minor proteolytic digestion. Finally, these heterodimers are formed during the terminal stage of esophageal epithelial differentiation (Fig. 9). Their formation is therefore a late event, most likely designed to stabilize the final filament network/bundle. In this regard, it is important to distinguish the in vivo cystine-stabilized heterodimers that we describe here from the heterodimer involved in the initial step of filament formation. For example, Hatzfeld and Weber (1990) and Steinert (1990) recently reported that the formation of a heterodimer is an obligatory first step during keratin filament formation. In the former study, Hatzfeld and Weber (1990) introduced by site-directed mutagenesis a cysteine residue in exactly the same relative positions of the alpha-helical domains of two simple epithelial keratins K8 and K18, both of which lack cysteine. Since these two cysteine residues are introduced into the “d” position of the heptad repeat pattern, they face each other within a coiled coil. Thus air-oxidation of a mixture of these two keratins results in the efficient formation of a cystine-stabilized heterodimer, which, interestingly, is fully capable of forming 10 nm filaments. The heterodimers that we describe here (Figs 9-11) are distinct from this type of artificially constructed, cystine-stabilized heterodimer, in that the former most likely involve the crosslinking between type I and type II keratins of different coiled coils. We make this speculation because cysteines of type I and II keratins of all coexpressed keratin “pairs”, including the mouse K4 and K13 (Rentrop et al., 1986; Knapp et al., 1986), reside in very different locations of the helical domains; this makes it impossible for the two keratin helices, lying parallel and in perfect register in an (initial) heterodimer (Ip et al., 1985; Parry et al., 1985; Steinert and Roop, 1988), to form disulfide bonds. Another reason is that a great majority of cysteines of the coiled coil domains are not in the “a” and “d” positions of the heptad, and are therefore pointing outwards away from the area of coiled coil interactions. For these reasons, we strongly suspect that the cystine-stabilized heterodimers that we describe here are primarily designed to stabilize the longer-range interactions between neighboring coiled coils (Aebi et al., 1983).

The disulfide-stabilized homodimer of K13 has two interesting implications. First, recent data showed quite clearly that interactions between the two members of a homodimer are less stable and are therefore thermodynamically less important than those of heterodimers during filament initiation (i.e. the formation of the initial dimer; Hatzfeld and Weber, 1990; Steinert, 1990). However, the fact that some K13 keratins are cystine-stabilized as a homodimer in terminally differentiated esophageal epithelial cells raises the possibility that interactions involved in forming certain homodimers may be involved in stabilizing the overall filament structure during a late step of filament assembly. Second, the fact that we could not generate cystine-stabilized K13 homodimers using reconstituted filaments again raises questions about the normalcy of such filaments (Steinert, 1990). On one hand, the in vitro formation of the two heterodimers indicates that these reconstituted filaments must be quite normal in terms of the precise spatial alignment of the several pairs of sulfhydryl groups that are involved in heterodimer formation (Figs 12 and 13). On the other hand, the fact that no homodimer is formed at all suggests that certain higher-order structure normally stabilized by homodimers is probably totally missing in such filaments.

Since intermolecular disulfide-crosslinking of keratins occurs only in superficial squamous cells (Fig. 9) that are thought to be metabolically inert, it is possible that the cytoplasm of these cells is relatively aerobic and the crosslinking occurs spontaneously between cysteine residues that are spatially favorably oriented. This possibility is supported by our present observation that such a crosslinking pattern can be reproduced in part using in vitro reconstituted keratin filaments (Fig. 12).

We have established in this paper that K4 and K13, the two major differentiation-related keratins in esophageal epithelium, become crosslinked during an advanced stage of differentiation forming both heterodimers (K4/K13) and homodimers (K13-K13; Figs 9-11). In considering the functional significance of these disulfide-stabilized keratin dimers, it is interesting to note that the degrees of disulfide crosslinking of different keratin molecules vary roughly in proportion to the degree by which the epithelium forms a tough and physically durable protective superficial cell layer. Thus simple epithelial keratins contain either very few or no cysteine residues at all, and are usually not crosslinked; esophageal keratins are crosslinked mainly resulting in the formation of dimers which may be compatible with a rather flexible cornified layer; epidermal keratins, on the other hand, formed crosslinked polymers which may account for a tougher stratum corneum structure; finally, hair keratins contain numerous cysteines forming a highly crosslinked and extremely rigid and durable hair fiber. This indicates that the intermolecular crosslinking of keratins is a tissue-specific and terminal differentiation-related event whose function is to achieve various degrees of physical stabilization or hardening of the superficial squamous cells.

We thank Dr. James Rheinwald of Harvard Medical School for providing cultured human mesothelial cells, and Cong Xu for excellent technical assistance. This work was supported by NIH grants EY 4722, AR 34511 and AR 39749.