Lessons from Keratin 18 Knockout Mice: Formation of Novel Keratin Filaments, Secondary Loss of Keratin 7 and Accumulation of Liver-specific Keratin 8-Positive Aggregates

The targeting vector was constructed from a genomic keratin 18 λ phage kindly provided by T. Morita (Osaka University, Osaka, Japan; Ichinose et al., 1988). The targeting vector contains two arms of mouse keratin 18 genomic DNA; a 5′–1.2-kb StuI-EcoRV (positions 1,202–2,485 in the K18 gene sequence, Genbank/EMBL/DDBJ Accession No. M22832) and a 3′– 4.7-kb EcoRI (positions 3,542)-BamHI (positions 8,200) fragment, separated by the mouse phosphoglycerate-kinase (PGK) promoter–driven HPRT minigene (PGK/pDWM1; Magin et al., 1992). The HPRT minigene replaces most of exon 1 and part of exon 2. We prepared two vectors, one with the HPRT minigene in sense and one in the antisense orientation, with respect to the K18 transcription unit. This construct was inserted into a Bluescript II SK+ vector containing a Herpes simplex virus thymidine kinase gene also driven by the mouse PGK promoter. The targeting vector was linearized with ClaI, so that vector sequences were downstream of the thymidine kinase gene (Fig. 1). For electroporation, DNA was purified by Qiagen columns (Düsseldorf, Germany), followed by phenol/chloroform extraction and ethanol precipitation.

5′- and 3′-specific probes for murine K7, 8, 18, 19, and 20 were subcloned in Bluescript II SK+ (Stratagene, Heidelberg, Germany). Gel-purified inserts were random prime–labeled using a kit from Boehringer Mannheim (Mannheim, Germany) and 50 μCi of ³²P-dCTP (Amersham Buchler GmbH, Braunschweig, Germany).

Culture conditions for the embryonic stem cell line HM-1 and selection of targeted colonies were described previously (Stacey et al., 1994). Electroporation was done with 200 μg of linearized vector DNA at 800 V and 3 μF. Enrichment after double selection with HAT and ganciclovir was ∼15-fold.

Colonies surviving the HAT/ganciclovir selection regime were transferred into 24-well plates (Becton Dickinson, Heidelberg, Germany) and approximately half of each was processed for PCR analysis (Selfridge et al., 1992). Genomic DNA was purified with a kit (Qiagen Inc.). The primers used were specific for the sequence upstream of the StuI site of cytokeratin 18: GTCCTCAGCACCCTGTAACCTG (Genbank/EMBL/DDBJ Accession No. M22832, nucleotides 1,316–1,337) and for the HPRT minigene: AGCCTACCCTCTGGTAGATTGTCG (HPRT–cDNA sequence, positions 960–983, see (Selfridge et al., 1992). Reaction conditions were as described before (Selfridge et al., 1992) with the following time/temperature parameters: 25 s/95°C, 25 s/65°C and 1 min/72°C for 35 cycles. An Omnigene thermocycler (MWG, Ebersberg, Germany) was used. Taq polymerase was purchased from GIBCO BRL (Karlsruhe, Germany). Genomic DNA from colonies, which produced a PCR product, was isolated from confluent embryonic stem cell cultures by proteinase K and RNAase A (200 μg/ml of both) digestion at 37°C overnight, followed by phenol/chloroform extraction and ethanol precipitation. Genomic Southern blotting was carried out under stringent conditions (hybridization at 42°C in 50% formamide, 5× SSPE, 5× Denhardt, 100 μg/ml herring sperm DNA, 1% SDS, 10% dextran sulfate; four washes for 15 min each at 68°C in 0.1× SSPE, 0.1 SDS) to analyze the structure of the targeted allele by hybridization with three probes: (a) a 3′-EcoRI fragment outside the targeting vector; (b) the HPRT minigene; and (c) a 5′-EcoRI fragment located upstream of the targeting vector.

Generation of knockout mice was done as described before (Porter et al., 1996) using BALB/c and C57BL/6 blastocysts as recipients. Animals were bought from Harlan Winkelmann (Borken, Germany) with SPF status. Mice were genotyped using a PCR assay of mouse tail DNA, which included the two primers described for identifying ES cell colonies and an additional primer downstream of the EcoRV site: CGGTCTGGATTCCACCCATTC (Genbank/EMBL/DDBJ Accession No. M22832, positions 2,535–2,555). This PCR reaction gave two products; one specific for the wild type (1.3 kb) and one for the altered allele (1.5 kb).

Animals were killed by cervical dislocation and tissues were snap-frozen in liquid nitrogen immediately after dissection. Total RNA was extracted as described by Chomczynski and Sacchi (1987) with minor modifications. 5 mM vanadyl-ribonucleotide (GIBCO-BRL, Karlsruhe, Germany) was included in all purification steps. Aliquots of various total RNA preparations (30 μg) were loaded on 1.4% agarose gels in the presence of formamide/formaldehyde as denaturant and RNA was transferred to Genescreen membranes (New England Nuclear, Dreieich, Germany) by capillary transfer using 10× SSPE as buffer (Sambrook, 1989). Hybridization and washing conditions were as described above for Southern blotting.

Protein extracts enriched in cytoskeletal proteins were prepared from frozen tissue by homogenization as described previously (Magin et al., 1983). Total protein extracts were prepared as before (Schröder et al., 1997). The following proteinase inhibitors were included in all buffers: Aprotinin at 0.1 μg/ml, E-64 at 0.5 μg/ml and Pefabloc SC at 100 μg/ml (all from Boehringer Mannheim). One- and two-dimensional gel electrophoresis were carried out as described (Magin et al., 1983; Bader et al., 1991). Proteins were transferred electrophoretically onto nitrocellulose membranes (0.2 μm pore size; Schleicher & Schuell, Dassel, Germany) using 25 mM sodium borate, 1 mM EDTA, pH 9.2, as transfer buffer in a wet blot apparatus (BioRad Laboratories, München, Germany). Immunodetection was performed with peroxidase-coupled secondary antibodies (Dianova, Hamburg, Germany) and enhanced chemiluminescence with reagents from Pierce (St. Augustin, Germany) as described (Schröder et al., 1997). Primary antibody dilutions were as follows: Troma 1 and 3 (K8 and 19, respectively), 1:5,000; Ks 18.04 (K18; Progen, Heidelberg, Germany) diluted 1:8,000; Ks 20.10 (K20, Progen; and from R. Moll, Halle, Germany) diluted 1:1,000; RCK 105 and OVTL 12–30 (K7; Eurodiagnostics, Maastricht, The Netherlands; and from F. Ramaekers, University of Maastricht, Maastricht, The Netherlands) diluted 1:1,000. Antibodies to desmosomal proteins were kindly provided by W.W. Franke (German Cancer Research Center, Heidelberg, Germany) and D. Garrod (University of Manchester, Manchester, UK). Secondary antibodies were diluted 1:30,000.

Frozen sections (5 μm thick) were fixed in acetone at −20°C for 10 min and dried for a few hours before further processing. All antibodies were diluted in PBS containing 1% BSA. The primary monoclonal antibody dilutions were as follows: Troma 1 and 3 (K8 and 19; a kind gift of H. Baribault, Cancer Center, La Jolla, CA) 1:50, Ks 18.04 (K18; Progen) 1:30, RCK 105 and OVTL 12–30 (K7, kindly provided by F. Ramaekers and from Eurodiagnostics) 1:20, Ks 20.10 (K20) 1:10, II- 5F (antidesmoplakin, kindly provided by D. Garrod, University of Manchester, Manchester, UK) 1:10, mAB 121 (antiplectin, kindly provided by R. Owaribe, University of Nagoya, Nagoya, Japan) 1:200, M_M 120-1 (directed against MBs, provided by H. Denk, University of Graz, Austria) undiluted, Z 0458 (ubiquitin), diluted 1:100 (Dako Corp., Hamburg, Germany). FITC-, Texas red-, Cy2- or Cy3 labeled secondary antibodies were diluted as recommended by the manufacturers (Dianova, Medac, both Hamburg, Germany). Primary mouse monoclonal antibodies were detected with subclass-specific secondary antibodies (Medac) to minimize background. Slides were mounted in Mowiol (Calbiochem, Bad Soden, Germany) containing 0.1% Dabco (Sigma Chemical GmbH, Munich, Germany).

Fixation of tissue specimen and staining procedures were performed as described (Mertens et al., 1996; Ruiz et al., 1996).

The K18 targeting vector was constructed from a λ phage harboring K18 genomic sequences and contained 5.9 kb of overall homology (Fig. 1,A). We noted that the targeting vector containing the HPRT minigene in the same orientation as the K18 transcription unit yielded at least 10-fold less colonies than the one in the opposite orientation. ES cells, transfected with the former construct, also produced two- to threefold less HPRT- mRNA than those with the latter. Only colonies derived from the latter were analyzed further. Among 72 colonies analyzed after HAT/ganciclovir selection, nine proved to be positive by PCR and Southern blotting (Fig. 1, B and C). The three clones of HM-1 cells carrying the correctly targeted K18 allele gave 100% germline transmission when injected into BALB/c or C57Bl/6 blastocyts. Chimaeras were bred to BALB/c, Bl/6, 129/Sv, or MF-1 animals. Most results were obtained with intercrosses between 129/Sv and MF-1, or from animals on a 129/Sv background.

Heterozygous K18 males were mated to heterozygous and wild-type females. In all strains tested, the expected frequency of homozygotes was observed at birth. K18 null mice have a normal lifespan compared to wild-type animals. Up to the age of 4 mo, K18 mice of both genotypes were normal as judged by morphological or histological inspection. A consistent feature in 18-mo-old mice was a marked hepatomegaly. There were no other organ abnormalities. This is in contrast to K8 null mice, which suffered from an increased embryonic lethality, possibly due to liver hemorrhage on a C57Bl/6 background. With respect to the findings in K8 null fetuses, it can be concluded that the liver phenotype of the corresponding knockout mice is different. Whereas K8 null mice developed colorectal hyperplasia as adults in the strain FVB/N (Baribault et al., 1993, 1994), no anorectal prolapse was observed in K18 null animals. This highlights that type I and II keratins provide complementary but not identical functions.

In hepatocytes of adult liver, K18 and its type II partner K8 are the only two keratins expressed. Given that 50% of K8 knockout mice suffered embryonic lethality at day 12 and displayed liver hemorrhage, one might have suspected a similar phenotype in K18 null mice. To analyze whether a null allele had been generated, we tested total protein or cytoskeletal extracts from normal and K18 null mice for keratin expression by SDS-PAGE and Western blotting with specific keratin antibodies. No K18 was found in any of the protein extracts examined, indicating that the knockout for K18 was complete. Also, K8 was absent in hepatocytes of K18 null mice (Fig. 2) in agreement with the rapid degradation of single keratins in the absence of a partner keratin (Domenjoud et al., 1988; Kulesh et al., 1989; Magin et al., 1990). The absence of functional K18 was also confirmed by Western blotting with the broad range antibody IFA (not shown; Pruss et al., 1981), which binds to the COOH-terminal end of keratin rod domains (Magin et al., 1987). The amounts of K18 and its partner K8 as judged by SDS-PAGE analysis were slightly reduced but present in equimolar amounts in heterozygotes, indicating that in vivo the steady-state amount of keratins is determined by that of its corresponding partner (Fig. 2).

When total RNA prepared from normal and K18 null mouse livers was probed for the expression of K8 and 18, the mRNA for K18 was absent (Fig. 3), confirming the Western blot results. The mRNA level for K8 was unaltered. In agreement with findings from K8 null mice (Baribault et al., 1994), where the converse was found for K18 mRNA, we concluded that K8 and 18 are regulated independently at the transcriptional level, despite their close chromosomal localization (Waseem et al., 1990).

To examine other potential consequences of K18 ablation, we analyzed mRNA and protein expression in tissues with a more complex keratin expression pattern. The Northern blot analysis (Fig. 4) showed that, apart from the loss of K18, the level of K7, 8, 19, and 20 mRNAs remained largely unaltered. We noted a particularly strong expression of mouse K20, which was by far the most prominent keratin mRNA among those investigated. Interestingly, the high level of mRNA did not correlate with the low and rather restricted expression of the corresponding protein (not shown).

In cryosections of K18 null mouse liver, up to the age of 4 mo, we noted the complete absence of keratins by indirect immunofluorescence, in line with results from Northern and Western blotting. Histological analysis revealed the presence of small necrotic foci in liver sections from K18 null mice (Magin, T.M., unpublished observations). Our analysis included a panel of monoclonal antibodies and two polyclonal antisera directed against simple epithelial keratins, all of which failed to stain hepatocyte IF (Fig. 5). In control sections, antibodies to K8 and 19 strongly labeled bile duct epithelia known to express K7, 8, 18, and 19 (not shown). Here, the K8/19 staining was predominantly luminal, in agreement with a recent report on the participation of K19 in the formation of the apical cytoskeleton in cultured epithelial cells (Salas et al., 1997).

In epithelia, keratins are closely associated with desmosomes and hemidesmosomes (Koch and Franke, 1990; Schwarz et al., 1990). Whereas the loss of plakoglobin (Ruiz et al., 1996), BPAG-1 (Guo et al., 1995) and of plectin (McLean et al., 1996) is known to abolish the connection with keratin IF, the consequences of a keratin IF loss for the formation and/or distribution of desmosomes has not been studied before in vivo. We, therefore, stained frozen sections with antibodies to desmosomal proteins. As judged by conventional immunofluorescence microscopy, no difference in the number and distribution of desmosomes could be observed (Fig. 5). The additional analysis of ultrathin liver sections by electron microscopy confirmed the typical ultrastructure of desmosomes without attachment of keratin IF (Fig. 6). It is evident from these and other observations (Denk et al., 1985) that desmosomes form and are maintained in the absence of keratin filaments.

When frozen sections of 18-mo-old K18 null mice were stained, ∼20–30% of hepatocytes showed irregular K8-positive aggregates of variable size scattered throughout their cytoplasm. K8-aggregates were distributed in the cytoplasm and/or in the cell periphery of single or grouped hepatocytes (Fig. 7). K8-staining cells were often enlarged when compared to cells without K8 accumulation. Immunostaining for K7, 19, and 20, desmin and vimentin failed to detect any of these IF proteins in K8-positive or -negative hepatocytes (not shown). Double immunofluorescence analysis with antibodies to ubiquitin and the Mallory body antigen, M_M120-1, in combination with K8 antibodies revealed that the aggregates contained K8, ubiquitin, and the M_M120-1 protein, thus resembling MBs in human livers with alcoholic hepatitis (Fig. 7). In contrast to old K18 null mice, neither abnormal keratin structures nor ubiquitin- or M_M120-1–positive cytoplasmic depositions were found in control sections from age-matched wild-type mice.

Previous in vitro studies have shown that any type I and II keratins can assemble to heterotypic complexes or IF (Hatzfeld and Franke, 1985; Eichner et al., 1986; Magin et al., 1987; Paladini et al., 1996), whereas in vivo, they form specific pairs (Moll et al., 1982; Sun et al., 1985; Cooper and Sun, 1986). As many simple epithelia coexpress the type II K7 and 8 along with type I K18, 19, and 20 (Moll et al., 1982; Lane, 1993), our K18 null mice offered a unique opportunity to analyze possible consequences on keratin pairing.

When we compared frozen sections of uterus from normal and homozygous K18 mice, we noted not only the lack of K18, but also that of K7 (Fig. 8), which was in agreement with our Western blot analysis of cytoskeletal extracts (Fig. 9). The lack of K7 or its drastic reduction was not restricted to uterus but observed also in bile duct and kidney epithelia (not shown). In contrast, the expression and distribution of K8 and K19 seemed unaltered in the tissues examined (Figs. 8 and 10).

To see whether K7 might be soluble and was, therefore, missing in cytoskeletal extracts of K18 null mice, we also performed immunoblotting of total proteins with K7 antibodies. Here we detected trace amounts of K7 (not shown). At the same time, the level of K7 mRNA remained unchanged in all tissues examined (Fig. 4). The relative amount of K8 and 19 remained equal in K18 null mice, but was slightly reduced in homozygous compared to normal mice. The presence of K8 and K19 in the cytoskeletal fraction suggested that these two keratins were able to form insoluble structures in vivo. Therefore, we went on to study their distribution by light and electron microscopy.

It is well established that keratin head domains are essential for filament formation (Albers and Fuchs, 1987, 1989; Bader et al., 1991; Wilson et al., 1992; Hatzfeld and Burba, 1994). The role of the tail domain is more controversial. Whereas some experiments have documented typical IF formed from K8 and 19 in vitro and in cultured cells (Hatzfeld and Weber, 1990; Bader et al., 1991), others have suggested that the lack of its tail domain renders K19 unable to form IF (Lu and Lane, 1990).

In a number of tissues like kidney and uterine epithelium, the forced absence of K18 left K8 and the tail-less K19 as the only keratins expressed, giving us the unique opportunity to analyze K8/19 structures in greater detail. As judged by immune electron microscopy, they formed bona fide IF. Filaments appeared smooth and in typical loose arrays in wild type (Fig. 10,A), in K18 null mice (Fig. 10,B), and when attached to desmosomes in a manner indistinguishable from those observed with other keratin combinations. We also noted the insertion of keratin IF to the surface of mitochondria (inset in Fig. 10 A). The overall number of IF appeared similar in K18 null mice and in wild-type animals, in agreement with immunofluorescence data.

Immunofluorescence analysis of several additional tissues, including colon, small intestine, kidney, vagina, lung, and pancreas demonstrated that the overall appearance of keratin IF in those tissues of K18 null mice appeared similar to that of wild-type mice, despite the absence of K18. We concluded that in those tissues K19 and K20 seem to be functional equivalents of K18. This could explain why we have not observed a hyperplasia of the colon epithelium that was seen in K8 null mice. Our observations clearly show that in vivo K19 is a fully competent partner for K8, at least in simple epithelia of the mouse.

During the last few years, considerable progress has been made in understanding the cytoskeletal functions of keratins in skin. These were revealed by the discovery of point mutations in epidermal keratin genes (Fuchs, 1994; McLean and Lane, 1995) and by the establishment of conventional transgenic (Vassar et al., 1991) or knockout mice (Lloyd et al., 1995; Porter et al., 1996; for a review, see Magin, 1997). The function of simple epithelial keratins is less clear due to the small number of mouse models and to the lack of genetic disorders caused by mutations in these keratin genes (for a recently discovered K18 mutation see Ku et al., 1997). The overall phenotype of K8 (Baribault et al., 1993, 1994) and 18 knockout mice is less dramatic than that of K10 or 14 knockout animals (Lloyd et al., 1995; Porter et al., 1996; Reichelt et al., 1997).

Given that K8 and 18 are coexpressed and represent subunits of the same 10-nm filament throughout embryonic development and many adult epithelia (Jackson et al., 1980; for a review see Lane, 1993), one might have assumed a similar phenotype for K8 and K18 knockout mice. Surprisingly, K18 null mice are viable, fertile, and have a normal life expectancy. However, they develop a specific liver pathology defined by the accumulation of K8-positive Mallory bodies in old animals. Mid-gestational death accompanied by liver hemorrhage or colorectal hyperplasia in adults, which were the hallmarks of K8 null mice (Baribault et al., 1993, 1994), were not associated with the K18 knockout.

In the mouse, desmosomes form first in trophectodermal cells of the 32-cell stage embryo (Ducibella et al., 1975; Jackson et al., 1980), preceded by the presence of K8 and 18 (Chisholm and Houliston, 1987; Emerson, 1988; Lane, 1993), which at that stage are not well organized as a typical cytoskeleton. It has been proposed that the interaction of desmosomes with IF allows the transduction of forces among epithelial cells (Arnn and Staehelin, 1981). Regarding the formation of desmosomes, several authors have provided evidence that plaque components are assembled in the cytoplasm, attached to or in close association with keratin IF (Jones and Goldman, 1985; Green et al., 1987; Pasdar and Nelson, 1988; Pasdar et al., 1991), invoking a requirement of keratin IF for desmosome formation. In that respect K18 knockout mice have allowed us to analyze desmosomes in the absence of keratin IF ab origine. Clearly, hepatocyte desmosomes have a typical appearance and size distribution in the absence of IF as judged by electron microscopy. By light microscopy we have found no evidence for a reduced amount of desmosomes or an increased cytoplasmic staining of desmoplakin which would be indicative of an increased turnover rate or instability of desmosomes. With respect to cell–cell interaction, we have not detected areas of cell separation, which would indicate a reduced desmosome-mediated adhesion. Taken together one can conclude that at least in liver of K18 null mice the formation and maintenance of desmosomes can occur in the absence of keratin IF.

The most surprising finding in hepatocytes of old K18 null mice was the accumulation of K8-positive aggregates, which developed in the absence of other keratins, vimentin, or desmin. These aggregates were identified as Mallory bodies by double immunofluoresecence with antibodies directed against ubiquitin and the Mallory body-specific antigen M_M120-1 (Zatloukal et al., 1990). Based on in vitro assays (Steinert et al., 1976; Hatzfeld and Franke, 1985), as well as cell microinjection or transfection studies (Domenjoud et al., 1988; Kulesh et al., 1989; Magin et al., 1990; Bader et al., 1991), it has been demonstrated that single keratins do not form IF, are unstable, and become rapidly degraded in vivo. This was confirmed in vivo in K8 (Baribault et al., 1993) and young K18 null mice. At present, neither the triggers nor the point of time of K8 accumulation in hepatocytes are known. With regard to the observed alignment of K8 aggregates at the cell periphery or close to desmosomes, it is tempting to speculate that they could be initiated by the formation of homopolymeric K8 assemblies, which are initially protected against proteolysis. In due course, these initial K8 seeds would increase in size and redistribute also within the cytoplasm. These K8-containing aggregates might acquire anomalous tertiary structures, and become recognized by ubiquitin-targeting mechanisms (Hershko and Ciechanover, 1994; Haas, 1997). Due to their large size, however, they might be relatively protease resistant.

The slow process of keratin aggregation and the absence of a functional IF cytoskeleton in K18 null mice has also been found in a variety of diseases, like human alcoholic hepatitis (Denk et al., 1979), blistering skin disorders (Corden and McLean, 1996), or insulin-dependent diabetes in mice (Blessing et al., 1995). The development of K8 aggregates in old K18 null mice is the first example of the formation Mallory bodies, resulting simply from a keratin gene knockout.

Corroborative data on the regulation of keratin expression indicate that most genes are regulated at the transcriptional level (Jorcano et al., 1984; Oshima, 1992; Fuchs and Weber, 1994), with the possible exception of K6 in epidermis (Tyner and Fuchs, 1986). The continued transcription of K8 in our K18 knockout mice, as well as other studies (Knapp et al., 1989; Baribault et al., 1993, 1994; Lloyd et al., 1995; Porter et al., 1996; Reichelt et al., 1997), have not provided evidence for an interdependent control of types I and II keratin genes. Based on these studies, it rather seems that fine tuning of keratin expression involves the formation of heterotypic protein polymers. The rapid turnover or accumulation of single, unpaired keratins indicate a lack of translational feedback mechanisms in this gene family like those operating in the tubulin gene family (Yen et al., 1988).

In the context of these data, the absence or drastic reduction of K7 staining in uterine, kidney, and bile duct epithelia of K18 null mice was completely unexpected. We excluded epitope masking of K7 by using several K7 antibodies, all of which failed to react on frozen sections of K18 null mice. At least two possibilities can be envisaged to explain the lack or drastic reduction of K7 in several epithelia of K18 null mice: (a) K7 could be incapable of forming IF with K19 in vivo. This is unlikely given that these two keratins formed IF de novo upon gene transfer into 3T3 fibroblasts (Lu and Lane, 1990). It is nevertheless conceivable, that in natural epithelia, individual keratins have a distinct subcellular distribution making them unavailable for other than their “natural” partner. (b) K7 might be unable to bind to K19 because it is competed out by K8. Recent measurements using viscosimetry and plasmon surface resonance have indeed shown significant differences of keratin affinities in vitro (Hofmann and Franke, 1997), extending previous measurements based on “melting” keratin complexes in the presence of urea (Franke et al., 1983; Hatzfeld and Franke, 1985). Obviously heterotypic complexes between K8/19 seem to assemble faster and have a higher affinity than those between K7 and K19.

K19 has a unique structure with a short α-helical extension of nine amino acids instead of a typical tail domain (Bader et al., 1986). It is coexpressed with simple epithelial as well as epidermal keratins and has been proposed to be a marker of epithelial stem cells (Moll et al., 1982; Bartek et al., 1985; Michel et al., 1996). Based on cell transfection experiments, it has been proposed that the lack of a proper tail domain rendered K19 unable to form regular IF (Lu and Lane, 1990), supporting the idea that the IF cytoskeleton of K19 expressing cells might be more flexible than that built of other keratins. This view has been challenged by other in vitro and cell transfection studies, showing that K19 would form typical IF (Hatzfeld and Weber, 1990; Bader et al., 1991).

The ablation of K18 had given us a unique opportunity to analyze the properties of K19 in vivo. A striking feature of tissues like kidney, bile duct, and uterus, which in K18 knockout mice transcribed only K8 and 19 genes, was the positive K19 immunoreactivity. Immune electron microscopy of uterine epithelium revealed beyond doubt that K8 and 19 formed IF of normal appearance with an intracellular distribution similar to that of IF in normal mice. K8/19 IF interacted with desmosomes in a classical manner (Schwarz et al., 1990) and were attached to the surface of mitochondria (for an early observation see Mose-Larsen et al., 1982).

This strongly argues in favor of a normal filament-forming capacity of the murine K19 in vivo. Furthermore, these findings are in line with a previous report showing K8/19 filaments in Wilms tumor (Ishii et al., 1989) as well as in vitro studies (Herrmann et al., 1996), demonstrating that desmin and vimentin (type III IF) form IF without tail domains. In functional terms, it still has to be proven that K8/ 19 IF are equivalent to their tail-containing counterparts.

In summary, the generation of K18 null mice has revealed the following major findings: (a) In contrast to K8 null mice, our animals are viable, fertile, and have a normal life expectancy. (b) Desmosomes form, are maintained, and retain full adhesive properties in the absence of keratin IF. (c) Old mice displayed enlarged hepatocytes with K8 aggregates that were identified as Mallory bodies. The development of K8 aggregates in old K18 null mice is the first example of the formation of Mallory bodies resulting simply from a keratin gene knockout. (d) K7 was absent or drastically reduced in several tissues despite continued transcription of its mRNA, suggesting that K8 preferentially assembles with K19 in the absence of K18. (e) The tail-less keratin 19 forms bona fide IF with K8 in vivo and seems able to compensate for the regular K8/18/ 19 IF in functional terms.

We thank E.B. Lane (Dundee, UK), D. Garrod, W.W. Franke, K. Owaribe, and H. Baribault (La Jolla, CA) for gifts of antibodies and R. Leube (University of Mainz, Mainz, Germany) for the K20 probe. T. Morita is thanked for the murine K18 gene. T.M. Magin is most grateful to Conny Stumptner (University of Austria, Graz, Austria) for initiating the analysis of old K18 null mice during a stay in his lab.

IF

intermediate filaments

MBs

Mallory bodies

PGK

phosphoglycerate kinase