Apoptosis Generates Stable Fragments of Human Type I Keratins*

Type I and II keratins help maintain the structural integrity of epithelial cells. Since apoptosis involves progressive cell breakdown, we examined its effect on human keratin polypeptides 8, 18, and 19 (K8, K18, K19) that are expressed in simple-type epithelia as noncovalent type I (K18, K19) and type II (K8) heteropolymers. Apoptosis induces rapid hyperphosphorylation of most known K8/18 phosphorylation sites and delayed formation of K18 and K19 stable fragments. In contrast, K8 is resistant to proteolysis and remains associated with the K18 fragments. Transfection of phosphorylation/glycosylation-mutant K8 and K18 does not alter fragment formation. The protein domains of the keratin fragments were determined using epitope-defined antibodies, and microsequencing indicated that K18 cleavage occurs at a conserved caspase-specific aspartic acid. The fragments are found preferentially within the detergent-insoluble pool and can be generated, in a phosphorylation-independent manner, by incubating keratins with caspase-3 or with detergent lysates of apoptotic cells but not with lysates of nonapoptotic cells. Our results indicate that type I keratins are targets of apoptosis-activated caspases, which is likely a general feature of keratins in most if not all epithelial cells undergoing apoptosis. Keratin hyperphosphorylation occurs early but does not render the keratins better substrates of the downstream caspases.

Intermediate filament (IF) 1 proteins encompass the nuclear lamins and a large family of tissue-specific cytoplasmic proteins that include keratins in epithelial cells, desmin in muscle, neurofilaments in neuronal cells, and vimentin in mesenchymal cells (reviewed in Refs. [1][2][3]. Keratins are the largest IF protein subgroup and consist of more than 20 polypeptides (K1-K20) that are divided into relatively acidic type I (K9 -K20) and basic type II (K1-K8) keratins (4,5). All epithelial cells typically express at least one type I and one type II keratin, as noncovalent obligate heteropolymers, in an epithelial cell-type specific manner. For example, simple-type epithelia preferentially express K8/18 with various levels of K19 and K20 (6 -10), keratinocytes express K1/10 and/or K5/14 depend-ing on their differentiation state within the epidermal layer, and corneal epithelial cells express K3/12 (4). Although the full scope of keratin function is not known, one clear keratin function is to help maintain epithelial cell integrity particularly upon cell stress. This role is supported by several animal studies and a growing list of human epidermal, oral, and ocular diseases that result from keratin mutations (11)(12)(13)(14)(15)(16).
Keratins undergo several modifications that are likely involved in regulating their function (reviewed in Ref. 17), with phosphorylation being the most studied (reviewed in 18,19). For K8/18, the known in vivo phosphorylation sites include Ser-52/Ser-33 of K18 (20) 2 and Ser-23/Ser-431/Ser-73 of K8 (22,23). Of note, keratin phosphorylation is highly dynamic (24 -27) and is modulated during several physiological states, including mitosis and cell stress (17,19). The use of phosphoepitope-specific antibodies has facilitated the study of IF protein phosphorylation (17,28). For example, antibody LJ4, which recognizes Ser(P)-73 of K8, does not stain nondividing cells by immunofluorescence but exhibits a strong signal upon induction of apoptosis using anisomycin or etoposide or during mitosis as noted in cultured cells or in mouse liver after partial hepatectomy (23).
Apoptosis is asssociated with a number of ordered morphological and biochemical events that ultimately lead to cell death (29,30). Central to this process is the family of "caspase" proteases that cleave at aspartic acid in the context of preferred motifs (31)(32)(33). The importance of caspases is highlighted by the ability of short peptides, whose sequences mimic the sequence context of caspase substrates, to selectively inhibit apoptosis. Identification of in vivo caspase substrates is highly relevant in terms of understanding apoptosis in general and determining the relative importance of these substrates. The list of the caspase substrates is growing and includes procaspases, poly(ADP-ribose) polymerase (PARP), nuclear lamins, protein kinase C-␦, MEKK-1, topoisomerases, and fodrin (31)(32)(33)(34)(35). In this report, we describe the apoptosisassociated biochemical events that involve keratins. We show that apoptosis is associated with early keratin hyperphosphorylation that is later followed by preferential caspase-mediated proteolysis of type I (K18 and K19) but not type II (K8) keratins. The apoptosis-induced keratin phosphorylation, which involves most but not all known K8/18 phosphorylation sites, does not affect keratin susceptibility to degradation by the downstream caspases. In addition, K8 and K18 that are mutated at several known phosphorylation and glycosylation sites are equally susceptible to degradation in transfected cells. Our results suggest that apoptosis-induced fragment formation of K18 and K19 is likely to occur in other cytoplasmic IF proteins that have similar predicted caspase-sensitive sequences.
In Vitro Proteolysis of K8/18 Immunoprecipitates with CPP32 or with Detergent Cell Lysates-Cells were cultured in the presence or absence of An (16 h), followed by solubilization with 1% Nonidet P-40 in phosphate-buffered saline (pH 7.4) containing okadaic acid (0.5 g/ml), phenylmethylsulfonyl fluoride (0.1 mM), leupeptin (10 M), pepstatin (10 M), aprotinin (25 g/ml), and EDTA (10 mM) (buffer A) (37). After centrifugation (16,000 ϫ g, 30 min), lysates were used for immunoprecipitation of K8/18 with mAb L2A1-protein A-agarose beads. Cells (Ϯ FIG. 1. Characterization of apoptosis-induced and transfection-generated K18 and K19 fragments. Panel A, HT29 cells were cultured in the presence or absence of An. Cells (adherent and floater) were solubilized with 1% Nonidet P-40 followed by immunoprecipitation of K8/18 using mAb L2A1. An treatment is associated with K8 Ser-73 phosphorylation, which generates a distinct K8 species termed HK8 (23). Panels B and C, K8/18 immunoprecipitates were obtained from BHK cells (B) that were co-transfected with K8 and K18 and from An-treated HT29 cells (H) and then analyzed by SDS-PAGE and Coomassie staining (panel B). Identical precipitates were transferred to PVDF membranes and then blotted with antibodies that specifically recognize the indicated K8 or K18 phosphorylation sites (panel C). Arrow points to immunoglobulin band under nonreducing conditions. Panel D, total cell lysates were obtained from HT29 (Ϯ An) and K8/18-transfected (ϩ) or untranstected (Ϫ) BHK cells followed by analysis by SDS-PAGE, transfer to a PVDF membrane, and then immunoblotting with anti-PARP antibody. Panel E, K8/18 immunoprecipitates were obtained from HT29 cells (Ϯ An). No floater cells were present in ϪAn cells, while the ϩAn cells were separated into floater (lane 3) and adherent (lane 2) cells prior to solubilization and immunoprecipitation. Panel F, K8/18 (lane 1), K8/19 (lane 2), or control agarose-protein A (lane 3) immunoprecipitates were obtained from An-treated HT29 cells followed by SDS-PAGE analysis. Duplicate samples were also transferred to PVDF membranes followed by blotting with anti-K19 mAb KA4 or B/A2. An) were similarly solubilized with 1 mM dithiothreitol in buffer A, and this lysate was used as a protease source after pelleting. Immunoprecipitates were then incubated (3 h, 37°C) with CPP32 buffer (25 mM Hepes, 1 mM dithiothreitol, pH 7.5) in the presence or absence of CPP32, or with the 1% Nonidet P-40 detergent lysates that were obtained from Ϯ An-treated cells, followed by analysis by SDS-PAGE and then staining with Coomassie Blue. Duplicate samples were transferred to a PVDF membrane followed by immunoblotting with antikeratin antibodies.

Apoptosis Induces Fragmentation of K18 -Previously, we
showed that treatment of HT29 cells with An induces apoptosis (23). Immunoprecipitation of K8/18 from An-treated HT29 cells co-precipitates two major bands of 29 and 23 kDa (termed p29 and p23), which were also noted upon cotransfection of BHK cells with K8 and K18 cDNA (Fig. 1, A and B). Purification of p29 and p23 from An-treated HT29 cells or from K8/18-transfected BHK cells followed by N-terminal and internal microsequencing showed that both polypeptides represent K18 fragments with p23 being generated by cleavage at K18 Asp-237 (Fig. 2). Digestion at Asp and the context of the sequence (i.e. 234 VEVD) suggest that apoptosis-activated caspases are likely responsible for generating the K18 fragments. This is supported by inhibition of p29/p23 formation if An-treated HT29 cells are coincubated with cell-permeable caspase-3 (i.e. CPP32) but not with caspase-1 or calpain II inhibitors (not shown). Caspase-mediated digestion of PARP into an 85-kDa fragment (31)(32)(33)(34) was also noted in K8/18-transfected BHK and An-treated HT29 cells (Fig. 1D). Anti-PARP reactivity with hamster (i.e. in BHK cells) PARP was less than that of human (i.e. in HT29 cells) (Fig. 1D), and a similar pattern was noted using several other commercial antibodies tested (not shown). The N-terminal acetylation of keratins (39) coupled with our inability to obtain N-terminal sequencing of p29 suggest that p29 represents the N-terminal domain of K18. This is supported by sequencing p29 internal fragments (Fig. 2) and by immunoblotting of K8/18 precipitates that were obtained from An-treated HT29 or transfected BHK cells with K18 epitopespecific antibodies that recognize Ser(P)-33/Ser-52 (Fig. 1C). In contrast, antibodies that recognize K8 phosphorylation sites in the head (pS73) and tail (pS431) domains did not recognize p29 or p23 (Fig. 1C), and a separate anti-K18 mAb recognized p23 but not p29 (Fig. 5B, panel b).
Anisomycin treatment of adherent HT29 cells generates floater cells that are primarily apoptotic (as determined by annexin V and propidium iodide staining, not shown) while the remaining adherent cells are Ͻ10% apoptotic. The p29/p23 K18 fragments, that are associated with K8/18 precipitates, are found preferentially in the floater-cell apoptosis-enriched pool (Fig. 1E, compare lane 2 with lane 3). Of note, although most of K18 is cut, it still remains associated with K8/HK8 (Fig. 1E,  lane 3). A similar immunoprecipitation pattern to that in Fig.  1E (lane 3) was also obtained using four other antibodies (three antibodies recognize K8 and one recognizes K18, not shown), thereby indicating that the p29/p23 fragments can indeed remain associated with K8.
Apoptosis Also Induces Fragmentation of K19 -Apoptosisinduced digestion of K18 prompted us to test if similar digestion occurs in K19, which is also expressed in HT29 cells and has a similar 233 SVEVD potential caspase-cleavable sequence within the L1-2 domain. Comparison of K8/19 and K8/18 immunoprecipitates, that were obtained from An-treated HT29 cells, showed several unique polypeptides that are preferentially associated with K8/19 precipitates (Fig. 1F). Of these, two major polypeptides (p28 and p20) corresponded to K19 fragments as determined using the epitope-defined anti-K19 antibodies KA4 and B/A2 (Fig. 1F) which recognize K19 domains IB and II, respectively (see Fig. 2B for keratin domains). The KA4 mAb recognized p28 (Fig. 1F, double asterisk), and mAb B/A2 recognized p20 (and K19) while neither antibody recognized the p29/p23 K18 fragments (Fig. 1F). Similarly, a panel of antibodies that recognize K8 and/or K18 did not recognize K19 or its p28/p20 fragments (not shown). The band that is indicated by a single asterisk (Fig. 1F), which was recognized by KA4 and B/A2 antibodies, was not investigated and may correspond to an overlapping fragment or may be nonspecific since it was not seen in blots of total cell lysates (Fig. 3, panel c). Similar K19 fragments were also noted after transfection of BHK cells with K8/19 and were confirmed by blotting with mAbs KA4 and B/A2 (not shown).
Relationship of Keratin Phosphorylation to Caspase-mediated Keratin Digestion-We compared the time course of apoptosis-induced keratin proteolysis and hyperphosphorylation. Immunoblotting of total cell lysates with epitope-specific anti-phosphokeratin antibodies showed that K8 (Ser-73/ 431) and K18 (Ser-52) hyperphosphorylation occurs within 30 min of An exposure, whereas K18 Ser-33 phosphorylation is not significantly altered (Fig. 3, panels d-g). In contrast, K18 and K19 fragments begin to appear, as analyzed in the total cell lysate or in association with K8/18 immunoprecipitates, after 4 h of An treatment (Fig. 3, panels a-c). This separation of keratin hyperphosphorylation and digestion suggests that if keratin phosphorylation plays a role in its caspase-medi-

FIG. 2. Microsequencing analysis of K18 p29/p23 fragments.
Panel A, K8/18 immunoprecipitates were obtained from An-treated HT29 cells or from K8/18-transfected BHK cells followed by SDS-PAGE analysis, transfer to PVDF membranes, Coomassie Blue staining, and then direct N-terminal sequencing of p29 and p23. Alternatively, gel pieces containing individual p29 or p23 bands were digested with Lys-C protease, followed by HPLC separation of the released peptides, and then microsequencing of internal peptides. All derived sequences had 100% identity with human K18. Panel B, IF proteins share the structural features of N-and C-terminal globular "head" and "tail" domains, respectively, that are separated by an ␣-helical "rod" domain. The rod domain is divided into subdomains that are separated by short "linker" (L) regions. The K18 amino acid location of the domains, the location of Asp-237 K18 caspase cleavage, and the internally sequenced peptides of p29/p23 are indicated. ated digestion then it may be by signaling downstream events and/or by improving the ability of keratins to act as caspase substrates.
The role of keratin phosphorylation in its caspase-mediated digestion was tested by subjecting K8/18 precipitates that were obtained from nonapoptotic or apoptotic cells (i.e. endogenous basal and hyperphosphorylated keratins, respectively) to in vitro digestion by CPP32 or by cell lysates that were obtained from apoptotic or nonapoptotic cells. As shown in Fig. 4A, cell lysates from nonapoptotic cells do not cleave K18 (lanes 5 and 5Ј) while cell lysates from apoptotic cells cleave K18 regardless of its phosphorylation state (lanes 6 and 6Ј). Also, CPP32 digested K18 equally well (Fig. 4A, lanes 3 and 3Ј) regardless of its source. As expected, control precipitates that were obtained from An-treated cells (Fig. 4A, lanes 2Ј and 4Ј) or that were treated with a nonapoptotic cell lysate (lane 5Ј) had basally associated p29/p23. The role of keratin phosphorylation in ker-atin fragmentation was also examined in the BHK transfection cell system by transfecting a variety of K8 and K18 mutants that correspond to known phosphorylation and glycosylation sites (20,22,23,40) followed by testing for the presence of the p29/p23 fragments. As shown in Fig. 4B, transfection with several phosphorylation and glycosylation K8 and K18 mutants afforded similar K18 fragments to those noted upon transfection of wild-type K8/18.
Distribution of K18 Fragments within the Cytosolic and Detergent-soluble and -insoluble Compartments-We examined the distribution of p29/p23 K18 fragments in the cytosolic, Nonidet P-40, Emp, and remaining pellet fractions of apoptotic HT29 cells. The majority of the K8/18 pool is associated with the Emp and pellet fractions (Fig. 5, A and B) and so is p29/p23. Almost all pellet-associated K18 is degraded as detected by Coomassie staining and immunoblotting (Fig. 5, A and B), as is K18 in floater cells (Fig. 1E). In addition, p23 is distributed in FIG. 3. Time course of keratin phosphorylation and proteolysis. HT29 cells were treated with An (in Me 2 SO (DMSO)) or with an equivalent volume (0.1% of culture media volume) of Me 2 SO for the indicated hours. Floater and adherent cells were pooled, and a fraction was solubilized in 2% SDS-containing sample buffer while the remaining was solubilized in 1% Nonidet P-40 followed by immunoprecipitation of K8/18. Equivalent fractions were separated by SDS-PAGE and then Coomassie stained (a and b) or were transferred to PVDF membranes then blotted with the indicated antibodies to K18, K8, or K19 (mAb KA4). Similar K18/p23 and K19/p20 patterns to the corresponding K18/p29 and K19/p28 were obtained after blotting with mAb D5 or B/A2 (anti-K19/p20), respectively (not shown).
the Emp and pellet fractions (Fig. 5B, panel b), whereas p29 is found predominantly in the pellet fraction (Fig. 5B, panels a and d).

DISCUSSION
The findings of this report are summarized in Fig. 5C. Apoptosis results in selective caspase-mediated proteolysis of type I keratins, K18 and K19, while relatively sparing their partner type II keratin, K8. Although one K18 caspase site was defined (Asp-237) (Fig. 2), it is possible that other cut sites also occur with the potential presence of very small fragments. Given that similar caspase digestion motifs are found in other type I keratins and cytoplasmic IF proteins, it is likely that such proteolysis may be a generalized phenomenon. The time course of keratin degradation was similar to that of PARP (not shown). A significant fraction of K18 p29/p23 fragments remains associated with K8, and the majority of the K18 fragments are insoluble (Fig. 5), as normally found for K8/18 in nonapoptotic cells (17). However, although the cytosolic and Nonidet P-40 fractions have nearly equal amounts of p29/p23, the Emp and pellet fractions contain different proportions of p29 versus p23. This difference may reflect enhanced Emp extractibility of p23 versus p29 or may indicate an association of p23 with a cellular protein that is Emp-extractable.
The significance of keratin fragmentation remains to be determined. The apoptosis-associated morphological changes suggest that cytoskeletal reorganization is likely to be important in facilitating these changes. Most apoptotic cells (as determined by nuclear staining) manifested keratin filament disruption (not shown), although it remains to be determined if the disruption is due to keratin cleavage and/or hyperphosphorylation, particularly since both modalities can independently cause filament reorganization. For example, keratin filament reorganization may occur in association with a variety of hyperphosphorylated states (17,19), or after adenovirus infection and subsequent cleavage of the K18 head domain at Met-73 by the viral L3 proteinase (41). Of note, a similar finding of K18 caspase-mediated fragmentation was noted in apoptotic human SNG-M endometrial and mouse HR-9 parietal cells (42). Deletion analysis coupled with the known lamin caspase cleavage site (43,44) pointed to Asp-237 as a cleavage site that was verified by the inability of caspases to cleave K18 Asp-237 3 Glu (42). Interestingly, keratin fragments have been identified in sera of patients who harbor any one of a variety of epithelial tumors (45,46) and may have diagnostic potential (47).
Keratin hyperphosphorylation occurred early after induction of apoptosis while fragmentation was a late event. Increased keratin phosphorylation involved at least three of four known K8/18 sites (i.e. K8 S73/431 and K18 S52, but not K18 S33). The function of K8 pS73 is not known, but K18 pS52 is important for filament reorganization (20) and occurs with K8 pS431 FIG. 4. In vitro reconstitution of K18 proteolysis and effect of K8/18 phosphorylation mutants on keratin proteolysis. Panel A, K8/18 precipitates and detergent lysates (1% Nonidet P-40) were obtained from Anϩ and AnϪ cells as described under "Experimental Procedures." K8/18 precipitates were incubated with CPP32 buffer (lanes 2 and 2Ј), buffer ϩ CPP32 (lanes 3 and 3Ј), no additions while keeping over ice (lanes 4 and 4Ј), or with detergent lysates from ϪAn-treated (lanes 5 and 5Ј) or ϩAN-treated (lanes 6 and 6Ј) cells. Lane 1 shows CPP32 alone. After 3 h (37°C), equivalent fractions were separated by SDS-PAGE and Coomassie stained or were blotted with anti-pS33 K18 (anti-K18/p29) or with mAb D5 (anti-K18/p23). Note that CPP32 also cleaves K18 into a slightly faster migrating species than undigested K18 (lanes 3 and 3Ј). Panel B, BHK cells were transiently transfected with vector alone, wild-type K8/18 or the indicated K8 and K18 mutant constructs. After 3 days, cells were solubilized, followed by immmunoprecipitation of K8/18, analysis by SDS-PAGE, and then Coomassie staining. The glycosylation sites of K18 (Ser-29, -30, and -48) have been previously described (40). during mitosis and growth factor receptor activation (22,26). Hyperphosphorylation did not involve K18 Ser-33, which regulates binding to 14-3-3 proteins (37), 2 and no binding of 14-3-3 proteins to K8/18 was observed during apoptosis (not shown). It remains to be determined if keratin degradation and hyperphosphorylation are obligate, important, or a bystander phenomenon. Transfection of K8/18 phosphorylation mutants into BHK cells did not decrease fragment formation (Fig. 4B), which indicates that phosphorylation of these sites is not essential for proteolysis. In addition, in vitro reconstitution of caspase-mediated K8/18 digestion appears independent of keratin phosphorylation (Fig. 4A), which indicates that keratin hyperphosphorylation does not increase keratin susceptibility to caspase digestion.
Although the most common locations of human keratin mutations that have been identified to date are at the beginning and end of the rod domain (11)(12)(13)(14) (termed helix initiation and termination domains, respectively, Fig. 2B), mutations within the L1-2 domain of K14 have been described in a few patients. Such mutations are in immediate proximity to the potential K14 caspase cleavage site, based on sequence similarity of this K14 region (48) with K18 and K19 (e.g. K18 and K19 have 234 VEVD while K14 has 270 VEMD sequences). For example, Met-272 3 Arg (49) and Val-270 3 Met (50) mutations have been reported in patients with the Koebner and Weber-Cockayne forms, respectively, of epidermolysis bullosa simplex. Our results raise the questions of whether K14 undergoes caspasemediated proteolysis, and do the K14 Met-272 3 Arg (49) and Val-270 3 Met mutations interfere with such proteolysis. One clear relevance of these questions is that a decrease in keratin degradation could play a role in the pathophysiology of these mutations.