Mammalian Tolloid Metalloproteinase, and Not Matrix Metalloprotease 2 or Membrane Type 1 Metalloprotease, Processes Laminin-5 in Keratinocytes and Skin*

a major adhesive ligand for epithelial cells, undergoes processing of its (cid:1) 2 and (cid:2) 3 chains. This study investigated the mechanism of laminin-5 processing by keratinocytes. BI-1 (BMP-1 isoenzyme inhibitor-1), a selective inhibitor of a small group of astacin-like metal-loproteinases, which includes bone morphogenetic protein 1 (BMP-1), mammalian Tolloid (mTLD), mammalian Tolloid-like 1 (mTLL-1), and mammalian Tolloid-like 2 (mTLL-2), inhibited the processing of laminin-5 (cid:1) 2 and (cid:2) 3 chains in keratinocyte cultures in a dose-dependent manner. In a proteinase survey, all BMP-1 isoenzymes processed human laminin-5 (cid:1)

Several proteases have been implicated in laminin-5 processing. Exogenous addition of matrix metalloprotease 2 (MMP-2) 1 cleaved the ␥2 subunit of rat laminin-5 (12). A subsequent study suggested that membrane type 1 matrix metalloprotease (MT1-MMP) may play a role in cleaving laminin-5 (13). Cleavage of laminin-5 by plasmin converted the ␣3 chain into the 165-kDa form observed in human breast and rat epithelial cells and capable of nucleating hemidesmosomes (14). Bone morphogenetic protein 1 (BMP-1) has also been implicated in laminin-5 proteolysis. N-terminal sequencing of the 105-kDa ␥2 chain obtained from human keratinocytes revealed a cleavage site that matched the minimal consensus sequence of this metalloprotease (15). In vitro studies demonstrated that BMP-1 cleaved the recombinant ␥2 short arm at the predicted site and that the enzyme cleaved both the ␣3 and ␥2 chains of whole laminin-5 to generate characteristic 165-and 105-kDa fragments, respectively (15). Thus, of the proteases reported to cleave laminin-5, only BMP-1 was shown to process both chains, whereas plasmin was reported to process only the ␣3 chain, and MMP-2 and/or MT1-MMP were reported to process only the ␥2 chain of rat laminin-5.
BMP-1, first identified in osteogenic extracts of bone (16), is a metalloprotease of the astacin family (17). It has substantial homology with proteins involved in morphogenetic patterning such as Tolloid in Drosophila and Xolloid in Xenopus, and the mammalian enzyme may function both in patterning and in modifying components of the extracellular matrix (18). The first identified activity for BMP-1 was the cleavage of the C-terminal propeptide from procollagen types I-III in the process of collagen deposition in fibrillogenesis (19). Three additional mammalian enzymes highly homologous to BMP-1 have been reported: mammalian Tolloid, an alternatively spliced variant of the Bmp1 gene (mTLD) (20); mammalian Tolloid-like 1 (mTLL-1) (21) and mammalian Tolloid-like 2 (mTLL-2) (18). All of these BMP-1 isoenzymes share a similar domain structure comprising an N-terminal prodomain, an astacin-like metalloprotease domain, and several EGF (epidermal growth factor) and CUB domains (a widespread module found in developmentally regulated proteins), thought to be involved in mediating protein-protein interactions (17) (Fig. 1). mTLD, mTLL-1, and mTLL-2 also have additional C-terminal EGF and CUB domains not found in BMP-1 (18,20,21) (Fig. 1). The overall sequences of these three proteins have between 72 and 75% identity to one another, and identity is even higher in conserved regions such as the metalloprotease domain (18). In addition to procollagen C-proteinase activity, the BMP-1 isoenzymes have been reported to process prolysyl oxidase, probiglycan (22), and procollagen VII to their mature forms (23,24), and to cleave Chordin (18).
In the present study, we have utilized a novel inhibitor of the BMP-1 family to investigate processing of laminin-5 by the BMP-1 isoenzymes expressed in keratinocytes. We demonstrate that treatment of keratinocytes with this inhibitor significantly reduced the cleavage of both the ␣3 and ␥2 chains. A comparison of the enzymatic activities of the four BMP-1 isoenzymes on laminin-5 revealed that all forms are able to cleave both the ␣3 and ␥2 chains. However, only mTLD was present in significant quantities, at both the RNA and protein levels, in cultured keratinocytes under the conditions used in the study of laminin-5 cleavage. In addition, a defect of laminin-5 processing was demonstrated in skin lacking mTLD/BMP-1. Interestingly, neither MMP-2 nor MT1-MMP was able to cleave the ␥2 chain of human laminin-5, although they cleaved the rat chain, most likely reflecting sequence differences between the two proteins. Together, these data suggest a novel role for mTLD in processing laminin-5 in keratinocytes and the skin.

EXPERIMENTAL PROCEDURES
Enzymes and Inhibitors-Plasmin (United States Biological, Swampscott, MA or Calbiochem) and purified MMP-2 and MT1-MMP catalytic domains (Chemicon, Temecula, CA) were purchased from commercially available sources. BI-1 (a kind gift of FibroGen, Inc.), a selective hydroxamic acid inhibitor of BMP-1 and its isoenzymes, was dissolved in Me 2 SO.
A fluorogenic peptide containing a consensus BMP-1 cleavage site (49) was used to assess the inhibitory activity of BI-1 on BMP-1. In a similar assay, a fluorogenic peptide with the sequence Mca-P-L-G-L-Dpa(Dnp)-A-R-NH 2 (where Mca is 7-methoxycoumarin-4-acetic acid and Dpa is dinitrophenylalanine) (excitation 360 nm, emission 465 nm; Bachem, catalog No. M-1895) was used to measure the inhibitory activity of BI-1 on MMP-1, MMP-2, and MMP-9. The latter enzymes were activated according to the manufacturer's instructions using p-aminophenylmercuric acetate. BMP-1, and activated MMP-1, -2, and -9 were incubated for 1 h at 37°C in the presence of BI-1 and the appropriate fluorogenic peptide substrate (50 M); the increase in fluorescence was used to determine initial rates. MMP-2 activity was also verified by examining its ability to digest chicken type I [ 3 H]procollagen at 25°C as described previously for rat type I collagen (25), except that products were detected by autoradiography instead of Coomassie staining. Type 1 [ 3 H]procollagen was a kind gift from FibroGen, Inc. The effect of BI-1 on this reaction was examined. The activity of MT1-MMP and the effect of BI-1 on this enzyme were verified using pro-MMP-2 as a substrate and examining its conversion to the mature form. The digestions were analyzed by Western blot as below using antibody Ab-3 (Calbiochem) against MMP-2, which recognizes both the 66-and 72-kDa forms. All control reactions were run in buffer consisting of 50 mM Tris, pH 7.5, 150 mM NaCl, 5 mM CaCl 2 , 0.02% Brij 35.
Protein Analysis-Proteins were separated by SDS-PAGE using NOVEX gels (Invitrogen) and analyzed by Western blot as described previously (26). Goat anti-rabbit or goat anti-mouse secondary antibodies were from Amersham Biosciences and were used according to the manufacturer's instructions. Blots were developed using a standard ECL kit (Amersham Biosciences) unless indicated otherwise.
Recombinant BMP-1 Isoenzymes-Purified BMP-1 and conditioned media from cell lines stably expressing FLAG-tagged mTLD, mTLL-1, and mTLL-2 were the kind gifts of FibroGen, Inc. The mTLD, mTLL-1, and mTLL-2 proteins were affinity purified over a FLAG-Sepharose column essentially as described (18). Concentrations of purified enzymes were assessed using BCA assays and by direct comparison of Coomassie-stained band intensity with protein standards of known concentration. The activity of the purified enzymes was assayed by measuring the cleavage of a fluorogenic peptide containing a consensus BMP-1 cleavage site (49). The ability of BMP-1, mTLD, and mTLL-1 to cleave type I [ 3 H]procollagen was verified as described previously (18).
Antibodies-A laminin-5 polyclonal rabbit antibody (pKal) was raised against laminin-5, purified from human keratinocyte conditioned media by antibody affinity chromatography as described previously (5). Monoclonal antibody 19562 against the ␥2 chain of human laminin-5 was from Chemicon. The rabbit polyclonal antibody 1084 directed at domains IV and V of the mouse laminin ␥2 chain (31), was the kind gift of Dr. Rupert Timpl, Martinsreid, Germany. The polyclonal antisera S.E.-85 and S.E.-144 directed against the ␣3 and ␥2 chains of laminin-5, respectively (32), were a kind gift from Dr. Gim Meneguzzi, Nice, France.
The following affinity purified rabbit polyclonal antibodies raised against peptides from BMP-1 isoenymes were kind gifts of FibroGen, Inc.: pAb 585 against the BMP-1 C terminus (33), pAb 586 against the mTLD C terminus (33), pAb 531 against the mTLL-1 catalytic domain, pAb 517 against the BMP-1 and mTLD catalytic domain, and pAb TLL2 against the mTLL-2 catalytic domain. The reactivity of each antibody against 0.25 g of BMP-1, mTLD, mTLL-1, and mTLL-2 was assessed by Western blot. To test the presence of BMP-1 isoenzymes in keratinocyte medium, 90% confluent cultures of keratinocytes were washed three times to remove residual pituitary extract supplements and grown overnight in supplement-free medium (Keratinocyte SFM, Invitrogen). The conditioned media were collected, centrifuged at 4000 ϫ g to remove cellular debris, and concentrated 100 -200-fold using Centricon Plus 20 centrifugal concentrators (Millipore). Concentrated conditioned media derived from the culture of 1.5-2.5 ϫ 10 6 cells were electrophoresed per lane alongside 0.2 g of the purified BMP-1 isoenzymes as positive controls and analyzed by Western blot with the panel of anti-BMP-1 isoenzyme antibodies.
Immunoprecipitation of Unprocessed Laminin-5 and Enzyme Assays-For immunoprecipitation, human primary keratinocytes were grown to 70% confluency in six 225-cm 2 flasks, washed with Hanks' balanced salt solution (BioWhittaker) and Cys/Met-free keratinocyte growth medium (KGM; BioWhittaker), and then starved in the same medium for 30 min at 37°C. Cells were labeled for 24 h in Cys/Met-free KGM supplemented with 0.1 mCi/ml EasyTag EXPRESS (PerkinElmer Life Sciences) protein labeling mix. After labeling, cells were lysed with ice-cold radioimmunoprecipitation assay buffer (5). Cell debris was removed by centrifugation at 14000 ϫ g for 30 min at 4°C. The supernatant was snap-frozen and stored at Ϫ80°C until its use in immunoprecipitation.
For immunoprecipitation the supernatant was precleared with gelatin-Sepharose 4B (Amersham Biosciences) overnight at 4°C using 100 l of Sepharose/each ml of supernatant. Precleared supernatant was incubated overnight at 4°C with protein G-Sepharose beads to which K140 had been preadsorbed. Beads were pelleted, washed, and resus- Soluble, partially processed laminin-5 was purified from the conditioned media of LSV5 cells transfected with the wild-type ␥2 chain, as described previously, using antibody affinity chromatography (5). 1 g of purified protein was digested in a total volume of 20 l of digestion buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 5 mM CaCl 2 ) for 4 h at 37°C.
To produce unprocessed rat laminin-5 for cleavage assays, 804G cells were seeded in 48-well plates at a density of 2 ϫ 10 5 cells/well in the presence of 20 M BI-1 inhibitor and grown for 16 h. Cells were removed using 20 mM NH 4 OH and the remaining matrix washed as described previously (34). Laminin-5 was digested in a total volume of 100 l of digestion buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 5 mM CaCl 2 ) containing proteases for 4 h at 37°C. Following digestion, the matrix was washed twice with phosphate-buffered saline and solubilized as described previously (34). The processing state of laminin-5 was analyzed by Western blot using 1:1000 pKal with the signal detected using ECL Plus chemiluminescent reagent (Amersham Biosciences).
RNA Isolation and cDNA Synthesis-Cytoplasmic RNA was isolated from cells using RNeasy reagent kit (Qiagen). RNA was subsequently treated with RNase-free DNase using an "on column" DNase digestion kit (Qiagen). RNA concentration was determined with RiboGreen (Molecular Probes). cDNA synthesis was performed on 200 ng of total RNA using Omniscript™ reverse transcriptase (Qiagen), priming with oligo(dT) at a concentration of 1 M. Placental RNA was obtained from Clontech.
Quantitative PCR-Levels of gene expression were determined using a LightCycler instrument (Roche Applied Science) using the FastStart DNA Master SYBR Green (Roche Applied Science) with a final concentration of 4.1 mM MgCl 2 . The thermal cycling conditions began with polymerase activation at 94°C for 6 min followed by 42 cycles at 95°C for 15 s, 60°C for 5 s, and 72°C for 10 s. A standard curve comprising the gel-purified and PicoGreen (Molecular Probes)-quantitated amplicon from a previous PCR was run concurrently with each sample. Primers, used at a final concentration of 500 nM, are indicated in Table  I. The equivalent cDNA made from 4 ng of total RNA was used in each PCR reaction.
Other Methods-Indirect immunofluorescent (IDIF) microscopy and immunoelectron microscopy of skin samples were performed as described previously (35).

Inhibition of Laminin-5 processing in Keratinocytes Using a BMP-1 Isoenzyme
Inhibitor-Extracellular processing of laminin-5 by keratinocytes was shown to be inhibited by orthophenanthroline, an inhibitor of metalloproteases, and sequence analysis of the cleavage site of the ␥2 chain implicated BMP-1 in this activity (15). Moreover, BMP-1 cleaved both the ␣3 and ␥2 chains of radiolabeled, immunoprecipitated laminin-5 (15). In this study we have utilized BI-1, a hydroxamic acid-based selective inhibitor of BMP-1 isoenzymes, to investigate the role of these metalloproteases in laminin-5 processing. BI-1 has IC 50 values of 6, 2, and 4 nM for BMP-1, mTLL-1, and mTLL-2 respectively compared with IC 50 values of greater then 50 M for MMP-2 and MMP9. 2 Under the culture medium calcium concentrations used in this study, keratinocytes produced a mixture of unprocessed and processed forms of both the ␣3 and ␥2 chains of laminin-5 in the extracellular matrix ( Fig. 2A, i). When cells were seeded and grown in the presence of increasing concentrations of BI-1 for 72 h, the appearance of the unprocessed 200-kDa ␣3 and 155-kDa ␥2 bands indicated that the inhibition was dose-dependent ( Fig. 2A, i). The kinetics of cleavage differed for the two chains. Cleavage of the ␥2 chain was 85% inhibited at 0.5 M BI-1 ( Fig. 2A, iii). In contrast, processing of the ␣3 chain appeared more complex. In the absence of the inhibitor, three bands were apparent: the 200-kDa unprocessed form, the 165-kDa form processed at the C-terminal end of the chain, and a 145-kDa form, which had undergone additional processing in domain IIIa of the N-terminal region ( Fig. 2A, ii) (5,15). These forms comprise 14, 52, and 34%, respectively, of the total ␣3 (Fig. 2B). The addition of BI-1 inhibits both of these cleavage events, with the 145-kDa band reduced to 9% of the total at 5 M BI-1, whereas the 165-kDa band still comprised 28% of the total ␣3 at 50 M BI-1 (Fig. 2B). This may indicate that the cleavage site in the G-domain, which generates the 165-kDa fragment, is particularly labile and sensitive to proteolysis.
Laminin-5 ␣3 and ␥2 Chains Are Substrates for all BMP-1 Isoenzymes-BMP-1 has been reported to cleave both the ␣3 and ␥2 chains of radiolabeled human laminin-5 immunoprecipitated from keratinocyte cell lysates (15). To determine whether other BMP-1 isoenzymes were also capable of catalyzing this reaction, we tested the ability of the panel of BMP-1 isoenzymes to cleave unprocessed laminin-5. Purified recombinant BMP-1, mTLD, mTLL-1, and mTLL-2 used in the digestions were present as single bands on a Coomassiestained SDS-polyacrylamide gel with molecular weights corresponding to those of the mature proteins from which proregions had been proteolytically removed (18) (Fig. 3A). Their proteolytic activity was assayed using a fluorogenic peptide containing a consensus BMP-1 cleavage sequence, and all enzymes were found to have similar activity (Fig. 3A). In addition, the ability of BMP-1, mTLD, and mTLL-1 to cleave type I procollagen as reported previously (18) was confirmed (data not shown). mTLL-2 also cleaved type I procollagen at a low efficacy, similar to the efficiencies of mTLD and mTLL-1 (data not shown).
Radiolabeled, unprocessed intracellular human laminin-5 was immunoprecipitated from keratinocytes and used as a substrate for digestion. Under the conditions of the assay, unprocessed laminin-5 was completely stable in the absence of added enzyme (Fig. 3B). Digestion of unprocessed laminin-5 with each BMP-1 isoenzyme at 100 nM resulted in the appearance of the 165-kDa processed ␣3 chain and the 105-kDa processed ␥2 band. No smaller molecular weight ␥2 chain fragments were observed. The laminin-5 cleavage activities using 100 nM BMP-1, mTLD, mTLL-1, and mTLL-2 were abolished in the presence of 10 M BI-1, and only the unprocessed ␣3 and ␥2 forms were observed. mTLD, mTLL-1, and mTLL-2 Cleave the ␥2 Chain of Laminin-5 at the BMP-1 Site-BMP-1 has been shown to cleave the human laminin-5 ␥2 chain at a site identical to that which generates the 105-kDa fragment found physiologically (15). To determine whether other BMP-1 isoenzymes also cleaved this chain at an identical site, we used a mutant ␥2 chain, ␥2-pNC, in which the four-amino acid BMP-1 site (YSGD) was deleted (28). This construct was stably transfected into keratinocytes lacking the ␥2 chain (LSV5) (27) and was used to produce laminin-5 containing the mutant ␥2 chain. Likewise, wild-type ␥2 was generated by transfection of LSV5 cells with a plasmid encoding the wild-type ␥2 chain. Mutant and wild-type laminin-5 were radiolabeled and immunoprecipitated from these cells and used as substrates for digestion. All BMP-1 iso-2 M. Brenner, personal communication.
mTLD Processes Epidermal Laminin-5 enzymes cleaved the ␥2 chain of the wild-type laminin-5, whereas none of the enzymes cleaved the ␥2 chain of the mutant laminin-5, indicating that all BMP-1 isoenzymes cleave the ␥2 chain at the established BMP-1 site (Fig. 3C).
Cleavage of Laminin-5 by MMP-2, MT1-MMP, and Plasmin-In addition to BMP-1, the metalloproteases MMP-2 and MT1-MMP have been reported to cleave the ␥2 chain of laminin-5 (12,13). To clarify the role of these enzymes in laminin-5 processing and to compare their activity with that of BMP-1, we studied their cleavage activity on both human and rat laminin-5. To verify the activity of MT1-MMP and the specificity of BI-1, we digested pro-MMP-2 to its active form with MT1-MMP in the presence and absence of the inhibitor. Likewise, MMP-2 activity was tested using chicken type I procollagen as a substrate, because MMP-2 had previously been shown to cleave rat type I collagen (25). Neither MMP-2 cleavage by MT1-MMP nor procollagen cleavage by MMP-2 was inhibited by 10 M BI-1 (Fig. 4A). To directly compare the abilities of MMP-2 and MT1-MMP to cleave laminin-5 with those of the BMP-1 isoenzymes, we tested the activities of these proteases by digesting labeled immunoprecipitated human laminin-5 as described above. Surprisingly, neither MMP-2 nor MT1-MMP cleaved the ␥2 chain to either the 105-kDa or 80-kDa forms, even at enzyme concentrations as high as 300 nM (Fig. 4B).
However, both enzymes cleaved the ␣3 chain to the 165-kDa form, and these processing events were not inhibited by 10 M BI-1 (Fig. 4B).
Recently, it was reported that digestion of soluble human laminin-5 ␥2 chain with MT1-MMP generated several bands smaller than 105 kDa (36). In our studies with immunoprecipitated laminin-5, the protein used in enzyme digestion reactions is complexed to Sepharose beads. We therefore examined the ability of MT1-MMP to digest soluble laminin-5. No cleavage of the ␥2 chain was seen at molar ratios of MT1-MMP to laminin-5 up to 3:1 (Fig. 4C), confirming our results obtained with the immunoprecipitated laminin-5. We were also unable to demonstrate MMP-2-or MT1-MMP-mediated cleavage of laminin-5 deposited upon tissue culture plastic by human keratinocytes, a substrate that is cleaved by BMP-1 isoenzymes (data not shown).
The lack of a clear 80-kDa (␥2pЈ) laminin-5 band in any of our MMP-2 and MT1-MMP digestions is contrary to the findings of previous studies but may be explained by the use of human instead of rat laminin-5, as the two share no identity at the MMP-2 cleavage site (residues 582-589 in rat and 601-608 in human). We therefore examined rat laminin-5 from the extracellular matrix of rat bladder epithelial 804G cells, which contains a laminin ␥2 chain typically processed to a mixture of 105

FIG. 3. BMP-1 isoenzymes digest ␣3
and ␥2 chains of laminin-5. A, BMP-1 isoenzymes were purified as described previously (18). The activity of 0.1 g of protein was determined using a fluorogenic peptide containing a consensus BMP-1 cleavage site and expressed as % cleavage/h Ϯ S.D. B, radiolabeled, immunoprecipitated human laminin-5 was incubated with 100 nM BMP-1 isoenzymes for 2 h in the presence and absence of 10 M BI-1. Bands were detected by autoradiography. C, radiolabeled human laminin-5 was immunoprecipitated from LSV5 cells containing either the wild-type ␥2 chain (g2) or the BMP-1 site mutant ␥2 chain (pNC). Wild-type and mutant proteins were digested for 2 h with 100 nM BMP-1 isoenzymes, and the cleaved ␥2 bands (105 kDa) were detected by autoradiography.
mTLD Processes Epidermal Laminin-5 (␥2p)-and 80 (␥2pЈ)-kDa fragments. When 804G cells were cultured in the presence of BI-1, only the processed 80-kDa ␥2 band was detected; the addition of MMP-2 or MT1-MMP to this material resulted in the full conversion of unprocessed ␥2 to the 80-kDa band. No ␥2p band at 105 kDa was detected (Fig 4D). The addition of BMP-1 generated the ␥2p but not the ␥2pЈ fragments, as observed with the human substrate. These results suggest that both BMP-1 isoenzymes and MMPs are active in processing laminin-5 in this rat cell line, and that MMP-2 and MT1-MMP directly convert the 155-kDa ␥2 chain to the 80-kDa ␥2pЈ form without cleavage to an intermediate form.
The serine protease, plasmin, was reported to cleave the ␣3 chain of laminin-5 (14). The addition of 100 nM plasmin cleaved the ␣3 chain of laminin-5 to the characteristic 165-kDa band (Fig. 4B). The size of this band is indistinguishable from that seen for laminin-5 processed by keratinocytes or for laminin-5 digested by BMP-1, mTLL-1, or mTLL-2. Plasmin also appeared to cleave the ␥2 chain to the 105-kDa form. The addition of 10 M BI-1 did not inhibit the digestion activity of plasmin on either the ␣3 or ␥2 chains (Fig. 4B).
The Major Secreted BMP-1 Isoenzyme in Cultured Human Keratinocytes Is mTLD-To identify which BMP-1 isoenzyme is secreted by human keratinocytes, we analyzed keratinocyte conditioned media by Western blot. Keratinocytes have been reported to secrete predominantly mTLD in its proenzyme form in response to TGF␤1 (33). However, neither mTLL-1 nor mTLL-2 was examined in this previous report. To study all BMP-1 isoenzymes, we initially determined the reactivities of a panel of antibodies with BMP-1, mTLD, mTLL-1, and mTLL-2. pAb 517, raised against the catalytic domain of BMP1/mTLD, cross-reacted strongly with all four isoenzymes (Fig. 5A, lanes  6 -9). pAb 585, raised against the C terminus of BMP-1, recognized only this enzyme, whereas pAb 586, raised against the C terminus of mTLD, recognized this isoenzyme and also mTLL-2. pAb 531, raised against the catalytic domain of mTLL-1, was specific for mTLL-1, and pAb Tll-2, raised against the analogous sequence in the catalytic domain of mTLL-2 to pAb 531, recognized both mTll-2 and mTll-1 (Fig. 5B). Thus BMP-1, mTLD, mTLL-1, and mTLL-2 each have a unique pattern of reactivities with this panel of antibodies.
When concentrated conditioned keratinocyte medium was analyzed using these antibodies, a doublet at ϳ130 and 140 kDa was detected with pAb 517 and pAb 586 antibodies only (Fig. 5A, lanes 1 and 3). No bands were detected with any of the other antibodies that were not also present in the serum only control (Fig. 5A, lanes 2, 4, and 5). From this pattern of reactivity, we conclude that mTLD is the major BMP-1 isoenzyme found in keratinocyte conditioned media. The molecular masses of the two bands are in the range of the proenzyme, predicted from sequence data to be 143 kDa, and the mature form of mTLD, reported to be 130 kDa (18). A minor 100-kDa band detected with pAb 517 and pAb 586 is likely a degradation product of mTLD.
Expression of RNA levels for the BMP-1 isoenzymes in primary keratinocytes was also assessed using quantitative PCR, with placental and fibroblastic RNA as positive controls for BMP-1 and mTLD. A suitable positive control RNA could not be mTLD Processes Epidermal Laminin-5 found for mTll-2, but primers were validated using a plasmid containing the mTLL-2 cDNA. In agreement with the detection of only mTLD in conditioned keratinocyte media, this isoenzyme was the most abundant form in keratinocytes, expressed at levels approximately two orders of magnitude greater than BMP-1 and mTLL-1 (Table II). Human dermal fibroblasts also expressed significant levels of mTLD. No mTLL-2 was detected in any of the samples analyzed.
mTLD/BMP-1-deficient Skin Shows Defective Laminin-5 Processing-Given that all BMP-1 isoenzymes are capable of cleaving both the ␣3 and ␥2 chains of laminin-5, the presence of mTLD in keratinocyte conditioned media suggests that mTLD may be the enzyme responsible for cleaving laminin-5 in skin. To further evaluate this possibility, we examined mTLD/BMP-1-deficient skin prepared from embryos homozygous null for the Bmp1 gene (37), which encodes both proteases, using pAb 1084 directed at domains IV and V of the laminin ␥2 chain. This antibody recognizes only those laminin-5 molecules that contain an unprocessed ␥2 chain (38). Increased unprocessed laminin ␥2 chain was noted at the dermal-epidermal junction in mTLD/BMP-1-deficient skin compared with wild-type skin in E18 mouse embryos by IDIF microscopy (Fig. 6, A and C). In contrast, expression of total laminin-5, detected by the pKal pAb, was similar in both samples (Fig. 6, B and D). As a next step, we examined the localization of laminin-5 in mTLD/BMP-1-deficient and wild-type skin using immuno-electron microscopy. As with IDIF studies, increased unprocessed extracellular laminin ␥2 chain was noted in mTLD/BMP-1-deficient skin (Fig. 6, F and G) relative to wild-type skin (Fig. 6E). Extracellular unprocessed laminin-5 appeared to localize predominantly to the lamina densa. Although blister formation was not clinically apparent in mTLD/BMP-1-deficient mouse skin, at the ultrastructural level the lamina densa was often not well opposed to overlying hemidesmosomes, and hemidesmosomes were often rudimentary in appearance in mTLD/BMP-1-deficient skin (e.g. Fig. 6F). When complete separation of the epidermis and dermis was noted in deficient skin, it appeared to occur in the plane of the lamina lucida with unprocessed laminin ␥2 chain localizing to the dermal side of the skin separation (e.g. Fig. 6G).

DISCUSSION
The selective inhibitor BI-1 has facilitated the study of laminin-5 processing in vitro. BI-1, at the concentrations used, inhibits the BMP-1 isoenzymes but does not inhibit other metalloproteases such as MMP-2 and MT1-MMP, and thus it functions as a tool for distinguishing the effect of these enzymes in cellular processes. In vitro digestion assays of unprocessed laminin-5 with BMP-1 isoenzymes, analysis of conditioned keratinocyte media with a panel of antibodies against these isoenzymes, and quantitative PCR studies together suggest that mTLD mediates laminin-5 cleavage in primary keratinocytes.
Interestingly, the ␣3 and ␥2 chains are not equivalent in terms of sensitivity to proteolysis. Processing of the ␥2 chain was inhibited at substantially lower concentrations of BI-1 than either of the ␣3 chain cleavages. Similarly, in cell culture there is a lag in ␥2 processing relative to ␣3 processing (5,15). The kinetics of this cleavage may reflect a greater accessibility of the protease to the ␣3 chain cleavage site than to the ␥2 cleavage site or the fact that cleavage of the ␣3 chain cooperatively enhances ␥2 chain cleavage. The physiological significance of these kinetics may be linked to the differing roles played by the individual chains in influencing cellular processes such as adhesion, migration, and matrix assembly.
BMP-1 isoenzymes have been shown to be generally similar in their substrate specificities, with all enzymes capable of processing pro-lysyl oxidase (18) and procollagen (16) (and this report) to their mature forms. It is probable that these similarities reflect the need for redundancy in the biological functions of the BMP-1 isoenzymes. This was demonstrated previously in Bmp1 knockout mice in which deficiency of BMP-1 and mTLD  did not prevent processing of type I and type VII procollagens, indicating the involvement of several isoforms in these activities (24,37). The similarity of the BMP-1 isoenzyme activities is likewise indicated by the fact that all enzymes are processed at the ␥2-chain at the same established BMP-1 cleavage site.
The presence of mTLD, and the apparent absence of the other BMP-1 isoforms in keratinocyte media by Western blot analysis, as well as the predominant expression of mTLD in primary keratinocytes and fibroblasts by quantitative PCR analysis all suggest that mTLD is likely active in cleaving laminin-5 in vivo. TGF␤1 stimulation was not required to detect the mTLD bands in keratinocyte media in this study despite a previous report that the unprocessed form of mTLD is detected only in response to TGF␤1 (33). The differing results could reflect the 3-5-fold greater amount of conditioned media loaded in the current study than in the previous report (33), as no specific bands were detected when less than this amount of media was analyzed.
BMP-1 and laminin-5 were previously shown to co-localize at the dermal-epidermal junction of fetal calf skin (15). The polyclonal antibody used in that study was raised against BMP-1 but is reactive against both BMP-1 and mTLD. 3 Furthermore, BMP-1/mTLD localized to basal keratinocytes in normal human skin, whereas mTLL-1 expression was absent in basal keratinocytes (24). These results are consistent with a role for mTLD in the dermal-epidermal basement membrane and, coupled with our findings of defective laminin-5 processing in mTLD/BMP-1-deficient mice, support the hypothesis that mTLD is involved with the processing of the laminin-5 ␥2 and ␣3 chains in vivo.
Previous studies have shown that MMP-2 (12) and MT1-MMP (13) each cleaved the laminin-5 ␥2 chain, in apparent contradiction to our study. However, these previous studies used rat laminin-5, whereas our study examined human laminin-5. The sequence of the rat ␥2 chain at the MMP-2 cleavage site (residues 582-589) (12) has very low identity with the corresponding region of the human protein (residues 601-608), which likely explains the observed lack of cleavage of human laminin ␥2 chain by MMP-2 in our experiments. It is quite possible that MMP-2 and MT1-MMP cleave the same unconserved site on the rat laminin ␥2 chain, because each enzyme yields a similar 80-kDa product; however, the sequence of the MT1-MMP cleavage site has not been reported. Although MMP-2 and MT1-MMP cleaved rat laminin ␥2 chain to the 80-kDa fragment, neither enzyme, even at concentrations as high as 300 nM, could convert laminin-5 ␥2 chain to the 105-kDa fragment (12,13), which is the only processed product of the laminin ␥2 chain found in human tissues (5).
MMP-2 and MT1-MMP are known to be expressed in human keratinocytes (39 -42). However, laminin-5 processing in keratinocyte cultures was completely inhibited by BI-1, which we have shown to have no effect on MMP-2 or MT1-MMP activity. These findings, coupled with the observation that the sequence of the human ␥2 chain in vivo cleavage site exactly matches that of the BMP-1-cleaved ␥2 chain in vitro (15), suggest that neither MMP-2 nor MT1-MMP is likely to play a role in processing the human laminin ␥2 chain. Interestingly, the addition of BI-1 to rat 804G cells inhibited the generation of the ␥2p but not ␥2pЈ fragments, which suggests that one of the BMP-1 isoenzymes may be cleaving rat laminin-5 in addition to MMP-2 or MT1-MMP in these cells. The relative contribution of each of these enzymes to laminin-5 cleavage in the rat model is an area deserving of further investigation.
Although neither MMP-2 nor MT1-MMP cleaved the human laminin ␥2 chain, both enzymes, as well as all other proteases examined in this and other reports, were able to cleave the ␣3 chain to a fragment indistinguishable in size from the 165-kDa product observed in the extracellular matrix of cultured keratinocytes. This panel included a serine protease, plasmin, and members of the astacin and matrix metalloprotease families. This suggests that the ␣3 cleavage site lies on a portion of the chain that is highly sensitive to proteolytic attack, consistent with electron micrograph data that places the cleavage site in an accessible spacer region between globular domains G3 and G4 (43). Recently, the C-terminal peptide released after cleavage of the 200-kDa ␣3 chain was isolated from the conditioned medium of human keratinocytes (44). The N-terminal sequence of this fragment suggested that a member of the MMP, astacin, or ADAM families could act as the processing enzyme of this chain (44). In a separate report, partial sequencing of two fragments of 35 and 37 kDa released by cleavage of the ␣3 chain indicated that cleavage occurred in the region between the G3 and G4 domains (45). Although MMP-2 and MT1-MMP do cleave the ␣3 chain in our in vitro assays, this may not be physiologically relevant, as these enzymes are not inhibited by BI-1, which prevents most laminin-5 processing when added to keratinocyte cultures. This observation does not preclude a role for these enzymes in the upstream activation of another protease involved in cleavage of the ␣3 chain. Our results hint at a heretofore unsuspected complexity of laminin-5 processing in which different enzymes may cleave the ␣3 and ␥2 chains and in which ␣3 chain processing itself may involve more than one protease. The involvement of proteases of the MMP family and plasmin in modifying the extracellular matrix in the process of carcinoma invasion is well established (1,2). It has been suggested that members of the astacin family may also play a role in these processes (1). Proteolytic modification of laminin-5 structure has been shown to be an important process underlying keratinocyte migration, wound healing, and tumor invasion. In particular, increased expression of laminin-5 is often observed at the margins of squamous cell carcinoma tumors (46 -48), and thus processing of such laminin-5 by BMP-1 isoenzymes may be a critical step in invasion. Therefore, specific inhibitors of BMP-1 isoenzymes, such as BI-1, may prove to be invaluable in the study of laminin-5 processing in squamous cell carcinoma invasion.