Limited Redundancy of the Proprotein Convertase Furin in Mouse Liver*

Furin is an endoprotease of the family of mammalian proprotein convertases and is involved in the activation of a large variety of regulatory proteins by cleavage at basic motifs. A large number of substrates have been attributed to furin on the basis of in vitro and ex vivo data. However, no physiological substrates have been confirmed directly in a mammalian model system, and early embryonic lethality of a furin knock-out mouse model has precluded in vivo verification of most candidate substrates. Here, we report the generation and characterization of an interferon inducible Mx-Cre/loxP furin knock-out mouse model. Induction resulted in near-complete ablation of the floxed fur exon in liver. In sharp contrast with the general furin knock-out mouse model, no obvious adverse effects were observed in the transgenic mice after induction. Histological analysis of the liver did not reveal any overt deviations from normal morphology. Analysis of candidate substrates in liver revealed complete redundancy for the processing of the insulin receptor. Variable degrees of redundancy were observed for the processing of albumin, α5 integrin, lipoprotein receptor-related protein, vitronectin and α1-microglobulin/bikunin. None of the tested substrates displayed a complete block of processing. The absence of a severe phenotype raises the possibility of using furin as a local therapeutic target in the treatment of pathologies like cancer and viral infections, although the observed redundancy may require combination therapy or the development of a more broad spectrum convertase inhibitor.

seven closely related subtilisin-like serine proteases (furin, PC1 (also known as PC3), PC2, PC4, PC5 (also known as PC6), PACE4 and LPC (also known as PC7 and PC8)) (1-3) and two more distantly related members (SKI-1/S1P and NARC-1/ PCSK9) (2,4), with different, albeit partially overlapping, expression patterns and subcellular localization. Since the former seven PCs have similar specificities (basic amino acid motifs) it is impossible to predict physiological enzyme-substrate pairs, and most attempts at identification have relied on in vitro studies. However, co-overexpression studies of substrates with a PC are prone to false positive results, due to non-physiologically high enzyme levels, often in a heterologous cellular context. Direct evidence for enzyme-substrate pairs in vivo has been obtained from knock-out mouse models and human patients (recently reviewed in Ref. 5). Most informative have been the knock-out mouse models for PC1, PC2, and PC4, which have a restricted expression pattern. Inactivation of PC1 and PC2, which are restricted to neuroendocrine tissues, causes multiple endocrine peptide-processing defects that result in severe but viable phenotypes (6,7). The differences between PC1 deficiency in mouse and humans are striking, the former is characterized by dwarfism, the latter by obesity (8,9). Inactivation of germ cell-specific PC4 results in reduced fertility, at least in part due to blocked processing of propituitary adenylate cyclase-activating polypeptide (PACAP) (10,11). Altogether, these studies have indicated that some substrates can be cleaved only by one PC, whereas other substrates are cleaved by multiple PCs. Depending on the co-expression of other PCs, redundancy may be cell type-specific.
The knock-out mouse model of the ubiquitously expressed convertase LPC has no reported phenotype suggesting complete redundancy (12). Of course, it is possible that a more specialized function of LPC has resulted in a phenotype too subtle to detect by standard analysis. The null phenotypes of the other broadly expressed enzymes furin, PACE4 and PC6, feature early embryonic lethality, indicating vital, non-redundant functions of these enzymes at early developmental stages (5,13,14). However, direct analysis of potential substrates involved in embryogenesis is difficult because of their low expression levels. Only BMP-4 cleavage has been examined directly and was shown to occur normally even in furin/PACE4 double knock-out embryos (15). Indirect evidence was given for other substrates such as Nodal. The embryonic lethality precludes the analysis of many substrates expressed only in adult life or well differentiated tissues.
This study describes the first conditional knock-out mouse model available for PCs. The inducible nature of the model has allowed us to investigate adult mice. We have focused on processing of substrates in liver for three reasons. First of all, the inducible knock-out system used here results in near-complete inactivation of the fur gene in liver but is less efficient in other tissues. Second, this biosynthetically very active organ produces a variety of potential substrates. Finally, the expression of high levels of furin in the presence of other PCs such as PACE4, PC6, and LPC provides an excellent opportunity to study whether specific enzyme-substrate relations exist in the presence of an extensive processing back-up potential.

MATERIALS AND METHODS
Derivation of fur flox/flox Mice-The same 6-kb NotI fragment comprising exon 1A to intron 6 murine fur sequences as used for the construction of a targeting vector for the general fur knock-out mouse (14) was used for the construction of a floxed targeting vector to generate a conditional fur knock-out mouse. A 2-kb BamHI fragment encoding the hygromycin B phophotransferase (hygB) gene, fused to the phosphoglycerate kinase promoter and flanked by loxP sites, was cloned into a BamHI site upstream of exon 2. A third loxP site was introduced in a BamHI site downstream of exon 2 (Fig. 1A). The introduced loxP sites consisted out of a central 34-bp loxP recognition sequence flanked by additional sequences, which were used to generate, by means of PCR, fragments with suitable restriction enzyme recognition sites at their ends. The total size of each loxP site insertion was 92 bp. The NotI insert of the targeting construct was excised and electroporated into E14 embryonic stem cells (16). Hygromycin B-resistant colonies were analyzed by Southern blot analysis. PCR amplification was used to confirm the presence of the third loxP in the recombinants. To remove the HygB selection marker cassette flanked by loxP sites from the floxed fur allele in vitro, Cre recombinase was transiently expressed in the ES cell line after electroporation of the pGK-CreNLSbpA plasmid (kindly provided by W. Mü ller, University of Cologne). To obtain ES cell lines with the desired deletion of the HygB selection marker cassette resulting from partial recombination by Cre recombinase (Fig. 1B), the protocol as described by Torres and Kü hn (17) was applied. Correct, partial Cre recombination was confirmed by PCR amplification, and these ES cell clones were injected into C57Bl6/J blastocysts using standard procedures. The resulting chimeras were mated with C57Bl6/J mice. Southern blotting and PCR analysis of tail DNA samples were used to confirm transmission of the fur flox allele to the offspring. Heterozygous offspring were intercrossed to obtain homozygous fur flox mice.
Breeding with Cre Recombinase Transgenic Mice-To obtain in vivo inactivation of the fur flox allele by Cre recombinase homozygous fur flox/flox mice were crossed with two Cre transgenic mice. The pGK-Cre transgene mouse is a general deleter strain, which shows an early and uniform Cre expression (18). In the Mx-Cre mouse (strain Mx-Cre31a), Cre expression can be transiently induced by application of interferon (interferon-␣ or interferon-␤) or synthetic double-stranded RNA (e.g. polyinosinic-polycytidylic acid (pI-pC), an interferon inducer) (19). To induce expression of Cre in mice with the inducible Mx-Cre allele, 250 g of pI-pC (ICN Pharmaceuticals; solution of 1 mg/ml in water) was injected five times intraperitoneally at 2-day intervals. Two days later (10 days after the first injection), the animals were analyzed.
Probes and PCR Analysis for Genotyping of ES Cell Lines and Mice-For Southern blot analysis, both fur genomic and cDNA probes were used (Fig. 1A). cDNA probe 1 is a 0.6-kbp KpnI subfragment of the fur cDNA described previously (20) corresponding to exon 1A (only 3Ј part) and exons 2-4 (nucleotides 176 -761). cDNA probe 2 was obtained by PCR amplification of nucleotides 970 -1231 corresponding to exons 7 and 8 sequences. Genomic probe 3 encompasses a 0.5-kbp BamHI-KpnI fragment located immediately downstream of the BamHI site used to introduce the third loxP site, and encoding intron sequences and exon 3 and exon 4. The HygB probe, a 529-bp fragment of the hygromycin B phosphotransferase gene, was generated by PCR amplification using primers 5Ј-CAGCGAGAGCCTGACCTATTGC-3Ј and 5Ј-CGATCCT-GCAAGCTCCGGATG-3Ј.
Primers P1 and P2 are localized upstream and downstream respectively, of the BamHI site used to insert the HygB cassette flanked by loxP sites. This pair gives an amplimer size of 219 or 311 bp, depending on the presence of the loxP site. Primers P3 and P4 are localized upstream and downstream, respectively, of the BamHI site used to insert the third loxP site. This pair gives an amplimer size of 390 or 482 bp, depending on the presence of the loxP site. Combination of primer P1 and P4 gives fragments of 1442 and 1258 bp on the fur flox and fur wt allele, respectively. After deletion by Cre recombinase, this primer set gives a fragment of 322 bp on the inactivated fur ⌬flox allele.
For analysis of transmission of the Cre transgenes to the offspring, two additional primer pairs were used. Primer pair 5Ј-CCGGGCTGC-CACGACCAA-3Ј and 5Ј-GGCGCGGCAACACCATTTTT-3Ј amplified an internal Cre fragment of 445 bp.
Primer pair 5Ј-GCGGAGCCAGCACTATTTA-3Ј and 5Ј-CCGGCAT-CAACGTTTTCTTTT-3Ј amplified a fragment of ϳ450 bp corresponding to Mx-promoter sequences up to sequences in the Cre coding region.
Northern Blot Analysis and Quantitative Reverse Transcription PCR-Total RNA was extracted from liver tissue from induced wildtype and transgenic mice. For Northern blot analysis 15 g of RNA was size-fractionated on a 1% agarose gel, blotted, and hybridized according to standard procedures as described (20). For fur an exon 2-specific cDNA probe of 380 bp was used and a 2.8-kb cDNA probe spanning the entire coding sequence was also used. In addition, a 1.7-kb cDNA probe for PACE4, a 3-kb PC5 cDNA probe (containing the sequences common in PC5A and PC5B), a 2.5-kb LPC cDNA probe, and a 350-bp actin probe were used. Quantitative reverse transcription PCR was performed using TaqMan real-time PCR. cDNA was prepared using the SuperScript first-strand synthesis system for reverse transcription PCR (Invitrogen). Real-time PCR was performed using the qPCR core kit (Eurogentec) and was carried out using the ABI Prism 7700 sequence detection system (Applied Biosystems). Probes and primers for fur and GAPDH (glyceraldehyde-3-phosphate dehydrogenase gene), the reference gene, were designed by Primer Express 2.0 (Applied Biosystems). The probes were labeled on the 5Ј-end by 6-carboxyfluorescein and the 3Ј-end by 6-carboxytetramethylrhodamine. The following primers were used: fur exon 2 forward primer, 5Ј-CAGAAGCATGGCTTCC-ACAAC-3Ј; fur exon 3 reverse primer, 5Ј-TGTCACTGCTCTGTGCCA-GAA-3Ј; fur exon 2-3 probe, 5Ј-TGGGCCAGATCTTCGGTGACTATTA-CCA-3Ј; GAPDH forward primer, 5Ј-ATGGCCTTCCGTGTTCCT-3Ј; GAPDH reverse primer, 5Ј-CAGGCGGCACGTCAGAT-3Ј; GAPDH probe, 5Ј-CCCCCAATGTGTCCGTCGTG-3Ј. The relative gene expression was quantified using the comparative threshold cycle method (Applied Biosystems).
Histology-For histological analysis, liver tissues from both induced and non-induced wild type and transgenic mice were processed in parallel to provide appropriate controls also for the induction procedure itself.
Paraffin blocks were serially sectioned at 7 m on a rotary microtome (SLEE, Mainz, Germany) and sections further processed for alternating routine stains (hematoxylin and eosine) and immunohistochemistry. Tissues embedded in Araldite were cut on a LEICA Ultracut UCT at 1 m and serial sections were stained with either toluidine blue or paraphenylene diamine.
Western Blotting-Processing of candidate substrates was analyzed by Western blotting. Postnuclear supernatant of liver tissue was centrifuged for 1 h at 100,000 ϫ g to obtain soluble and membrane fractions. SDS-PAGE and isoelectric focusing were performed as described previously (21); equal amounts of protein from liver extracts from mice of the indicated genotype were loaded. Immunodetection of the insulin receptor was performed by a combination of immunoprecipitation using rabbit antibody CO-10 followed by Western blotting using monoclonal antibody CT-1 as described (22). Both antibodies were kindly provided by Ken Siddle (Cambridge, UK). Anti-␣ 1 -microglobulin antibody was a generous gift of Cecilia Falkenberg (Lund, Sweden). Anti-␣ 5 integrin antibody was purchased from Chemicon, anti-mouse albumin was from Biotrend, and anti-vitronectin was from Santa Cruz Biotechnology. Anti-lipoprotein receptor-related protein (LRP) antibody, directed against the C terminus of the LRP precursor and the ␤-subunit, has been described before (23).

Derivation of ES Cells with a FUR flox Allele-
The targeting construct for the generation of a floxed fur (fur flox ) allele, enabling conditional inactivation, contained the hygromycin B phophotransferase (HygB) gene, flanked by loxP sites, and was cloned upstream of exon 2 (Fig. 1A). A third loxP site was introduced downstream of exon 2. Deletion of exon 2 sequences, encoding the translation initiation site, the signal peptide, and part of the prodomain of furin, was expected to inactivate the fur gene. Following electroporation into E14 ES cells individual drug-resistant colonies were analyzed by Southern blotting using external cDNA probe 2 (corresponding to exons 7 and 8).
Of 180 drug-resistant colonies analyzed, 24 clones appeared to result from homologous recombination. PCR analysis showed that in 6 out of 24 clones, recombination at the 3Ј-end of the construct had occurred downstream of the third loxP site. Southern blot analysis of these six clones confirmed that the targeting had indeed occurred correctly (data not shown).
Since the presence of the floxed drug-resistance gene in intron 1 (fur floxϩHygB allele, Fig. 1, A and B) could have some deleterious effect on the expression of the floxed fur gene, this cassette was removed in vitro by partial loxP recombination. To achieve this, a Cre recombinase plasmid was transiently expressed in one of the ES cell lines. To obtain ES cell lines in which Cre recombination resulted in only partial deletion, the protocol as described (17) was applied. After choosing 100 individual clones and splitting into two pools, one of the pools was analyzed for HygB sensitivity. Out of 40 clones, which had regained HygB sensitivity, three showed the proper partial deletion of the HygB cassette (fur flox allele), whereas the others resulted from complete deletion (fur ⌬flox allele) (Fig. 1B). Fig.  2A shows a Southern blot analysis confirming the partial or complete deletion, which was further substantiated by PCR (data not shown).
Generation of fur flox/flox Mice and Conversion of the Floxed fur Allele (fur flox ) into a fur Null Allele (fur ⌬flox ) by Cre Recombinase-All three fur flox ES cell lines gave rise to germ line transmission after mating of chimeras with C57Bl/6J mice. Heterozygous (C57Bl/6J ϫ 129) F1 offspring were intercrossed, and viable offspring were genotyped by PCR. Of a progeny of 94 pups, 24 (26%) were wild type (fur wt/wt ), 43 (46%) were heterozygotes (fur flox/wt ), and 27 (29%) were homozygous for the floxed allele (fur flox/flox ). The homozygous fur flox/flox mice were, as expected, normal, viable, and fertile. Southern blot analysis was used to confirm the genotyping results as obtained by PCR analysis (Fig. 2B). Northern blot analysis of liver samples of littermates revealed similar fur mRNA expression levels for all three genotypes (data not shown).
To demonstrate that the floxed fur allele (fur flox ) could be converted into a fur null allele (fur ⌬flox ) via inactivation by Cre recombinase in vivo, homozygous fur flox/flox mice were crossed with a general deleter pGK-Cre transgenic mouse (18). All heterozygote fur flox/wt offspring carrying the pGK-Cre transgene showed the expected general deletion of fur exon 2 sequences as monitored by PCR and Southern blot analysis (Fig.  2B). These heterozygotes (fur ⌬flox/wt ) were intercrossed and viable offspring was genotyped by PCR. Only heterozygote (fur ⌬flox/wt ) (42 pups, 65%) and wild type (fur wt/wt ) (23 pups, 35%) offspring was identified. This ratio of about 2 to 1 is in accordance with the expectation, that the fur ⌬flox/flox genotype is embryonically lethal like the general fur knock-out mouse (14). It should be noted that these mice carried zero, one, or two copies of the pGK-Cre allele, as both the parental mice carried one copy of the pGK-Cre allele. However, since the null allele (fur ⌬flox ) is germ line-transmitted, it is independent of continued expression of Cre recombinase. To exclude any interference of the presence of the pGK-Cre transgene, these heterozygotes (fur ⌬flox/wt ) were crossed with wild type C57Bl/6J mice. Heterozygotes (fur ⌬flox/wt ) without the pGK-Cre transgene were identified by PCR analysis and intercrossed. Again no viable homozygous fur ⌬flox/⌬flox mice were born. To determine the timing of the developmental arrest, embryos from heterozygous intercross matings were analyzed macroscopically and genotyped by PCR. As described in detail for the classical fur knockout embryos (14), no homozygous fur ⌬flox/⌬flox mutant embryos were recovered from 11.5 days postcoitus onwards, although at this stage some deciduas contained remnants of resorbed tissue. At 9.5 and 10.5 days postcoitus about 25% of the embryos analyzed were homozygous fur ⌬flox/⌬flox mutant. Macroscopically these mutant embryos showed a similar phenotype as the classical fur knock-out embryos: defects in ventral closure and axial rotation (data not shown). The fur ⌬flox allele (deletion of exon 2 sequences) and the classical fur knock-out allele (targeted disruption of exon 4 sequences by a selection marker gene) can both be considered genuine fur null alleles.
Conditional Inactivation of the fur flox Allele in the Liver-To inactivate the fur flox allele conditionally, the fur flox/flox mice were crossed with Mx-Cre transgenic mice. In this transgenic line, Cre expression can be induced by pI-pC. This has been shown to result in complete deletion of reporter genes in liver; near-complete deletion in spleen; partial deletion in duodenum, heart, lung, uterus, thymus, and kidney; and inefficient deletion in muscle, tail, and brain (19). Offspring heterozygous for the floxed fur allele and the Mx-Cre transgene were subsequently intercrossed to obtain homozygous fur flox/flox mice carrying in addition one or two Mx-Cre alleles. After induction, Cre recombination of the floxed fur allele was analyzed in the liver and in brain tissue. In brain very little recombination could be detected, whereas near-complete inactivation of the floxed fur allele was observed in liver, consistent with previous results obtained with Mx-Cre mice (8 and 100%, respectively (19)). However, besides the inactivated fur ⌬flox allele and the non-recombined fur flox allele, a third kind of fur allele could be detected, albeit at very low amounts. Analysis with an exon 2 specific probe revealed that this third fur allele had arisen from integration of the excised fragment into the other allele in the same cell (data not shown). Obviously, as long as the excised circular DNA fragment remains present, it can participate in a second Cre recombination event. To avoid this problem of reintegration resulting in a third kind of fur allele with unknown impact, a different crossing scheme was designed. The Mx-Cre allele was crossed into mice heterozygous for the fur ⌬flox allele obtained initially by crossing with a pGK-Cre transgene (see above). Subsequently, mice heterozygous for both the fur ⌬flox allele and the Mx-Cre transgene were crossed with homozygous fur flox/flox mice (Fig. 3). Since one of the fur alleles is already inactivated, reintegration will just restore one of the two inactivated alleles into a fur flox allele with a zero net effect. The resulting offspring was used to inactivate the fur flox allele conditionally in the liver. As shown in Fig. 4, in fur ⌬flox/flox and fur ⌬flox/wt mice, heterozygous for the Mx-Cre allele, the fur flox allele was recombined into a fur ⌬flox allele in the liver with high efficiency, whereas in brain hardly any Cre recombination could be detected. From these analyses it can be concluded that in fur ⌬flox/flox mice, heterozygous for the Mx-Cre allele, the fur gene can be efficiently inactivated in liver upon induction of Cre expression. This was further substantiated by Northern blot analysis and quantitative PCR using RNA isolated after pI-pC induction from livers from fur ⌬flox/flox mice, heterozygous for the Mx-Cre allele (Fig. 5 and Table I). This results demonstrated that ϳ1% of normal levels of wild type fur mRNA was  A and B), and LPC did not reveal any major compensatory up-regulation in the fur-deficient livers (Fig. 5).
Histological Analysis of Furin-deficient Mouse Liver-Liver architecture was analyzed in four sets of mice (wild type induced and not induced, transgenic mice induced and not induced) by both routine stains of paraffin-and plastic-embedded tissues as well as by immunohistochemical and lectin histochemical analysis (Fig. 6). The latter was employed to screen for abnormalities in vascular and bile duct architecture and macrophage or leukocyte reactions that could point toward less overt pathological changes of liver tissue.
On routine stains, we occasionally observed a mild congestion of the liver as evidenced by distended blood vessels filled with packed erythrocytes. This inconstant effect was, however, linked to the pI-pC induction procedure and was independent of genotype.
Differences in liver lobe architecture or hepatocyte morphology were neither observed by routine hematoxylin and eosin stains on paraffin material nor by toluidine blue stains on Karnovsky-fixed 1-m resin sections. The architecture of blood vessels and bile ducts was analyzed by PNA and Dolichos bifloris lectin histochemistry and was found to be unaffected.
Likewise, a screen for leukocytes and macrophages by CD45 and F4/80 immunohistochemistry did not reveal any changes in distribution, density, or morphology of these cells. Finally, no increase of RCA-1 stainability of macrophages, which is an early marker of phagocyte activation upon pathologic stimuli, was recorded. In summary, no evidence was found for pathologic changes of the liver.
Biochemical Analysis of Potential Furin Substrates in Liver-To obtain insight into physiological furin substrates and potential redundancy, mice carrying zero or one functional fur allele in the liver, generated as described above, were analyzed. Three membrane-anchored and three soluble proteins, with expression levels ranging from low to high, were selected (Fig.  7). All candidate substrates have been suggested to be furin substrates on the basis of in vitro or ex vivo experiments.
The insulin receptor is a heterotetrameric cell surface protein composed of two ␣ and two ␤ chains of 135 and 90 kDa, respectively (22). Both chains are derived from a single precursor by cleavage at a R-P-S-R-K-R-R2S-L sequence. Endoproteolytic cleavage of the proinsulin receptor was studied using antibodies directed against epitopes in the ␣ chain. As is shown in Fig. 7, processing of the proinsulin receptor was unaffected by functional inactivation of furin, suggesting complete redundancy or that furin is not the physiological convertase.
Integrins are cell surface transmembrane glycoproteins that function as adhesion receptors linking the extracellular matrix proteins to the cytoskeleton. All integrins are heterodimers consisting of 1 of 18 ␣-subunits and 1 of 8 ␤-subunits. 9 of the 18 ␣-subunits are internally cleaved into a membrane-bound light chain and an amino-terminal heavy chain that remain associated. The ␣ 5 integrin precursor is cleaved at a H-H-L-Q-K-R2E-A sequence and is expressed at high levels in liver (24). Under steady state conditions, virtually all ␣ 5 integrin is in its processed form as demonstrated with an antibody directed against a carboxyl-terminal epitope, which detects mainly the processed 35-kDa light chain and only trace amounts of the 140-kDa precursor. However, in the absence of furin the majority remains unprocessed, indicating an important for role for furin in processing of ␣ 5 integrin. On the other hand, the presence of (sub)normal amounts of processed ␣ 5 integrin indicate that at least one other enzyme can cleave pro-␣ 5 integrin in liver.
The low density LRP is a multifunctional receptor with multiple ligands. The mature receptor is derived from a 600-kDa precursor by cleavage at a S-N-R-H-R-R2Q-I site to generate 515-and 85-kDa fragments that remain associated non-covalently (25). Using an antibody directed against the 85-kDa fragment, the processing of LRP was studied in the conditional furin knock-out mouse. Under steady state conditions, virtually all immunodetectable LRP in all samples was in the processed form indicating redundancy of furin. However, longer exposure (upper panel) revealed some high molecular weight precursor. Moreover, additional bands were observed in the furin-deficient liver membranes with slightly different electrophoretic mobilities (indicated with an asterisk). Although the nature of these bands was not further investigated, their persistent appearance in only the furin-deficient samples suggests that it is the consequence of a slightly altered maturation process.
The most abundant plasma protein is albumin, which is produced in the liver from its precursor proalbumin by cleavage at a R-G-V-F-R-R2E-A site (26). Since the propeptide is only FIG. 3. Crossing scheme of conditional fur knock-out mice. Schematic representation of the crossing scheme used for the analysis. Mice with the correct genotype were induced with pI-pC.

FIG. 4. Induction with pI-pC result in efficient recombination in liver but not in brain.
Mice of different genotypes were induced with pI-pC, and genomic DNA from liver and brain was digested with KpnI and analyzed by Southern blotting using genomic probe 3. Near-complete recombination of the floxed allele occurs in the liver after induction of Cre recombinase (compare middle panel with right panel). L ϭ liver; B ϭ brain. 5. Northern blot analysis of fur, PACE4, PC5, and LPC in transgenic mice. RNA was isolated from liver after pI-pC induction and used for Northern blotting. Wild type fur mRNA was detected using an exon two specific probe, whereas a larger probe was used to detect both wildtype mRNA and mRNA lacking exon 2 sequences. After hybridization with PACE4, PC5, and LPC, the blots were stripped and reprobed with actin for normalization. ⌬flox/flox indicates fur ⌬flox/flox mice without Mx-Cre allele, and ⌬flox/⌬flox indicates fur ⌬flox/flox mice containing one Cre allele, resulting in the knock-out genotype after induction. six amino acids, the precursor cannot be separated easily from the processed form by SDS-PAGE. The presence of three arginines in the propeptide, on the other hand, facilitates the separation by isoelectric focusing. Inactivation of both fur alleles resulted in an increase of proalbumin. It should be noted, however, that the vast majority was normally processed indicating substantial redundancy for processing of proalbumin.

FIG.
Two additional abundant plasma proteins synthesized in liver are ␣ 1 -microglobulin and bikunin, which are derived from the same precursor by cleavage at a I-A-R-A-R-R2A-V sequence (27). ␣ 1 -Microglobulin is an immunosuppressive protein that belongs to the lipocalin superfamily and bikunin is a member of the Kunitz-type protease inhibitor superfamily. Processing of the ␣ 1 -microglobulin/bikunin precursor was studied using an antibody directed against ␣ 1 -microglobulin. Inac-tivation of both fur alleles resulted in a strong reduction in processing, although mature ␣ 1 -microglobulin remained detectable in (sub)normal amounts.
Finally, the processing of vitronectin was studied. Vitronectin is an adhesive glycoprotein in the extracellular matrix and in plasma. It is composed of two chains of 65 and 10 kDa, which are generated by cleavage of the precursor at a R-R-S-S-R2S-I site (28). As is shown in Fig. 7, furin is largely redundant for the processing of vitronectin since ablation of active furin results in only a minor increase in precursor.
Taken together, these data show that in liver of adult mice there is (limited) redundancy of furin for every candidate substrate tested, although the level of redundancy for different substrates is highly variable.

DISCUSSION
In this study we have examined the inducible inactivation of the fur gene. No adverse effect was observed from treatment with pI-pC per se, with the exception of occasional mild congestion of liver. Despite the near-complete inactivation in liver, only a mild phenotype was observed. No morphological abnormalities could be found, and processing of various proproteins appeared to be unaffected or impaired, but never completely blocked. Accumulation of substantial amounts of precursors in mice carrying one intact fur allele was never observed. Altogether, these data indicate considerable redundancy of proprotein processing activity in liver. This biosynthetically very active organ expresses high levels of furin as well as other PCs such as PACE4, PC6, and LPC (29,30), which are likely to provide this processing redundancy. It is possible that the physiological role of furin in liver is underappreciated in this model due to overcapacity of processing activity provided by other PCs. Redundancy was not acquired by up-regulation of compensatory PCs. These results are in sharp contrast with the early embryonic lethality of the general fur knock-out (14) which indicate a non-redundant function during early embryogenesis.  It also contrasts with a number of processing studies performed in furin-deficient cell lines, which have suggested a crucial role for furin for many substrates. These discrepancies can be due to a number of reasons. First, in heterologous expression experiments misleading results can arise from the co-expression of substrates with convertases, which are normally never expressed in the same cell. In addition, redundancy may be cell type-dependent, as was for instance shown for PACAP, which is cleaved exclusively by PC4 in testis (10), whereas in brain PC4 is absent, and PACAP may be cleaved by PC1 or PC2 instead (31).
Processing of the insulin receptor by furin has been suggested by several studies, based on lack of processing in furindeficient cells (32) or biochemical studies using purified furin (33). The former study was later complemented by a study showing partial rescue by PACE4 (34). Our results show complete redundancy in the knock-out mice, although this does not exclude that in wild type mice furin is partly responsible for precursor cleavage. Normal processing was anticipated given the severe phenotype of the liver-specific insulin receptor knock-out and the requirement of processing for activation (35,36).
Processing of ␣ 5 integrin has been studied in furin-deficient cells and by overexpression studies, which indicated involvement of furin and PC6A but not PACE4 or LPC (24). Our results are consistent with these observations, although we cannot determine which enzyme(s) is (are) responsible for the redundancy.
Processing of LRP, which does not seem to be essential for its endocytic funtions (37), has been studied in furin-deficient cells that display severely impaired processing (25). These results contrast with our observation that virtually all LRP was processed under steady state conditions, and only traces of precursors could be detected. The small differences in electrophoretic mobility of the precursors in furin-deficient liver (indicated with asterisks in Fig. 7) might be the consequence of a slight changes in timing of processing relative to other posttranslational modifications such as complex glycosylation.
Proalbumin processing has been well studied in both human and in animal model systems. Naturally occurring human proalbumin genetic variants have been described in which mutations in the cleavage site result in impaired cleavage without significant clinical consequences (38). To identify the physiological processing enzyme(s) many different approaches have been taken implicating furin but also other enzymes like PACE4 and LPC (26,39,40). Here, we demonstrate that in the absence of furin, the majority of proalbumin is normally cleaved in mouse liver. The clear increase in proalbumin, on the other hand, indicates a significant involvement of furin under physiological conditions.
Processing of the precursor of ␣ 1 -microglobulin/bikunin has also been studied in furin-deficient cells without differences in degree of processing as compared with the parental cell line. This suggests processing by another PC, although furin was able to cleave the substrate in overexpression studies (27). Here we find an important role for furin in the processing of ␣ 1 -microglobulin/bikunin, although partial cleavage was still occurring in its absence.
Intracellular processing of vitronectin has only recently been demonstrated (28). In that study furin was put forward as a candidate for processing on the basis of in vitro cleavage. Although furin might be involved in this cleavage step, it clearly does not play an essential role as virtually all immunoreactive vitronectin in furin-deficient liver was cleaved to Vn65.
It is difficult to understand why there is complete redundancy of furin for some but not other substrates. It is clearly not related to expression levels as (near) complete redundancy was observed for both albumin, the most abundant secretory protein in liver, and the non-abundant insulin receptor. It is also not correlated with membrane association, since both ␣ 5 integrin and ␣ 1 -microglobulin/bikunin seemed, to a certain extent, to depend on furin for cleavage, whereas for LRP and albumin processing, furin is much less important. We have also analyzed the substrates using a recently described method for prediction of cleavage sites specific for furin or PCs in general, based on artificial neural networks (Table II) (41). Again, no correlation was observed. In contrast, the cleavage site of ␣ 5 integrin was predicted to be a poor furin site but a good general PC cleavage site. Albumin on the other hand was predicted to be a furin substrate but not a general PC substrate. These data show that the program is a powerful tool for predicting cleavage sites but not for predicting physiological PCs. The poor performance of this program in predicting the furin specificity revealed in this study probably reflects the influence of coexpression studies on the data set used to develop the program. Whether or not a substrate can be cleaved by furin in vivo is probably determined by factors other than linear sequence alone, such as subcellular localization, trafficking, and secondary structure exposure of the site (1,42). The indispensability of furin, as studied in this knock-out model, is in addition also dependent on coexpressed compensatory PCs.
The consequence of the observations made in this study for therapeutic targeting of furin is 2-fold. The mild phenotype of this inducible knock-out model might be indicative for a lack of severe side effects when applying furin inhibitors in the treatment of pathologies such as cancer, viral infections, or bacterial toxins (5,43). On the other hand, inhibition of furin should have a severe impact on processing of the target substrate(s) to exert a therapeutic effect. The feasibility of using furin as a therapeutic target has recently been demonstrated in mice injected with bacterial toxins (44,45). Inhibition of furin with hexa-D-arginine provided protection against an otherwise lethal dose of toxin. It remains to be established, however, whether furin is a useful therapeutic target in the treatment of human pathologies.