Sorting of Furin at the Trans-Golgi Network

The eukaryotic subtilisin-like endoprotease furin is found predominantly in the trans-Golgi network (TGN) and cycles between this compartment, the cell surface, and the endosomes. There is experimental evidence for endocytosis from the plasma membrane and transport from endosomes to the TGN, but direct exit from the TGN to endosomes via clathrin-coated vesicles has only been discussed but not directly shown so far. Here we present data showing that expression of furin promotes the first step of clathrin-coat assembly at the TGN, the recruitment of the Golgi-specific assembly protein AP-1 on Golgi membranes. Further, we report that furin indeed is present in isolated clathrin-coated vesicles. Packaging into clathrin-coated vesicles requires signal components in the furin cytoplasmic domain which can be recognized by AP-1 assembly proteins. We found that besides depending on the phosphorylation state of a casein kinase II site, interaction of the furin tail with AP-1 and its μ1subunit is mediated by a tyrosine motif and to less extent by a leucine-isoleucine signal, whereas a monophenylalanine motif is only involved in binding to the intact AP-1 complex. This study implies that high affinity interaction of AP-1 or μ1 with the cytoplasmic tail of furin needs a complex interplay of signal components rather than one distinct signal.

The trans-Golgi network (TGN) 1 constitutes the sorting station where soluble and membrane proteins, targeted to different post-Golgi compartments, are packed into distinct carrier vesicles. Three TGN exit routes are known: constitutive transport to the cell surface, transport to secretory granules in cells with a regulated secretory pathway, and transport to endosomes (for review, see Ref. 1). The last pathway is followed primarily by lysosomal enzymes bound to the mannose 6-phos-phate receptors (MPRs) by their mannose 6-phosphate residues (for review, see Ref. 2). Vesicles leaving the TGN for subsequent transport to the endosomes are clathrin-coated. Formation of these clathrin-coated vesicles (CCVs) involves binding of AP-1 Golgi-specific assembly proteins to the TGN membrane. AP-1 binding is regulated by the ADP-ribosylation factor ARF-1, a small GTP-binding protein (3,4), and requires the presence of transmembrane proteins, among which MPRs are the major constituents (5)(6)(7). CCVs are transport intermediates of vesicular traffic not only from the TGN to endosomes, but also from the plasma membrane to endosomes. Golgi-and plasma membrane-derived CCVs can be distinguished by their different sets of assembly proteins; AP-1 complexes are restricted to coated buds and vesicles of the TGN, whereas AP-2 complexes act in endocytosis at the plasma membrane. Both adaptor complexes are heterotetrameric. The AP-1 complex is composed of two 100-kDa subunits, ␥ and ␤1 adaptin, a medium subunit 1 (47 kDa), and a small subunit 1 (19 kDa). The AP-2 complex is of similar size and composition with two large subunits, ␣ and ␤2 adaptin (100 kDa), in association with a 2 (50 kDa) and a 2 subunit (17 kDa) (for review, see Ref. 8). A major function of both adaptor complexes is to promote clathrin cage assembly onto the respective membrane via their ␤ subunits (9). A second important function of AP-1 and AP-2 is the binding to cytoplasmic domains of membrane proteins (for review, see Ref. 10) by interaction with their tyrosine-based motifs (11)(12)(13)(14)(15) or dileucine-based sorting signals (13, 16 -19). Both motifs are known to be important for endocytosis at the plasma membrane as well as targeting to various endosomal compartments, to lysosomes, and to the basolateral plasma membrane in polarized cells (for review see Refs. 20 and 21). With the yeast two-hybrid system and a number of in vitro assays it was shown that different types of tyrosine-based sorting signals (15,(22)(23)(24) and also dileucine motifs (25,26) can be recognized by the chains of AP-1 and AP-2.
Furin is a membrane-associated, subtilisin-like eukaryotic endoprotease that proteolytically cleaves a large number of proproteins C-terminally at the consensus sequence RXK/RR. Among these substrates are membrane-bound and secretory proteins such as prohormones, proproteins of receptors, plasma proteins, growth factors, and also bacterial toxins and many surface glycoproteins of enveloped viruses (27; for review, see Refs. 28 and 29). Furin is concentrated in the TGN and cycles between this compartment, the plasma membrane, and endosomes. The responsible routing and TGN localization signals reside in the cytoplasmic domain of furin (30 -33). The sorting signals include the tyrosine-based signal YKGL 765 , which mediates endocytosis, and an acidic cluster CPSDSEEDEG 783 , which is required for TGN localization (33)(34)(35)(36). Deletion anal-ysis and mutagenesis of the furin tail revealed a leucine-isoleucine (LI 760 ) sorting signal and a critical phenylalanine (F 790 ), which separately mediate internalization of chimeric furin proteins. 2 The two serine residues of the acidic cluster are phosphorylated by casein kinase II (CKII) in vivo and in vitro. Phosphorylation is assumed to regulate trafficking of furin in endosomes (35,36). Recently, a cytosolic connector protein, PACS-1 (phosphofurin acidic cluster-sorting proteins), was identified, which directs transport of furin from the endosomes back to the TGN by binding to the phosphorylated acidic cluster and connecting the endoprotease to the clathrin sorting machinery (38). In the regulated secretory pathway of neuroendocrine cells, retrieval of furin from clathrin-coated immature secretory granules depends on the interaction of AP-1 with CKII-phosphorylated furin tail (39). However, the routing of furin at the exit of the TGN and especially the motifs required at this sorting station are not well understood. In this study we show that furin is able to promote AP-1 recruitment, the first step of clathrin coat assembly at the TGN, and that furin is present in isolated CCVs. Using in vitro binding and peptide competition assays, we have examined in detail the interaction of the furin cytoplasmic tail sorting signals with the 1 subunit and the whole AP-1 complex.

EXPERIMENTAL PROCEDURES
Raising an Antiserum against the Cytoplasmic Domain of Bovine Furin-A cDNA fragment coding for the furin tail (amino acids 741-797) was amplified by polymerase chain reaction from pSG5:bfur (33) and ligated into the EcoRI-HindIII sites of the Escherichia coli expression vector pMAL TM -c2 (New England Biolabs, Germany). The plasmid was transfected into the E. coli TB1 strain. The furin tail containing the maltose-binding protein was purified by affinity chromatography according to the protocol provided by Biolabs (Germany), dialyzed against PBS, and used in 200-g portions for the immunization of rabbits by four injections at 4-week intervals. The serum was tested by enzymelinked immunosorbent assay, Western blotting, immunoprecipitation, and immunofluorescence.
Recombinant DNA Methods-The 1 subunit of TGN adaptor AP-1 (AP47) was cloned by polymerase chain reaction from human lung gt10 cDNA library (CLONTECH, USA). The primers were derived from the mouse sequence (M62419) (40): forward primer, 5Ј-CCTTC-CAAAGCCGCCATGTCCGCCAGCGCCGTC-3Ј; reverse primer, 5Ј-CA-GAGGCCTCTCACTGGGTCCGGAGCTGATAATC-3Ј. The polymerase chain reaction product was ligated in the TA vector (Invitrogen) and sequenced by the dideoxy chain method using the ABI PRISM TM Dye Terminator Cycle Sequencing Ready Reaction Kit (Perkin-Elmer). After excision by EcoRI, the cDNA of the 1chain was inserted into the multiple cloning site of pBluescript vector (Stratagene). For generation of the different furin tail mutants the pSG5:bfur construct, described by Schä fer et al. (33), was used. Substitutions in the cytoplasmic tail were introduced by a polymerase chain reaction-based approach using pSG5: bfur as a cDNA template and synthetic oligonucleotides that contained the desired mutations within their sequence. Generation of the deletion mutants was done as described by Schä fer et al. (33). The sequences of all of the tail mutants were verified by dideoxy sequencing.
GST Fusion Protein Production-The cDNA coding for the cytoplasmic domains of wild type furin and furin tail mutants was ligated in-frame to the COOH terminus of GST using the pGEX-5X-1 vector (Pharmacia, Germany). The fusion proteins were expressed in E. coli strain BL21, and protein purification was carried out according to the manufacturer's instructions (Pharmacia, Germany).
Preparation of Bovine Brain Cytosol-Bovine brain cytosol was prepared as described (41). Briefly, bovine brains were homogenized in buffer A (0.1 M MES, 0.5 mM MgCl 2 , 1 mM EGTA, pH 7.0) containing 0.1 mM phenylmethylsulfonyl fluoride and 0.2 mM dithiothreitol. Nuclei, membranes, and other sedimental components were removed by centrifugation at 8,000 ϫ g for 50 min at 4°C followed by centrifugation at 130,000 ϫ g for 1 h at 4°C. The protein concentration of the cytosol supernatant was determined by protein assay (Pierce Chemical Co.) using BSA as a standard.
Isolation of CCVs-CCVs were isolated from MPR-deficient mouse fibroblasts (42) stably expressing furin by a standard procedure (43,7). After three washes with PBS containing 0.1 mM CaCl 2 and 1 mM MgCl 2 (PBS-Ca/Mg), cells were scraped and spun for 10 min at 2,000 ϫ g. The cells were resuspended in vesicle buffer (140 mM sucrose, 75 mM potassium acetate, 10 mM MES, pH 6.7, 1 mM EGTA, 0.5 mM magnesium acetate) containing 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, and protease inhibitor mixture (Boehringer Mannheim) and homogenized. From a postnuclear supernatant, a postmitochondrial supernatant was prepared, which was fractionated by two gradient centrifugations using a 10-ml continuous gradient of 10 -90% (w/v) 2 H 2 O (Sigma) in vesicle buffer and then a 10-ml continuous gradient from 2% (w/v) Ficoll, 9% (w/v) 2 H 2 O to 20% Ficoll, 90% 2 H 2 O in vesicle buffer. After centrifugation of the second gradient to equilibrium, 1-ml fractions were collected from the bottom, and aliquots of each fraction were precipitated by trichloroacetic acid (10% final concentration) for Western blotting analysis.
Generation of MPR-deficient Mouse Fibroblasts Stably Expressing Furin--The bovine furin subcloned into the pSG5 vector was stably expressed in MPR-negative fibroblasts as described (18). Briefly, a calcium phosphate precipitate was made containing 20 g of the linearized vector and 1 g of pBhygr, a plasmid containing the hygromycin resistance gene. The cells were incubated with the precipitate for 24 h, split, and plated in the presence of 350 g/ml hygromycin B (Boehringer Mannheim). After 12-14 days of selection, colonies were picked, expended, and tested for expression of furin by immunofluorescence.
Cell Culture and Transfection-HeLa cells or immortalized MPRdeficient mouse fibroblasts (42) and MPR-deficient mouse fibroblasts stably expressing bovine furin were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 units/ml penicillin, and 0.1 mg/ml streptomycin. For transient expression, HeLa cells were transfected with pSG5:bfur constructs by the calcium phosphate method (44). At 7 h post-transfection protein expression was stimulated by the addition of sodium butyrate (5 mM final concentration) to the culture medium.
Binding of AP-1 to Golgi Membranes-The in vitro binding of AP-1 was performed as described (5,6,17), except for using a different permeabilization technique (45). MPR-negative mouse fibroblasts stably expressing furin and MPR-negative mouse fibroblasts as control cells were cocultured on coverslips for 24 h to 90% confluence. Cells were washed twice with cold PBS-Ca/Mg and KOAc buffer (25 mM Hepes/KOH, pH 7.0, 111.5 mM potassium acetate, 2.5 mM MgCl 2 ), permeabilized by immersion in ice-cold KOAc buffer containing 40 g/ml digitonin for 4 min, and then washed twice with cold KOAc buffer again. For AP-1 binding the permeabilized cells were incubated with 200 l of bovine brain cytosol (9 mg of protein/ml in KOAc buffer) at 37°C for 10 min and replaced on ice for two washes with cold KOAc buffer. Then the cells were subjected to immunofluorescence analysis.
MPR-deficient mouse fibroblasts were fixed with 3% paraformaldehyde in PBS for 15 min, quenched with 50 mM NH 4 Cl in PBS, and then incubated with primary and secondary antibodies, diluted in PBS containing 1% BSA, as described for HeLa cells.
The intensity of the perinuclear ␥ adaptin fluorescence signal was quantitated on the single cell level. Digitized images were recorded using a laser scanning microscope (Carl Zeiss, Oberkochen, Germany) working with the blue line (408 nm wavelength) of an argon laser. For each sample, images of about 100 furin-expressing cells and as control a similar number of furin negative cells were recorded and saved as TIFF files. The intensity of perinuclear staining for ␥ adaptin was quantitated using the PC BAS software (Raytest, Germany).
In Vitro Binding Assays-The 1 chain was translated in vitro in the presence of 35 S-labeled methionine (Amersham, Germany) using the pBluescript 1 DNA as template and a coupled in vitro transcription translation kit (Promega). In vitro translated samples were centrifuged at 15,000 ϫ g for 10 min to remove insoluble materials. Glutathione-Sepharose 4B beads (Pharmacia Biotech, Sweden) loaded with 30 g of wild type or mutated GST-furin fusion protein were incubated with or without CKII (25 units/l) (Biolabs, Germany) and 2 mM ATP in 30 l of CKII buffer (20 mM Tris-HCl, 50 mM KCl, 10 mM MgCl 2 , pH 7.5) for 1 h at 30°C. Then precleared in vitro translated 1 was added followed by incubation for 2 h at 4°C in 500 l of binding buffer (0.05% Nonidet P-40, 50 mM Hepes, pH 7.3, 10% glycerol, 0.1% BSA, 200 mM NaCl). Sample buffer was added to washed beads, and the eluted material was subjected to SDS-PAGE. The radiolabeled bands were detected by fluorography. Quantitation was done by using a BioImager (Raytest, Germany).
For binding of AP-1 to immobilized furin tails 30 g of GST-furin fusion protein were adsorbed to glutathione-Sepharose 4B and incubated with or without CKII as described above. The immobilized fusion proteins were suspended in 400 l of bovine brain cytosol (6 mg of protein/ ml in buffer A) containing 1 mM phenylmethylsulfonyl fluoride and protease inhibitor mixture and incubated for 1 h at 4°C. Sepharose beads were washed three times with cold buffer A (see above) and then resuspended in sample buffer. The bound material was separated by SDS-PAGE and subjected to immunoblot analysis with ␥ adaptin-specific monoclonal antibodies.
In Vitro Competition Assay-Oligopeptides homologous to the furin tail were synthesized by using an ABI 432A peptide synthesizer (Applied Biosystems, Inc.). In vitro translated, 35 S-labeled 1 chain was preincubated in 100 l of binding buffer in the presence or absence of peptides for 15 min on ice. Then immobilized GST-furin fusion protein was added and incubated for 2 h at 4°C. For AP-1 competition, 100 l of bovine brain cytosol (1 mg of protein/ml in buffer A) was preincubated with or without peptides for 15 min on ice followed by incubation with immobilized GST-furin fusion protein for 1 h at 4°C. Detection of bound AP-1 or 1 was done as described above.
Western Blotting-Electrophoretic transfer of proteins to nitrocellulose membranes was done by standard procedure (46). Incubations with primary antibodies diluted in PBS containing 0.1% Tween 20 were performed for 1 h at room temperature. Primary antibodies were detected using horseradish peroxidase-labeled anti-mouse antibody from rabbit (Dako, Denmark), and the secondary antibodies were detected using the Super Signal system (Pierce). When blots were probed sequentially with different antibodies, the membranes were stripped in 62.5 mM Tris-HCl, pH 6.8, 2% SDS, and 100 mM ␤-mercaptoethanol for 30 min at 50°C after each immunodetection. Quantitation of the signals was done with a BioImager.

RESULTS
Furin Recruits AP-1 onto Golgi Membranes-Proteins like the MPRs, which are transported from the TGN to the endosomes via CCVs, have been shown to promote AP-1 recruitment onto Golgi membranes (5). To prove sorting of furin along the TGN/endosomal pathway we investigated first whether furin also promotes AP-1 translocation. For this purpose, wild type furin (wt) and furin tail mutants lacking the tyrosine-motif (dYKGL) or the cytoplasmic domain (d743-797) or one that contains a nonphosphorylatable acidic signal (S776A/S778A) (Fig. 1A) were expressed in HeLa cells. To discriminate adaptin recruited by furin from adaptin recruited by other Golgi resident proteins, furin expression was stimulated by the addition Protein expression was stimulated by the addition of sodium butyrate to the culture medium. Cells were permeabilized, fixed, and double labeled with an antiserum raised against the cytoplasmic tail of furin (a, c, e, g) and with ␥ adaptin-specific monoclonal antibody 100/3 (b, d, f, and h) and immunostained. Panel C, the intensity of the perinuclear ␥ adaptin fluorescence signal was quantitated for about 100 transfected and a similar number of nontransfected cells. The ratio of these two fluorescence intensities was taken as the measure of increase in adaptin signal. Three independent experiments were performed. Error bars represent the S.D. of sodium butyrate. Then the cells were double labeled with a polyclonal anti-furin antibody to discriminate transfected from nontransfected cells, and with a monoclonal antibody directed against ␥ adaptin to identify the AP-1 complex. Quantitation of AP-1 recruitment by measuring the fluorescence associated with the Golgi region (Fig. 1C) and qualitative analysis of AP-1 recruitment (Fig. 1B, a-h) have shown that overexpression of wild type furin correlates with an increase of ␥ adaptin staining in the perinuclear region. Cells expressing the furin mutants dYKGL and S776A/S778A show only a slight increase in a perinuclear ␥ adaptin signal, whereas the tail-less furin mutant (d743-797), which was also found partly in the perinuclear region because of its overexpression, remains without any significant effect on adaptin binding. As a control, highly expressed influenza virus hemagglutinin, a protein known to follow the constitutive secretory pathway toward the plasma membrane, showed no specific increase in the perinuclear ␥ adaptin signal (data not shown).
In a second assay we investigated recruitment of AP-1 by furin at a lower and thus more physiological level of furin expression. We took advantage of the MPR-negative fibroblasts generated and described by Ludwig et al. (42). These cells exhibit a low capacity for recruiting AP-1, which can be restored by reexpressing physiological levels of the MPRs (6). MPR-deficient mouse fibroblasts stably expressing furin and MPR-deficient mouse fibroblasts as control were cocultured, permeabilized with digitonin, and incubated with bovine brain cytosol. Recruitment of bovine AP-1 was examined by indirect immunofluorescence using a polyclonal anti-furin antibody and monoclonal antibody 100/3 against ␥ adaptin. As shown in Fig.  2, cells expressing furin exhibit a markedly enhanced perinuclear ␥ adaptin signal compared with nontransfected fibroblasts (Fig. 2, a and b). When bovine brain cytosol was omitted from the incubation, no AP-1 was detected because the antibody does not recognize mouse adaptin (Fig. 2, c and d). Quantitation of the ␥ adaptin fluorescence signal associated with the Golgi region revealed a 40 -55% increase in adaptin binding for furin expressing MPR-deficient fibroblasts compared with furin-negative control cells.
These results imply that furin, which is predominantly localized in the TGN, is able to recruit AP-1. Recruitment seems to be dependent on the presence of YKGL 765 and the phosphorylated acidic motif CPSDSEEDEG 783 .
Furin Localizes to CCVs-The observed recruitment of AP-1 on membranes of the Golgi region by furin is consistent with the sorting of furin into CCVs at the TGN. However AP-1 binding does not necessarily prove that furin is also sorted into CCVs in vivo. Therefore, using a standard fractionation protocol (7,43), CCVs were prepared from the MPR-deficient mouse fibroblasts (42) stably expressing furin. The last density gradient of the purification procedure was analyzed by Western blotting. Fig. 3 shows the typical distribution of different marker proteins throughout this gradient. As described (7), the dense fractions contained clathrin heavy chain, ␥ adaptin, and ␣ adaptin, whereas ␤-COP, a subunit of the coatomer chosen as a marker for vesicles of the early secretory pathway, was detected in the lighter fractions of the gradient. We detected furin in the same dense fractions also containing the marker proteins for CCVs. Compared with CCVs isolated from bovine brain by electron microscopy, the dense fractions contained the same spherical structures of 50 -100 nm in diameter (not shown). The results demonstrate that furin is sorted into CCVs in vivo.
The Cytoplasmic Tail Signals YKGL 765 , LI 760 , F 790 , and Clustered Acidic Amino Acids Are Required for Furin Interaction with AP-1 Golgi-specific Assembly Proteins in Vitro-Interaction of the cytoplasmic domain of different proteins with adaptor complexes and their medium () chains has been shown. Most of these interactions are mediated by tyrosinebased motifs and also leucine-based motifs. As shown by others (39) the TGN adaptor AP-1 binds to the cytoplasmic tail of furin in vitro dependent on the phosphorylation state of a CKII site within an acidic cluster. Our recruitment experiments described above show that furin expression promotes AP-1 binding to Golgi membranes. They indicate further that the phosphorylated motif CPSDSEEDEG 783 and in addition the tyrosine-based motif YKGL 765 are determinants for AP-1 binding in vivo. It has been shown previously that, besides the tyrosine and the CKII acidic motif (33)(34)(35)(36), two other determinants, a leucine-isoleucine sorting signal LI 760 and a monophenylalanine motif F 790 (see Footnote 2) are critical for correct intracellular sorting of furin. Therefore we tested the role of these four signals for interaction with AP-1 or its 1 component  in vitro. A set of furin tail mutants was produced substituted in the tyrosine, the LI, or the phenylalanine motif, and also in the cluster of acidic amino acids. These mutants have been analyzed in the following experiments.
AP-1 Binding to the Furin Tail Requires a Tyrosine-based Motif, a LI Sorting Signal, and, in Addition, a Monophenylalanine Motif-We tested whether the AP-1 adaptor complex can bind to the cytoplasmic tail of furin and furin tail mutants (Fig.  4A). GST-furin fusion proteins (GST-F) containing the cytoplasmic tail of wild type (wt) furin or mutated furin were used for this purpose. GST-F were immobilized on glutathione-Sepharose and incubated either with or without CKII to obtain phosphorylated or nonphosphorylated versions of the furin tail. Then bovine brain cytosol was added as the source of adaptor complexes, and the amount of AP-1 bound was measured by immunoblotting with the bovine-specific anti ␥ adaptin antibody 100/3. Confirming the results of Dittié and co-workers (39), nonphosphorylated GST-Fwt bound low amounts of ␥ adaptin, but after phosphorylation by CKII, binding increased (Fig. 4B). Several new findings obtained from our furin tail mutants now demonstrate that the situation is more complex (Fig. 4B). Mutants AKGA and LI/AN, substituted in the ty- Substituted amino acids are marked by arrowheads. Panel B, GST, GST-Fwt, and GST-furin tail mutants were immobilized on glutathione-Sepharose 4B beads, phosphorylated (ϩ) or not (Ϫ) with CKII, and then incubated with bovine brain cytosol. After several washes of the beads the bound material was separated by SDS-PAGE, and the amount of AP-1 bound was detected by immunoblotting with ␥ adaptin-specific monoclonal antibody 100/3 and quantitated by bioimager analysis. Panel C, GST and GST-F loaded glutathione beads were incubated with in vitro translated and [ 35 S]methionine labeled 1 chain, and the bound material was separated by SDS-PAGE and quantitated by bioimager analysis. The amount of bound 1 chain and AP-1, respectively, is expressed as a percentage of binding to CKII phosphorylated wild type furin tail (GST-Fwt-P). Images of representative experiments are shown in panels B and C. At least three independent experiments were performed. Error bars represent the S.E. rosine-based motif YKGL 765 or the leucine-isoleucine signal LI 760 , respectively, showed a significant decrease in ␥ adaptin binding compared with wild type furin. Surprisingly, mutation of the phenylalanine 790 (mutant F/N) resulted in a total loss of AP-1 interaction. Consistently, furin tail mutants substituted in more than one motif (mutants AKGA F/N, LI/AN F/N, LI/AN Y/A, LI/AN Y/A F/N) were also incapable of interacting with AP-1 adaptor complexes. Mutant LI/AN is the only mutant in which the binding affinity for AP-1 is still increased by phosphorylation, but only to about 60% of wild type. The results of this experiment demonstrate that a tyrosine-based motif, a LI sorting motif, and a single phenylalanine residue determine interaction of AP-1 with the cytoplasmic tail of furin. The phosphorylated acidic motif is important for high affinity binding but is functional only in the presence of an intact tyrosine motif and phenylalanine 790. 1 Binding to the Furin Tail Requires a Tyrosine-based Motif and a LI Sorting Motif-Next we tested binding of the 1 chain of AP-1 to the furin tail. Again we probed GST-F containing the cytoplasmic tail of wild type and mutated furin (Fig. 4A) in the phosphorylated or nonphosphorylated version. In vitro translated, [ 35 S]methionine-labeled 1 chain was added, and the amount of bound 1 was measured by autoradiography. As shown in Fig. 4C the medium chain 1 bound to GST-Fwt but not to GST alone. Used as a negative control, in vitro translated influenza virus NS1 did not interact with GST-Fwt (data not shown). Further, furin tail lacking either the tyrosine motif or the LI motif, mutants AKGA and LI/AN, bound 3-fold or 2-fold less efficiently 1 than wild type. When the phenylalanine 790 had been mutated to asparagine (mutant F/N) interaction with 1 was as efficient as for the furin wild type. This is in contrast to the AP-1 binding behavior with the F/N mutant (see Fig. 4B). The double mutants AKGA F/N and LI/AN F/N interacted with 1 like the single mutants AKGA and LI/AN, also indicating that substitution of phenylalanine 790 does not influence the binding of 1 chain. After substitution of both the tyrosinebased motif and the LI motif (mutant LI/AN Y/A) and as expected, when all three motifs were destroyed (mutant LI/AN Y/A F/N), practically no interaction was measured. As already found for AP-1 binding, we observed that phosphorylation of wild type and mutated furin resulted in an about 2-fold increase in 1 binding. To ensure that increased 1 binding to the furin tail is really caused by phosphorylation, we dephosphorylated CKII-treated GST-Fwt with alkaline phosphatase. After phosphatase treatment the 1 binding was reduced again, showing that stronger interaction attributes to CKII phosphorylation (data not shown). From these experiments it was apparent that the tyrosine-based motif and the LI sorting motif determine interaction of the furin tail with the medium chain 1. Phosphorylation of Ser 776 and Ser 778 by CKII enhances this interaction.
The Number of Negative Charges in the Acidic Cluster Is Crucial for the Affinity of AP-1 Binding-The effect of CKII phosphorylation on AP-1 binding raises the question of whether phosphates at a distinct site or an overall negative charge is critical for the enhancement of AP-1 binding affinity. Furin is phoshorylated by CKII both in vivo and in vitro on Ser 776 and Ser 778 , and exchange of both serine residues to aspartic acid mimics the diphosphorylated state of furin (35,39). As expected, such a GST-furin mutant (mutant S776D/ S778D, Fig. 5A) bound ␥ adaptin as efficiently as CKII-phosphorylated GST-Fwt (Fig. 5B). Therefore we tested whether introduction of negative charges in the form of aspartic acid also increases the affinity for AP-1 binding when they are placed at different positions in the furin tail. Mutant D798/799, a GST-F in which two aspartate residues were added to the carboxyl terminus of the furin tail, interacted less efficiently with ␥ adaptin than the mutant S776D/S778D but has a moderately higher AP-1 interaction than nonphosphorylated wild type. Interestingly, the R784D/G785D mutant, in which Arg 784 and Gly 785 were mutated to aspartic acid, bound to AP-1 as much as mutant S776D/S778D (see Fig. 5, A and B). These observations imply that for efficient AP-1 binding it makes no difference whether the negative charges are introduced by phosphorylation of Ser 776 and Ser 778 within the acidic sequence or by additional acidic amino acids in this region. However, negative charges added far from the acidic sequence have much less influence on ␥ adaptin binding. Therefore high affinity AP-1 binding seems to be dependent on the number of negative charges in or near the acidic cluster.
Furin Tail Peptides Specifically Compete for Furin AP-1 and 1 Interaction-Although the specificity of AP-1 and 1 binding to the furin tail were apparent from the effect of the single amino acid substitutions, we confirmed the specificity and the requirement of the different motifs by peptide competition assays. Therefore we tested the ability of furin tail homologous peptides (Fig. 6A) to inhibit the interaction of AP-1 or 1 with GST-Fwt (see Fig. 4A) in our in vitro binding assays. After a preincubation of bovine brain cytosol or in vitro translated 1 with increasing amounts of competing peptides, we determined the amount of AP-1 or 1 that was still able to bind to GST-Fwt. The competition curves in Fig. 6, B and C, show that the phosphorylated 41-amino acid tail peptide Arg 757 -Leu 797 inhibits the binding of ␥ adaptin and 1 to GST-Fwt most efficiently. Inhibition was dose-dependent and occurred at micromolar concentrations of peptide. As expected, inhibition of 1 binding to GST-F S776D/S778D, imitating the diphosphorylated state of furin, by the peptide Arg 757 -Leu 797 was much weaker (data not shown). To narrow down the sequence required for the competitive inhibition of adaptor binding, a short furin tail peptide Asp 756 -Pro 767 containing only the tyrosine motif and the LI sorting motif was tested. The peptide Asp 756 -Pro 767 inhibited binding of AP-1 and its 1 component, but to less extent than the 41-amino acid tail peptide. On the other hand, no competition was observed with two short peptides Asp 756 -Pro 767 AKGA and Asp 756 -Pro 767 LI/AN, substituted either in the tyrosine motif or the LI sorting motif. This underlines the specificity and cooperation of both motifs for ␥ adaptin and 1 binding. Taken together, the peptide competition assays have shown that the tyrosine motif and the LI sorting motif are critical determinants for competition of furin interaction with AP-1 and 1.

DISCUSSION
Furin is found predominantly in the TGN at steady state and appears to cycle between this compartment, the plasma membrane, and the endosomes. Experimental evidence for recycling has been obtained from immunofluorescence-based assays indicating that furin is endocytosed from the plasma membrane and travels from endosomes to the TGN (30 -33). Furin can enter the regulated pathway of secretory cells but is subsequently removed from immature secretory granules by AP-1 containing CCVs (39). However, the direct exit from the TGN to endosomes in nonsecretory cells has only been discussed (33,34,38), but not directly shown so far. This pathway would probably involve AP-1-containing CCVs budding from the TGN. Furthermore, as a specific rather than a default pathway of furin transport, it may require signal components in the furin cytoplasmic domain which mediate the packaging into these CCVs. This prompted us, first, to see whether there was evidence for AP-1/clathrin-dependent sorting of furin at the TGN and to analyze then in detail the binding sites in the cytoplasmic domain required for the interaction with AP-1 assembly proteins.
In this study we report that furin expression promotes the translocation of AP-1 assembly proteins onto Golgi membranes of HeLa cells and MPR-deficient mouse fibroblasts. AP-1 recruitment has been shown for the MPRs (5-7), which follow the TGN/endosomal pathway (47) and determine the amount of membrane-bound AP-1 and the number of TGN-derived CCVs (7). Minor components in transit through the TGN, the major histocompatibility complex class II and the varicella-zoster virus glycoprotein I, have also been demonstrated to trigger AP-1 recruitment on Golgi membranes, suggesting that they can potentially be packed in the same TGN-derived CCVs as the MPRs (17,48). Of course, AP-1 recruitment does not necessarily prove the sorting into CCVs in vivo. However, we show here that furin is present in CCVs isolated from furin-expressing MPR-deficient mouse fibroblasts. The applied method for CCV isolation does not permit distinguishing CCVs derived from the plasma membrane from those derived from the TGN. But together with the finding that furin also promotes AP-1 recruitment in the MPR-deficient fibroblasts, it is unlikely that furin is only a component of plasma membrane-derived CCVs. Thus, our results now strongly suggest that furin is packaged in AP-1-containing CCVs derived from the TGN. This notion is also supported by previous morphological studies suggesting that furin localizes to clathrin-coated regions of the TGN (33,34).
The AP-1 recruitment experiments suggested that the tyrosine motif and the phosphorylated acidic signal play a role in AP-1/furin binding in vivo. For exact analysis of the requirement of furin cytoplasmic tail sorting signals YKGL 765 , LI 760 , F 790 , and phosphorylated acidic cluster CPSDSEEDEG 783 for interaction with AP-1 and its 1 subunit we made use of an in vitro binding assay. Tyrosine or dileucine motifs in the cytoplasmic tails of membrane proteins have been reported to interact with AP-1 and AP-2 in similar experiments (49,14,13). The chains of adaptor complexes were also shown to recognize a variety of tyrosine-based motifs (15,(22)(23)(24) or dileucine motifs (25,26) in the yeast two-hybrid system and a number of in vitro assays. On the other hand, dileucine motifs have been reported to interact with ␤1 adaptin (37); but, until now little was known about adaptor binding to proteins like furin, which contain multiple sorting components in their cytoplasmic tail. It was therefore interesting to investigate if each of the motifs would be able to contribute to AP-1 binding, and, if so, whether they would function independently or in combination. Dittié and co-workers (39) showed in vitro that AP-1 binding to the cytoplasmic domain of furin depends on CKII phosphorylation of the acidic cluster. We demonstrate here that the situation for furin tail interaction with AP-1 and its 1 subunit is more complex.
Our results show, first, that furin/AP-1 and 1 association strongly depends on the tyrosine motif YKGL 765 and to less extent on the leucine-isoleucine signal LI 760 . The LI motif may therefore be different from classical dileucine motifs that have been shown to interact with ␤1 adaptin (52). In fact, the leucine-isoleucine motif LI 760 is very close to the tyrosine motif YKGL 765 and could well be part of it. Further evidence for the requirement of YKGL 765 and LI 760 for AP-1 and 1 binding is given by peptide competition. As also found by others (14,25,12,23), inhibition required a relatively large molar excess of soluble peptides. This is probably because only a minor fraction of the free peptides has the proper conformation, or this may reflect differences in the affinity of 1 and AP-1 for soluble versus immobilized peptides. It is noteworthy that association of the furin-derived tyrosine motif ISYKGL and 1 was not observed in the yeast two-hybrid system (23), thus demonstrating the importance of analyzing the binding components in the context of the whole cytoplasmic domain.
Second, we found that the monophenylalanine motif F 790 is only involved in binding to the intact AP-1 complex. Sorting signals that contain phenylalanine are rarely described, and their participation in adaptor binding has not been reported previously. Substitution of phenylalanine 790 does not affect the interaction between 1 and the furin tail. The reason for this discrepancy is unclear at present. Two hypotheses could be envisaged. First, it is possible that 1 is involved in binding F 790 , but only when present in the context of the complete AP-1 complex. An alternative explanation could be that F 790 binds to a different subunit on the AP-1 complex. It has been suggested that the ␤ subunits may be involved in the recognition of receptor tails, based on studies in which the asialoglycoprotein receptor (50) and the epidermal growth factor receptor (51) were found to bind to ␤ adaptin. In addition, Rapoport and co-workers (37) reported recently binding of dileucine motifs to ␤1 adaptin.
Further, our data confirm that phosphorylation of the two CKII sites within the acidic cluster is important for high affinity AP-1 binding to the furin tail (39), but we clearly demonstrate that phosphorylation enhances binding only in the presence of an intact tyrosine and phenylalanine motif. We show that phosphorylation also correlates with an increase in 1 binding. Neither the phosphorylated nor the nonphosphorylated furin tail is capable of binding to 1 after the loss of YKGL 765 and LI 760 . These data suggest that phosphorylation of the acidic cluster modulates the interaction between furin and AP-1/1 rather than providing a specific binding site. The two phosphate groups may function directly in interaction with AP-1 or indirectly by regulating the accessibility of sorting motifs, for example YKGL 765 or LI 760 . Alternatively, it is possible that phosphorylation of the CKII sites in the cluster of acidic amino acids increases the affinity for AP-1 by providing additional negative charges. It has been demonstrated that exchange of both serine residues, Ser 776 and Ser 778 , for aspartic acid mimics the diphosphorylated state of furin (35,39). We show now that for efficient AP-1 binding it is irrelevant whether negative charges are introduced by substitution of Ser 776 and Ser 778 to aspartic acid or by adding acidic amino acid residues to other positions within the acidic cluster. Our results argue that high affinity AP-1 binding is dependent on the number of negative charges within the acidic sequence but not on the overall charge of the furin tail. Mauxion et al. (18) reported for the cation-dependent MPR that the entire CKII phosphorylation site in the cytoplasmic domain is a dominant TGN sorting determinant for efficient packaging of the receptor in Golgi-derived vesicles. Similar to our observations these authors found that replacement of three negatively charged amino acids surrounding the serine at position 57 by alanine residues impaired AP-1 binding to membranes. On the other hand, replacement of the positively charged arginine at posi-tion ϩ3 of the serine 57 in the cation-dependent MPR tail by an aspartic acid resulted in an increased affinity of AP-1 for membranes.
Recently, Wan and co-workers (38) identified a new cytosolic connector protein, PACS-1, which binds specifically to the phosphorylated acidic cluster and connects furin to the AP-1/ clathrin sorting machinery for subsequent transport from the endosomes back to the TGN. Our in vitro data suggest that there is also a direct interaction of the furin cytoplasmic domain with AP-1 and 1. The interaction requires combination of the furin tail sorting signals YKGL 765 , LI 760 , F 790 , and phosphorylated acidic cluster rather than one distinct signal. However, it is worth mentioning that the bovine brain cytosol used in our study as source of AP-1 complexes might contain PACS-1, which thus mediates the furin-AP-1 interaction by binding to the phosphorylated acidic cluster. The in vitro translated samples could also contain some PACS-1 derived from the reticulocyte lysate of the translation kit. But, if so, we would expect that destruction of YKGL 765 , LI 760 , or F 790 would not affect AP-1 or 1 binding. Therefore, two different mechanisms of furin/AP-1 association may exist: a direct interaction mediated by combination of the different furin tail sorting signals at the TGN and a PACS-1-mediated interaction dependent on phosphorylation of the two CKII sites at endosomes.
In summary, our studies have shown that besides the phosphorylation state of a CKII site, interaction of the furin tail with AP-1 or 1 depends on a tyrosine motif and to less extent on a leucine-isoleucine signal, whereas a phenylalanine motif (F 790 ) is only involved in binding to the intact AP-1 complex. Whether in vivo all signals are utilized in TGN sorting of furin and whether additional cytosolic proteins are involved in furin-AP-1 interaction at the exit of the TGN remain to be determined. However, the results imply that high affinity interaction of AP-1 with the cytoplasmic tail of furin needs a complex interplay of signal components rather than one distinct signal.