Analysis of Mammalian Peroxin Interactions Using a Non-transcription-based Bacterial Two-hybrid Assay*

In recent years, substantial progress has been made in the identification of proteins involved in peroxisome biogenesis. However, with the exception of the peroxisome-targeting signal receptors and the receptor docking proteins, the function of most of these proteins, called peroxins, remains largely unknown. One step toward elucidating the function of a protein is to identify its interacting partners. We have used a non-transcription-based bacterial two-hybrid system to analyze the interactions among a set of 12 mammalian peroxins and a yeast protein three-hybrid system to investigate whether proteins that interact with the same peroxin and have overlapping binding sites cooperate or compete for this site. Here we report a detailed interaction map of these peroxins and demonstrate that (i) farnesylation, and not the CAAX motif, of Pex19p strongly enhances its affinity for Pex13p; (ii) the CAAXmotif, and not farnesylation, of Pex19p strongly enhances its affinity for Pex11pβ; and (iii) the C3HC4 RING (really interesting new gene) finger domain of Pex12p does not alter the binding properties of Pex5p for the C-terminal peroxisome-targeting signal PTS1. Finally, we show that the Pex5p-Pex13p interaction is bridged by Pex14p and that the latter molecule exists predominantly as a dimer in vivo. Collectively, as demonstrated in this work with peroxins, these results indicate that the bacterial two-hybrid system is an attractive complementary approach to the conventional transcription-based yeast two-hybrid system for studying protein-protein interactions.

reported recently that in S. cerevisiae Pex11p plays a primary role in the medium-chain fatty acid oxidation pathway and suggested the alternative view that this pathway regulates the level of a signaling molecule that actually modulates peroxisome number and size.
Defining the physical interactions between peroxins is an essential step toward the understanding of how these proteins function and how proteins are translocated through the peroxisomal membrane. Currently, the best established strategy to find connections between peroxins is the transcriptionbased yeast two-hybrid system (y-2HS) (17). Strategies thus far used to identify or to confirm physical interactions between peroxins are co(immuno)precipitation assays, far western and surface plasmon resonance analyses, blue native polyacrylamide gel electrophoresis, a transcription-based mammalian two-hybrid assay, and the use of cleavable cross-linkers (5, 18 -21).
Here we report a detailed interaction map of 12 mammalian peroxins that are available currently in our laboratory. This peroxin linkage map was obtained by using the classical y-2HS, as well as a non-transcription-based bacterial twohybrid system (b-2HS). The latter system, developed by Ladant and co-workers (22), exploits the functional modularity of the catalytic domain of Bordetella pertussis adenylate cyclase. That is, proteins of interest are fused genetically to T18 and T25, two complementary fragments that are both essential for adenylate cyclase activity. If the corresponding fusion proteins interact, cAMP will be produced. In an Escherichia coli strain deficient in endogenous adenylate cyclase, this functional complementation can be monitored easily as several resident genes, such as the LacZ or Mal genes, are activated transcriptionally and give rise to a selectable phenotype. We show that the non-transcription-based b-2HS and the transcription-based y-2HS are complementary approaches and compare our data with existing published interactions. In addition, our results support the observation that farnesylation of Pex19p enhances strongly its affinity for Pex13p (23) and uncover the potential of human Pex14p to homodimerize in vivo and to bridge the Pex5p-Pex13p interaction in vitro. Finally, we have delineated the different protein interaction domains of Pex5p and Pex14p and show that Pex5p has partially overlapping but distinct binding sites for the C 3 HC 4 RING finger domain of Pex12p and PTS1 molecules.

EXPERIMENTAL PROCEDURES
Strains and Cell Lines-The E. coli strain Top10FЈ (Invitrogen) was used for all DNA manipulations, as well as for the expression of recombinant proteins. The adenylate cyclase-deficient E. coli strain BTH101 (Hybrigenics) was used for the bacterial two-hybrid assay. Yeast two-hybrid and protein three-hybrid assays were performed using the S. cerevisiae strain SFY526 (CLONTECH). Chinese hamster ovary (CHO) cells were cultured in MEM␣ medium (BioWhittaker) supplemented with 10% (v/v) fetal calf serum (Invitrogen), 2 mM GlutaMAX-I (Invitrogen), and antibiotic/antimycotic (100 g/ml peni-cillin G, 100 g/ml streptomycin sulfate, and 0.25 g/ml amphotericin B) solution (Invitrogen) in a humidified 37°C, 5% CO 2 incubator.
Plasmid Constructions-The cDNA fragments encoding the analyzed proteins were amplified by PCR with the appropriate primers (Invitrogen) and cloned into the yeast two-hybrid vectors pGBT9 or pGAD424 (CLONTECH), the yeast protein three-hybrid vector pBridge (CLONTECH), or the bacterial two-hybrid vectors pKT25 or pUT18C (Hybrigenics) using standard molecular biology procedures (24). PCRs were performed routinely using Pfx DNA polymerase (Invitrogen). To facilitate the cloning of cDNAs into the bacterial twohybrid vectors, the multiple cloning sites of pKT25 and pUT18C were modified. For pKT25, the PstI-BamHI fragment, which contains restriction sites for SalI and XbaI, was replaced by a DNA cassette (primer pair 5Ј-GGCTCGAGGCGGCCGCGG-3Ј/5Ј-GATCCCGCGG-CCGCCTCGAGCCTGCA-3Ј) containing unique restriction sites for XhoI and NotI. For pUT18C, the BamHI-EcoRI fragment, which contains restriction sites for SmaI, KpnI, and SacI, was replaced by a DNA cassette (primer pair 5Ј-GATCCGAATCCGGTACCGCGGCCGC-3Ј/5Ј-AATTGCGGCCGCGGTACCGGATTCG-3Ј) containing unique restriction sites for KpnI and NotI. The corresponding derivatives of pKT25 and pUT18C were termed pMF413 and pMF424, respectively. The detailed cloning procedures (including the list of oligonucleotides) of the extensive number of other constructs can be obtained from the corresponding authors. The identities of essential constructs were confirmed by DNA sequencing.
Two-hybrid and Protein Three-hybrid Analyses-The cotransformation of plasmids in competent SFY526 yeast cells and the colony lift ␤-galactosidase filter assay were performed as described by the manufacturer (CLONTECH). The interactions were indicated as strong (ϩ) or weak (Ϯ), depending on the colony staining times (Ͻ5 and Ͻ24 h, respectively). The liquid culture ␤-galactosidase assay with o-nitrophenyl-␤-D-galactopyranoside as substrate was also performed as described by the manufacturer (CLONTECH). However, for this assay the yeast cells were grown for 72 h in minimal dropout medium without leucine and tryptophan (y-2HS) or in minimal dropout medium without leucine and tryptophan in the absence or presence of 1 mM methionine (y-3HS). Bacterial two-hybrid assays were performed as described (25) using the E. coli strain BTH101 (Hybrigenics). Transformed bacterial cells were plated on indicator plates (either on Mac-Conkey medium (Difco Laboratories) supplemented with maltose (Acros Organics) or on Luria Bertani medium supplemented with 40 g/ml X-gal (MBI Fermentas) and 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside (MBI Fermentas)) containing 50 g/ml ampicillin (Roche Molecular Biochemicals) and 25 g/ml kanamycin (Fluka). The interactions were indicated as strong (ϩ) or weak (Ϯ) depending on the colony staining times (Ͻ28 and Ͻ52 h, respectively). Quantification of the functional complementation mediated by interaction between two proteins was done by measuring ␤-galactosidase activities in liquid cultures (25).
Cross-linking Experiments-90% confluent CHO cells, grown in 100-mm tissue culture dishes (Sarstedt), were washed twice with 10 ml of phosphate-buffered saline. One ml of phosphate-buffered saline, supplemented with 50 l of Me 2 SO containing different concentrations of dithiobis(succinimidyl propionate (DSP; Perbio)) (0 to 2.5 mM final concentration) was added to each plate. After incubation for 60 min at room temperature, the cells were harvested by scraping firmly with a stiff rubber policeman. The cell suspensions were transferred to a microfuge tube, and the proteins were precipitated by the addition of 100 l of 0.15% (w/v) deoxycholate and 100 l of 72% (v/v) trichloroacetic acid. After washing the pellets once with 1 ml of acetone, pellets were resuspended in 300 l of sample buffer (3% (w/v) SDS, 8.7% (v/v) glycerol, 0.004% (w/v) bromphenol blue, and 60 mM Tris/HCl, pH 6.8) in the absence or presence of 5% (v/v) 2-mercaptoethanol. The samples were boiled for 5 min, subjected to SDS-PAGE (ExcelGel SDS (gradient 8 -18); Amersham Biosciences), transferred to nitrocellulose, and analyzed with the appropriate antisera.
In Vitro Binding Assay-For the microtiter plate assay, 1 g of human serum albumin cross-linked to the peptide CSYHKHLK-PLQSKL (PTS1) (26) was coated (50 mM sodium carbonate buffer, pH 9.6) overnight onto a microtiter plate. After washing the wells five times with TBST, the wells were blocked with 5% (w/v) Protifar (Nutricia) in TBST for 1 h. Subsequently, each well was incubated with 200 l of bacterial lysate, prepared as described before (27), containing no recombinant protein, (His) 6 -Pex5p, (His) 6 -Pex14p, or (His) 6 -Pex13p/SH3 (18), as schematically indicated in Table II. After extensive washings with TBST, bound Pex5p, Pex14p, and Pex13p were detected photometrically using peroxin-specific rabbit antisera, antirabbit IgGs coupled to alkaline phosphatase, and p-nitrophenyl phosphate. The optical densities were measured at 405 nm (filter: 690 nm). Data were corrected by performing simultaneously control experiments using 1 g of human serum albumin cross-linked to the peptide CVHESYHKHLKPLQ (26) as immobilized ligand.

Analysis of Interactions between Full-length Peroxins
Using the Gal4p-based Yeast Two-hybrid System-As a first step toward the development of a reliable peroxin network, we performed a matrix experiment (28) in which 12 full-length mammalian peroxins were tested systematically for pairwise interactions in the Gal4p-based yeast two-hybrid system (Fig.  1). Because the BD-Pex5p and BD-Pex19p fusion proteins self-activated the LacZ reporter gene, 24 pairwise combinations were not tested. Of the remaining 120 combinations, seven protein pairs were found to interact; the PTS1 receptor, Pex5p, interacts with its docking protein, Pex14p, whereas Pex19p, a predominantly cytosolic protein that contains a CAAX farnesylation consensus sequence, binds the peroxisomal integral membrane proteins Pex3p, Pex10p, Pex11p␤, Pex12p, Pex13p, and Pex16p. These results confirm previous observations that these protein pairs interact in the y-2HS (23, 29 -31). However, six other interactions (Pex5p-Pex5p, Pex7p-Pex5pL, Pex12-Pex5p, Pex12p-Pex10p, Pex14p-Pex14p, and Pex14p-Pex19p) were not found. These peroxin pairs were demonstrated previously to interact using other techniques including blot overlay assays, sizing chromatography, electron microscopy, and immunoprecipitation assays (7,20,29,32,33).
Analysis of Interactions between Full-length Peroxins Using a Non-transcription-based Bacterial Two-hybrid System-Because two peroxins self-activated the LacZ reporter gene, and more than 40% of the previously reported interactions between full-length peroxins were not detected in the y-2HS, we changed our strategy and performed a new matrix experiment using a non-transcription-based b-2HS (22). In this system, which exploits a cAMP signaling cascade in E. coli cya cells (i.e. deficient in endogenous adenylate cyclase), none of the 12 peroxins autoactivated the lacZ or Mal reporter genes, and six peroxin pairs interacted efficiently and specifically in reciprocal combinations of the two-hybrid moieties (Fig. 1). Four of these interactions (Pex14p-Pex5p, Pex3p-Pex19p, Pe11p␤-Pex19p, and Pex16p-Pex19p) were also detected in the y-2HS. The two remaining interactions, Pex14p-Pex14p and Pex14p-Pex19p, were not found in the y-2HS. The fact that Pex14p and Pex19p bind other peroxins in the y-2HS ( Fig. 1) largely eliminates the possibility that these peroxins are expressed poorly or fail to be targeted to the yeast nucleus. However, it may be that, because of a different folding environment, distinct domains of human Pex14p are exposed in S. cerevisiae versus E. coli. Three interactions (Pex10p-Pex19p, Pex12p-Pex19p, and Pex13p-Pex19p) found in the y-2HS were not detected in the b-2HS. Again, this may be related to a different folding environment or to expression problems. Unfortunately, as we currently lack specific antibodies to T18, T25, Pex10p, and Pex12p, we were unable to verify the expression levels of Pex10p and Pex12p. Consequently, we have no conclusive evidence whether the Pex19p-Pex10p and the Pex19p-Pex12p interactions occur in the b-THS. However, using an antiserum raised against (His) 6 -HsPex13p/SH3 (23), we were able to demonstrate that Pex13p was expressed correctly (Fig. 2). Therefore, in combination with the observation that the CAAX prenylation motif of Pex19p strongly enhances its affinity for Pex13p (23), we hypothesize that the observed lack of interaction between these two peroxins is caused by the fact that Pex19p is not prenylated in E. coli. Interestingly, Pex13p V164E , a mutant displaying a 20-fold enhanced affinity for Pex19p over the wild-type molecule in the y-2HS (23), does display a detectable binding to Pex19p in the b-2HS (Table I). Concerning the Pex19p-Pex11p␤ interaction, we found that the CAAX motif, and not prenylation, is an important determinant in the affinity of Pex19p for Pex11p␤. Indeed, Pex19p and Pex11p␤ were found to interact in the y-2HS and in the b-2HS, and deleting the prenylation motif from Pex19p (Pex19p⌬CAAX) resulted, in both systems, in a strongly reduced binding affinity for Pex11p␤ (see Ref. 23 and Table I). In this context, it is important to mention that Pex19p and Pex19p⌬CAAX are expressed similarly (Fig. 2) and display equivalent affinities for Pex3p, Pex14p, and Pex16p (Table I).
Two-hybrid Analysis of Interactions between Peroxin Domains-Although the systematic two-hybrid analysis of 12 full-length mammalian peroxins recapitulated ϳ70% of the interactions known from previous studies, a substantial number of reported interactions between full-length peroxins (Pex5p-Pex5p, Pex5pL-Pex7p, Pex5p-Pex12p, and Pex10p-Pex12p) could not be detected. However, as the expression of full-length mammalian peroxins in yeast and bacteria may result in variable misfolding, we investigated whether additional interactions could be detected using peroxin domains instead of full-length open reading frames. Using this approach, we were able to confirm that the C 3 HC 4 domain of Pex12p, but not the full-length molecule, interacts with Pex5p and Pex10p in the y-2HS (29, 31, 34) (Fig. 1). Evidence for the Pex5p-Pex5p (20,35) and Pex5pL-Pex7p (7) interactions could not be found. However, using the domain approach, we could confirm that the SH3 domain of Pex13p binds to Pex14p (Fig. 1), an interaction observed previously in a blot overlay assay (18). Interestingly, this domain of Pex13p was  Exponentially growing bacterial cells (strain HB101), transformed with plasmids encoding one of the indicated T18 fusion proteins (T18-hybrid) and T25 fused to either Pex19p or Pex19p⌬CaaX, were assayed for ␤-galactosidase using o-nitrophenyl-␤-D-galactopyranoside as substrate (46). Formed o-nitrophenol was measured at 420 nm and normalized for culture densities (optical density at 600 nm ϭ 1) and time (1 h). The values given are the mean (Ϯ the S.D.) of at least three measurements performed on independent single colonies and corrected for the blank (T18 and T25 only, 3.3 Ϯ 0.2). When the corresponding double transformants where grown on MacConkey/ maltose agar plates, they appeared as white (not underlined) or red (underlined) colonies. Pex13p V164E represents a Pex13p mutant in which the valine amino acid residue at position 164 is altered to glutamate (23 also found to interact weakly with itself, at least in the y-2HS (Fig. 1). Finally, after using the b-2HS to delineate the domains of Pex14p responsible for homodimerization and Pex19p binding (see below), we could show that the domains comprising amino acids 138 -200 and 1-81, but not the full-length Pex14p molecule, also interact with Pex14p and Pex19p, respectively, in the y-2HS (Fig. 1).
Pex14p Exists Predominantly as a Dimer in Vivo-Based on (i) our two-hybrid results that suggest that Pex14p, but not Pex5p, may homodimerize in vivo; and (ii) other reports suggesting that Pex5p functions as a monomeric (36), a homodimeric (35) or a homotetrameric protein (20), we further investigated the actual oligomeric state of these peroxins in mammalian cells. To stabilize protein-protein interactions, we performed chemical cross-linking experiments on CHO cells using the thiol-cleavable, homobifunctional, and amine-reactive cross-linker DSP. Immunoblot analysis of the cross-linked products after SDS-PAGE of the samples revealed that Pex14p, but not Pex5p, exists mainly as a multimeric complex in vivo (Fig. 3). Several lines of evidence suggest that this complex is in fact a homodimer of Pex14p. (i) Pex14p can homodimerize when it is expressed in E. coli (Fig. 1); (ii) the cross-linked Pex14p-containing complex of CHO cells migrates during SDS-PAGE to a similar distance as the dimerized fraction of bacterially expressed (His) 6 -Pex14p (data not shown); and (iii) the cross-linked Pex14p-containing complex is not recognized by antisera raised against the Pex14pinteracting peroxins Pex5p, Pex13p, and Pex19p (data not shown). Interestingly, even in the absence of cross-linker a small but significant fraction of Pex14p migrated as a dimer on SDS-PAGE under non-reducing conditions (Fig. 3). However, deletion analysis studies revealed that the dimeric Pex14p structure is not maintained by disulfide bonding between the subunits but by its coiled-coil structure (see below). The ease of disulfide bridge formation and its specificity support further the view of a very close, non-covalent homotypic interaction in vivo (37). Whether dimerization mutants of Pex14p (i) have an altered affinity for the other interacting peroxins, and (ii) are able to complement peroxisome biogenesis in a cell line deficient in Pex14p remains to be investigated.
The Pex5p-Pex13p Interaction Is Bridged by Pex14p-In view of published reports that the SH3 domain of yeast Pex13p functions as a peroxisomal membrane receptor for Pex5p (38 -40), our inability to detect an interaction between the corresponding human orthologues in the bacterial and yeast two-hybrid systems may seem surprising (Fig. 1). However, in vitro reconstitution experiments of a putative peroxisomal protein import complex consisting of PTS1 ligand, Pex5p, Pex14p, and the SH3 domain of Pex13p (amino acids 236 to 403) demonstrated that Pex14p, but not the SH3 domain of Pex13p, interacts with Pex5p loaded with PTS1 cargo (Table II). In addition, these experiments show that Pex14p can bind simultaneously to Pex5p and Pex13p/SH3

TABLE II In vitro reconstitution of a Pex5p-containing peroxisomal protein import subcomplex
Microtiter plate wells, coated with 1 g of human serum albumin cross-linked to the peptide CSYHKHLKPLQSKL (PTS1), were exposed in three subsequent incubations to bacterial lysates containing no recombinant protein (Ϫ), (His) 6 -Pex5p, (His) 6 -Pex14p, and (His) 6 -Pex13p/SH3 as indicated schematically. After extensive washings, bound Pex5p, Pex14p, and Pex13p were detected photometrically using peroxin-specific rabbit antisera and anti-rabbit IgGs coupled to alkaline phosphatase and p-nitrophenyl phosphate. To substract the background of the anti-Pex13p antiserum due to a weak cross-reactivity with human serum albumin, data were corrected by performing simultaneously control experiments using one microgram of human serum albumin cross-linked to the peptide CVHESYHKHLKPLQ as immobilized ligand. The experiments were performed in triplicate. The positive interactions are underlined.  (Table II). Summarized, these results suggest that in mammals (i) Pex14p, and not Pex13p, functions as the docking protein for cargo-loaded Pex5p, and (ii) the Pex5p-Pex13p interaction is bridged by Pex14p (Table II).
Mapping the Interaction Domains of Pex5p and Pex14p-Pex5p and Pex14p interact with each other and a number of other proteins. We employed the yeast and bacterial twohybrid assays to further delineate the specific domains involved in the binding of these peroxins to their respective partners. For Pex5p, we mapped the site required for PTS1 and Pex12p interaction to the tetratricopeptide repeat domains labeled 2-8 and 2-9, respectively (Fig. 4). In addition, we found that Pex5p has multiple binding sites for Pex14p, and these binding sites are localized in the N-terminal portion of the molecule (Fig. 4). Similar observations were published by Schliebs et al. (20). In addition, these authors proposed that the di-aromatic WXXX(F/Y) pentapeptide repeat motifs in the N-terminal portion of Pex5p are the determinants for the interaction with Pex14p. To verify this hypothesis, we altered in the deletion mutant Pex5p (222-602) , which still interacts with Pex14p while containing only one di-aromatic pentapeptide motif, the amino acid residue tryptophan to glycine at position 271. The observation that the binding affinity of Pex5p (222-602)(W271G) for Pex14p, but not for PTS1 and Pex12p  , decreases nearly to background levels (Table III), confirms that the di-aromatic peptide repeats are indeed important for Pex14p binding. With respect to Pex14p, we have conducted a series of experiments to map the Pex5p-binding site, the Pex19p-binding site, and the domain that is capable of self-association (Fig. 5). Surprisingly, the Pex14p-binding sites for Pex5p and Pex19p could not be separated physically and encompass the amino acids 22 to 81 (Fig. 5). The putative coiled-coil region was sufficient and required for homodimerization (Fig. 5).

The C 3 HC 4 RING Finger Domain of Pex12p Has No Regulatory Role in the PTS1 Binding Properties of Pex5p-
The fact that (i) the Pex12p (275-359) -binding domain and the PTS1binding domain of Pex5p are overlapping (Fig. 4), and (ii) Pex12p acts downstream of the Pex5p-receptor docking event (11), suggests that Pex12p may have a regulatory role in the PTS1 binding properties of Pex5p. To investigate whether this is indeed the case, we employed y-3HS. In this system, which can be seen as an extension of the y-2HS, the effect of a third protein (in our case HA-tagged Pex12p  or HA-tagged PTS1) on the binding properties of two other (interacting) proteins (in our case the Gal4p DNA-BD-PTS1 fusion protein and the Gal4p activation domain (AD)-Pex5p Exponentially growing cells (yeast strain SFY526), transformed with plasmids encoding one of the bait proteins (PTS1, Pex12p (275-359) , or Pex14p, fused to the Gal4p DNA-binding domain) and one of the prey proteins (Pex5p (222-602) , Pex5p (222-602)(W271G) , or Pex5p (332-602) , fused to the Gal4p activation domain), were assayed for ␤-galactosidase using o-nitrophenyl-␤-D-galactopyranoside as substrate (see "Experimental Procedures"). The optical densities were measured at 420 nm, corrected for the blank ("empty" plasmids), and normalized for culture densities (optical density at 600 nm ϭ 10) and time (24 h). The values given are the mean (Ϯ the S.D.) of at least three measurements performed on cultures derived from independent single yeast colonies.

Bait
Optical density  4. Mapping of the interaction domains of Pex5p. Pex5p deletion mutants, fused to the Gal4p AD (y-2HS) or T18 (b-2HS), were tested for interaction with PTS1, Pex12p (275-359) , and Pex14p, fused to the Gal4p DNA-BD (y-2HS) or T25 (b-2HS), in the yeast (y) or bacterial (b) two-hybrid system. Double transformants expressing one of the bait (BD and T25, respectively) and one of the prey (AD and T18, respectively) fusion proteins were selected, and ␤-galactosidase activity was monitored as described under "Experimental Procedures." The binding affinities of the tested protein pairs are indicated as follows: strong interaction (ϩ), weak interaction (Ϯ), or no interaction (Ϫ). The numbers on the left indicate the amino acid residues present in the corresponding Pex5p deletion mutants, and X represents the alteration of the tryptophan residue to glycine at position 271. The tetratricopeptide repeat motifs are shaded in gray, whereas the WXXX(F/Y) repeats are indicated by vertical arrows. The delineated interaction domains are indicated above the schematic representation of the full-length Pex5p molecule. fusion protein) can be investigated. Notice that in the y-3HS we used, the expression levels of the third protein could be regulated by the conditional Met25 methionine promoter (41). As shown in Table IV, the PTS1 binding properties of Pex5p were not altered when the HA-tagged Pex12p (275-359) molecule was expressed at different levels (Fig. 2). On the other hand, HA-tagged PTS1 did compete with BD-PTS1 for Pex5p binding (Table IV). These results, combined with the observation that Pex12p (275-359) interacts with Pex5p in the y-2HS (see Refs. 29 and 34, and see Fig. 4), suggest that Pex5p has distinct binding sites for the C 3 HC 4 RING finger domain of Pex12p and PTS1 proteins and that Pex5p and Pex5p-Pex12p (275-359) have similar PTS1 binding properties. DISCUSSION To decipher the peroxisomal protein import process, it is essential to identify the physical interactions between peroxins. Currently, the y-2HS is the most commonly used tool to dissect these interactions. However, as this assay may yield false positive results (42)(43)(44) and false negative results (44,45), it is important that identified interactions (or non-interactions) are confirmed by at least one alternative, non-yeastbased assay. Techniques used frequently to confirm yeast two-hybrid interactions include blot overlay assays, sizing chromatography, and co(immuno)precipitation assays. However, these in vitro assays are often laborious, require optimized experimental conditions, and sometimes require specialized reagents (e.g. purified proteins, 35 S-labeled proteins, or highly specific antibodies of sufficient titer). As an alternative to these in vitro assays, the b-2HS was evaluated (22,46). This in vivo interaction system differs basically from the y-2HS in that (i) a bacterial cell, and not a yeast cell, is used as host organism, and (ii) it is based on the reconstitution of a cAMP-dependent signal transduction pathway and not on the reconstitution of a functional transcriptional activator. Hence, there is no need for nuclear localization signals on the fusion proteins.
As a first step to validate the b-2HS, we performed a matrix experiment in which 12 full-length mammalian peroxins were tested systematically for pairwise interactions and compared these results with data obtained from similar yeast two-hybrid experiments. This validation study yielded 9 different interacting peroxin pairs; four pairs (Pex14p-Pex5p, Pex19p-Pex3p, Pex19p-Pex16p, and Pex19p-Pex11p␤) were found to interact in both two-hybrid systems, three pairs (Pex19p-Pex10p, Pex19p-Pex12p, and Pex19p-Pex13p) were only detected in  Yeast cells (strain SFY526), transformed with plasmids coding for a Gal4p DNA-binding domain-PTS1 fusion protein, a Gal4p activation domain-Pex5p fusion protein, and a third (conditionally expressed from the Met25 promotor) HA-tagged fusion protein (PTS1 or Pex12p (275-359) , were grown for 72 h in minimal dropout medium without leucine and tryptophan, in the absence (Ϫ) or presence (ϩ) or 1 nM methionine. The cells were assayed for ␤-galactosidase using o-nitrophenyl-␤-D-galactopyranoside as the substrate (see "Experimental Procedures"). The optical densities were measured at 420 nm and normalized for culture densities (optical density at 600 nm ϭ 10) and time (24 h). The values given are the mean (Ϯ the S.D.) of three measurements performed on each of four separate transformants. Notice that the HA-tagged fusion protein is repressed in the presence of 1 mM methionine and expressed in the absence of methionine.  (18,19,30) and that Pex19p, a molecule required for peroxisome membrane synthesis, binds to a broad spectrum of peroxisomal membrane proteins (12). With respect to the interactions that were observed only in the y-2HS, it is essential to notice that it has been reported that the CAAX prenylation motif of Pex19p is an important determinant in the affinity of Pex19p for Pex10p, Pex11p␤, Pex12p, and Pex13p (23). In combination with the fact that protein prenylation is a typical eukaryotic post-translational modification process (47), our data suggest that prenylation, and not the CAAX motif itself, is an important determinant in the affinity of Pex19p for Pex10p, Pex12p, and Pex13p. However, with respect to the Pex19p-Pex11p␤ interaction, we provide evidence that the CAAX motif, and not prenylation, of Pex19p seems to be important. Whether the CAAX tetrapeptide is required directly for binding to Pex11p␤, or whether the corresponding deletion changes the folding of Pex19p in such a manner that binding to distinct sites is affected, is not currently clear. With respect to the interactions that were only detected in the b-2HS, it has to be mentioned that, depending on the host organism, full-length proteins might prevent the identification of several interactions because of various intrinsic folding problems (45,48). For example, the finding that the coiled-coil structure, but not the full-length version, of Pex14p interacts with itself in the y-2HS supports this hypothesis. Protein subdomains may be revealed by conformational changes (e.g. caused by post-translational modifications and/or interactions with other molecules) in the holomolecule. Furthermore, these newly exposed domains may mediate interactions not necessarily found with the parent protein. As an example of this phenomenon, we tested systematically Pex12p (275-359) and Pex13p (236 -403) to uncover additional interactions. Importantly, this approach yielded one novel (Pex13p (236 -403) -Pex13p) and three known (Pex5p-Pex12p (275-359) , Pex10p-Pex12p (275-359) , and Pex13p (236 -403) -Pex14p) peroxin pairs (18,29,34).
Two reported peroxin pairs, Pex5p-Pex5p (20, 35) and Pex5pL-Pex7p (35), could not be detected in the yeast and bacterial two-hybrid systems. However, we have currently indirect evidence that human Pex7p is not properly folded when it is expressed in yeast or bacteria. 2 When expressed in a mammalian expression system, Pex7p does interact with Pex5pL (49).
In another series of experiments, we employed the bacterial and yeast two-hybrid assays to delineate the specific interaction domains of Pex5p and Pex14p that are involved in the binding to their respective partners. With respect to Pex5p, we obtained experimental evidence for the hypothesis that the WXXX(F/Y) repeat motifs of this molecule are important determinants for Pex14p binding (20), and showed that the Pex12p-binding domain and the PTS1-binding domain overlap physically. While this work was in preparation, similar conclusions were reached by Dodt et al. (49) and Saidowsky et al. (50). However, by using a y-3HS, we could also show that PTS1 ligands and Pex12p (275-359) do not compete for the same binding site on Pex5p. With respect to Pex14p, we also found that the Pex14p-binding sites for Pex5p and Pex19p could not be separated physically. Unfortunately, as Pex19p did not interact with Pex14p in the y-2HS, we were not able to perform a yeast protein three-hybrid assay to investigate whether Pex19p is able to alter the Pex5p binding properties of Pex14p.
The experiments described above demonstrate that the bacterial and yeast two-hybrid systems are complementary approaches for the analysis of protein interactions. In addition, the b-2HS has, compared with the y-2HS, a lower false positive rate. That is, acidic proteins with amphipathic ␣-helices like Pex5p and Pex19p, which in the y-2HS act frequently as transcriptional activators when they are fused to the Gal4p DNA-binding domain (51), do not self-activate the LacZ and Mal reporter genes in the b-2HS. This finding underscores one of the benefits of the non-transcription-based two-hybrid systems; specifically, the spatial separation of the association between the hybrid proteins and the transcriptional activation readout.
To investigate the actual oligomeric state of Pex5p and Pex14p, we performed an in vivo cross-linking approach on CHO cells. These experiments provided physical evidence that Pex14p exists predominantly as a homodimer in its native configuration. In addition, our results also suggest that Pex5p exists predominantly as a monomer in mammalian cells. Although these results are in agreement with our two-hybrid data, they do not exclude the possibility that Pex5p can transiently form oligomers. In this context, it is interesting to note that purified, bacterially expressed (His) 6 -tagged Pex5p does form homotetramers (20). However, the fact that the same molecule tends to form filamentous aggregates in low salt buffer (20) indicates that these in vitro experiments have to be interpreted with care.
Finally, as we did not observe an interaction between Pex5p and Pex13p in both two-hybrid systems, we investigated further the role of the SH3 domain of Pex13p in the Pex5preceptor docking event. Therefore, we developed an in vitro binding assay in which a putative peroxisomal protein import complex consisting of PTS1 ligand, Pex5p, Pex14p, and Pex13p/SH3 was reconstituted. These experiments revealed that Pex14p, and not Pex13p, functions as the initial docking site for cargo-loaded Pex5p. These findings are supported further by the fact that Pex5p accumulates on the peroxiso-mal membrane in mammalian cells that are deficient in Pex13p or overexpress Pex14p and remains cytosolic in cells deficient in Pex14p (35).