A Region from the Medium Chain Adaptor Subunit (μ) Recognizes Leucine- and Tyrosine-based Sorting Signals*

Tyrosine-based sorting signals in the cytosolic tails of membrane proteins have been found to bind directly to the medium chain subunit (μ) of the adaptor complexes AP-1 and AP-2. For the leucine-based signals, an interaction with AP-1 and AP-2 has been reported, but no specific interacting subunit has been demonstrated. After searching for molecules interacting with the leucine-based sorting signals within the cytosolic tail of the major histocompatibility complex class II-associated invariant chain using a phage display approach, we identified phage clones with homology to a conserved region of the AP-1 and AP-2 μ chains. To investigate the relevance of these findings, we have expressed regions of mouse μ1 and μ2 chains on phage gene product III and investigated the binding to tail sequences from various transmembrane proteins with known endosomal targeting signals. Enzyme-linked immunosorbent binding assays showed that these phages specifically recognized peptides containing functional leucine- and tyrosine-based sorting signals, suggesting that these regions of the μ1 and μ2 chains interact with both types of sorting signals.

Tyrosine-based sorting signals in the cytosolic tails of membrane proteins have been found to bind directly to the medium chain subunit () of the adaptor complexes AP-1 and AP-2. For the leucine-based signals, an interaction with AP-1 and AP-2 has been reported, but no specific interacting subunit has been demonstrated. After searching for molecules interacting with the leucinebased sorting signals within the cytosolic tail of the major histocompatibility complex class II-associated invariant chain using a phage display approach, we identified phage clones with homology to a conserved region of the AP-1 and AP-2 chains. To investigate the relevance of these findings, we have expressed regions of mouse 1 and 2 chains on phage gene product III and investigated the binding to tail sequences from various transmembrane proteins with known endosomal targeting signals. Enzyme-linked immunosorbent binding assays showed that these phages specifically recognized peptides containing functional leucine-and tyrosinebased sorting signals, suggesting that these regions of the 1 and 2 chains interact with both types of sorting signals.
The cytosolic tails of membrane proteins contain the information for targeting to various intracellular destinations. Two main classes of endosomal sorting signals have been identified: one class characterized by an essential tyrosine and one by a double leucine (LL) motif or variants thereof (1,2). There is increasing evidence that both classes of sorting signals interact directly with the adaptor complexes AP-1 and AP-2 for clathrindependent sorting from the trans-Golgi network (TGN) 1 or the plasma membrane, respectively (3)(4)(5)(6)(7)(8)(9)(10)(11). Studies on adaptor complexes binding to tail columns indicate a multivalent attachment of aggregated complexes with a binding constant in the micromolar range (3,4,7). This is confirmed by recent data obtained by surface plasmon resonance (9), suggesting a model where binding to the signals is associated with clustering of the adaptors. In fact, adaptor complexes tend to aggregate and drive the formation of coats in vitro (12), and a requirement for AP clustering is in agreement with the current model where adaptors are first recruited to the membrane by a docking protein and then associate to form a coat to where the membrane proteins diffuse laterally.
A specific interacting subunit of the AP complexes was recently demonstrated for the tyrosine-based signals as they were shown to interact with both 1 and 2 , the medium chain subunits of AP-1 and AP-2, respectively (10,13). No such interaction was demonstrated for the leucine class signals in those assays, although there are data indicating that these also mediate binding to the AP complexes (9).
The latest information about how signals may be recognized by the sorting machinery was obtained using the yeast twohybrid system (13), a library-based method for studying protein-protein interactions. We have used phage display technology as an alternative library-based approach to obtain more information about protein sequences recognizing leucine-based signals. Phage display of random peptides has in several cases been successfully applied to obtain detailed information about protein interactions (14). When using a 15-mer synthetic peptide encompassing the LI sorting motif of the major histocompatibility complex class II-associated invariant chain (Ii) (15)(16)(17) as a target for a random phage display library, we obtained phage clones with homology to a conserved region of the medium chain adaptor subunits (). In this report, we investigated these findings by expressing the corresponding domains at the phage surface and found that they are able to discriminate between wild-type and mutant leucine-and tyrosine-based signals.

EXPERIMENTAL PROCEDURES
Phage Library and Bacteria-The 10-mer fUSE5 library with a complexity of 2 ϫ 10 7 primary phage clones has been described previously (18). The fUSE5 and fUSE2 phages and the Escherichia coli bacterial strains K91K and MC1061 were generous gifts from G. P. Smith (University of Missouri).
Enrichment of fUSE5 Phage Binding to an Invariant Chain Tail Peptide-The screening procedure was modified from that of Smith and Scott (19) and has been discussed in detail elsewhere (20). Briefly, the invariant chain tail peptide 1 MDDQRDLISNNEQL 14 K was biotinylated before binding to SoftLink avidin resin (Promega), a methacrylate polymeric matrix with covalently bound monomeric avidin. One round of panning was performed as follows. The SoftLink-bound peptide was blocked for 1 h in PBS, pH 7.4, containing 1% BSA before incubation with the fUSE5 phage for 1 h at room temperature. After extensive washing in PBS containing 0.1% BSA and 0.02% Tween (7-10 times), the bound phages were eluted with glycine HCl, pH 2.2. The eluate was amplified in K91K bacteria and polyethylene glycol-precipitated before a new round of panning was performed. Single clones were grown in LB medium containing 20 g/ml tetracycline and examined for binding to the invariant chain peptide by ELISA (see below).
Cloning of Inserts into fUSE5-The 1 phage was made by making the oligonucleotide 5Ј-tcggccgacggggcc cga gac aac ttt gtc atc atc tac gag ctg cta gat gag ctc atg gac ttt ggg tac ccg cag acc act gac ggg gccgctggggccgaaact-3Ј double-stranded by polymerase chain reaction, cutting with SfiI, and cloning into a SfiI site in the phage gpIII gene (underlined regions are flanking sequences, and coding region are shown as triplets). The 2 and 2 12N phages and the 1 alanine mutant phage were made the same way with the sequences 5Ј-aaa aat aat ttt ctt att tat gag ctt ctt gat gag att ctt gat ttt ggt tat cct cag aat agt gaa ggg-3Ј, 5Ј-aaa aat aat ttt ctt att tat gag ctt ctt gat gag-3Ј, and 5Ј-cga gac aac ttt gtc atc atc tac gag ctg cta gat gag ctc atg gac ttt gcc gcc gcc gcc acc act gac-3Ј, respectively, flanked by the same sequences as underlined above. All oligonucleotides were synthesized at the Biotechnology Center (Oslo, Norway).
fUSE5-Peptide Binding Assay-MaxiSorp wells (Nunc) were coated with peptide in PBS, pH 7.4, overnight at 4°C or at room temperature at concentrations of 500 g/ml for free peptides and 250 g/ml for BSA-coupled peptides before blocking in PBS containing 1% BSA for 1 h at room temperature. 20% methanol was added to free peptides during coating as this seemed to improve both binding and specificity in the assays for some of the peptides. Phage was added in PBS and 1% BSA for 2 h at room temperature at different concentrations. 10 10 phages, calculated as transducing units on K91K bacteria, correspond to a phage concentration (in transducing units) of 133 pM. Phage concentrations above 300 pM were not used in these assays as they only resulted in a marked increase in background problems. The plates were washed extensively in PBS containing 0.05% Tween 20 before horseradish peroxidase-conjugated anti-M13 IgG (Amersham Pharmacia Biotech) was added for 1 h at room temperature (1:5000 in PBS and 1% BSA). Another extensive wash was performed in PBS/Tween 20 before the addition of 100 l of substrate (1 mg/ml ABTS (Boehringer Mannhein) in citrate buffer, pH 4.0), and the plates were read at 405 nm in a Titertek Multiskan plate reader.
Peptides- Table I gives an overview of all peptides used and their abbreviations and sequences. The invariant chain peptides Ii LI wt (where wt is wild type) and Ii DQ-AA/LI-AA were synthesized at the Biotechnology Center. The BSA-coupled sequences from the tail of the transferrin receptor (TfR wt and TfR Y-A), TGN38 (TGN38 wt and TGN38 Y-A), and the invariant chain (Ii ML wt and Ii ML-AA) as well as the BSA-conjugated C5 control sequence and unconjugated C1 (10) and C2 (15) were kindly provided by Dr. G. Banting (Bristol, United Kingdom). The Lamp-1 (lysosomal associated membrane protein 1) peptides Lamp-1 wt and Lamp-1 Y-A were gifts from Dr. W. Hunziker (Lausanne, Switzerland), and the CD3␥ peptides CD3␥ wt and CD3␥ LL-AA were gifts from Dr. C. Geisler (Copenhagen, Denmark). The control peptides C3 and C4 were synthesized by Genosys Biotechnologies Inc. (Cambridge, MA). Amino acid analysis of the BSA conjugates was performed by Dr. K. Sletten (Biotechnology Center), showing that similar amounts of peptide were coupled per BSA for the wild type and its corresponding mutant. This was done to ensure that lack of/reduced binding in the assays not was due to different amounts of peptides. To ensure that free peptides equally coated the MaxiSorp wells and gave similar results as BSA-coupled peptides, we performed a binding assay with free Ii ML wt and Ii ML-AA peptides. These were biotinylated and HPLC-purified (Dr. G. Banting), and the coating efficiency for the wild type compared with the mutant was measured using horseradish peroxidase-streptavidin. These experiments showed that similar binding curves were obtained with free and BSA-coupled peptides and that the two peptides equally coated the MaxiSorp wells. This was as expected since large differences in coating efficiency due to small differences in sequence are usually a problem only with very short peptides (5-6 residues).
Conjugation of Peptides to Alkaline Phosphatase and Binding Assay of Tail Peptides-Conjugation of peptides to alkaline phosphatase (ALP) was essentially done as described previously (18). Briefly, the peptides 10 and 15 and the control peptide C3 were dissolved to 140 g/ml in PBS, pH 7.4, before ALP (25 mg/ml) was added to a molar ratio of 1:20 (ALP/peptide) for 1 h at room temperature. 50 l of glutaraldehyde in PBS was added, and the tube was further incubated for 2 h at room temperature before adding glycine (1 M). The conjugates were purified by filtration with Centricon 100 spin columns (Amicon, Inc., Beverly, MA) and concentrated to a final concentration of 0.25 mg/ml. Amino acid analysis of the conjugates showed that, on average, seven peptides were coupled per ALP molecule. The binding assay was performed as follows. MaxiSorp plates were coated and blocked as described for phage binding assays. ALP-peptide conjugate was added in PBS and 1% BSA to a final concentration of 10 g/ml. Free ALP and ALP-C3 were added at the same concentration to control wells.
Competition Assay with Biotinylated Phages-Phages were biotinylated with a large molar excess of biotin for 3-4 h at room temperature and then dialyzed overnight in PBS, pH 7.4, to remove free biotin. Biotin-labeled phages at a fixed concentration were allowed to interact with MaxiSorp plates coated with tail sequences in the presence or absence of an excess of unlabeled phage (up to 300 pM; see above). Binding of biotin-labeled phage was detected using horseradish peroxidase-streptavidin and horseradish peroxidase substrate.

Identification of a fUSE5 Clone Binding to an Invariant
Chain Tail Peptide-After screening a random fUSE5 display library against a peptide derived from the 15 N-terminal amino acids of Ii including an LI sorting signal, we identified phagedisplayed peptides with homology to a region of the medium chain adaptor subunits () (20). One phage clone expressing the residues GFPQ in its gpIII insert (Fig. 1, Phage clone 1) was found to interact strongly with the Ii peptide used in the screening (Ii LI wt), suggesting that this clone could recognize the LI-type endosomal sorting signal of Ii. Homology searches in the SwissProt data base with sequences derived from the interacting phage led us to a conserved region of the medium chain adaptor subunit () (Fig. 1). Other isolated phage clones from the screening also showed similarity to the same part of the chain region (Fig. 1, Phage clones 2-4).
fUSE5 Clones Expressing Chain Regions of 24 Amino Acids as gpIII Fusion Proteins-We wanted to investigate the relevance of the homologies between the phage clones and the chains by expressing a region from mouse 1 in fUSE5 gpIII and testing its ability to recognize the invariant chain LI signal. A secondary structure prediction on mouse 1 using the Chou and Fasman method (Ref. 22; see Ref. 21) and the SSPRED program 2 indicated an amphipathic helix preceding a turn containing the GYPQ motif. By running the ANTIGENIC algorithm (23) on the protein, the region was predicted to be antigenic, suggesting that it is surface-exposed. Based on the sequence comparisons and these structural predictions, region 104 -127 from mouse 1 could form a functional domain at the phage surface. The phage clone was made by inserting an FIG. 1. Alignment of phage clones with a region from the medium chain adaptor subunits (). Several fUSE5 clones displaying peptides with homology to a region of the medium chain adaptor subunit () were isolated after screening for phage recognizing the LI sorting signal of the major histocompatibility complex class II-associated invariant chain. These were aligned with the corresponding region of various chains. The selected region is highly conserved between chain subunits from both Golgi AP-1 ( 1 ) and plasma membrane AP-2 ( 2 ). A similar region is also found in the chain homologue 3 of AP-3, with which some of the selected phage clones also could be aligned.
oligonucleotide encoding the region 104 -127 of mouse 1 into phage gene III as described under "Experimental Procedures" (referred to as the 1 clone; see Table II). As shown in Fig. 2A, the 1 clone recognized the wild-type LI-containing peptide in a saturation-dependent fashion, whereas binding to the alanine mutant was not significantly higher than to BSA alone. Interestingly, the 1 clone discriminated between the two peptides better than the GFPQ clone isolated after screening (data not shown), indicating that the 1 clone is more adapted to the leucine signal than the clone from a limited random library. To test if the 1 clone could bind to a broader range of leucinebased sorting signals, we also used peptides based on the second leucine signal in the Ii cytosolic tail (Ii ML wt) (15,17) and the LL signal in the CD3␥ tail (24). As shown in Fig. 2 (B and  C), the 1 sequence could also discriminate these wild-type tail sequences from their corresponding alanine mutants. The 1 clone showed a higher interaction above background levels with these mutants compared with the Ii DQ-AA/LI-AA peptide. This may be due to interactions with additional amino acids in peptides where only the crucial leucine core motifs are changed to alanine (Table I).
Neither a fUSE2 control phage nor irrelevant fUSE5 clones gave signals in ELISA, even at high titers, confirming that other domains on the phage surface do not bind tail sequences. Microtiter wells coated with irrelevant control peptides gave signals below the BSA background level, even though some contained double leucines, tyrosines, or other residues usually present in signals (Table I). This shows that the context of the residues is important for binding and that the binding is not due to nonspecific interactions with bulky hydrophobic residues.
The Tyr-based sorting motifs are the most studied, and it is well established that proteins with these motifs bind adaptors. We therefore decided to include some Tyr-based signals in our assay, and we found that the 1 phage clone could discriminate wild-type Lamp-1 (25), TGN38 (26), and TfR (27) from their corresponding alanine mutants (Fig. 2, D-F). The binding curves showed saturation characteristics at the higher concentration of phages.
The 1 region (amino acids 104 -127) is very similar to the corresponding region in 2 (amino acids 102-125), with the differences being mostly conservative changes (Fig. 1). To test the mouse 2 homologue, region 102-125 was cloned into phage gpIII ( 2 clone) ( Table II). As shown in Fig. 3, the characteristics of the binding between the 2 clone and the tail sequences are almost identical to what we obtained for the 1 clone. This suggests that the region we have identified is involved in signal recognition by both AP-1 1 and AP-2 2 .
NMR data on the Ii tail has earlier demonstrated that a nascent ␣-helix in the N-terminal part of the Ii tail is stabilized in 20% methanol (17). Other peptides have also been shown to stabilize their folding in the presence of organic cosolvents (28,29). Interestingly, we found that methanol added to 20% during the coating of wells with unconjugated peptides increased the specificity of binding between free tail peptides and the clones (data not shown). This was observed in particular for the CD3␥ peptides, for which we observed clear differences in binding only when methanol was present during coating. The signal obtained against Ii wt seemed to increase with methanol added during the coating, whereas the lack of binding to the mutant Ii peptide was very clear both with and without methanol. Based on these observations, we suggest that methanol could  Table I for abbreviations and sequences). Abs, absorbance.
give the peptides a more native conformation when locked on the surface of the wells. Methanol added to the BSA conjugates seemed to enhance nonspecific interactions, possibly due to misfolding of these proteins.
BSA coupling is an approach used to ensure equal binding when coating a solid phase with small proteins or peptides (30). Amino acid analysis of our BSA conjugates showed that similar amounts of peptides were coupled for a wild type and its corresponding mutant. We can therefore rule out that the lack of/reduced binding to mutant peptides was due to the lack of primary attachment of these to the MaxiSorp wells or lack of peptide on the conjugates. Furthermore, Ii ML wt and Ii ML-AA peptides not coupled to BSA gave binding curves similar to those shown for Ii ML peptides in Fig. 2B. The free Ii ML peptides (which were biotinylated and HPLC-purified) were found to coat the MaxiSorp wells at equal efficiencies (see "Experimental Procedures"), showing that changing the ML residues to alanine did not affect the ability to attach to Maxi-Sorp wells.
Binding of Alkaline Phosphatase Conjugates of Chainderived Peptides to Tail Sequences-To investigate whether binding could also be obtained with a shorter sequence containing the GYPQ motif and when not expressed as a phage fusion product, we coupled the peptides 10 and 15 (Table I) to alkaline phosphatase as described. Amino acid analysis of the ALP-conjugate showed that, on average, seven peptides were coupled to each ALP molecule, giving a multivalence corresponding to that in the fUSE5 phage (four copies of gpIII). Fig.  4 (A and B) shows ALP-10 binding to the tail sequences Ii ML wt and TGN38 wt, to their corresponding mutants, and to BSA. Similar results were obtained for other tail sequences tested, both for ALP-10 and ALP-15 (data not shown). As shown, neither unconjugated ALP nor ALP coupled to an irrelevant control peptide (C3 in Table II) interacted with the tail peptides. In conclusion, the ALP conjugates bound to both tyrosine-and leucine-based signals, confirming that shorter sequences than the 24-residue region in our phage clones containing the GFPQ sequence could recognize the signals and that the peptide sequences do not necessarily have to be expressed on the phage.
Role of the GYPQ Sequence in the Interaction between the Domain and Sorting Signals-To investigate if the GYPQ residues are involved in binding to the tail sequences, we constructed a fUSE5 1 clone with these 4 residues changed to alanines ( 1 AAAA) ( Table II). The ELISA binding curves for the wild-type 1 clone versus the alanine mutant showed a reduced binding for the alanine mutant compared with the wild type (Fig. 5A). To quantitate the difference in binding between the wild-type 1 clone and 1 AAAA, we performed a titration at a point on the binding curve at the first point of saturation

ALP-(no peptide)
Ϫ a n denotes the average number of peptides coupled per ALP molecule and was estimated by amino acid analysis to be in the range 5-9 for all conjugates.

BSA-CKNLVLPADLFPNQEK
None Ϫ (corresponding to a phage concentration of 20 pM in Fig. 5A).
Phages were bound to wells coated with the Ii peptide, eluted by glycine HCl, and then titrated by means of counting transducing units. The average results from four titrations with S.D. values are shown in Fig. 5B. When equal amounts of the different phage clones (as calculated from transducing units) were coated onto MaxiSorp wells and detected by horseradish peroxidase-conjugated anti-phage antibody, equal ELISA signals were obtained, showing that the number of phages as calculated from transducing units may be used as an accurate measure of the actual number of phage particles (data not shown). This was checked because different phage clones might have different efficiencies of infection, giving different titers at the same number of actual particles. According to the titration data, the binding was 60% reduced at saturation for 1 AAAA compared with the wild-type 1 clone. Since the binding was not completely abolished, there may be other determinants in the cloned region that also contribute to binding or facilitate a functional structure. A fUSE5 clone expressing the first half of the 24-amino acid motif ( 2 12N) (Table II) was made to check whether the highly conserved upstream residues alone could mediate binding. This clone, however, gave no signal above the BSA level when tested against the various signal peptides (data not shown). By developing a competition assay based on biotinylated 1 phages, we also showed that when adding a 10-fold excess of unlabeled 1 phage, we observed a significant inhibition in binding, but not when adding the same amount of unlabeled 1 AAAA phage (Fig. 6). These results strongly indicate that the GYPQ residues are important for recognition of sorting signals.

DISCUSSION
While the tyrosine-based signals are shown to interact with the medium chain adaptor subunits () as well as with intact AP complexes (9 -11, 13), a protein candidate for interaction with leucine-based signals present in, for example, Man-6-P receptors, CD3␥, IgG Fc receptor, and major histocompatibility complex class II-associated Ii has not yet been defined. A dileucine signal transferred to the hemagglutinin tail has been shown to bind both AP-1 and AP-2 (9), as has the leucinecontaining IgG Fc receptor tail 3 and the CD3␥ tail (31). Segregation of the Man-6-P receptors into Golgi-derived clathrincoated pits is thought to involve a double leucine (4,32,33), although this interaction seems to be more complex as it also involves a casein kinase II site that may alone mediate binding to AP-1 (34). These data strongly suggest that some subunit of the AP complexes also binds leucine-based signals.
Our results especially indicate a role for the chain in adaptor binding to both leucine-and tyrosine-based sorting signals, as a conserved region of the various chains expressed at the surface of a fUSE5 fusion phage specifically recognizes peptides with leucine-and tyrosine-based signals. We suggest that the GYPQ sequence of the mouse 1 and 2 chains in particular is important for recognition of the sorting signals, as binding was significantly reduced when these residues were changed to alanine. Other sequenced mammalian chains such as the human and rat 2 chains also contain the same GYPQ sequence, whereas the 3 chains have GFPL in the corresponding site (Fig. 1). The 3 chain is a subunit of the recently identified AP-3 complex (35, 36) that seems to be involved in protein trafficking via a novel clathrin-independent pathway. Furthermore, 3 chains were recently shown to also recognize sorting signals of the tyrosine class (37). We have not expressed 3 regions on phage to test for binding to signals, but interestingly, we found enriched phage clones from the library expressing both GFPI and GFPV motifs. This may indicate that a GFPL motif can be functionally similar to a GYPQ motif during recognition of signals by the 3 chains.
Only leucines or tyrosines in a proper context were recognized by the cloned regions. This is consistent with our knowledge about sorting signals; leucines or tyrosines present in a tail sequence are not themselves able to function as sorting signals, but depend on the surrounding context (1,2,15,17). Sorting signals seem to comprise short, interchangeable, and self-determined structural motifs interacting with soluble proteins independent of orientation, as signals from type I mem-3 S. Honing and W. Hunziker, personal communication.  Table I for abbreviations and sequences). Abs, absorbance.
brane proteins have been shown to be active in type II membrane proteins and vice versa (27,38,39). Consistent with this, the phage clones in our study were shown to recognize sorting signals from both type I and II proteins.
The hypothesis that adaptor preclustering is required for binding tail sequences (12) is in agreement with results from in vitro studies of adaptor complexes binding to tail columns (3,4,7,9). These data indicate a multivalent attachment of aggregated complexes with a binding constant in the micromolar range and the requirement of a large (1000-fold or more) excess of competing peptide. In our system, it is possible that the multivalence of the expressed domains on the phage surface with four to five copies of the gpIII fusions per phage contributes to an enhanced binding to the signals.
The number of binding sites for AP complexes on a chip used for surface plasmon resonance was estimated in one study to be 500-fold lower than the number of peptides coupled, indicating that these sites are made up of more than one peptide simultaneously interacting with the AP complexes (9). This observation is in agreement with the requirements for relatively high concentrations of peptides in our phage-peptide assays. Peptides in the required arrangement may be rare on the surface of our microtiter wells, as is also supposed to be the case on the chips used for surface plasmon resonance.
Not much is known about the structure of the chains, but comparison of 2 chains from different species suggests that the protein is divided into two regions of similar length, but with different characteristics (40). Region I (230 amino acids) is more conserved and is separated from region II (190 amino acids, containing several sequences specific to 2 interspersed with other strongly conserved sections) by a linker region of variable length (10 -42 amino acids, linking the two functional domains together). Limited tryptic cleavage of AP-2 complexes bound to clathrin coats results in the cleavage and degradation of the 2 C terminus (41). Since the N terminus of ϳ24 kDa remains intact, it has been suggested that the interactions determining the association with the other subunits may be located in this part (region I). Our results suggest that there is also a recognition site for sorting signals in region I, which is the more conserved. This does not rule out the possibility that other regions of the chains also interact with the same sorting signals. The specificity in binding of full-length chains to the various signals may be determined by more than one region, and our results suggest that one binding motif is common or general. Other binding sites may reside in the less conserved C-terminal part of the chains, which also could give different recognition characteristics for 1 and 2 . A role for a more C-terminal interaction site is supported by the fact that not only the full-length mouse 2 clone, but also a truncated mouse 2 clone (⌬ 2 ) expressing residues 121-435 has been shown to interact with tyrosine signals (10,13). The sequence we have studied spans residues 102-125 of 2 , meaning that most of this region is deleted in their ⌬ 2 clone.
Using the two-hybrid approach, no binding was demonstrated between the chains and the LL signal of CD3␥ or the NPXY subclass of tyrosine signals (10,13). One explanation for this could be that a proper folding is not obtained in the yeast nucleus or that the interaction is below the detection limit in the two-hybrid system. This limit is usually at interactions in the micromolar range (42), and the binding constant between tyrosine signals and the chains was estimated to be as low as 10 M (13). By expressing chains in E. coli, we have, however, been able to show binding of both full-length 1 and 2 chains to leucine-based sorting signals. 4 Phage display libraries have been used to study a large variety of protein-protein interactions during the past years (14). In our case, this approach led to identification of a conserved region of the chains that recognizes tail sequences with a variety of endosomal sorting signals in vitro. However, it still remains to be proven that the site we have identified is actually used when the AP complexes interact with signals in vivo. The detailed mechanisms and kinetics of the recruitment of clathrin, adaptors, and proteins into coated pits and the following sorting of vesicles are not known. A chain-binding site for functional endosomal sorting signals may be required at any of the steps in this process. Sorting, which involves signals, clathrin, and adaptors, is known to take place at TGN and the plasma membrane and most likely also at other intracellular localizations (1,43,44). To account for this complexity and to assure specific sorting of the various molecules, an endosomal sorting signal might have more than one site of interaction with the cellular sorting machinery. As the site we have identified is conserved between chains and recognizes a relative broad spectrum of tyrosine-and leucine-based signals, it is tempting to speculate that this interaction takes place at one of the earlier stages in the sorting process. However, only further elucidation of this intricate process, which must be solved both in time and space, will allow us to make any firm conclusions.