Signal-binding specificity of the mu4 subunit of the adaptor protein complex AP-4.

The medium (mu) chains of the adaptor protein (AP) complexes AP-1, AP-2, and AP-3 recognize distinct subsets of tyrosine-based (YXXphi) sorting signals found within the cytoplasmic domains of integral membrane proteins. Here, we describe the signal-binding specificity and affinity of the medium subunit mu4 of the recently described adaptor protein complex AP-4. To elucidate the determinants of specificity, we screened a two-hybrid combinatorial peptide library using mu4 as a selector protein. Statistical analyses of the results revealed that mu4 prefers aspartic acid at position Y+1, proline or arginine at Y+2, and phenylalanine at Y-1 and Y+3 (phi). In addition, we examined the interaction of mu4 with naturally occurring YXXphi signals by both two-hybrid and in vitro binding analyses. These experiments showed that mu4 recognized the tyrosine signal from the human lysosomal protein LAMP-2, HTGYEQF. Using surface plasmon resonance measurements, we determined the apparent dissociation constant for the mu4-YXXphi interaction to be in the micromolar range. To gain insight into a possible role of AP-4 in intracellular trafficking, we constructed a Tac chimera bearing a mu4-specific YXXphi signal. This chimera was targeted to the endosomal-lysosomal system without being internalized from the plasma membrane.

The heterotetrameric adaptor protein (AP) 1 complexes AP-1, AP-2, AP-3, and AP-4 are components of protein coats that associate with the cytosolic face of organelles of the secretory and endocytic pathways (reviewed in Refs. [1][2][3][4]. AP-2 is associated with the plasma membrane and mediates rapid internalization of endocytic receptors, whereas AP-1, AP-3, and AP-4 are associated with the trans-Golgi network and/or endo-somes and mediate intracellular sorting events. AP complexes are thought to participate in protein sorting by inducing the formation of coated vesicles as well as concentration of cargo molecules within the vesicles. Concentration of integral membrane proteins is mediated by direct interaction of the AP complexes with sorting signals present within the cytosolic tails of the proteins. Several types of cytosolic sorting signals have been described, the most common of which are referred to as "tyrosine-based" or "dileucine-based" depending on which residues are critical for activity (5,6).
Our laboratory has been particularly interested in the role of the chains in signal recognition. We have previously demonstrated that 1 and 2 display a bipartite structure, with the amino-terminal one-third being involved in interactions with the corresponding ␤ chains and the C-terminal two-thirds being involved in recognition of YXXØ-type signals (23). X-ray crystallography revealed that the YXXØ-binding domain of 2 consists of a banana-shaped all-␤ structure to which the signals bind in an extended conformation (19). The Tyr and Ø residues fit into hydrophobic pockets on this domain. Both crystallographic (19) and binding (13)(14)(15)(16)(17)(18) studies have suggested that the identities of the Ø residue and the residues surrounding the critical Tyr residue are important determinants of the specificity of interaction. Although the subsets of YXXØ signals recognized by 1, 2, and 3A overlap to a significant extent, each chain nonetheless exhibits certain preferences for residues neighboring the critical Tyr residue (14). For example, 1, 2, and 3A prefer Leu, Leu, and Ile residues at the Ø positions and neutral, basic, and acidic residues at the X positions, respectively. We have argued that these preferences alone are unlikely to account for the functional specificity of each AP complex (14). However, they probably contribute to the selectivity and efficiency of specific signal recognition events.
Although much has been done to characterize the signalbinding specificity of 1, 2, and 3A, little is known about sequence preferences for the more recently described 4 (also known as -ARP2) (24). Previous studies have shown that 4 interacts weakly with YXXØ signals from the lysosomal membrane proteins LAMP-1 (AGYQTI) (18) and CD63 (SGYEVM) (25) and the trans-Golgi network protein TGN38 (SDYQRL) (18). To determine whether 4 might be able to recognize with higher affinity a defined subset of YXXØ signals, we have undertaken a yeast two-hybrid screening of a combinatorial YXXØ library. The results show that 4 prefers signals with Phe at position YϪ1, Asp at Yϩ1, Pro or Arg at Yϩ2, and Phe at Yϩ3 (Ø). A signal that fits this latter preference is found in the lysosomal membrane protein LAMP-2, and indeed, we found that the LAMP-2 signal binds to 4 both in the yeast two-hybrid system and in vitro. We also found that a reporter integral membrane protein bearing a 4-specific YXXØ signal is delivered to the endosomal-lysosomal system without being internalized from the plasma membrane.

EXPERIMENTAL PROCEDURES
Recombinant DNA Constructs-The constructs Gal4AD-1, Gal4AD-2, and Gal4AD-3A in the pACTII(LEU2) plasmid (CLONTECH, Palo Alto, CA) have been described previously (12,13). The Gal4AD-4 construct was prepared by ligating a BamHI-SacI polymerase chain reaction fragment corresponding to the 5Ј-part of 4 and a SacI-PstI cDNA fragment corresponding to the 3Ј-part of 4 into the BamHI-XhoI sites of the pACTII(LEU2) vector using a PstI-XhoI adaptor. As previously described (14), a DNA fragment encoding the 33-amino acid cytoplasmic tail of TGN38 engineered to contain an EagI site (by introduction of silent mutations in place of the codons for Arg 21 and Pro 22 from the TGN38 cytoplasmic tail) was used to prepare the pGBT9-TGN⌬-EagI construct by ligation into the EcoRI and XhoI sites of the pGBT9(TRP1) vector (CLONTECH). Oligonucleotides encoding either a combinatorial XXXYXXØ peptide library (14) or different YXXØ-type signals were digested with EagI and PstI and then ligated into pGBT-9-TGN⌬-EagI cut with EagI and PstI. The amino acid sequence encoded by the resulting constructs was Gal4BD-HNKRKIIAFALEGKRSKVT-RRPKXXXYXXØ. The construct Gal4BD-␤2 was kindly provided by Dr. M. S. Robinson (University of Cambridge, Cambridge, United Kingdom). All of the other two-hybrid constructs were made by ligation of polymerase chain reaction products into the pGBT9 or pACTII vector. The construct pET28a-4-(156 -453) was obtained by cloning nucleotides 466 -1362 of the coding sequence of 4 into pET28a (Invitrogen, Carlsbad, CA) using NheI and HindIII restriction sites. pET28a-4-(156 -453) was digested with NdeI and BstEII to release a 1066-base pair fragment containing the amino-terminal His 6 tag and ligated with the NdeI-BstEII fragment of vector pET16b (Invitrogen) containing the His 10 tag. The resulting construct was named pET28a-His 10 -4-(156 -453). Interleukin-2 receptor ␣ subunit (Tac) chimeric constructs were prepared by ligation of complementary oligonucleotides (coding for the PLSYTRF, DLYYDPM, and DLYADPM sequences) between an XbaI site inserted at the 3Ј-end of the Tac cDNA and the BamHI site from the expression vector pCDL-SR␣ (26).
Yeast Culture, Transformation, and Two-hybrid Assays-The Saccharomyces cerevisiae strain HF7c (MATa, ura3-52, HIS3-200, lys2-801, ade2-101, trp1-901, leu2-3,112, GAL4-542, gal80-538, LYS2::GAL1-HIS3, URA3::(GAL4 17-mers)3-CYC1-lacZ) (CLON-TECH) was maintained on yeast extract/peptone/dextrose-agar plates. Transformations were done by the lithium acetate procedure as described in the instructions for the MATCHMAKER two-hybrid kit (CLONTECH). For colony growth assays, HF7c transformants were streaked on plates lacking leucine, tryptophan, and histidine and allowed to grow at 30°C, usually for 4 -5 days, until colonies were visible. For two-hybrid screening of the combinatorial library, the yeast cells were first transformed with Gal4AD-4 and plated onto yeast dropout agar plates lacking leucine as described in the protocol for the MATCHMAKER two-hybrid system. Transformants were retransformed with the combinatorial DNA library and selected on plates lacking leucine and tryptophan for selection of co-transformants and lacking histidine for selection of interacting clones; Leu ϩ Trp ϩ and His ϩ colonies were then tested for ␤-galactosidase activity. Colonies expressing ␤-galactosidase were cultured in dropout medium containing leucine but lacking tryptophan to obtain cells carrying only the library plasmid and not the medium subunit plasmid. The resulting cells were then mated with the yeast strain Y187 (MATa) transformed with Gal4AD-4 constructs or with pTD1-1 (SV40 large-T antigen cDNA in pACTII; negative control for histidine FIG. 1. Two-hybrid screening of a combinatorial peptide library. A, sequences of XXXYXXØ clones selected by 4. A Gal4BD-XXXYXXØ library was coexpressed with a Gal4AD-4 construct in yeast cells. Co-transformants expressing interacting Gal4BD and Gal4AD constructs were selected in medium lacking tryptophan, leucine, and histidine and tested for expression of ␤-galactosidase activity. A list of the sequences obtained from library plasmids isolated from those clones is shown (see "Experimental Procedures" for details). B, statistical analysis of the library screening results. The preferences of 4 for residues within the XXXYXXØ sequence were inferred from the ⌬F values in S.E. units (y axis; see "Experimental Procedures" for details) at each position (panels YϪ3 to Yϩ3). Levels of significance are indicated by different gray tones, with the darkest representing the most significant (Ն2 S.E. also indicated with **). NS, not significant. C, cross-reactivity analysis of some sequences selected by 4. To test the binding specificity of signals selected by 4 (indicated at the top of each panel), the corresponding Gal4BD constructs were co-transformed with different Gal4ADconstructs and tested for complementation of histidine auxotrophy. Cell growth in liquid medium lacking histidine was measured as turbidity at 600 nm. auxotrophy and ␤-galactosidase activity) to test the binding specificity of library clones.
Cell Culture and Transfection-HeLa cells (American Type Culture Collection, Manassas, VA) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% (v/v) fetal bovine serum, 100 units/ml penicillin, and 100 g/ml streptomycin (Biofluids, Inc., Rockville, MD) (regular medium). Primary cultures of skin fibroblasts from AP-3-deficient mocha mice (Jackson Laboratory, Bar Harbor, ME) were obtained as previously described (27) and maintained in regular medium. The night before transfection, cells were seeded onto 6-well plates (Costar Corp., Corning, NY) in 2 ml of regular medium. The following day, the cells were cotransfected with the Tac constructs and pCI-NEO (Promega, Madison, WI) using Fugene-6 reagent (Roche Molecular Biochemicals). To obtain stable transfectant clones, the regular medium from HeLa cells was replaced with fresh medium containing 1 mg/ml G418 (Calbiochem) 24 hours after transfection. The clones obtained were analyzed for expression of the Tac constructs by immunofluorescence microscopy.
Statistical Analyses-The experimental (observed) frequency for each residue at each position of the XXXYXXØ sequence was calculated using the sequences selected by the 4 subunit from the combinatorial library. Preferences were evaluated by calculating the difference between the observed and expected frequencies (⌬F) in standard error units (14). Any ⌬F value above 1 (i.e. favored) or below Ϫ1 (i.e. disfavored) was considered to be significantly different from 0 (random).
Site-directed Mutagenesis-Single amino acid substitutions were made using the QuickChange mutagenesis kit (Stratagene, La Jolla, CA). Briefly, 50 ng of plasmid carrying the target cDNA was incubated with two complementary primers (2 mM each) containing the desired mutation in the presence of 2 mM dNTP mixture and 2.5 units of Pfu DNA polymerase for 16 cycles according to the following temperature profile: 0.5 min at 95°C, 1 min at 55°C, and 8 or 16 min at 68°C. After replication of both vector strands, the methylated parental DNA was digested for 1 h at 37°C with 10 units of DpnI endonuclease, and the nicked vector with the desired mutation was transformed into Escherichia coli.
In Vitro Binding Assays-35 S-labeled 4-(156 -453) protein was obtained by in vitro transcription/translation using the TNT T7 Quick coupled transcription/translation system (Promega) and Easytag TM expression protein labeling mixture (PerkinElmer Life Sciences) according to the manufacturers' instructions. In brief, 500 ng of the pET28a-4-(156 -453) construct was incubated with 20 l of TNT Quick Master Mix and 11 Ci of [ 35 S]methionine in a total volume of 25 l at 30°C for 90 min. The transcription/translation reaction mixture (containing 35 Slabeled 4-(156 -453)) was diluted 1:100 in binding buffer and centrifuged (180,000 ϫ g, 15 min, 4°C). 500 l of supernatant was applied to peptide-coupled beads and incubated for 12 h at 4°C. The beads were washed three times at 4°C with binding buffer without bovine serum albumin, boiled in Laemmli sample buffer, and separated by SDSpolyacrylamide gel electrophoresis. The SDS gel was soaked in sodium salicylate and subjected to autoradiography.
Expression and Purification of 4-(156 -453)-E. coli BL21(DE3) cells were transformed with pET28a-His 10 -4-(156 -453); a single colony was picked; and the presence of the construct was verified. 2 liters of LB/kanamycin medium was inoculated with 100 ml of preculture and grown at 37°C until A 600 reached 1.6. Protein expression then was induced by the addition of isopropyl-␤-D-thiogalactopyranoside to a final concentration of 3 mM, and the cells were incubated at 37°C for another 4 h. The cells were harvested, resuspended in buffer A (20 mM Tris-HCl (pH 8.0), 250 mM NaCl, and 5 mM imidazole), and sonicated. After centrifugation, the supernatant was loaded onto a Ni 2ϩ -nitrilotriacetic acid Superflow column (QIAGEN Inc., Valencia, CA), and the recombinant protein was eluted using buffer A with 1 M imidazole. Preparation of Peptide-coupled Beads and Surface Plasmon Resonance Sensor Chips-The following peptides were obtained from Zymed Laboratories Inc. (South San Francisco, CA): CWKRHHTGYEQF, CWKRHHTGAEQF, CWKRHHTGYEQA, CWRPKETLYRRF, CWRP-KETLARRF, and CWRPKETLYRRA. Peptide-coupled beads for in vitro binding assays were prepared by coupling the Cys residue of the peptides to EZ-Link TM PEO-maleimide-activated biotin (Pierce) in phosphate-buffered saline (pH 6.9) at peptide and biotin concentrations of 1 and 1.67 mM, respectively. The reaction was quenched by the addition of ␤-mercaptoethanol to a final concentration of 10 mM. 50 l of Immu-noPure immobilized streptavidin beads (Pierce) was washed twice with phosphate-buffered saline (pH 6.9), incubated overnight with 300 l of biotinylation reaction, and washed three times with binding buffer (0.05% (w/v) Triton X-100, 50 mM HEPES (pH 7.3), 10% (v/v) glycerol, 100 mM KCl, 2 mM MgCl 2 , 0.1 mM CaCl 2 , 50 M dithiothreitol, and 0.1% bovine serum albumin). Surface plasmon resonance experiments were carried out on a BIAcore 1000 instrument (BIAcore AB, Uppsala) at 25°C using SA sensor chips with streptavidin covalently immobilized on a carboxymethylated dextran matrix. The chips were conditioned by 10 consecutive 1-min injections of 1 M NaCl, 50 mM NaOH, and 0.25% (w/v) SDS at a flow rate of 10 l/min and washed extensively with Tris-buffered saline (20 mM Tris-HCl (pH 7.0), 250 mM NaCl, 5 mM EDTA, and 0.005% (v/v) polysorbate 20). Biotinylated peptides were injected at a concentration of 500 nM in Tris-buffered saline running buffer at a flow rate of 2 l/min onto the chip surface until the desired level of immobilization (ϳ150 response unit) was achieved. Unoccupied streptavidin was blocked by biotin (30 l of a 10 M solution at a flow rate of 5 l/min). The sensor chip was then washed by five consecutive 1-min injections of regeneration solution (25 mM NaOH, 500 mM NaCl, and 0.0005% (w/v) SDS). Flow cell 1 (with biotin-treated streptavidin) was left blank and used as a reference surface.
Surface Plasmon Resonance Spectroscopy-Surface plasmon resonance permits, in a label-free mode, real-time detection of binding events on the chip surface and estimation of binding parameters (28). 10 l of 4-(156 -453) at the indicated concentrations was injected onto sensor chip surfaces. Dissociation of bound protein was carried out for 10 min, and then the surface was regenerated by two 30-s injections of regeneration solution and by two 30-s injections of running buffer. All experiments were repeated twice on two different chips. Data transformation and overlay sensorgrams were prepared using BIAevaluation Version 3.0 software (BIAcore AB). The response from the reference surface was subtracted from the other three flow cells to correct for refractive index changes, matrix effects, nonspecific binding, injection noise, and base-line drift. Using nonlinear least-squares fitting, the equilibrium dissociation constant (K D ) was evaluated by fitting data to a single site interaction model (Equation 1), where RU eq is the steady-state response level, RU max is the maximal capacity of the surface (which was floated during the fitting procedure), and C is the concentration of 4 in micromolar.
Antibodies and Immunofluorescence Microscopy-Immunofluorescence microscopy of fixed permeabilized cells and antibody internalization microscopy experiments were done as previously described (27,29). The following monoclonal antibodies were used: anti-mouse LAMP-2 monoclonal antibody ABL-93, anti-human LAMP-2 monoclonal antibody H4B4, and anti-human CD63 monoclonal antibody H5C6 (Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA). A polyclonal antiserum to recombinant Tac was raised in rabbits. Alexa 448 and Cy3-conjugated secondary antibodies were from Jackson Im-munoResearch Laboratories, Inc. (West Grove, PA).

Signal-binding Specificity of 4 Determined by Screening of a Combinatorial XXXYXXØ Yeast Two-hybrid Library-We
have previously analyzed the signal-binding specificity of 1, 2, and 3 (A and B isoforms) by screening a Gal4AD-XXXYXXØ combinatorial library using the yeast two-hybrid system (14). Here, we have used the same method to define the signal-binding specificity of the 4 subunit of AP-4. To this end, the combinatorial library was coexpressed with a Gal4AD-4 construct in yeast cells. Twenty clones that grew in medium lacking histidine and that tested positive for ␤-galactosidase activity were isolated, and their amino acid sequences were deduced from DNA sequencing (Fig. 1A). A statistical analysis of the residues found at each position is shown in Fig. 1B; positive or negative ⌬F values correspond to residues that were favored or disfavored, respectively. Only residues with ⌬F values equal to or greater than 1 or equal to or lower than Ϫ1 were considered significant. Overall, 4 seemed to have a distinct preference for aromatic amino acids at several positions of the XXXYXXØ sequence. The most commonly found amino acids at each position were Cys, Tyr, Phe, Tyr, Asp, Pro, and Phe, respectively. Tyr was also favored at YϪ1 and Arg at Yϩ2. Among the residues at the Ø position, the only preference was for Phe, whereas Val was strongly disfavored (Fig. 1B).
Some of the sequences selected by 4 (DLYYDPM, ET-LYRRF, DFYYERL, and DYCYDRF) were tested for their ability to interact with other subunits (Fig. 1C). The results showed that the DLYYDPM sequence was specific for 4, whereas the ETLYRRF and DYCYDRF sequences interacted with 2 and 4, and the DFYYERL sequence interacted with all four chains (Fig. 1B). Thus, 4 shares with the other chains the ability to interact with distinct but overlapping sets of YXXØ-type sequences.
Interaction of 4 with Naturally Occurring Tyrosine-based Sorting Signals-To further characterize the interactions of 4 with YXXØ motifs, we used the yeast two-hybrid system to test for interactions with YXXØ signals found in the cytosolic tails of some transmembrane proteins. The YXXØ signal of TGN38 was replaced by the analogous signals from LAMP-1, CD68, CD63, and LAMP-2, and interactions with chains were tested using the yeast two-hybrid system. A qualitative assay for growth on histidine-deficient plates revealed that 4 interacted only with the YXXØ signal from human LAMP-2 (HTGYEQF) (Fig. 2A). The LAMP-2 signal was not recognized only by 4 though, as it bound even more strongly to 2 and 3A (Fig. 2,  A and B). A salient feature of this signal is the presence of Phe at the Ø position, which fits the 4 preferences deduced from the combinatorial analyses. The Tyr-to-Ala and Phe-to-Ala variants of the LAMP-2 signal (HTGAEQF and HTGYEQA, respectively) were unable to interact with 4 or with any other chain (Fig. 2B).
To verify the yeast two-hybrid results, we performed a binding assay using in vitro transcribed/translated 4-(156 -453) and chemically synthesized and biotinylated LAMP-2 peptides. The peptides were bound to streptavidin beads and incubated with radioactively labeled 4-(156 -453). Bound 4 was revealed by SDS-polyacrylamide gel electrophoresis and fluorography. As shown in Fig. 2C, 4-(156 -453) bound well to the wild-type LAMP-2 sequence (HTGYEQF), but only barely to the Tyr-to-Ala (HTGAEQF) and Phe-to-Ala (HTGYEQA) variants of the sequence.
The resolution of the crystal structure of the 2 signalbinding domain allowed identification of residues that are directly involved in interactions with the critical tyrosine residue of the signals (19). Several of those residues are conserved in the other chains, including 4 (3). To determine whether interactions of 4 with YXXØ signals involved conserved residues in the tyrosine-binding pocket, we mutated the conserved Asp 190 or Lys 438 residue of 4 to Ala. Two-hybrid assays revealed that these mutations abrogated interactions of 4 with the tyrosine-based signal from LAMP-2 (Fig. 2D). Thus, the structural bases for the recognition of YXXØ signals by 4 appear to be similar to those of 2.
Characterization of 4-YXXØ Interactions by Surface Plasmon Resonance Spectroscopy-4-YXXØ interactions were further characterized by surface plasmon resonance spectroscopy. In these studies, we used three biotinylated peptides: CWRP-KETLYRRF, corresponding to one of the sequences selected from the combinatorial library (Fig. 1B), and its Tyr-to-Ala (CWRPKETLARRF) and Phe-to-Ala (CWRPKETLYRRA) variants. Preliminary in vitro binding experiments showed that the CWRPKETLYRRF peptide bound radiolabeled 4-(156 -453) (Fig. 3A) in a concentration-dependent manner (Fig. 3B), whereas CWRPKETLARRF and CWRPKETLYRRA did not (Fig. 3, A and B). The three biotinylated peptides were loaded onto separate flow cells of a streptavidin-coated chip. Recombinant 4-(156 -453) was then applied, and binding of the protein was measured by an increase in response units. The signal for the CWRPKETLYRRF peptide at a concentration of 13.4 M reached a plateau at ϳ1000 response units, whereas that of the two variant peptides only reached 100 -150 response units. This was in the range of the nonspecific binding of 4-(156 -453) to the biotinylated streptavidin surface without any peptide bound, as shown by the blank curve. After ending the injection of protein solution at 5 min, the value for the signal dropped sharply for all samples, indicating that the binding process was mostly reversible. However, ϳ20% of the binding could not be reversed even after washing for 10 min (data not shown). We performed an analysis of the interaction of different concentrations of 4-(156 -453) with the CWRP-KETLYRRF peptide (Fig. 4A). As expected, the signal amplitude was dependent on the amount of 4-(156 -453) applied. The response approached a plateau value (a steady-state level, RU eq (Equation 1)) after ϳ4.5 min. A plot of RU eq against the concentration of 4 is presented in Fig. 4B. Nonlinear regression analysis of these data yielded an apparent equilibrium dissociation constant of 7.0 Ϯ 2.5 M and a maximum binding capacity (RU max ) of the surface of 1550 Ϯ 165 response units. Although these values should be considered only estimates, it is nonetheless clear that the interactions are of low affinity.
Intracellular Localization of a Chimeric Protein Bearing a 4-specific Signal-To gain insights into the possible function of AP-4, we took advantage of the identification of a YXXØ signal (DLYYDPM) that was apparently specific for 4 (Fig.  1B). This signal, as well as its corresponding Tyr-to-Ala mutant (DLYADPM), was appended to the cytosolic tail of the transmembrane protein Tac (29). The constructs were stably expressed in HeLa cells, and their intracellular distribution at steady state was examined by immunofluorescence microscopy using antibodies to the Tac luminal domain. We observed that the Tac-DLYYDPM chimera was present in the Golgi complex and plasma membrane (Fig. 5A). Treatment with the lysosomal inhibitor leupeptin, however, resulted in accumulation of Tac-DLYYDPM in intracellular vesicles (Fig. 5B). Some of these vesicles colocalized with the lysosomal transmembrane proteins CD63 (Fig. 5, D-F) and LAMP-2 (Fig. 5, G-I), suggesting that a fraction of the Tac-DLYYDPM chimera was transported to late endosomes or lysosomes. The vesicular staining and colocalization of the chimera with LAMP-2 were not affected by the absence of the AP-3 complex in cells from the mocha mouse strain (Fig. 5, J-L) (30), consistent with the observation that the DLYYDPM signal does not interact with 3A (Fig. 1C). The DLYYDPM signal did not mediate internalization of the chimera from the cell surface (Fig. 6, A and B), whereas a PL-SYTRF signal derived from the transferrin receptor did (Fig. 6, E and F). As expected, the Tyr-to-Ala mutant chimera (DLY-ADPM) and a Tac construct without any tyrosine-based sorting signal were not significantly internalized (Fig. 6, C and D, and J and K, respectively). These observations were in agreement with the inability of the DLYYDPM signal to interact with 2 and suggested that the vesicular localization of the Tac-DLYY-DPM chimera was not the result of internalization from the cell surface.

DISCUSSION
The results of the experiments reported here show that 4 shares, with other members of the family of AP subunits, the ability to recognize a subset of YXXØ sorting signals. As is the case for other chains, interactions of 4 with YXXØ signals require the Tyr and Ø residues (Figs. 2 and 3) and are saturable (Fig. 4). These properties emphasize the remarkable structural conservation of the chain family of proteins. Indeed, of 15 residues in 2 known to be involved in interactions with Tyr and Ø residues (19), 14 are identical in 4 (3), with the remaining one being a conservative Leu 173 (2)-to-Val 187 (4) substitution. Mutation of one of two of the identical amino acids, Asp 190 or Lys 438 , to Ala abrogates interaction of 4 with the signals (Fig. 2D), confirming that 2 and 4 recognize YXXØ signals in a similar fashion.
These structural similarities notwithstanding, the subset of YXXØ signals recognized by 4 exhibits some characteristic features that distinguish it from that of other chains. The most salient feature of 4 specificity is the preference for aromatic residues (Phe or Tyr) at various positions neighboring the critical Tyr residue. None of the other chains characterized to date exhibits this preference (14). The preference for Phe residues is particularly strong at the YϪ1 and Yϩ3 (Ø) positions. In the case of the Ø position, this might be explained by the Leu 173 (2)-to-Val 187 (4) substitution. The smaller Val 187 residue lining the hydrophobic pocket could allow accommodation of the large aromatic side chain of Phe while disfavoring binding of the smaller Val side chain. Another preference specific for 4 is Asp at position Yϩ1, whereas other preferences are similar to those of other chains. For instance, the selectivity for Pro at Yϩ2 appears to be a general characteristic of all the chains. This suggests that a bend in the polypeptide chain imposed by Pro stabilizes the conformation of the signals for interaction with chains. 4 also favors Arg at Yϩ2, a preference shared only with 2 (14). In the case of 2, this preference for Arg is due to the establishment of hydrophobic interactions of the Arg side chain with Trp 421 and Ile 419 of 2 and a hydrogen bond between the guanidinium group of Arg and Lys 420 of 2 (19). Two of these residues in 2 (Trp 421 and Lys 420 ) are conserved in 4 (Trp 429 and Lys 440 , respectively), but not in the other chains (3), which probably explains why only 2 and 4 favor Arg at Yϩ2.
Despite the fact that 4 prefers certain residues at positions neighboring the critical Tyr residue, the subset of YXXØ signals recognized by 4 overlaps to a significant extent with those recognized by other chains (Fig. 1C). This further strengthens the previous conclusion that chains recognize distinct but overlapping sets of YXXØ signals (14). Therefore, the involvement of AP complexes in specific sorting events cannot depend solely on the specificity of signal recognition by their chains. Rather, the role of signal preferences is likely to "fine-tune" the efficiency of sorting.
A screening of several naturally occurring YXXØ signals revealed that the lysosomal targeting signal from LAMP-2 (HTGYEQF) (30) interacts with 4 (Fig. 2). This signal has a Phe residue at the Ø position, which could explain why it binds to 4 (Fig. 1B). Previous studies had demonstrated weak interactions of 4 with two other lysosomal membrane proteins, LAMP-1 (18) and CD63 (25). Taken together, these observations suggest a possible role for the AP-4 complex in sorting to lysosomes. However, the signals from all of these lysosomal membrane proteins interact better with 2 and 3A than with 4 (Fig. 2, A and B). To gain insight into the potential function of AP-4, we took advantage of the identification of a signal (DLYYDPM) that interacts exclusively with 4 (Fig. 1C). This signal was placed at the cytosolic carboxyl terminus of a Tac chimeric construct devoid of other sorting signals (13). The resulting Tac-DLYYDPM chimera was expressed by stable transfection into HeLa cells, and its localization was determined by indirect immunofluorescence microscopy. In the absence of protease inhibitors, the protein exhibited a steadystate localization to the Golgi complex and plasma membrane. However, incubation with the lysosomal inhibitor leupeptin resulted in accumulation of the protein in lysosomes, as shown by colocalization with LAMP-2 (Fig. 5). This indicated that the Tac-DLYYDPM chimera is transported to and degraded in lysosomes. As expected, this accumulation was dependent on the critical Tyr residue of the signal. The Tac-DLYYDPM chimera was not efficiently internalized from the plasma membrane (Fig. 6), in accordance with its inability to interact with 2 ( Fig. 1C). In addition, the Tac-DLYYDPM chimera was still targeted to lysosomes in AP-3-deficient mocha cells, further demonstrating that AP-3 does not play a role in the recognition of the DLYYDPM signal. Even though our two-hybrid results indicated that there is no interaction between the DLYYDPM signal and 1 (Fig. 1C), we cannot rule out the possibility that AP-1 could somehow be involved in sorting of the Tac-DLYY-DPM chimera. However, Meyer et al. (31) have suggested that targeting of proteins to lysosomes is not affected in 1-deficient cells. These observations are consistent with the possibility that 4 and, by extension, the AP-4 complex are involved in targeting proteins from the trans-Golgi network to the endosomal-lysosomal system. This involvement could provide an alternative means of sorting proteins to lysosomes, the existence of which has been suggested by previous studies (31)(32)(33)(34). The evidence for a role of AP-4 in targeting to the endosomallysosomal system presented here, however, is indirect and should be considered tentative until it becomes possible to study protein sorting in AP-4-deficient cells. Attempts to ablate expression of this complex in mice are underway.