INTRODUCTION

The clathrin-associated adaptors AP-1 and AP-2 are heterotetrameric complexes that mediate attachment of clathrin to membranes and recruitment of integral membrane proteins for incorporation into clathrin-coated vesicles (reviewed in Refs. 1 and 2). AP-1 is localized to the trans-Golgi network, where it mediates biosynthetic protein transport to the endosomal/lysosomal system, whereas AP-2 is localized to the plasma membrane and mediates rapid internalization of endocytic receptors. AP-1 is composed of four chains termed  (~91 kDa), 1 (~101 kDa), µ1 (~47 kDa), and 1 (~19 kDa). The four chains of AP-2 are homologous to those of AP-1 and are known as  (~104-108 kDa), 2 (~104 kDa), µ2 (~50 kDa), and 2 (~17 kDa), respectively. The homologous chains of each complex are thought to subserve similar structural and functional roles. Two of the "large" chains,  and , mediate specific targeting of the corresponding adaptors to the trans-Golgi network and the plasma membrane, respectively (3, 4). The other large chains, 1 and 2, have been implicated in clathrin binding (5-8). A role for the  chain in clathrin binding has also been proposed (9). Recent work has demonstrated that µ1 and µ2 (known as the "medium" chains) bind tyrosine-based sorting signals present within the cytosolic domain of some integral membrane proteins and thus probably mediate the capture of these proteins into clathrin-coated areas of the membranes (10-14). The function of the 1 and 2 chains (referred to as the "small" chains) is currently unknown.

The overall structure of the adaptor complexes and the arrangement of the different chains within them have been studied using various approaches. Analyses by deep-etch electron microscopy have revealed that the AP-2 complex has a brick-shaped core of ~8 nm ("head") with two smaller appendages of ~2-3 nm ("ears") separated from the core by thin linker strands ("hinges") (15). Mild proteolysis of AP-2 resulted in the release of both ears, leaving a head composed of ~60-65-kDa amino-terminal fragments of  and 2, and intact µ2 and 2 (6, 16-18). The division of the adaptor large chains into head, hinge, and ear domains suggested by the combination of electron microscopy and limited proteolysis analyses was further supported by sequence comparisons with structural homologs of  (i.e.  (19) and  (20, 21)) and 2 (i.e. 1 (18, 22), 3A (23, 21), and 3B (originally referred to as -NAP, Ref. 24)), which revealed three distinct regions with different degrees of sequence conservation. Although the AP-1 complex has not yet been visualized by electron microscopy, limited proteolysis and sequence analyses of its chains suggest that it has an overall structure similar to AP-2 (6, 19).

Additional insight into the structure of the AP-1 and AP-2 complexes was obtained from analyses of subunit interactions using the yeast two-hybrid system (4). This approach revealed the following pairwise interactions for AP-1 subunits: -1, -1, and 1-µ1; the analogous interactions observed for AP-2 were -2, -2, and 2-µ2. Interactions were also detected between the and µ chains of different AP complexes (i.e. 1-µ2 and 2-µ1), which suggests that 1 and 2 may be interchangeable within the complexes. This is not surprising, since 1 and 2 are very closely related (85% identity and 92% similarity; Refs. 18, 22, and 25). As expected from previous structural studies, all interactions involving the large chains (, , 1, and 2) occurred at the level of their head domains (4).

In contrast to the detailed information that is now available about the structure of the large chains, very little is known about the domain organization of the µ chains, as well as their arrangement within the AP complexes. These chains have been implicated in two types of interaction: assembly with the  chains (4) and recognition of tyrosine-based sorting signals (10-12). The regions of the molecules that are responsible for these functions and the relationship of these regions to the rest of the complex have not been established. In this study, we present a structure-function analysis of the µ chains aimed at identifying the domains that interact with the  chains and with tyrosine-based sorting signals. These analyses reveal that both µ1 and µ2 can be functionally dissected into two parts: an amino-terminal one-third (residues 1-145 of µ1 and µ2) that interacts with the  chains and the carboxyl-terminal two-thirds (residues 147-423 of µ1 and 164-435 of µ2) that bind tyrosine-based sorting signals. These observations suggest a model in which the amino-terminal one-third of µ2 is embedded within the AP-2 head via interaction with 2, whereas the carboxyl-terminal two-thirds of the protein project outward from the head, placing this domain in a position to interact with tyrosine-based sorting signals.

EXPERIMENTAL PROCEDURES

The constructs GAL4 DNA-binding domain (GAL4bd)¹-TGN38 Tail 1 (YQRL), GAL4bd-TGN38 Tail 1 (AQRL), GAL4ad-µ1, and GAL4ad-µ2 have been described previously (10, 11). The construct GAL4bd-2 was kindly provided by Dr. M. S. Robinson (Department of Clinical Biochemistry, 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 vectors (CLONTECH, Palo Alto, CA). The nucleotide sequences of all the recombinant constructs were confirmed by dideoxy sequencing.

The Saccharomyces cerevisiae strain HF7c (MATa, ura3-52, HIS3-200, lys 2-801, ade2-101, trp1-901, leu2-3, 112, gal4-542, gal80-538, LYS2::GAL1-HIS3, URA3::(GAL4 17-mers)₃-CYC1-lacZ) (CLONTECH) was maintained on YPD agar plates. Transformation was 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 large enough for further assays. Quantitative growth assays were carried out as follows; 3-5 colonies of each HF7c transformant were added to 20 ml of liquid medium lacking histidine (His medium) and grown at 30 °C to 1.0-1.2 OD₆₀₀/ml. Cultures (3 × 10³ OD₆₀₀ units, ~3 µl) were then inoculated into 20 ml of His medium in the absence or presence of several concentrations of 3AT (3-amino-1,2,4-triazole, Fluka Chemie AG, Buchs, Switzerland). After 2 days of incubation at 30 °C, the OD₆₀₀ of triplicates were measured. Results were expressed as the ratio of the signal obtained in the presence of 3AT and the signal obtained in its absence (control). -Galactosidase assays of HF7c transformants in liquid culture (5-7 colonies/culture) were done using a chemiluminescent -galactosidase assay kit (CLONTECH). Briefly, yeast cells were resuspended at 10 OD₆₀₀/ml in lysis buffer (100 mM sodium phosphate, 1 mM dithiothreitol, pH 7.4) and broken by vortexing in the presence of glass beads. After centrifugation at 4 °C for 15 min in a microcentrifuge, 10-50 µl of the lysate was added onto 200 µl of kit reaction buffer, incubated for 1 h at room temperature, and the light emission recorded as a 5-s integral in a tube luminometer (Monolight 2010, Analytical Luminescence Laboratory, San Diego, CA). Results were normalized by protein concentration and expressed as the mean ± standard deviation of three independent determinations.

Single amino acid substitutions to alanine in µ2 were made using the QuickChange site-directed mutagenesis kit (Stratagene, La Jolla, CA). Briefly, 50 ng of pGBT9 or pACTII plasmids carrying the µ2 chain cDNA were incubated with two complementary primers (2 µM each) containing the desired mutation in the presence of 2 mM dNTP mix 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 having the desired mutation was transformed into Escherichia coli.

Proteolytic cleavage of adaptor complexes was performed by incubating clathrin-coated vesicles (total protein concentration: 500 µg/ml; vesicles were kind gifts of E. Eisenberg and L. Green, NHLBI, National Institutes of Health and T. Kirchhausen, Harvard Medical School, Boston, MA) with different concentrations (1-1250 µg/ml) of trypsin (Sigma) for 10 min at 37 °C in phosphate-buffered saline, 5 mM MgCl₂, 1 mM dithiothreitol, pH 7.4. The reaction was stopped by addition of 50 mM soybean trypsin inhibitor (Sigma), and the digested samples were analyzed by SDS-PAGE (4-20% precast gradient gels; Novex, San Diego, CA) and electroblotted onto nitrocellulose. After incubation with primary antibodies to AP-2 chains and horseradish peroxidase-conjugated secondary antibodies, bands were detected using the ECL system (Amersham).

A rabbit polyclonal antiserum (R11-29) to an amino-terminal sequence of the µ2 chain was obtained by immunization with a peptide corresponding to residues 11-29 of mouse µ2 (KGEVLISRVYRDDIGRNAV). This antibody was specific for µ2 and did not recognize the related protein µ1 (data not shown). Other antibodies used were: AC1-M11 (anti-, Ref. 26, kindly provided by M. S. Robinson, Department of Clinical Biochemistry, University of Cambridge, Cambridge, UK), 100/2 (anti-, Ref. 27, obtained from Sigma), 100/1 (anti-, Ref. 27, obtained from Sigma), and C420-A9 (anti-, kindly provided by T. Kirchhausen).

Sequences from different µ chains were compared using the PLOTSIMILARITY program from the Wisconsin Package (version 8.1-UNIX, Genetics Computer Group, Madison, WI), and µ2 loop probability was calculated using the neural network prediction system PHD (28-30).

RESULTS

Interactions of µ1 and µ2 with 2 and with a Tyrosine-based Sorting Signal

Previous studies have demonstrated that the clathrin-associated adaptor medium chains µ1 and µ2 interact with the 2 chain of AP-2 (4) and with YXXØ-type² tyrosine-based sorting signals (10-12). To further characterize these interactions, we used a yeast two-hybrid system in which the interacting polypeptides were expressed as fusions with the activation or binding domains of the transcriptional activator GAL4 (31) (Fig. 1). Association of the two GAL4 domains driven by interacting polypeptides leads to activation of transcription of the HIS3 gene, which allows growth in defined media lacking histidine (His), and of the lacZ gene, which results in expression of -galactosidase activity (Fig. 1).

[View Larger Version of this Image (28K GIF file)]

A qualitative assay for growth on His plates confirmed that both µ1 and µ2 interacted with 2 and with the YQRL signal (Fig. 2A). Interactions with the signal were dependent upon the critical tyrosine residue (Fig. 2A). To compare the apparent avidities of these interactions, we first monitored the growth of the transformed yeast strains in His liquid medium (Fig. 2B). In this assay, stronger interactions result in higher rates of growth. In addition, we developed a more discriminating assay in which the growth of transformed yeast strains in His liquid medium was measured in the presence of varying concentrations of 3AT, an inhibitor of the HIS3-encoded enzyme imidazole-glycerol-phosphate dehydratase (Fig. 2C). In this assay, the stronger the interactions between the two-hybrid partners, the higher the concentration of 3AT needed to inhibit growth. Both quantitative growth assays indicated that µ2-2 interactions were somewhat stronger than µ1-2 interactions (Fig. 2, B and C), consistent with previously published data (4). The assays also revealed that µ2-YQRL interactions were much stronger than µ1-YQRL interactions, as evidenced by the marked differences of growth in liquid medium (Fig. 2B) and by the shift (2 orders of magnitude) of the 3AT inhibition curves (Fig. 2C).

Both µ1 and µ2 interact with 2 and with the tyrosine-based signal YQRL.

His

[View Larger Version of this Image (36K GIF file)]

Delineation of Regions of µ2 Involved in Interactions with 2 and with the YQRL Signal

To delineate the region of µ2 involved in interactions with 2, we constructed a series of µ2 deletion mutants (Fig. 3A) and evaluated them by two-hybrid assays for growth on His plates (Fig. 3A), resistance to growth inhibition by 3AT (Fig. 3A), and -galactosidase activity (Fig. 3B). All assays gave similar results. Deletion of the amino-terminal 120 amino acids of µ2 (construct 121-435) abrogated interaction with 2. In contrast, constructs with deletions of up to 290 amino acids from the carboxyl terminus of µ2 (e.g. construct 1-145) were able to interact with 2. A further deletion leaving the amino-terminal 70 amino acids of µ2 (construct 1-70), however, completely abolished interaction with 2. These observations suggested that the binding site for 2 is contained within an amino-terminal segment of µ2 comprising amino acids 1-145.

Analysis of the interaction of 2 with µ2 deletion mutants.

[View Larger Version of this Image (27K GIF file)]

The same assays were used to examine interaction of µ2 deletion mutants with the YQRL signal (Fig. 4, A and B). We found that deletion of up to 163 amino acids from the amino terminus of µ2 (e.g. construct 164-435) had little or no effect on the ability of the protein to bind YQRL. Deletion of an additional 19 amino acids (construct 183-435), however, prevented interaction with the signal. Truncation of the last 23 amino acids from the carboxyl terminus of µ2 (construct 1-412) did not prevent interaction with the signal, but removal of 25 amino acids from the carboxyl terminus of a construct starting at amino acid 164 (construct 164-410) led to a complete loss of interaction. From these observations, we concluded that the YXXØ-binding region is contained within a carboxyl-terminal segment spanning amino acids 164-435, with the suggestion that the last 23 amino acids are dispensable for interactions.

[View Larger Version of this Image (24K GIF file)]

Taken together, the above experiments indicate that the ability to interact with 2 resides within the amino-terminal one-third of µ2, whereas interactions with YXXØ-type signals are a function of the carboxyl-terminal two-thirds of the protein (see scheme in Fig. 8).

[View Larger Version of this Image (27K GIF file)]

Effects of Single Amino Acid Substitutions on Interactions of µ2 with 2 and the YQRL Signal

To further define the regions of µ2 involved in interactions with 2 and with the YQRL signal, we introduced single amino acid substitutions (to alanine) within segments of µ2 that are highly homologous to segments of µ1 (Table I; see Fig. 8 for position of the residues). Most of the amino acids targeted for mutagenesis were identical in µ2 and µ1 and were contained within either the 2-binding region (Leu-82, Tyr-83, Ile-96, and Glu-98) or the YXXØ-binding region of µ2 (Asp-176, Met-209, Phe-265, Asp-269, and Gly-270). We expected that the conservation of these amino acids would reflect an essential requirement for structure or function. The mutated µ2 constructs were tested for interactions using the 3AT growth inhibition assay (Table I) and the -galactosidase assay. The results of both assays demonstrated that mutation of Leu-82, Tyr-83, or Ile-96 significantly decreased interaction with 2 but had little or no effect on the interaction with the YQRL signal (Table I and Fig. 5, A and B). Conversely, mutation of either Asp-176, Met-209, Phe-265, Asp-269, or Gly-270 had no effect on interaction with 2 but significantly decreased interaction with the YQRL signal (Fig. 5, A and B). Mutation of Glu-98 resulted in slight but significant decreases in interactions with both 2 and YQRL, suggesting that this particular mutation has a more general effect on the structure of µ2. The differential effects of most of these mutations are consistent with the results of the deletion analysis (Figs. 3 and 4) and thus reaffirm the idea that the 2- and YXXØ-binding sites are contained within the segments 1-145 and 164-435, respectively. The fact that various mutations resulted in reduction of binding suggests that the conserved residues tested are either all involved in interactions with 2 or YXXØ or required for the conformational integrity of the 2- and YXXØ-binding domains.

Table I. Binding characteristics of µ2 point mutants analyzed by 3AT inhibition assays All residues targeted for mutagenesis were identical in µ2 and µ1, except for Ile-96 in µ2, which is replaced by a leucine in µ1. The 2 and YQRL binding activity of the different mutants was estimated by determining the ratio of half-maximal inhibitory concentrations of 3AT according to the formula: (IC50)mutant/(IC50)WT. The (IC50)WT values for µ2 binding to 2 and YQRL were 34 ± 2 mM and 23 ± 2 mM, respectively. The results are expressed as the mean ± standard deviation of three determinations. Mutants (IC50)mutant/(IC50)WT  2 binding YQRL binding L82A 0.30  ± 0.05 0.78  ± 0.20 Y83A 0.58  ± 0.15 0.80  ± 0.14 I96A 0.11  ± 0.03 0.74  ± 0.18 E98A 0.70  ± 0.20 0.80  ± 0.10 D176A 1.03  ± 0.10 0.60  ± 0.10 M209A 0.98  ± 0.12 0.02  ± 0.01 F265A 0.96  ± 0.15 0.20  ± 0.07 D269A 0.95  ± 0.10 0.20  ± 0.12 G270A 1.10  ± 0.20 0.20  ± 0.06

Analysis of the interaction of µ2 point mutants with 2 and the YQRL signal.

[View Larger Version of this Image (30K GIF file)]

Mapping of the Regions of µ1 Involved in Interactions with 2 and with the YQRL Signal

As shown in Fig. 2, µ1 is also capable of interacting with 2 and YXXØ-type signals (4, 10-12). This observation, in conjunction with the fact that µ1 and µ2 share 40% identity and 64% similarity over the entire length of their polypeptide chains (32), suggests that µ1 and µ2 may have a similar domain organization. To test this hypothesis, we constructed a small number of µ1 deletion mutants (Fig. 6) that were analogous to key µ2 constructs. Using various yeast two-hybrid assays, we found that segments of µ1 spanning amino acids 1-145 and 147-423 interacted with 2 and YQRL, respectively (Fig. 6). This observation indicated that µ1 and µ2 have a similar, bipartite domain organization.

Analysis of the interaction of µ1 deletion mutants with 2 and the YQRL signal.

[View Larger Version of this Image (25K GIF file)]

All the experiments described thus far were performed with isolated medium chains using the yeast two-hybrid system. To examine how these results correlated with the domain organization of µ2 in the context of the complete AP-2 complex, we performed proteolytic digestion experiments. To this end, we treated purified brain clathrin-coated vesicles (CCVs), which contain clathrin-AP-2 assemblies, with trypsin. Digestions were performed on CCVs rather than purified AP-2 because µ2 was found to be more sensitive to proteolysis in CCVs (33). The trypsin-digested samples were analyzed by SDS-PAGE and immunoblotting using antibodies to AP-2 subunits (Fig. 7). As expected from previous work (6, 16-18), the  and  chains were cleaved into ~60-kDa and ~40-kDa fragments, which corresponded to the head and ear domains, respectively (Fig. 7, A-D).

[View Larger Version of this Image (47K GIF file)]

The cleavage pattern of µ2 was investigated using an antibody to residues 11-29 of the protein (Fig. 7F); thus, all the µ2 fragments identified in the immunoblots contained the amino terminus of the protein. Treatment with low concentrations of trypsin resulted in cleavage of full-length µ2 (apparent molecular mass ~50 kDa) into a species of 27 kDa; with increasing trypsin concentrations, the 27-kDa species disappeared concomitant with the appearance of a 25-kDa species (Fig. 7F). These observations are in agreement with previous work of Matsui and Kirchhausen (33), who demonstrated that a similarly generated 25-kDa fragment of µ2 corresponds approximately to the amino-terminal half of the protein. Treatment with higher concentrations of trypsin resulted in conversion of the 27-25-kDa species into a 17-kDa fragment and eventually a 15-kDa fragment (Fig. 7F). No fragments smaller than 15 kDa were observed, even at the highest concentration of trypsin tested (1.25 mg/ml); at this concentration, virtually all of the µ2 was cleaved into the 15-kDa fragment (Fig. 7F). Thus, these experiments demonstrated the existence of trypsin-sensitive sites at about one-third and one-half of the µ2 polypeptide chain, from the amino terminus. The amino-terminal one-third of µ2 likely represents an independent folding domain of µ2 and roughly corresponds to the region of µ2 involved in 2 binding. The resistance of this µ2 domain to proteolysis may be due to its location within the AP-2 core. On the other hand, the carboxyl-terminal two-thirds of µ2 (which contains the YXXØ-binding region) appear to be accessible to the protease in the context of the AP-2 complex.

DISCUSSION

The results of the two-hybrid analyses presented here suggest that the µ2 chain of AP-2 has a bipartite structure, with the amino-terminal one-third of the molecule (amino acids 1-145) being involved in assembly with the 2 chain and the carboxyl-terminal two-thirds (amino acids 164-435) being responsible for the recognition of YXXØ-type sorting signals (Fig. 8). A more limited analysis of µ1 suggests that this protein has a similar bipartite structure. The two functional regions of µ2 are completely separable and remain fully active in the absence of the other part of the molecule. Moreover, point mutations can be introduced within each of the two regions that impair one type of interaction without affecting the other. The functional domain organization of µ2 suggested by the two-hybrid analyses is consistent with the location of a major trypsin-sensitive site at about one-third of the polypeptide chain relative to the amino terminus (Fig. 7). The amino-terminal one-third of the molecule is very resistant to proteolysis, which may be due to its interaction with 2 within the AP-2 core.

The 2- and YXXØ-binding domains are separated by an intervening region (at a minimum comprising amino acids 146-163) that does not appear to be required for either interaction. Comparison of adaptor medium chain sequences reveals that the 146-163 segment corresponds to a stretch of low similarity (Fig. 8, bracket in upper graph). These observations suggest that the segment encompassing amino acids 146-163 may act as a linker between the 2- and YXXØ-binding domains of µ2. Linker sequences often appear as loops in the three-dimensional structure of proteins (34). This may also be the case for the 146-163 segment, as analysis of secondary structure predicts a high probability of a loop in that region (Fig. 8, bracket in lower graph).

Another region of low sequence similarity in the adaptor medium chains occurs at amino acids 220-250 of µ2 (Fig. 8, upper graph). This region also contains a trypsin-sensitive site that was reported previously by Matsui and Kirchhausen (33) and confirmed in this study (Fig. 7). We speculate that this region might constitute another linker connecting two subdomains, both of which are required for YXXØ binding. The inability to trim further the YXXØ-binding region (residues 164-412) without losing binding activity may be due to the involvement of both of these subdomains in interactions with signals.

The µ1 and µ2 chains are members of a growing family of homologous coat proteins that also includes µ3A and µ3B (formerly known as p47A and p47B; Ref. 35), -COP (36-38), and ARP-2 (39). µ3A and µ3B are components of the recently described AP-3 complex (21, 23, 40, 41) and share ~30% identity with µ1 and µ2 (35). Like µ1 and µ2, µ3A and µ3B bind YXXØ-type signals (11, 40). Also by analogy to µ1 and µ2, we expect that µ3A and µ3B will interact with the 3A (21, 23) and 3B (-NAP; Ref. 24) chains of AP-3, both of which are homologous to 1 and 2. -COP is a component of the COPI coat (38) and exhibits ~20% identity to µ1 and µ2. It is currently unknown whether -COP interacts with any signals; however, two-hybrid analyses have shown that it interacts with -COP, the chain of COPI related to 1 and 2 (38). While still fragmentary, all of this evidence suggests that the µ and  chain homologs of various coat protein complexes might be involved in similar types of interaction. Given that the homology among µ chain family members extends throughout their entire polypeptide chains, we anticipate that they will have a similar functional domain organization as well as a similar arrangement within their respective complexes.

The mode of assembly exemplified by µ- interactions may in fact apply to other interactions within the adaptor complexes. Indeed, the small chains 1 and 2 have been shown to interact specifically with the  and  chains of AP-1 and AP-2, respectively (4). Strikingly, all members of the small chain family described to date (1, 2, 3A, 3B, and -COP) bear low but significant homology to the amino-terminal half of the medium chains (37, 40, 42). This suggests that a large portion of the small chains may be dedicated to interactions with a member of the -- family, perhaps leaving a short carboxyl-terminal extension available for some other function.

On the basis of the results presented here, we propose that the µ2 chain is anchored to the AP-2 core via interaction of 2 with an amino-terminal segment comprising approximately one-third of the µ2 polypeptide chain. The remaining two-thirds of µ2 likely project outward from the AP-2 core, placing this domain of µ2 in a position to interact with tyrosine-based sorting signals. Our data also suggest that µ1 may be similarly arranged within the AP-1 complex, and that this structural organization of the medium chains may be a general feature of all adaptor complexes.