Conservation and Diversification of Dileucine Signal Recognition by Adaptor Protein (AP) Complex Variants*

The clathrin-associated, heterotetrameric adaptor protein (AP) complexes, AP-1, AP-2, and AP-3, recognize signals in the cytosolic domains of transmembrane proteins, leading to their sorting to endosomes, lysosomes, lysosome-related organelles, and/or the basolateral membrane of polarized epithelial cells. One type of signal, referred to as “dileucine-based,” fits the consensus motif (D/E)XXXL(L/I). Previous biochemical analyses showed that (D/E)XXXL(L/I) signals bind to a combination of two subunits of each AP complex, namely the AP-1 γ-σ1, AP-2 α-σ2, and AP-3 δ-σ3 hemicomplexes, and structural studies revealed that an imperfect variant of this motif lacking the (D/E) residue binds to a site straddling the interface of α and σ2. Herein, we report mutational and binding analyses showing that canonical (D/E)XXXL(L/I) signals bind to this same site on AP-2, and to similar sites on AP-1 and AP-3. The strength and amino acid requirements of different interactions depend on the specific signals and AP complexes involved. We also demonstrate the occurrence of diverse AP-1 heterotetramers by combinatorial assembly of various γ and σ1 subunit isoforms encoded by different genes. These AP-1 variants bind (D/E)XXXL(L/I) signals with marked preferences for certain sequences, implying that they are not functionally equivalent. Our results thus demonstrate that different AP complexes share a conserved binding site for (D/E)XXXL(L/I) signals. However, the characteristics of the binding site on each complex vary, providing for the specific recognition of a diverse repertoire of (D/E)XXXL(L/I) signals.

X-ray crystallographic analyses have shed light on the structural basis for the interactions of YXXØ and (D/ E)XXXL(L/I) signals with the AP-2 complex (18 -20). Both binding sites are located on the AP-2 "core," a domain formed by the amino-terminal regions of ␣ and ␤2, and the entire 2 and 2 subunits (Fig. 1). The YXXØ-binding site comprises two hydrophobic pockets on 2 that accommodate the Y and Ø residues of the signals (18). The binding site for (D/ E)XXXL(L/I) signals likely corresponds to that of a dileucinecontaining "Q-peptide" from CD4, (RMpSQIKRLLSE), which was recently identified by Kelly et al. (20). The Q-peptide does not strictly conform to the definition of a (D/E)XXXL(L/I) signal, although the Gln residue at position Ϫ4 or the phos-phorylated Ser residue at position Ϫ5 in the CD4 peptide could behave similar to the Asp or Glu residue at position Ϫ4 in the canonical signals (the first critical leucine is considered position 0). This site consists of hydrophobic pockets on 2 that fit the Leu and (L/I) residues, and a basic patch straddling the boundary of ␣ and 2 that might bind the Gln, phosphorylated Ser, or (D/E) residues of the signals (20). The AP-2 core occurs in two conformations: an inactive conformation in which both signal-binding sites are occluded by interaction with different parts of ␤2 and an active conformation in which both binding sites are exposed on a surface that is coplanar with a PtdIns-4,5-P 2 -binding site on ␣ (21). This conformational change is triggered by phosphorylation of a threonine residue in 2 (22) and results in increased affinity of AP-2 for PtdIns-4,5-P 2 -enriched domains of the plasma membrane, thus allowing simultaneous recognition of both types of signal (23). The exact location of the signal-binding sites on AP-1 and AP-3, and the mechanistic details of the interactions have not been determined.
Herein, we report the use of site-directed mutagenesis and Y3H assays to map the binding sites for (D/E)XXXL(L/I) signals on AP-1 and AP-3. We find that these AP complexes have a (D/E)XXXL(L/I)-binding site similar to that of AP-2. Analysis of the fine specificity of interactions of various (D/ E)XXXL(L/I) signals with AP-1, AP-2, and AP-3 subunits, however, reveals both signal-and AP-complex-dependent differences. We also investigated the assembly of different AP-1 subunit isoforms in cells and the ability of these isoforms to recognize (D/E)XXXL(L/I) signals. We demonstrate that heterotetramers containing all possible combinations of ␥ and 1 isoforms, with the exception of ␥2-␤1-1-1C, are assembled in vivo. Finally, we show that the AP-1 ␥1-1A, ␥1-1B, and ␥1-1C hemicomplexes recognize all (D/ E)XXXL(L/I) signals tested, whereas ␥2-1A and ␥2-1B have a more restricted specificity. These findings indicate that AP-1, AP-2, and AP-3 share a conserved binding site for (D/E)XXXL(L/I) signals, albeit with distinct specificity determined by the exact nature of the signal, as well as the particular AP complex and subunit isoforms involved in the interactions. Based on these observations, we argue that a proper definition of AP complexes must include the specific composition of subunit isoforms.
Northern Blot Analysis-Two commercial nylon membranes (Human 12-lane MTN TM Blot and Human MTN TM Blot III, Clontech, Mountain View, CA) with immobilized human poly(A) ϩ RNA (1 g) from different tissues were used for Northern blot analysis. Membranes were hybridized with a 32 P-labeled human 1C probe (splice variant 1) prepared using the Megaprime TM DNA labeling system (GE Healthcare) and [␣-32 P]dCTP (GE Healthcare).
Y3H Analysis-Y3H vector construction and assays were performed as described previously (see Fig. 2A) (15,27). Double transformants were selected in medium lacking leucine, tryptophan, and methionine but containing histidine (ϩHis) and subsequently plated in ϩHis medium, as well as in the same medium lacking histidine (ϪHis). Interactions between dileucine signals and AP subunits result in activation of the GAL4 promoter and activation of HIS3 gene transcription as evidenced by growth on ϪHis plates. Experiments were also performed on ϪHis plates containing 3-amino-1,2,4-triazole (3-AT), a competitive inhibitor of the His3 protein to minimize background growth due to nonspecific interactions and also to detect differences in the avidity of interactions. Parallel plating on ϩHis plates provided a control for loading and viability of double transformants. The schematic shows the subunit composition and isoforms of the four AP complexes (for review, see Ref. 1). Combinatorial assembly of the various subunit isoforms could result in up to twelve AP-1 complexes, four AP-2, eight AP-3, and one AP-4. The inclusion of AP-1 ␤1 as an AP-2 subunit is based on the observed formation of ␤1-containing AP-2 complexes upon knockdown of AP-2 ␤2 (5) or disruption of the corresponding gene (6). The AP complexes have been represented according to the structures of the AP-1 and AP-2 core complexes (47,50) and of the ear domains of AP-1 ␥ (51, 52), AP-2 ␣ (53-54) and AP-2 ␤2 (55). The schematic depicts a core comprising the trunk domains of the large subunits (␥, ␣, ␦, or ⑀ and ␤1-␤4 for AP-1, -2, -3, or -4, respectively) together with the corresponding medium () and small subunits (). The hinge and ear domains of the large subunits are shown protruding from the core of the complexes (see the AP-4 schematic). The depiction of two subdomains (an N-terminal IgG-like ␤ sandwich and a C-terminal platform) in the ear domains of AP-1 ␤1, AP-3 ␦, AP-3 ␤3, and AP-4 ⑀ is based on alignment with AP-2 ␣ and ␤2 subunits and secondary structure predictions. The prediction of a single C-terminal platform in AP-4 ␤4 is based on the lack of conservation of the N-terminal IgG-like ␤ sandwich in this subunit as compared with the corresponding subunits in other AP complexes.
Metabolic Labeling and Immunoprecipitation-recapture Analysis-Transfected cells (two 150-mm plates per condition) were subjected to metabolic labeling for 12-15 h using EasyTag Express TM 35 S protein labeling mix (PerkinElmer Life Science, Waltham, MA). Preparation of Triton X-100 extracts and immunoprecipitation-recapture experiments were performed as described previously (28,29). The composition of the solubilization and immunoprecipitation buffer was 50 mM Tris-HCl, pH 7.4, 1% Triton X-100, 300 mM NaCl, 5 mM EDTA supplemented before use with 10 mM iodoacetamide, 2 g/ml leupeptin, and 1 mM 4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride. The immunoprecipitates bound to protein A-or protein G-Sepharose beads were denatured in 100 mM Tris-HCl pH 7.4, 1% SDS, 10 mM DTT, diluted ϳ20-fold with immunoprecipitation buffer and subjected to an additional round of immunoprecipitation (recapture step). The recapture beads were dissolved in Laemmli sample buffer and analyzed by SDS-PAGE and phosphorautoradiography (Typhoon 9200 PhophorImager, GE Healthcare).
Mutational Mapping of Binding Site for Canonical (D/E)XXXL(L/I) Signals on 2-We next examined the structural determinants for interaction of AP-2 with the (D/ E)XXXL(L/I) signals described above. The crystal structure of the AP-2 core in complex with the dileucine-containing Qpeptide from CD4 showed that most of the residues involved in the interaction were on the 2 subunit ( Fig. 3A) (20). To determine whether these 2 residues were required for interaction with canonical (D/E)XXXL(L/I) signals, we mutated them and tested for interaction of the mutant proteins with the Nef and tyrosinase signals using the Y3H assay (Fig. 3B). The results showed that several 2 mutations, including A63D, V88D, E100A, and L103S, largely abolished the interaction of ␣C-2 with both signals (Fig. 3B). Other 2 mutations affected the interactions with Nef and tyrosinase differently. For instance, the N92A and L101A mutations had no binding domain vector pBridge, whereas the small AP subunits (1, 2, 3, or 4) were subcloned into MCS2 of the same vector. The sequences encoding full-length mouse ␥1, human ␥2, rat ␣C, human ␦, or human ⑀ were subcloned into the GAL4 activation domain vector pGADT7. ori, origin of replication; Amp, ampicillin resistance gene; NLS, nuclear localization signal. B, sequences of the HIV-1 Nef (NL4 -3 variant) flexible loop and mouse tyrosinase and human LIMP-II cytosolic tails with signals conforming to the (D/ E)XXXL(L/I) are underlined. The diaspartate motif at ϩ10 and ϩ11 from the HIV-1 Nef ENTSLL dileucine signal that is required for binding to AP-2 (35) is also underlined. The mouse tyrosinase cytosolic tail contains a second (D/E)XXXL(L/I) motif (DYHSLL) C-terminal to the ERQPLL shown in the schematic. Although this second sequence is present in mouse and rat tyrosinase, it is not conserved in other species such as humans, is not involved in lysosomal/melanosomal sorting (56), and is not required for interaction with AP complexes (34). C, all (D/E)XXXL(L/I) signals tested interact with AP-1, AP-2, and AP-3 but not AP-4. Double transformants were plated in medium lacking histidine, leucine, tryptophan, and methionine (ϪHis), to detect interaction among constructs, and in medium lacking only leucine, tryptophan, and methionine (ϩHis) as a control for loading and viability of double transformants. In this experimental set, the ϪHis plates were supplemented with a low concentration (0.2 mM) of 3-AT (a competitive inhibitor of the His3 protein) to minimize background growth due to nonspecific interactions. The lack of interaction of AP-4 with the various (D/E)XXXL(L/I) signals was also observed in ϪHis plates lacking 3-AT. The image shown represents a composite of different plates from the same experiment. Results shown are representative of at least three experiments with similar results. Tyrase, tyrosinase. For details, see "Experimental Procedures." effect on the interaction with the Nef signal but decreased the interaction with the tyrosinase signal, particularly as seen in the presence of 1 mM 3-AT (Fig. 3B). From these results, we concluded that canonical (D/E)XXXL(L/I) signals bind to the same site as the CD4 Q-peptide but exhibit distinct requirements for specific residues within the binding site. The lower sensitivity of the Nef dileucine signal to 2 substitutions may be due to a stabilizing effect of the previously described diaspartate motif at ϩ10 and ϩ11 from the ENTSLL (Fig. 2B), which is absent in the tyrosinase signal and has been proposed to interact with a basic patch comprising Lys 297 and Arg 340 on the ␣ subunit ( Fig. 3A) (27).
Identification and Characterization of (D/E)XXXL(L/I)binding Site on 1 and 3-We next sought to determine whether AP-1 and AP-3 have a (D/E)XXXL(L/I)-binding site similar to that on AP-2. The 2 residues that participate in the interaction with (D/E)XXXL(L/I) signals are conserved on 1A and 3A (Fig. 4A). We therefore tested the effect of mutating these 1A and 3A residues on the interaction with the Nef and tyrosinase signals in Y3H assays. We found that most of these mutations affected the interactions of the corresponding ␥1-1A and ␦-3A hemicomplexes with both signals (Fig. 5, A and B). The specific requirements for interaction, however, were signal-and adaptor-dependent. As shown above for ␣C-2, the interactions of ␥1-1A and ␦-3A with the Nef signal (Fig. 5, A and B, left panels) were generally less sensitive to substitutions than the corresponding interactions with the tyrosinase signal (Fig. 5, A and B, right panels). In addition, the interaction of both the Nef and tyrosinase signals with ␥1-1A (Fig. 5A) was stronger and less sensitive to substitutions than the corresponding interactions with ␦-3A (Fig. 5B). In this context, whereas the interaction with the signals was reduced by most substitutions in 3A with the exception of L107A (for Nef) and D98A (for tyrosinase) (Fig.  5B), the interaction with 1A was abolished only by V88D and I103S (for Nef) and also by A63D (for tyrosinase) (Fig.  5A). Among the substitutions that did not abolish binding of the Nef and tyrosinase signals to the ␥1-1A hemicomplex was 1A R15E (an effect on Nef was only detected in the presence of 1 mM 3-AT) (Fig. 5A). This substitution involves a basic residue equivalent to 2 Arg 15 which, along with ␣C  (20). To this end, we compared the binding of the Nef, tyrosinase, and LIMP-II signals to the ␥1 R15E, ␣C R21E, and ␦ R26E mutants and the corresponding wild-type proteins as hemicomplexes with 1A, 2, and 3A, respectively. The results showed that substitution of these basic residues generally decreased interactions with the signals (Fig. 6). The effects were more marked for the ␥1 R15E and ␦ R26E mutations than for the ␣C R21E mutation and were also more noticeable for the interaction with the Nef and LIMP-II signals than with the tyrosinase signal (Fig. 6). In addition, we observed that the effects were smaller for the ␣C R21E mutation than for the 2 R15E mutation (Fig. 6B), whereas similar reductions were observed when comparing the ␦ R26E and 3A R15E mutations (Fig. 6C). This suggests that basic residues on both the ␣ and 2 subunits of AP-2 and the ␦ and 3 subunits of AP-3 contribute to the electrostatic interaction with the acidic residue at position Ϫ4 of the signals. This is in line with the observation that mutation of both basic residues in the patch (␣C Arg 21 and 2 Arg 15 ) is necessary to decrease binding of the AP-2 core to the CD4 Q-peptide (20). In contrast, mutation of ␥1 R15E caused a much greater reduction in binding to the (D/E)XXXL(L/I) signals than mutation of 1A

. Analysis of AP-2 residues involved in the interaction with Nef and tyrosinase (D/E)XXXL(L/I) signals.
A, residues in the ␣ (Arg 21 ) and 2 (Ala 63 , Val 88 , Asn 92 , Glu 100 , Leu 101 , and Leu 103 ) subunits that were subjected to mutagenesis are shown in black on the surface representation of the three-dimensional structure of the AP-2 core complex (Protein Data Bank codes 1GW5 and 2VGL) (50). The ␣, ␤2, 2, and 2 subunits are depicted in light blue, green, magenta, and gold, respectively. Shown in red on the ␣ subunit are residues Lys 297 and Arg 340 , which are also required for the binding of the AP-2 ␣-2 hemicomplex to HIV-1 Nef (27). B, Y3H assays showing the effect of 2 substitutions on the interaction of the AP-2 ␣-2 hemicomplex with HIV-1 Nef and tyrosinase (Tyrase) (D/E)XXXL(L/I) signals. Experiments were performed as indicated in the legend to Fig. 2. Positive controls included the interaction of (D/E)XXXL(L/I) signals with the AP-1 ␥-1 and AP-2 ␣-2 hemicomplexes, whereas double transformants expressing (D/E)XXXL(L/I) signals and discordant ␥ and 2 pairs were used as negative controls. Double transformants were plated on ϪHis and ϪHis plus 1 mM 3-AT medium to analyze the interactions at different levels of stringency and on ϩHis medium as a control for loading and viability. The image shown represents a composite of different plates from the same experiment.
Arg 15 (Fig. 6A), indicating that the basic residue contributed by ␥1 is the most critical for interaction with the (D/E) residue of the signals. A summary of the results of the mutational analyses is shown in Fig. 7.
Assembly of AP-1 Complexes Containing Different ␥ and Subunit Isoforms-Combinatorial assembly of AP subunit isoforms could generate additional diversity in the recognition of sorting signals. This is particularly the case for the AP-1 complex, which could occur in at least 12 varieties (Fig.  1). Of all of the AP-1 subunit isoforms, 1B is the only one that is restricted to a particular cell type, polarized epithelial cells (36), whereas the other isoforms appear to be widely expressed in mammalian tissues and cells (36 -39). Thus, most cells could have up to six different AP-1 complexes containing common ␤1 and 1A subunits and variable ␥1/␥2 and 1A/ 1B/1C isoforms. This potential heterogeneity begs the questions of whether all of these combinations do assemble in vivo and whether they exhibit any differences in the recognition of (D/E)XXXL(L/I) signals.
We used a biochemical approach to examine the formation of multiple AP-1 variants comprising different ␥ and 1 isoforms. To this end, we stably transfected M1 human fibroblasts with HA epitope-tagged 1A, 1B, or 1C constructs. Following metabolic labeling with [ 35 S]methionine, cell lysates were subjected to immunoprecipitation with antibody to the HA epitope and recapture of the solubilized immunoprecipitates with antibodies to other AP subunits. The results of these experiments showed that all three 1-HA isoforms are incorporated into AP-1 complexes containing ␥1, ␤1, and 1 subunits but not into complexes including the ␣C subunit of AP-2, the ␤3A subunit of AP-3, or the ⑀ subunit of AP-4 (Fig.  8, A-C). A similar analysis of M1 cells stably transfected with HA-tagged ␥2 showed that this subunit is also specifically incorporated into AP-1 complexes (Fig. 8D). In addition, we subjected these HA-␥2 stable transfectants to transient transfection with Myc-epitope-tagged 1A, 1B, or 1C. The immunoprecipitation-recapture analysis showed that, whereas the ␥1 isoform was incorporated into AP-1 complexes containing either 1A, 1B, or 1C, the ␥2 isoform was incorporated into AP-1 complexes containing 1A or 1B, but not 1C (Fig. 8E). These experiments thus demonstrated that, of the six possible AP-1 complexes arising from combinations of different ␥ and 1 subunits, five can be formed in the transfected M1 cells.
Interaction of (D/E)XXXL(L/I) Signals with AP-1 Hemicomplexes Containing Different ␥ and Isoforms-We next analyzed the ability of hemicomplexes that have different combinations of ␥ and 1 subunit isoforms to interact with (D/E)XXXL(L/I) signals using Y3H assays. The results showed that all AP-1 hemicomplexes containing ␥1 (␥1-1A, ␥1-1B, and ␥1-1C) interact with similar avidities with the Nef, tyrosinase, and LIMP-II signals (Fig. 9). In contrast, the ␥2-1A and ␥2-1B hemicomplexes displayed signal-dependent interactions; they interacted very weakly with Nef, robustly (similar to the activity of the ␥1-1Aand ␥1-1B-containing complexes) with the tyrosinase signal, and not at all with the LIMP-II signal (Fig. 9). The absence of interactions with (D/ E)XXXL(L/I) signals observed in yeast transformed with vectors encoding ␥2 and 1C (Fig. 9) is consistent with the biochemical experiments demonstrating the lack of assembly of complexes containing this combination of subunits (Fig. 8E).

DISCUSSION
The results of our analyses demonstrate that canonical (D/ E)XXXL(L/I) signals bind to the same site on AP-2 as the noncanonical CD4 Q-peptide (20) and that AP-1 and AP-3 have a  (20). Alignments were generated with CLC Sequence Viewer; decreasing conservation of residues is shown by red to blue rainbow coloring. similar binding site. This is evidenced by the loss of signal binding by the 2 V88D or L103S substitutions and the homologous 1A V88D and I103S and 3A V94D and L109S substitutions. Nonetheless, there are also differences in the interactions that are dependent on both the signals and adaptors (see Fig. 7). The approach that we used to assess differences in (D/E)XXXL(L/I) signal recognition is the substitution of residues on AP-1 and AP-3 that are equivalent to those important for the interaction of AP-2 with the CD4 Q-peptide. Our results show that the binding of these signals to AP-1 is less sensitive to substitution of single residues in its putative binding site than the corresponding interactions with AP-2 and AP-3. Two residues that exemplify these differences are 1A Arg 15 and Leu 101 , which can be substituted with relatively little impact on the ability of ␥1-1A to recognize (D/E)XXXL(L/I) signals. In contrast, substitution of the corresponding residues in 3A (Arg 15 and Leu 107 ) decreased the interaction of (D/E)XXXL(L/I) signals with ␦-3A (especially in the case of tyrosinase). A mixed outcome, inhibition by 2 R15E and (D/E)XXXL(L/I) signal-dependent effects with 2 L101A, was observed when analyzing the homologous substitutions in 2.
The differences in the interaction of (D/E)XXXL(L/I) signals with AP complexes also extend to the interaction of the   . The effect of substitutions is depicted ranging from red (binding completely abolished) to violet (no effect) in a rainbow gradient (see relative color gradient below C). A "blue shift" (lower sensitivity to mutations) can be observed for 1A when comparing the effects of substitutions on this subunit with those at equivalent positions in 2 or 3A. C, relative effect of ␥ R15E, ␣C R21E, and ␦ R26E substitutions on the interaction with HIV-1 Nef, tyrosinase, and LIMP-II signals based on the results in Fig. 6. The effect of substitutions is depicted as indicated for A and B. Tyrase, tyrosinase.
(D/E) residue with basic residues on both subunits of the hemicomplexes. We found that interaction with ␥1-1A depends mainly on ␥1 Arg 15 , whereas interactions with the other hemicomplexes involve basic residues on both subunits, namely ␣ Arg 21 and 2 Arg 15 in ␣-2 and ␦ Arg 26 and 3A Arg 15 in ␦-3A (Fig. 6). The requirement of residues in two subunits of each complex (␥ and 1 for AP-1, ␣ and 2 for AP-2, and ␦ and 3 for AP-3) explains why it was necessary to use Y3H assays to detect interactions of (D/E)XXXL(L/I) signals with AP subunits (15). From these experiments, we con-clude that although AP-1, AP-2, and AP-3 share a similar binding site, residues in both the signals and the adaptors make different contributions to the interactions, thereby defining the fine specificity of signal recognition events.
Genetic studies have begun to address the physiologic significance of the existence of AP subunit isoforms. AP-1 exhibits the greatest diversification of subunit isoforms because vertebrates express two ␥ (␥1 and ␥2; both ubiquitous), two 1 (1A and 1B; the first ubiquitous, the second epithelialspecific), and three (1A, 1B, and 1C) paralogs (1). Targeted disruption of the 1A gene in mouse causes embryonic lethality at day 13.5 (during mid-organogenesis) and only polarized epithelial cells from early embryos exhibit membrane binding of AP-1, likely due to the expression of 1B (40). Inactivation of the ␥1 gene is also lethal and results in death of mouse embryos prior to implantation (41), indicating that ␥2 is unable to compensate for the lack of ␥1 expression. Recent findings also support specific requirements for the three 1 isoforms (supplemental Fig. 1A). A mutation in the human 1A gene leading to premature translation termination causes the neurocutaneous MEDNIK syndrome (mental retardation, enteropathy, deafness, peripheral neuropathy, ichthyosis, and keratoderma) (42). On the other hand, human 1B deficiency causes an X-linked mental retardation syndrome characterized by basal ganglia calcifications and elevated protein levels in cerebrospinal fluid (43)(44)(45). Moreover, mice deficient in 1B are viable and fertile but display altered synaptic vesicle recycling in hippocampal synapses, reduced motor coordination and long-term spatial learning and memory deficiencies (39). The 1C isoform was first identified in a bioinformatics analysis (1) and is also expressed in a variety of mouse (39) and human tissues (supplemental Fig. 1C). Interestingly, three splice variants of 1C exhibiting differences at their C termini Cell lysates were subjected to immunoprecipitation (IP) with anti-HA, and the immunoprecipitates were denatured, diluted, and subjected to additional rounds of immunoprecipitation (RC, recapture) using antibodies against the HA epitope, or 1, ␤1, ␥1, ␣, ␤3, or ⑀ subunits (A-C) or 1A, 1, ␤1, HA, ␥1, ␣, ␤3, or ⑀ subunits (D). The recaptured immunoprecipitates were subjected to SDS-PAGE and fluorography. The analysis demonstrates that 1A, 1B, and 1C subunits are all incorporated into AP-1 complexes also containing 1, ␤1, and ␥1 subunits, but not into AP-2, AP-3, or AP-4 (lack of recapture of 1A, 1B, or 1C by either anti-␣, -␤3, or -⑀) (A-C). The ␥2 subunit also assembles into AP-1 but not AP-2, AP-3, or AP-4 (D). E, M1 cells stably transfected with HA-␥2 were subjected to transient transfection with vectors driving expression of 1A-Myc, 1B-Myc, or 1C-Myc (a Myc-tagged stonin 2 proline-rich domain (Stn2 PRD) was used as a negative control). Transfected cells were metabolically labeled and lysed, followed by immunoprecipitation with anti-Myc and recapture with either anti-␥1 or anti-HA (for detection of ␥2). Note that ␥1 is incorporated into AP-1 complexes containing any of the three isoforms of 1, whereas ␥2 can only assemble into AP-1 complexes containing either 1A or 1B. In contrast, ␥2-1A and ␥2-1B interact weakly with HIV-1 Nef, interact strongly with the tyrosinase tail, and do not recognize the LIMP-II signal. The lack of interaction detected for all double transformants expressing ␥2 and 1C is consistent with lack of assembly of AP-1 complexes comprising these two subunits, as evidenced by immunoprecipitation-recapture analysis of metabolically labeled cells (Fig. 8). This lack of interaction cannot be explained by lack of expression of ␥2 and 1C subunits in yeast given the binding of the tyrosinase signal to ␥2-1A or ␥2-1B and to ␥1-1C, respectively. The image shown represents a composite of different plates from the same experiment. SV40 L T-Ag, SV40 large T-antigen (negative control for interactions with pBridge constructs and positive control for interaction with p53).
have been identified (supplemental Fig. 1B), but there is no information on the alterations brought about by their lack of expression.
Although these genetic studies have provided information on the requirement of specific AP-1 subunit isoforms at the organismal level, the molecular and cellular bases for these requirements are not known. Moreover, except for 1A and 1B (46), the assembly of different subunit isoforms into AP-1 had not been demonstrated prior to our study. Our observations provide the first assessment of the in vivo assembly of AP-1 heterotetramers formed by different combinations of ␥ and subunits. We found that 1A, 1B, and 1C can all assemble into AP-1 heterotetramers. In addition, we observed that ␥1 is incorporated into complexes containing any 1 isoform, whereas ␥2 can assemble into complexes containing 1A or 1B but not 1C. The structural basis for the incompatibility of ␥2 with 1C is unknown. Although the crystal structure of an AP-1 core complex (trunk of ␥1 and ␤1 subunits along with 1A and 1A) has been solved (47), the contacts between the ␥1 trunk and 1A are too extensive to make inferences based on primary structure differences between ␥1/␥2 and between 1A/1C.
Our experiments also provide the first indication that certain combinations of subunit isoforms have intrinsically different recognition specificities, as exemplified by the differences in (D/E)XXXL(L/I) signal recognition by AP-1 hemicomplexes containing ␥1 or ␥2 (Fig. 9). The inability of ␥2-containing complexes to interact with signals that are recognized by ␥1-containing complexes might explain why ␥1deficient mice are inviable despite the ubiquitous expression of ␥2. What might then be the specific role of ␥2, given the ubiquitous expression of ␥1 and the strong interaction of ␥1based hemicomplexes with dileucine signals? The strong avidity of ␥2-1A and ␥2-1B hemicomplexes for the tyrosinase signal (as opposed to other (D/E)XXXL(L/I)-based signals) ( Fig. 9) suggests that ␥2-containing AP-1 complexes might sort specific cargo to specialized cellular compartments such as melanosomes, a process in which AP-1 has been implicated (48).
AP-1 complex variants containing 1A or 1B have been referred to as AP-1A or AP-1B, respectively (49). In light of the broader repertoire of AP subunit combinations that can be assembled and the differences in the activity of some of these combinations, we submit that, whenever relevant, the proper definition of an AP complex must include information on the specific subunits that constitute it (e.g. AP-1 (␥1-1A-␤1-1B) or AP-1 (␥2-1A-␤1-1A)).