Structural Requirements for Interactions between Leucine-sorting Signals and Clathrin-associated Adaptor Protein Complex AP3*

,

The adaptor protein complexes (APs) 1 AP1, AP2, AP3, and AP4 are important components of the intracellular sorting machinery. They associate with transport vesicles along the secretory and endocytic pathways and can interact with membrane proteins that contain signals for sorting into the appro-priate transport vesicles (1,2). Each AP contains four polypeptides, called adaptins: two large chains of ϳ100 kDa, a medium chain of ϳ50 kDa, and a small chain of ϳ25 kDa. AP1 complexes are associated with transport clathrin-coated vesicles (CCVs) derived from the trans-Golgi network (TGN), AP2 complexes are associated with the endocytic CCVs, whereas the functional localization of AP3 and AP4 is not clear yet. It seems, however, that AP3 mediates sorting to lysosomes and related organelles and mediates sorting on endosomes, whereas AP4 has been implicated in basolateral sorting (3)(4)(5)(6)(7)(8)(9).
Two best characterized types of protein-sorting signals are tyrosine-and leucine-based sorting signals (see Refs. 2, 3, and 10). It is believed that interactions between sorting signals and AP complexes is one of the determining steps in protein sorting, namely the sequestration of cargo membrane proteins into a specific type of transport vesicle for delivery to its intermediate or final destination. Interactions between various sorting signals and AP complexes have been demonstrated in a variety of experimental systems (reviewed in Refs. 1 and 11). All APs seem to recognize tyrosine-sorting signals via their respective medium chain, are associated with specific populations of transport vesicles, and confer distinct sorting properties onto these vesicles (1,3). Interactions between AP complexes and leucine signals were studied less extensively, and different AP subunits have been reported to interact with leucine signals. Kirchhausen and co-workers (12) has reported interactions between several leucine signals and a large (␤ 1 ) chain of the AP1 complex in a cross-linking assay. In our laboratories interactions between leucine signals and medium chains of AP1 and AP2 were observed in phage-display assays (13), in protein-protein interaction assays on magnetic beads (14), and in surface plasmon resonance assays (15).
Two molecules containing leucine-sorting signals are of particular interest for this study: the invariant chain (Ii) and lysosomal integral membrane protein II (LIMPII). Ii is a type II transmembrane protein associated with the major histocompatibility complex (MHC) class II in the endoplasmic reticulum (ER). It is believed that Ii has a multitude of functions that aid MHC class II in antigen presentation. Two of the main functions are to prevent endogenous polypeptides from binding to the MHC class II groove and mediate the sorting of the MHC class II to the endosomes either directly or via the plasma membrane (16 -18). Ii carries two leucine-sorting signals within its cytoplasmic tail, the membrane-distal signal (LI residues in positions 7 and 8), and the membrane-proximal signal (ML residues at positions 16 and 17). Ii is sorted to endosomes via the plasma membrane, and either signal is independently sufficient for endosomal localization of Ii (19,20). LIMPII is a type III transmembrane protein, and has one leucine signal in its 20-amino acid cytoplasmic tail. This signal is necessary and sufficient for direct sorting of LIMPII to endosomes/lysosomes (10,21).
It has previously been shown that sorting signals from Ii interact in vitro with AP1 and AP2, but not to AP3, which is consistent with the in vivo sorting of Ii to endosomes indirectly via the plasma membrane (15). On the other hand, the LI-motif of LIMPII preferentially binds to AP3 and in addition to this in vitro observation, LIMPII is known to be sorted to endosomes/ lysosomes via a direct transport route. which does not include appearance at the cell surface (10,21). Because leucine-based motifs of Ii and LIMPII appear to be remarkably similar to each other (leucine pair and double acidic residues in the positions Ϫ4 and Ϫ5), we investigated the influence of nearby residues on the specificity of interactions between those signals and AP1, AP2, and AP3. We constructed a set of GST fusion proteins containing various chimeric Ii-LIMPII constructs, which were then tested for adaptor binding by surface plasmon resonance. To corroborate our in vitro observations, we also analyzed the intracellular fate of the wild type LIMPII and LIMPII with the invariant chain signal, transfected into MDCK cells.

MATERIALS AND METHODS
DNA Constructs-For the in vitro studies, the cytoplasmic tails of the wild type Ii (Met 1 to Arg 30 ) and its various mutants and wild type LIMPII (Gly 460 to Thr 478 ) and its various mutants (see Fig. 1) were made by PCR from the cytoplasmic tail of both of the proteins. They were then fused in-frame to the C terminus of the GST protein using BamHI and EcoRI sites in pGEX-2T vector (Amersham Biosciences). For the in vivo studies, we received a full-length LIMPII cDNA from I. Sandoval (Madrid), and the construct was subcloned into pMEP4 vector to utilize its inducible promoter and stably transfected into MDCK cells. A LIMPII QRD mutant was constructed by PCR mutagenesis using full-length LIMPII as the template and the primer set, GCT TAG ACG CTA GCG ATG GCC CGA (upper primer) and CCA TTA AAT TGC GGC  CGC TTA GGT TCG TAT GAG GTC TCG TTG TTC ATC TGC AGT  (lower primer). The PCR product was then subcloned into pMEP4 using NheI and NotI sites and stably transfected into MDCK cells.
Antibodies-The mouse monoclonal antibody 29G10 (a kind gift from Dr. I. Sandoval, Universidad Autónoma de Madrid) was used to detect LIMPII as described elsewhere (23). Goat anti-mouse secondary antibodies coupled to Alexa 594 were obtained from Molecular Probes (Leiden, The Netherlands). A rabbit anti-mouse secondary antibody was obtained from DAKO AS (Denmark).
Expression and Purification of GST Fusion Proteins-All fusion proteins were expressed and purified as recommended by the manufacturer (Amersham Biosciences). Briefly, BL21 cells carrying the constructs of interest were induced with 0.25 M isopropyl-1-thio-␤-Dgalacropyranoside for 3 h and collected by centrifugation. The fusion proteins were released by a series of 15-s sonication steps or by Bugbuster kit (Novagen) and then purified on GST-Sepharose (Amersham Biosciences). The purity and size of the proteins were verified by SDS-PAGE. GST, which served as a negative control, and the chimeric constructs were used for adaptor binding measurements after dialysis against BIA buffer (20 mM HEPES pH 7, 150 mM NaCl, 10 mM KCl, 2 mM MgCl 2 , 0.2 mM dithiothreitol).
Stable Transfection of MDCK Cells and Clonal Selection-MDCK cells (strain II) pretransfected with the EEA1 binding domain fused to GFP (24) were stably transfected with various LIMPII constructs by the calcium phosphate procedure of Ref. 25. Clones expressing the DNA constructs in pMEP4 vector under control of metallothionein promoter were selected in the presence of hygromycin B (0.3 mg/ml). Resistant clones were selected and incubated with 25 mM CdCl 2 overnight to induce expression of the protein of interest. Clones that expressed constructs of interest were identified with anti-LIMPII antibodies. 3-4 positive clones with different expression levels of the protein of interest were selected in each case. MDCK cells expressing LIMPII constructs were grown in full growth medium (Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 2 mM glutamine, 25 units/ml penicillin, and 25 g/ml streptomycin) in 6% CO 2 in a 37°C incubator.
Radiolabeling and Immunoprecipitation-For metabolic labeling, transfected MDCK cells were grown to ϳ80% confluence in full medium supplemented with 25 M CdCl2. The cells were washed twice and incubated for 40 min at 37°C in cysteine/methionine-free Dulbecco's modified Eagle's medium medium (Bio Whittaker) and then incubated for 5 h with 50 Ci/ml of [ 35 S]cysteine/methionine. After metabolic labeling, the cells were washed with ice-cold PBS and lysed in ice-cold lysis buffer (1% Triton X-100, 0,1 M NaCl, 1 mM EDTA, 10mM Na 2 HPO 4 ) containing a mixture of protease inhibitors (1 g/ml leupeptin, 0,25 mM phenylmethylsulfonyl fluoride, 2 g/ml antipain, 1 g/ml aprotinin, 1 g pepstatin A, 1 g/ml E-64). The lysate was centrifuged at 13,000 rpm for 10 min at 4°C to remove cell debris. LIMPII was precipitated from the supernatant with 29G10 monoclonal antibody overnight at 4°C. Secondary antibodies, rabbit anti-mouse IgGs coupled to protein A-Sepharose beads (Amersham Biosciences) were added for 2 h at 4°C. Immunoprecipitates were extensively washed. The proteins were eluted from the beads by incubating in gel-loading buffer (50 mM Tris-HCl, pH 7.5, 1% SDS, 10% glycerol, 100 mM dithiothreitol, and 0.1% bromphenol blue) at 95°C for 5 min. Immunoprecipitates were resolved by 10% SDS-PAGE. The bands were detected with the Bio-Rad GS-250 PhosphorImager. The intensity of the bands was quantified with the Molecular Analyst 2.0.2 software (Bio-Rad).
Immunofluorescence Microscopy-Transfected MDCK cells expressing various LIMPII constructs were grown on glass cover slips and fixed in 3% paraformaldehyde (PFA) in PBS for 10 min at room temperature. Fixed cells were incubated for 30 min with a primary antibody followed by a 30-min labeling with a secondary antibody to label total LIMPII. Antibodies were diluted in PBS, 0.1% saponin to label total intracellular protein or in PBS containing no saponin for the experiments where only surface labeling was desired. To monitor internalization of anti LIMPII antibodies from the cell surface, antibodies were added to the cells for 30 min on ice. The cells were then washed with PBS chased in complete medium for different time periods before fixation. Secondary antibodies were then added in PBS with saponin. Fluorescence was detected, and images were acquired using a Leica TCS-NT digital scanning confocal microscope equipped with a 60/1.2 water immersion objective. The pinhole value was kept below 1. The images were processed for presentation with Adobe Photoshop software.
Preparation of AP1 and AP2-AP1 and AP2 were prepared from pig brain essentially as described elsewhere (26). Briefly, clathrin-coated vesicles were purified from brain after homogenization and differential centrifugation. The adaptor proteins were released from clathrin-coated vesicles with 0.5 M Tris/HCL, pH 7.0 and applied to a Superose-6 column (2.5 ϫ 75 cm, equilibrated in the same buffer) connected to an FPLC (Amersham Biosciences) system at a flow rate of 0.3 ml/min. Fractions containing AP1 and AP2 were identified by SDS-PAGE and separated from each other by subsequent hydroxylapatite chromatography as described elsewhere (27). Fractions containing either AP1 or AP2 were dialyzed against BIA buffer (see above), which was used for all experiments using surface plasmon resonance.
Preparation of AP3-For analyzing AP3 binding, pig brain cytosol was fractionated by gel filtration as described (22) using a 60-cm TSK SWXL 3000 size-exclusion column connected to a perfusion chromatography work station (Perseptive Biosystems). The AP3-containing fractions were shown to be devoid of AP1 and AP2 (for details, see Ref. 22).
Surface Plasmon Resonance-The interaction between the different Ii constructs and adaptors was analyzed in real time by surface plasmon resonance (28) using a BIAcore 3000 biosensor (BIAcore AB). Ii constructs were immobilized via their GST moiety of the GST-Ii chimera to the surface of a CM5 sensor chip coated with anti-GST antibodies. The subsequent interaction experiments were performed at a flow-rate of 20 l/min. Association was recorded for 2 min during which adaptor proteins at different concentrations were injected and followed by recording dissociation for 2 min during which buffer was perfused. A short pulse injection (15 s) of 20 mM NaOH, 0.5% SDS was used to regenerate the surface after each experimental cycle. The anti-GST surface retained its binding capacity for at least 15 cycles of association, dissociation, and regeneration. AP1 and AP2 were used at concentrations ranging from 20 to 200 nM. The cytosolic fractions enriched in AP3 were used at three dilutions.
Determination of Kinetic Rate Constants-The association and dissociation constants k a and k d for the interactions were calculated by using the evaluation software of the BIAcore 3000. The mathematical models used are described in more detail elsewhere (27) In brief, the association was determined after 15-20 s following the switch from buffer solution to adaptor solution to avoid distortions due to injection and mixing. The dissociation rate constants were determined after 5-10 s following the switch to buffer solution. After a rapid (ϳ30 s) dissociation phase of adaptor from Ii-GST, the dissociation kinetics decreased to a low rate. The association constant k a , the dissociation constant k d , and the calculation of the equilibrium constant K D ϭ k d /k a were determined by using the BIA evaluation software version 1.2, assuming a first order kinetic A ϩ B ϭ AB. Relative binding values were then calculated from the K D values.
Standard deviations (in percentage of the mean value) for AP1 and AP2 equilibrium constants (K D ), or AP3 affinity were measured relative to the wild type LIMPII construct using the same adaptor batch.

RESULTS
We have previously demonstrated that AP1 and AP2 adaptor protein complexes could interact with the invariant chain in an in vitro assay monitored with surface plasmon resonance technique, and that such interactions were dependent on the intact leucine-sorting signals in the invariant chain (15). We have also shown that two other molecules that contain similar leucine signals, tyrosinase and LIMPII, only poorly interact with AP1 or AP2, but strongly bind to AP3 (15,22). Furthermore, interactions between LIMPII and AP3 and the invariant chain and AP1 and AP2 were dependent on acidic residues upstream of the critical leucine signals (22,29). This latter finding suggested that residues in close apposition to the leucine-based motif could play a critical role in determining adaptor binding. We therefore investigated the role of other residues upstream of the leucine signals for their influence on AP1, AP2, and AP3 binding. To this end, we constructed a set of GST fusion constructs containing various chimeric constructs in which one or several amino acids upstream of the leucine signals were swapped between the invariant chain and LIMPII molecules (Fig. 1) and measured their interactions with the adaptor complexes. Since the invariant chain has two independent leucine signals, we made two sets of constructs. Each set had one of the leucine signals knocked out by an alanine substitution, and amino acid swap was performed upstream of the other, intact signal. Fig. 2, panel A shows the effect of substituting one or several amino acids in the invariant chain membrane-distal signal on AP3 binding. As reported earlier (15), the invariant chain does not interact with AP3, and here we did not observe any interaction between AP3 and the L17A mutant of the invariant chain. We then swapped the three residues between the critical acidic residues and leucine residues, QRD, for the three corresponding residues from the LIMPII molecule, RAP. The resulting construct, L17A, QRD 3 RAP exhibited significant AP3 binding, around 25% of the wild-type LIMPII construct. Next, we made three more invariant chain constructs in which two out of these three residues were swapped with the corresponding residues from the LIMPII molecule. All three of these constructs were able to interact with AP3 ( Fig. 2A). The L17A, QRD 3 RAD chimera retained the levels of AP3 binding exhibited by the triple substitution chimera (25% of wild type LIMPII). The L17A, QRD 3 QAP construct had somewhat decreased AP3 binding capacity, 13% of the wild-type LIMPII, whereas the L17A, QRD 3 RRP construct exhibited twice the amount of AP3 binding compared with the triple substitute construct (around 55% of the wild-type LIMPII binding).
All three single amino acid substitutions were also made. When asparatic acid residue at position Ϫ1 upstream of the critical leucines was substituted with proline (the L17A, QRD 3 QRP construct), this chimera showed some AP3 binding (14% of the wild-type LIMPII). The other two single substitution constructs did not bind to AP3 (0 -4% residual binding), suggesting an important role for the proline residue upstream of the leucine signal in AP3 binding.
We also investigated AP1 and AP2 binding to these L17A constructs. As shown in Fig. 2B, AP1 binding did not decrease dramatically in all but two chimeric constructs and stayed around 80% of that of the L17A construct. These two constructs, L17A, QRD 3 RAD, and L17A, QRD 3 RAP, exhibited AP1 binding of around 20 -25% of that of the L17A construct. Interestingly, the construct that lost most of the AP1 binding capacity, L17A, QRD 3 RRP, also gained the highest AP3 binding. AP2 binding (Fig. 2C) did not change for all but 2 constructs, L17A, QRD 3 QAP and L17A, QRD 3 QRP. These two constructs showed AP2 binding of ϳ60% of the L17A construct. There was no apparent correlation between the loss of AP2 binding and increase in AP3 binding for this set of constructs.
Next, we performed swaps of the residues upstream of the leucine pair between LIMPII and the second, membrane-proximal signal of the invariant chain. It is noteworthy, that in this case only two amino acids had to be swapped, since the membrane-proximal signal of the invariant chain also has a proline residue at the Ϫ1 position from the critical di-leucine (ML) signal. Fig. 3A shows AP3 binding to the set of constructs containing swaps around the membrane-proximal leucine signal of the invariant chain. We found weak AP3 binding to the L7A mutant itself (5% of the wild-type LIMPII binding). The L7A, QLP 3 RAP mutant that had two amino acids swapped had higher AP3 binding (10% of the wild-type LIMPII), and surprisingly, both constructs that had only one residue swapped, L7A, QLP 3 RLP, and L7A, QLP 3 QAP, possessed even higher AP3 binding (20 and 18% of the wild-type LIMPII, respectively). Since the double arginine-containing construct (L17A, QRD 3 RRP) had such a dramatic effect on AP3 binding to the membrane-distal signal, we created a similar swap in the membrane-proximal signal. The resulting construct, L7A, QLP 3 RRP had 55% binding of the wild-type LIMPII, confirming that the double arginine and a proline just upstream of leucine signal confer stronger AP3 binding irrespective of the leucine pair composition (LI or ML leucine pairs). Finally, we investigated the role of upstream acidic residues on AP3 binding. Both LIMPII and membrane-distal signal of the invariant chain contain two acidic residues in the positions Ϫ4 and Ϫ5 of the leucine pair (DE and DD, respectively), whereas the membrane-proximal signal of the invariant chain has only one acidic residue at the Ϫ4 position. We therefore introduced an extra acidic residue at the Ϫ5 position into the membraneproximal signal of the invariant chain together with the swap of residues in positions Ϫ2 and Ϫ3. The resulting construct, L7A, NEQLP 3 DERAP possessed much higher AP3 binding (80% of LIMPII) than the L7A, NERAP construct (also labeled as L7A, QLP 3 RAP) that had only 10% of LIMPII AP3 binding. Therefore, two acidic residues at the Ϫ4 and Ϫ5 positions seem to give rise to a better AP3 binding than a single acidic residue at the Ϫ4 position.
As with the membrane-distal signal, we studied the effect of the amino acid swap upstream of the membrane-proximal signal on AP1 and AP2 binding. As seen from comparing Fig. 3A and Fig. 3B, there is a strong correlation between the loss of AP1 binding and the increase in AP3 binding for all constructs studied. In the case of AP2 binding, a correlation between gain of AP3 binding and loss of AP2 binding was not significant, although two constructs that had the highest AP3 binding (L7A, QLP 3 RRP and L7A, NEQLP 3 DERAP) also showed some reduced AP2 binding (around 65% of the AP2 binding to the L7A construct).
To corroborate our findings on the importance of residues neighboring the leucine signal, we then analyzed amino acids swaps between LIMPII and the invariant chain on the ability of LIMPII to bind AP3, AP1, and AP2. As shown in Fig. 4A, when the RAP residues upstream of the leucine pair were swapped with the QRD residues from the invariant chain, the resulting construct exhibited a dramatic loss in AP3 binding (30% of the wild type LIMPII binding), a decrease comparable to that when the critical leucine residue was knocked out with an alanine (21% of the wild-type LIMPII AP3 binding). Therefore, we can conclude that the three residues upstream of the leucine pair are indeed part of the motif that mediates binding to AP3. The introduction of the QRD sequence into LIMPII molecule gave a slight increase in AP1 binding capacity (20%), compared with the wild-type LIMPII, but this was still only 12% of the wild-type Ii binding. When binding to AP2 was analyzed, the QRD introduction did not change the affinity for AP2 binding of the LIMPII mutant as compared with the wild type LIMPII, suggesting that binding of wild-type LIMPII to AP2 require also other structural elements than the amino acid changes we introduced.
Since the introduction of the QRD sequence into the LIMPII molecule significantly reduced its AP3 binding capacity but increased its AP1 binding capacity in vitro, we decided to investigate the in vivo fate of the respective mutant proteins. Wild type LIMPII has been shown to be transported from the TGN to lysosomes via endosomes (presumably via interactions with AP3 complex) and has not been detected on the plasma membrane (30). We speculated that the reduction in AP3 binding activity upon introduction of the QRD sequence into the wild-type LIMPII might result in the transport of LIMPII to the plasma membrane. Since several studies have shown that overexpression of certain proteins can lead to their general missorting and accumulation on the plasma membrane (31,32), we had to ensure that overexpressing our mutant LIMPII constructs does not cause unspecific cell surface delivery. Fig.  5A shows the immunoprecipitation of LIMPII molecules from the cell lines used for our studies. First, it is notable that expression of the mutant QRD LIMPII molecule was at the same level as the expression of the wild type LIMPII used for control experiments. Fig. 5B shows staining of the cells expressing these levels of the wild type LIMPII. To simplify the detection of endosomal structures, these cells were also transfected with the EEA1-GFP construct, which we have used as an excellent in vivo marker for early endosomes (24,54). For wild-type LIMPII, only intracellular staining was notable, while staining of LIMPII at the plasma membrane was not detectable, in line with the results from the groups of Sandoval and Hoflack (30,34). On the other hand, the LIMPII QRD mutant was detectable in significant amounts at the plasma membrane in addition to the intracellular staining (Fig. 5C). Furthermore, anti-LIMPII antibodies could be internalized from the plasma membrane of the cells expressing the LIMPII QRD protein (Fig. 5E), whereas no antibody uptake was observed in wild-type LIMPII-expressing cells (Fig. 5D). Finally, PFA-fixed, non-permeablized cells expressing either the wildtype LIMPII or the mutant LIMPII QRD construct were labeled with the anti-LIMPII antibody to reveal the surface staining only. As expected, no surface staining was detected in the cells expressing the wild-type LIMPII construct (Fig. 5F), whereas strong membrane staining was observed in cells expressing the mutant LIMPII QRD construct (Fig. 5G). The same results were obtained when these cells expressing the wild-type and the mutant LIMPII constructs were first labeled with the anti-LiMPII antibody on ice, and then fixed with PFA (not shown). In conclusion, the introduction of the QRD sequence into wild-type LIMPII indeed leads to its re-routing via the plasma membrane and internalization, most likely due to the loss of efficient AP3 binding whereas the capability to interact with PM adaptors are maintained. DISCUSSION In this study, we demonstrated that residues immediately upstream of the leucine pair in leucine-based sorting signals can be critical for sorting of the newly synthesized molecules to endosomal/lysosomal pathway. It has been believed for some time that the structure of the classic leucine sorting signal was (A)AXXXLL, where L is leucine, isoleucine, or methionine, A is an acidic residue, and X is any other amino acid residue. This study demonstrates that those "non-critical" residues may also contribute to the determination of adaptor binding, and in fact discriminate between, for example, AP1 and AP3 binding. This would then determine whether the molecules containing various leucine-based signals are sorted directly to endosomes/ lysosomes, or sorted via a different route via the plasma membrane.
According to the data presented in this report, the following residues in the proximity of the leucine pair are important for AP3 binding: proline residue at the Ϫ1 position from the leucine pair, a positively charged residue such as arginine at position Ϫ3, and negatively charged residues at positions Ϫ4 and Ϫ5. Importantly, effects of these residues on AP3 binding appear to be additive.
Indeed, the proline residue at Ϫ1 position confers some AP3 binding to the L7A set of constructs (Fig. 3A), and a single Asp to Pro substitution in the Ϫ1 position of the membrane-distal signal in the context of L17A grants significant (14% of the wild-type LIMPII) AP3 binding to the invariant chain construct. For comparison, the swap of 3 residues at positions Ϫ1, Ϫ2, and Ϫ3 (L17A QRD 3 RAP construct) grants the resulting chimera only 25% of wild-type LIMPII AP3 binding ( Fig. 2A).
A positively charged residue such as arginine at position Ϫ3 has an additive effect on AP3 binding when residues at positions Ϫ1 and Ϫ2 are also swapped, but may not grant AP3 binding properties by itself. Indeed, when arginine is introduced at position Ϫ3 in the membrane-distal signal of the invariant chain (L17A QRD 3 RRD), AP3 binding is not detectable ( Fig. 2A). However, when arginine is introduced in the Ϫ3 position of the membrane-proximal signal (L7A, QLP 3 FIG. 4. Panel A, binding of AP3 to the invariant chain and various LIMPII constructs. All numbers are shown relative to AP3 binding to the wild-type LIMPII-GST fusion construct. Panel B, binding of AP1 to the invariant chain and various LIMPII constructs. All numbers are shown relative to AP1 binding to the invariant chain-GST fusion construct. Panel C, binding of AP2 to the invariant chain and various LIMPII constructs. All numbers are shown relative to AP2 binding to the invariant chain-GST fusion construct. For all three panels, the numbers represent the average value obtained from 3-5 independent experiments and S.D. varied from 2 to 8%. RLP construct, Fig. 3A) significant AP3 binding is observed. We believe that this could be due to the additive effect of arginine at the Ϫ3 position and proline at the Ϫ1 position that is present in the native membrane-proximal signal of the invariant chain. This is further corroborated by introducing both proline at the Ϫ1 position and arginine at the Ϫ3 position of the membrane-distal signal (L17A, QRD 3 RRP construct, Fig.  2A). This construct exhibits much higher AP3 binding than the one in which only the proline residue is introduced at position Ϫ1 (L17A, QRD 3 QRP construct, Fig. 2A).
Interestingly, the invariant chain constructs that contain arginine at positions Ϫ3 and Ϫ2 together with proline at the position Ϫ1 in both membrane-proximal and membrane-distal signals bind AP3 significantly stronger than constructs containing original residues from LIMPII ( Figs. 2A and 3A). At the same time, when proline is not present at the Ϫ1 position, double arginines at the Ϫ2 and the Ϫ3 positions do not confer AP3 binding properties to the invariant chain construct ( Fig.  2A). It is thus possible that the double positive charge together with proline at the Ϫ1 position (which presumably gives the leucine pair flexibility to fit in the recognition site of AP3) is the preferred motif for AP3 recognition.
AP3 binding to the invariant chain chimeras seems to be significantly stronger when double negatively charged residues are present at positions Ϫ4 and Ϫ5 than if there is only one charged residue in the position Ϫ4 (Fig. 3A, compare constructs L7A, QLP 3 RAP and L7A, NEQLP 3 DERAP). It is also of notice that glutamic acid residue at the position Ϫ4 seems to cause stronger AP3 binding than an aspartic acid residue in the same position (compare constructs L17A, QRD 3 RAP, 25% relative AP3 binding and L7A, NEQLP 3 DERAP, 80% relative AP3 binding).
We compared the structure of other molecules containing leucine signals that are known to interact with AP3 complex, or are transported directly to endosomes/lysosomes/melanosomes/ vacuoles, presumably, via interactions with AP3 (Fig. 6, data for the figure are taken from Refs. 35 and 36 -39). The overwhelming majority of these molecules contain proline at position Ϫ1, such as tyrosinase and TRP1 families. A noticeable exception is yeast protein ALP that contains arginine at this position. Yeast adaptor protein complexes are somewhat different from mammalian APs, and it is possible that yeast AP3 recognition site differs from that of mammalian AP3s, although another yeast vacuolar protein, Vamp3p has a proline residue in its Ϫ1 position.
Interestingly, the majority of the depicted proteins contain a charged residue (in most cases arginine) at position Ϫ3 and a glutamic acid residue (rather than aspartic acid residue) at position Ϫ4. TRP1 does not have an acidic residue in position Ϫ4, but has glutamic acid residue at position Ϫ5 and another acidic residue at position Ϫ6 (Fig. 6). The general structure of a leucine signal that is recognized by mammalian AP3 is therefore likely to be (D/E)E(R/K)XPLL.
This study also revealed a strong correlation between the loss of AP1 binding and acquisition of AP3 binding in the invariant chain/LIMPII chimeric constructs (compare Figs. 2, A  with B, and 3, A with B). We therefore believe that the following may be the basis for sorting at the TGN: some signals with higher affinity for AP3 get sorted in the cargo vesicles destined for direct delivery to endosomes/lysosomes, and other signals have higher affinity for AP1 and are sorted to their destination in AP1-containing cargo vesicles. Previous studies (22) have also shown lack of correlation between the loss of AP2 binding and acquisition of AP3 binding for tyrosinase, and this agrees well with the results of this study. The overall picture of sorting at TGN is likely to be more complicated, as another set of FIG. 5. Sorting of wild type and mutant LIMPII. Panel A, equal expression of the wild-type LIMPII and LIMPII QRD constructs in MDCK cells. Cells expressing wild-type and mutant LIMPII were induced overnight with Cd 2ϩ to ensure the appropriate level of protein expression, starved for 30 min, and labeled for 5 h with [ 35 S]methionine/cysteine. Cells were then lysed, and total cellular protein was quantified with a beta-counter for each lysate. Equal amounts of total protein were then subjected to immunoprecipitation with anti-LIMPII antibody 29G10. Immunoprecipitates were subjected to SDS-PAGE. Lane 1, wild-type LIMPII; lane 2, control (non-transfected MDCK cells); lane 3, LIMPII QRD construct. Panels B-E, staining of cells expressing the wild-type LIMPII and LIMPII QRD construct. Stably double-transfected MDCK cells expressing EEA1-GFP (B-E) were grown on cover slips, and incubated with 25 M CdCl 2 overnight to express either wild-type LIMPII (B and D) or LIMPII QRD (C and E). B and C, cells were fixed in 3% PFA, made permeable with saponin, and labeled with 29G10 and subsequently with Alexa 594conjugated rabbit anti-mouse IgG. D and E, cells were incubated with 29G10 for 30 min on ice and then incubated 30 min at 37°C, fixed in 3% PFA, made permeable with saponin, and stained with Alexa 594-conjugated rabbit anti-mouse. F and G, cells were fixed in 3% PFA and labeled with 29G10 and subsequently with Alexa 594conjugated rabbit anti-mouse IgG. Only the plasma membrane staining was therefore detectable. Red channel, LIMPII; green channel, EEA1-GFP. molecules, so-called GGAs (40 -43) have recently been demonstrated to be involved at this sorting step as well. Importantly, GGAs have been recently demonstrated to interact with leucine signals (44 -48). AP1B and AP4 complexes may also be involved in sorting from the TGN, as they have been implicated in basolateral sorting of some molecules to in polarized cells (5,49). Further studies are required to determine which molecules are sorted via interactions with AP1, AP3, AP4, or GGAs.
At the same time, no correlation was apparent between the loss of AP2 binding and acquisition of AP3 binding in the same set of constructs (compare Figs. 2, A with B, and 3, A with B). Previous studies from our laboratories have also demonstrated relative non-specificity of AP2 for interactions with various sorting signals (29,50). A possible biological role of such relative unspecificity could be to ensure an efficient internalization of all proteins that contain leucine-based signals from the plasma membrane and thus guarantee the clearance of proteins that may have escaped to the cell surface as a result of missorting.
Our studies of substitutions in the invariant chain constructs were further corroborated by reverse substitution in LIMPII molecule. When residues at positions Ϫ1, Ϫ2, and Ϫ3 in LIM-PII were swapped for the appropriate residues from the membrane-distal signal of the invariant chain, the resulting construct exhibited increased AP1 binding and significantly reduced AP3 binding in vitro. In vivo, such construct appeared to be missorted to the plasma membrane instead of the direct delivery to endosomes/lysosomes.
It is important to mention the comprehensive studies performed by the groups of Bonifacino and co-workers (51,52), which revealed distinct preferences in recognition of various tyrosine-sorting signals by medium chains of the four AP complexes using the two-hybrid system. These studies demonstrated that different medium chains prefer tyrosine signals with different residues in positions ϩ1 and ϩ3 from the critical tyrosine residue, whereas the requirements for the residue in ϩ2 position were similar, as the medium chains of all four AP complexes seem to prefer proline residue at this position. Höning and co-workers (50) have recently reported the effect of swapping residues in the positions ϩ1 and ϩ2 between lysosomal proteins LAP and lamp I. The lamp 1 molecule has previously been shown to interact with both AP1 and AP2, whereas LAP interacted with AP2 only (53,33). Introduction of lamp I residues from the positions ϩ1 and ϩ2 into respective positions of LAP caused the latter molecule to interact with AP1 in vitro, and get transported to lysosomes directly rather than via recycling pathway, whereas reverse substitution into lamp I made it go into endosomes/lysosomes via recycling pathway instead of direct lysosomal pathway (50). In both types of sorting signals, leucine-and tyrosine-based, the critical surrounding residues play an essential role in the sorting events as they may determine the specific adaptor binding affinity. To classify sorting signals using the double leucine or tyrosine as a primary criteria may thus be rather misleading or an oversimplification as the tyrosine and leucine signals overlap in their binding ability to the same adaptors and the adaptor specificity is determined by surrounding residues.
FIG. 6. Molecules containing leucine-sorting signals that are known to interact with AP3 or be targeted directly to endosomes/lysosomes/melanosomes/vacuoles. Leucine signals from molecules that have been shown to interact with AP3 are shown in bold. Residues that may be preferred for AP3, but not for AP1 or AP2 binding, are shown in bold italic. Signals that have been shown to bind AP3 are marked with ϩ (based on data from Refs. 35 and 36 -39). The other constructs have not, to our knowledge, been tested for AP3 interactions. NT, not tested.