We employed a freely available on-line tool named MetaPocket 2.0 [23], which was designed to predict consensus sites in ranking by combining results of eight individually developed predictors including LIGSITE^csc [24], PASS [25], Q-SiteFinder [26], SURFNET [27], Fpocket [28], GHECOM [29], ConCavity [30] and POCASA [31]. We first examined efficacy of this combined computational approach by comparing the known binding sites in the crystal structure of Zn²⁺-bound form of des3-23ALG-2/ALIX ABS peptide complex (PDB ID: 2zne) with the binding sites predicted by MetaPocket 2.0 using the crystal structure of Ca²⁺-bound form of des3-20ALG-2 (PDB ID: 2zn9). As shown in Figure 1, the ALIX ABS peptides (panel A, orange spheres indicating PPYP and light orange spheres indicating YP, respectively) occupy two of the six predicted binding site clusters in chains A and B of ALG-2 dimer by MetaPocket 2.0 (panel B), demonstrating successful prediction. Two of the additionally predicted sites are formed at a crevice created between each molecule of ALG-2 dimer (chains A and B). Previous mutational analyses of ALG-2 showed that ALG-2^Y180A (substitution of Y180 with alanine) lost both the ability to form a homodimer [32] and the ability to bind to ALIX but retained the ability to bind to PLSCR3 and Sec31A [15]. Thus, an authentic binding site for type 2 motif should be unaffected by dimerization. Since the crevice formed between chains A and B of ALG-2 dimer can be excluded, two other predicted sites more proximal to N-terminal regions (cyan to green in the rainbow colors) are promising.

Prediction of potential binding sites for type 2 motif by MetaPocket 2.0 was further performed by using only chain A of the Ca²⁺-bound form of ALG-2 dimer as a query structure (PDB ID: 2zn9). The top five ranked sites (ID Nos. 1–5) are listed in Table 1. Amino acid residues are partially overlapped between ID Nos. 1 and 3 and between ID Nos. 2 and 4 (Table 1, underlined and double-underlined, respectively), suggesting juxtaposition of these sites. ID Nos. 1 and 3 contain residues known to interact with 3-PPYP and 11-YP of the ALIX peptide (1-QGPPYPTYPGYPGYSQ-16, interacting residues underlined) at previously named Pocket 1 and Pocket 2, respectively, in the crystal structure of the complex (Table 1, letters in magenta) [21]. As shown in the surface presentation in Figure 2, ID No. 1 (Pocket 1) and ID No. 3 (Pocket 2) are juxtaposed (left panels). Binding sites were also predicted in an area distantly located from these sites (bottom of front and side views). Residues of ID No. 2 (green) and ID No. 4 (cyan) partly merge (yellow) and create a pocket named Pocket 3, whose contour displayed at vertical section of line V₁–V₂ shows a concave shape (right panels). The relationship between ID numbers of the predicted potential binding sites and pocket numbers designated in this study are summarized in Table 2.

Side chains of amino acid residues comprising type 2 motif and those of neighboring residues are hydrophobic in PLSCR3 (41-QVPAPAPGFALFPS-54, ABM-2 underlined) and Sec31A (837-NPPPPGFIMH-846), suggesting hydrophobic interactions for binding between these ligands and ALG-2. As shown in the upper panel Figure 3, Pocket 3 has a high degree of hydrophobicity and it is created by amino acid residues with hydrophobic side chains (lower panels shown in stereoview). We constructed a ligand binding model in which Pocket 3 accepts N-acetyl-Pro¹Ala²Pro³Gly⁴Phe⁵-amide, a virtual penta-peptide of a type 2 motif sequence derived from PLSCR3, by using the docking simulation program AutoDock Vina [33]. The calculated lowest energy model (−7.8 kcal/mol; Supplementary file of DockModel_Scr3_ABM-2 in PDB format) was analyzed for protein-ligand interactions by LIGPLOT [34] and NCONT [35]. A schematic diagram is shown in Figure 4, and interactions are listed in Table 3. In this docking model, carbon atoms of Pro¹, Ala² and Phe⁵ of the ligand interact with side chains of L48, A51, L52, S53, F85, W89, I92, Q96 and F148. The carbon atom from the methyl moiety of the acetyl group also interacts hydrophobically with F99 and G108. On the other hand, hydrogen bonds are predicted to be formed between oxygen atoms of peptide bonds (Pro¹, Ala² and Phe⁵) and oxygen atoms and a nitrogen atom of Pocket 3 residues in ALG-2 (S53, T93 and Q96). Moreover, a nitrogen atom of the C-terminal amide forms a hydrogen bond with an oxygen atom in the side chain of Gln96 (Table 3, not shown in Figure 4). As shown in Figure 5, interacting residues are from three segments in the ALG-2 primary structure: (i) α-helix 2 (exiting helix of EF1) and the following loop (L48, A51, L52, S53); (ii) α-helix 4 that is continuous from the exiting helix of EF2 and entering helix of EF3 (F85, W89, I92, T93, Q96 and F99) and EF3 loop, and (iii) the start of α-helix 7 (F148). As shown in Figure 6, the model peptide fits well into Pocket 3. While Pro³ and Phe⁵ of the peptide settle to the bottom by interacting with L52 (colored in yellow) and F148 (colored in light green), the linker segment, Pro³Gly⁴, loops out from the cavity.

The crystal structure of ALG-2 in the Ca²⁺-bound form (PDB ID 2zn9) shows that Pocket 3 is largely covered with a crystallographic symmetry-related monomer (not the partner of dimer, Supplementary Figure S1). Inaccessibility of the ABS-2 peptide to Pocket 3 in the crystal might explain why we could not obtain the crystal structure of the complex between ALG-2 and the ABS-2 peptide. Masking of Pocket 3 by another ALG-2 molecule, however, may happen only in the crystal and in solution, if it happens at all, at a very high concentration of ALG-2. Pocket 3 should be opened to the ligand in the cell and in solution under experimentally controlled conditions.

To investigate the possibility that Pocket 3 is a genuine binding site for type 2 motif as predicted, we performed in vitro binding assays using amino acid-substituted mutants of ALG-2. Since a high concentration of ALG-2 (~5 μM) tends to aggregate in the presence of Ca²⁺ [36] and hampers analyses of protein-ligand interaction. Addition of non-ionic detergent alleviates the problem of aggregation but limits the methods to be employed for interaction analyses. Here we employed the glutathione-S-transferase (GST) pulldown method. Immobilization of the GST-fused ALG-2 proteins on glutathione Sepharose beads was thought to be effective to avoid the potential masking problem of Pocket 3 by neighboring ALG-2 molecule as described above. Among the predicted interacting residues of ALG-2, L52 and F148 were targeted for amino acid substitutions (L52, alanine; F148, serine) since they are located in the boundary of helices and loops and adverse effects on the secondary structures were presumed to be minimal (Figure 5). Supernatants of the lysates of cells expressing green fluorescent protein (GFP)-fused PLSCR3 of wild-type and mutants lacking either ABS-1 or ABS-2 were subjected to GST-pulldown assays in the presence of 100 μM CaCl² and 0.1% Triton X-100, and proteins bound to the glutathione Sepharose beads were analyzed by Western blotting using an anti-GFP antibody. As shown in Figure 7, while GFP-Scr3^ΔABS1 (retaining type 2 motif) was not detected in the pulldown fractions of GST-ALG-2^L52A and GST-ALG-2^F148S, GFP-Scr3^ΔABS2 (retaining type 1 motif) was detected in the case of these ALG-2 mutants. Retaining of the binding ability to type 1 motif suggests that structural adverse effects caused by amino acid substitutions of L52 and F148 are locally limited to Pocket 3 and, if any, its surroundings. In contrast, GST-ALG-2^ΔGF122 showed opposite results: binding to GFP-Scr3^ΔABS1 but not to GFP-Scr3^ΔABS2. Unexpectedly, wild-type GST-ALG-2 showed a poorer binding ability to GFP-Scr3^ΔABS2 than mutant ALG-2 proteins (2nd panel from the top). ABS-1 may have a weaker binding ability than ABS-2, but mutations of Pocket 3 might somehow favored ALG-2 with a stronger binding to ABS1, possibly by avoiding competition with endogenous type 2 motif-containing proteins including Sec31A. In other words, occupation of Pocket 3 might inhibit binding of type 1 motif-containing proteins to Pocket 1 and 2 in the same monomer molecule probably due to steric hindrance. No significant signals for GFP-Scr3 were detected in pulldown fractions of the wild-type and mutant GST-ALG-2 proteins in the presence of 5 mM EGTA, a Ca²⁺ chelator, (Supplementary Figure S2), indicating that mutations of L52 and F148 did not abrogate the Ca²⁺-dependency of interaction. In the second and third lowest energy models (−6.8 kcal/mol and −6.5 kcal/mol, respectively) obtained by the ligand binding simulation with AutoDock Vina, the peptide binds to a part of Pocket 2, and neither L52 nor F148 interacts with the model peptides (data not shown). Thus, the results strongly suggest that Pocket 3 is the most probable candidate for the type 2 motif-binding site and that the two types of ALG-2-binding motifs are accepted by different pockets in ALG-2.

Previous studies on crystal structures of the Ca²⁺-bound form of ALG-2 revealed binding of Ca²⁺ to EF1, EF3 and EF5 [21,37], among which the structure of EF1 was found not to be changed by Ca²⁺ binding in a comparison of the 3D structures with the metal-free form [21]. On the other hand, comparison of the metal-free form and the Ca²⁺-bound form of ALG-2 revealed a difference in configuration of the side chain of R125 that is present in the loop connecting EF3 and EF4 and interacts with the ALIX peptide in Pocket 1. Since EF5 has an incomplete Ca²⁺-coordination at the −z position (water molecule oxygen atom instead of bidentate side chain oxygen atoms of glutamic acid), its binding affinity seems to be lower than affinities of the other two EF-hands [21]. Assuming that binding of Ca²⁺ to EF3 is the most critical step for ALG-2 to accept ligands, we proposed a Ca²⁺/EF3-induced R125 switch model to explain Ca²⁺-dependent interaction between ALG-2 and ALIX [21]. This model, however, is not applicable to explain the Ca²⁺-dependent binding of type 2 motif peptides, and it remains to be clarified how the signal of binding of Ca²⁺ to EF3 is transduced to Pocket 3. Interestingly, some residues predicted to interact with the virtual type 2 motif peptide are localized in α-helix 4 that forms a part of EF2 and EF3 (Table 3, Figure 5), suggesting that EF3 is somehow involved in the Ca²⁺-dependency for interaction with type 2 motif-containing proteins. Although attempts of co-crystallization of ALG-2 in the complex with the type 2 motif peptides of PLSCR3 and Sec31A have failed, the new finding of Pocket 3 as the most probable candidate of the type 2 motif-binding site provides us a new idea for designing a ligand and/or mutant ALG-2 to facilitate crystallization of the complex for X-ray crystallography, and a definite solution to the Ca²⁺-dependent type 2 motif-binding mechanism will be obtained in the future.

A program of binding site predictor MetaPocket 2.0 [23] was run online [38]. A molecular visualization tool, PyMol [39], was used for structure representations and measurement of distances between atoms. For surface representation of hydrophobicity by PyMol, rTools 0.7.2 [40] was down-loaded and protscale was selected from Plugin. A 3D-structural model of the virtual peptide of N-acetyl-ProAlaProGlyPhe-amide containing type 2 motif was first constructed with COOT [41] and used for ligand docking simulation program AutoDock Vina [33,42]. The program was run by fixing the grids so that they covered entire Pocket 3 and also a part of Pocket 2, which is located at the back side of Pocket 3. Protein-ligand interactions were analyzed using the program LIGPLOT v.4.5.3 [34,43], and selected by NCONT in the CCP4 suite [35].

Mutations of nucleotides for amino acid substitutions in ALG-2 were performed by the PCR-based mutagenesis method using pairs of specific primers (L52A substitution: forward: 5′-GCTTCAGCAAGCTGCCTCCAACGGCACGTG-3′; reverse: 5′-CACGTGCCGTTGGAGGCAGC TTGCTGAAGC-3′; F148S substitution: forward: 5′-GGGCAGATCGCCAGCGACGACTTCATC CAG-3′; reverse: 5′-TGGATGAAGTCGTCGCTGGCGATCTGCCC-3′), an expression plasmid of GST-fused human ALG-2 (pGEX-4T-3/hALG-2) [12] as a template and KOD plus DNA polymerase (Toyobo, Osaka, Japan) essentially according to the instruction manual of a QuikChange Site Directed Mutagenesis Kit (Agilent Technologies, Santa Clara CA 95051, USA). After being digested with DpnI (Takara Bio Inc, Osaka, Japan), PCR products were used for transformation of Escherichia coli Top10. Isolated plasmid DNAs were subjected to DNA sequencing. After confirming nucleotide substitutions at the targeted sites but no changes in the rest of the ALG-2 cDNA sequence, Escherichia coli BL21 cells were transformed with the plasmids, and mutant ALG-2 proteins were expressed and affinity-purified using glutathione Sepharose beads as described elsewhere. Eluted ALG-2 proteins were dialyzed against 20 mM Tris-HCl, pH 7.5, 10 μM EDTA, 10 μM EGTA, and 100 mM NaCl and re-adsorbed to Sepharose beads.

An ALG-2 knocked down cell line (designated HEK293/ALG-2_KD) [32] that was established previously by the RNA interference method was used to minimize effects of competition on target binding between endogenous ALG-2 present in the cells and GST-ALG-2 immobilized on Sepharose beads. HEK293/ALG-2_KD cells were transfected with expression plasmids for GFP-fusion proteins of wild-type and deletion mutants of PLSCR3 by the conventional calcium phosphate precipitation method. One day after transfection, harvested cells were lysed with buffer A (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1.5 mM MgCl₂, 0.1% Triton X-100 and protease inhibitors), and supernatants obtained after centrifugation at 14,000 g for 10 min at 4 °C were subjected to GST-pulldown assays essentially as described previously after addition of CaCl₂ to 100 μM [10,12]. Briefly, Sepharose beads immobilizing GST-ALG-2 proteins were incubated with cleared cell lysates at 4 °C overnight. After Sepharose beads had been recovered by low-speed centrifugation (700 g) for 1 min and washed three times with buffer A containing 100 μM CaCl₂, proteins bound to the beads were subjected to SDS-PAGE followed by Western blot analysis. Proteins transferred to polyvinylidene difluoride (PVDF) membranes (Immobilon-P, Millipore, Bedford, MA, USA) were probed with an anti-GFP monoclonal antibody (clone B-2, Santa Cruz Biotechnology, Santa Cruz, CA, USA). Signals of Western blotting were detected by the chemiluminescence method using Super Signal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific Inc., IL, USA) and analyzed with LAS-3000mini (Fujifilm, Tokyo, Japan).