After Embedding in Membranes Antiapoptotic Bcl-XL Protein Binds Both Bcl-2 Homology Region 3 and Helix 1 of Proapoptotic Bax Protein to Inhibit Apoptotic Mitochondrial Permeabilization*

Background: Bcl-XL binds Bax at mitochondria, inhibiting Bax oligomerization and apoptosis. Results: Anchored in membranes, Bcl-XL·Bax heterodimer is formed from a rigid helix-in-groove interface plus a flexible helical dimer interface. Conclusion: Two interfaces contribute equally to the heterodimer stability required to inhibit Bax. Significance: This novel kind of protein-protein interaction stabilizes the membrane-bound heterodimer that is pivotal to apoptosis regulation. Bcl-XL binds to Bax, inhibiting Bax oligomerization required for mitochondrial outer membrane permeabilization (MOMP) during apoptosis. How Bcl-XL binds to Bax in the membrane is not known. Here, we investigated the structural organization of Bcl-XL·Bax complexes formed in the MOM, including the binding interface and membrane topology, using site-specific cross-linking, compartment-specific labeling, and computational modeling. We found that one heterodimer interface is formed by a specific interaction between the Bcl-2 homology 1–3 (BH1–3) groove of Bcl-XL and the BH3 helix of Bax, as defined previously by the crystal structure of a truncated Bcl-XL protein and a Bax BH3 peptide (Protein Data Bank entry 3PL7). We also discovered a novel interface in the heterodimer formed by equivalent interactions between the helix 1 regions of Bcl-XL and Bax when their helical axes are oriented either in parallel or antiparallel. The two interfaces are located on the cytosolic side of the MOM, whereas helix 9 of Bcl-XL is embedded in the membrane together with helices 5, 6, and 9 of Bax. Formation of the helix 1·helix 1 interface partially depends on the formation of the groove·BH3 interface because point mutations in the latter interface and the addition of ABT-737, a groove-binding BH3 mimetic, blocked the formation of both interfaces. The mutations and ABT-737 also prevented Bcl-XL from inhibiting Bax oligomerization and subsequent MOMP, suggesting that the structural organization in which interactions at both interfaces contribute to the overall stability and functionality of the complex represents antiapoptotic Bcl-XL·Bax complexes in the MOM.

Proteins of the B-cell lymphoma 2 (Bcl-2) 3 family can interact in the mitochondrial outer membrane (MOM) to regulate its permeability. These proteins contain one or more Bcl-2 homology (BH) motif and either promote or inhibit the MOM permeabilization (MOMP) that initiates apoptosis. Bcl-2-associated X (Bax) and Bcl-2 antagonist/killer 1 (Bak) contain three BH motifs (BH1-3) that are required for the MOMP and a six-residue structural motif similar to that found in the BH4 motif of antiapoptotic family members (1). Bcl-2, B-cell lymphoma-extra large (Bcl-XL), and other antiapoptotic family members contain four BH motifs (BH1-4) and function as both direct and indirect inhibitors of Bax and Bak. BH3-only proteins, such as BH3 interacting domain death agonist (Bid), Bcl-2-like protein 11 (Bim), and Bcl-2-associated death promoter (Bad), contain only the BH3 motif and promote the MOMP by activating Bax and Bak, inhibiting the antiapoptotic family members, or both (2)(3)(4).
Bax is inactive in the cytosol of normal cells but becomes active in apoptotic cells by changing conformation in a multistep process. First, Bax is recruited to the mitochondria by truncated Bid (tBid) in the MOM (5,6). The membrane-bound Bax recruits more soluble Bax to the MOM by autoactivation (7). These Bax proteins bind to each other to form an oligomeric pore in the membrane. The oligomerization most likely takes place after Bax changes to a multispanning conformation with helices 5, 6, and 9 embedding in the membrane (8). The structure of the oligomeric Bax pore in the membrane is largely unknown. However, the data from several recent studies suggest a possible consensus structure. Our photocross-linking study revealed two interdependent interfaces in the Bax oligomer that was formed in detergent micelles. The first one is formed by the BH1-3 regions, and the second one was formed by helix 1 plus the following loop and helix 6 (9). An electron paramagnetic resonance study of Bax oligomers that were formed in detergent micelles and liposomal membranes together with homology modeling of the Bax dimer using a Bak dimer model suggested an antiparallel helical dimer interface formed by the helix 2-3 region of neighboring Bax molecules in the oligomer (10,11). A disulfide cross-linking study indicated that the helix 2-3 interface was extended to include helix 4, which binds to other side of helix 2, resulting in a BH3-in-groove dimer interface (12). This interface was also observed in a crystallographic study using a Bax fragment containing the helix 2-5 region (13). The BH3-in-groove dimer interface together with another parallel helical dimer interface formed by helix 6 of neighboring Bax molecules could generate a Bax oligomer in the MOM (12).
To inhibit Bax oligomerization, Bcl-XL could inhibit any one or more of these Bax homointeractions, thereby interrupting the activation process. In addition, it could also sequester tBid, preventing tBid from binding to and activating Bax. Neither of these potential inhibitory functions of Bcl-XL is mutually exclusive. Thus, interaction with Bax could inhibit the conformational change in Bax, preventing it from adopting the multispanning state, the oligomerization of the multispanning monomers, or both. Early mutational studies indicate that the BH1-3 motifs of Bcl-XL and the BH3 motif of Bax mediate their interaction. For example, replacement of Asp 68 with arginine (D68R) in the BH3 region of Bax greatly impaired its interaction with antiapoptotic family members (14). In addition, replacement of Gly 138 with alanine (G138A) in the BH1 region of Bcl-XL abolished its binding to Bax and inhibition of apoptosis (15,16). The crystal structure of a truncated Bcl-XL protein in complex with a peptide containing the BH3 region of Bax shows that the peptide binds as an amphipathic ␣ helix to a hydrophobic groove formed by the BH1-3 regions of Bcl-XL (17). Although all of these studies provide clues, the complex structure formed by the full-length Bcl-XL and Bax proteins in membranes remains elusive.
Bcl-XL, like Bax, is a soluble cytosolic or peripheral membrane protein in normal cells. Upon activation by membranebound tBid or Bax, Bcl-XL integrates into the MOM (5). The membrane topology of the active Bcl-XL is not known. However, the homologous Bcl-2 changes conformation during apo-ptosis initiation, and this conformational change is required for its antiapoptotic activity (18 -20).
The following questions about Bcl-XL⅐Bax interaction are important and have not been answered. Does Bcl-XL binding to Bax prevent Bax from changing to the multispanning state, or does it prevent Bax oligomerization? Is the BH1-3 groove⅐BH3 helix interface in the crystal structure relevant to the Bcl-XL⅐Bax complex formed in the MOM? Are there additional regions of Bcl-XL and Bax involved in the complex formation, as we previously detected in the Bcl-2⅐Bax complex that was formed in detergent micelles (21)? If a second interface, in addition to the groove⅐BH3 interface, forms in the membranebound Bcl-XL⅐Bax complex, can ABT-737, a BH3 peptide mimetic that binds to the BH1-3 groove of Bcl-XL in solution (22), still disrupt the Bcl-XL⅐Bax complex in the membrane? To address these questions, we systematically mapped the interface and the topology of the Bcl-XL⅐Bax heterodimer that was formed in membranes. We built structural models for the heterodimer based on the experimental data, and we tested these models with additional experiments that used mutational and functional assays. We also monitored the effect of ABT-737 on the membrane-bound heterodimer, particularly its effect on the formation of a non-groove⅐BH3 interface. Together, these studies revealed critical structural features of a functional Bcl-XL⅐Bax heterodimer that was formed in the biological membranes. Our results demonstrate that the membrane-bound heterodimer has two interfaces, the conventional groove⅐BH3 interface seen in crystal and a novel helix1⅐helix 1 interface, which surprisingly can form in either a parallel or antiparallel fashion.

EXPERIMENTAL PROCEDURES
Materials-Phospholipids were purchased from Avanti Polar Lipids. The MOM liposomes consisting of MOM-characteristic lipids were made as described (7). The Bax Ϫ/Ϫ / Bak Ϫ/Ϫ mitochondria were isolated from the livers of Bak knock-out mice that also lack Bax as described (5). The Bax BH3 peptide that contains Bax residues 53-86, including the BH3 region, was synthesized by Abgent as described (7). The full-length human Bax protein with or without an N-terminal His 6 (6H) tag was expressed and purified as described (9,23). The plasmid for expression of the N-terminal 6H-tagged human Bcl-XL in Escherichia coli was modified from the pCYB3-Bcl-XL plasmid by inserting six histidine codons between the first two codons of Bcl-XL. The full-length human Bcl-XL protein with or without the N-terminal 6H tag was expressed and purified as described (5,23), except that the 6H-Bcl-XL eluted from the chitin column was purified using a Ni 2ϩ -nitrilotriacetic acid-agarose column, and the resulting protein was dialyzed in 20% (v/v) glycerol and 20 mM Tris/HCl, pH 8.0.
Single-cysteine and Single-lysine Bax and Bcl-XL Mutants-To construct plasmids for in vitro transcription and translation of Bax and Bcl-XL, we inserted the coding region of full-length human Bax or Bcl-XL into the vector pSPUTK (Stratagene). We created lysine-null (K0) Bax and Bcl-XL mutant plasmids by changing all of the lysine codons to arginine codons, and we created cysteine-null (C0) Bax and Bcl-XL mutant plasmids by changing all of the cysteine codons to alanine codons. We then created Bax and Bcl-XL mutant plasmids with a single lysine or cysteine codon at particular positions by mutating the corresponding codons in the K0 or C0 mutant plasmid to lysine or cysteine codon, respectively.
MOMP (Cytochrome c Release) Assay-The assay was modified from that described previously (24). Wild-type and mutant Bax and Bcl-XL proteins were synthesized using a transcription/translation-coupled SP6 RNA polymerase/reticulocyte lysate system (Promega). The resulting reaction containing the Bcl-XL protein (3 l), the Bax protein (3 l), or both was incubated with the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria (0.5 mg/ml total protein). Purified tBid protein (17 nM) (23) or Bax BH3 peptide (40 M) was added to the reactions to activate the Bcl-XL and Bax proteins. An adequate volume of buffer A (110 mM KOAc, 1 mM Mg(OAc) 2 , 25 mM HEPES, pH 7.5, and 2 mM glutathione) was added to each reaction to bring the total volume to 15 l. After incubation for 1 h at 37°C, the samples were centrifuged at 10,000 ϫ g for 10 min. The resulting pellet fractions were resuspended with 0.5% (v/v) Triton X-100 in phosphate-buffered saline (PBS, pH 7.4) to the same volume as the supernatant fractions and centrifuged again at 16,100 ϫ g for 10 min to obtain the second supernatant fractions as the detergent-solubilized mitochondrial pellet fractions. The amounts of cytochrome c in both supernatant and pellet fractions were measured using an enzyme-linked immunosorbent assay (ELISA) with the antibody against mouse cytochrome c from R&D Systems per its protocol. The fraction of cytochrome c release was calculated using the formula, [cytochrome c in supernatant]/ ([cytochrome c in supernatant] ϩ [cytochrome c in pellet]).
Apoptotic Activity of Bax Mutants in bax Ϫ/Ϫ /bak Ϫ/Ϫ Mouse Embryonic Fibroblasts (MEFs)-Phoenix cells were seeded in 100-mm dishes and MEFs in 96-well plates. The pBabe-MN-Bax-IRES-GFP plasmids containing wild-type Bax, Lys-null, or single-lysine mutants were constructed as described previously (9). Each plasmid (10 g) was transfected into the Phoenix cells with Exgene500 (Fermentas) to package the plasmid into a replication-incompetent murine virus. The media containing the virus were harvested 24 h after the transfection, filtered with a 0.2-m filter, and then added into the MEFs. After 48 h of infection, the MEFs were treated with 0.5 or 2 M etoposide for 24 h, and three-channel images (green fluorescent protein (GFP), annexin V R-PE, and DRAQ5) were collected from five fields of view in each well. In each field, cells were identified via the DRAQ5 imaging, and measurements of emission intensity of GFP and annexin V were taken per cell from their respective images. The total number of cells measured in the five fields of any well was used to calculate the annexin V-positive percentage score for that well. The fraction of annexin V-positive cells was determined for green cells (infected) and non-green cells (uninfected) separately. Expression of GFP alone from the internal ribosome entry site (IRES) sequence was somewhat toxic. The toxicity was shown in the bax Ϫ/Ϫ /bak Ϫ/Ϫ MEFs and poorly inhibited by Bcl-2 (data not shown); hence, the toxicity was probably not due to apoptosis. Because co-expression of wild-type Bax significantly increased the apoptosis, as detected by annexin V staining, the assay could clearly reveal the apopto-tic activity of the Bax mutants relative to the wild-type Bax that was assayed in parallel.
Disulfide and Bismaleimidohexane (BMH) Cross-linking-[ 35 S]Met-labeled single-cysteine Bax and Bcl-XL mutant proteins were synthesized in the wheat germ-based in vitro translation system as described (25). The resulting reaction containing the Bax protein (10 l), the Bcl-XL protein (10 l), or both was incubated at 37°C for 1.5 h with the MOM liposomes (4 mM total lipids) or for 1 h with the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria (0.5-0.7 mg/ml total proteins), 36 -40 M BH3 peptide or 1.2 nM tBid protein, and 1 mM dithiothreitol (DTT), and an adequate volume of buffer A to bring the total reaction volume to 40 l. A subset of the reactions included 50 M ABT-737. The resulting proteoliposomes were isolated by a sucrose gradient centrifugation as 250-l fractions from the top of the gradient (20,23). The mitochondria with Bax, Bcl-XL, or both bound were pelleted by centrifugation and resuspended in 100 l of buffer A as described (5). These membrane fractions were treated with 1 mM NaAsO 4 to decrease residual DTT and then with redox catalyst Cu(II)(1,10-phenanthroline) 3 (CuPhe; consisting of 0.3 mM CuSO 4 and 1 mM 1,10-phenanthroline) to induce disulfide cross-linking. After incubation on ice for 30 min, the oxidation reactions were quenched by 20 mM N-ethylmaleimide (NEM) and 100 mM EDTA. For the "0 min" controls, NEM and EDTA were added to the samples to block the disulfide cross-linking before the addition of CuPhe. To produce cross-linking with BMH, the proteoliposomes were incubated with 100 mM BMH at room temperature for 30 min, and then the cross-linking reactions were stopped by the addition of 50 mM ␤-mercaptoethanol. Both disulfide-and BMH-cross-linked samples were solubilized with 1% (v/v) Triton X-100 before precipitation with trichloroacetic acid. The resulting proteins were analyzed by SDS-PAGE under non-reducing or reducing conditions. The radioactive proteins and their adducts in the gels were detected by phosphorimaging with a Fuji FLA-9000 image scanner.
Photocross-linking-[ 35 S]Met-labeled Bax or Bcl-XL proteins with a single N ⑀ -(5-azido-2-nitrobenzoyl)-lysine (ANB-lysine) incorporated at specific locations were synthesized from RNA of the corresponding single-lysine Bax or Bcl-XL mutants using the wheat germ-based in vitro translation system as described (25,26). 10 l of the resulting Bax or Bcl-XL proteins was incubated at 37°C for 1.5 h with 2.2 M purified 6H-Bcl-XL or 6H-Bax protein, the MOM liposomes of 4 mM total lipids, and 36 -40 M BH3 peptide in a 20-l reaction adjusted by buffer A. The proteoliposomes were isolated as described above and photolyzed to induce cross-linking via the 5-azido-2-nitrobenzoyl (ANB) probe. The resulting photoadducts of the [ 35 S]Metlabeled protein and the 6H-tagged protein were solubilized in 0.25% (v/v) Triton X-100, enriched on Ni 2ϩ -nitrilotriacetic acid-agarose, eluted, and analyzed with reducing SDS-PAGE and phosphorimaging.
4-Acetamido-4Ј-((iodoacetyl)amino)stilbene-2,2Ј-disulfonic Acid (IASD) Labeling-[ 35 S]Met-labeled single-cysteine Bax or Bcl-XL mutant proteins were synthesized by using either the wheat germ-or reticulocyte lysate-based in vitro translation system as described (25,26). The resulting reactions containing the Bax protein (10 l), the Bcl-XL protein (10 l), or both were incubated at 37°C for 1.5 h with the MOM liposomes with 4 mM total lipids, 36 -40 M BH3 peptide, and an adequate volume of buffer A to bring the total reaction volume to 40 l. A subset of the reactions also included 0.5 M purified recombinant Bax or Bcl-XL protein. The resulting proteoliposomes were isolated as described above and labeled with 0.2 mM IASD at room temperature in the dark for 30 min in the absence or presence of 2% (w/v) CHAPS, 4 M urea, or both. The reactions were quenched by the addition of 10 mM ␤-mercaptoethanol. For the "0 min" controls, the ␤-mercaptoethanol was added before the addition of IASD to prevent the labeling. The radioactive IASD-labeled proteins were resolved from the unlabeled ones by isoelectric focusing (IEF) (8) and detected by phosphorimaging. The intensities of the IASD-labeled and unlabeled bands for each protein under each condition were quantified using the program ImageQuant from Fuji. The integrated intensities were used to calculate the fraction of IASD labeling according to the formula, intensity of labeled band(s)/(intensity of labeled band(s) ϩ intensity of unlabeled band(s)).
Modeling of Bcl-XL⅐Bax Heterodimer Interface and Membrane Topology-The BH1-3 groove⅐BH3 helix interface for the Bcl-XL⅐Bax heterodimer was modeled based on the crystal structure of a Bcl-XL protein⅐Bax BH3 peptide complex (PDB entry 3PL7) (17). The model was used to predict mutations that would disrupt the heterodimer interface, and the predictions were tested experimentally to verify the model.
The helix 1⅐helix 1 interface for the heterodimer was modeled based on the contact information of specific residues in these regions provided by the disulfide cross-linking experiments. The structure of Bax helix 1 was extracted from the NMR structure of a Bax monomer (PDB entry 1F16) and then positioned in a parallel orientation with the Bcl-XL helix 1 in the Bcl-XL protein⅐Bax BH3 peptide complex structure (PDB entry 3PL7) by interactive molecular modeling using the program PyMOL (Schrodinger, LLC). Bad contacts between the two helices were relieved by changing the rotameric state of several side chains by using the Dunbrack backbone-dependent rotamer library (27) via the mutagenesis wizard in PyMOL. The mutagenesis wizard was also used to change several residues to cysteines in order to fit the model to the disulfide cross-linking data. We used geometry regularization in the program COOT (28) to remove any distortions in the backbone that were introduced while changing the amino acids. The stereochemical quality of the backbone was checked with a Ramachandran plot, and the atom-atom clashes were scored using the program MOLPROBITY in the presence of hydrogen atoms added using the electron-cloud X-H bond lengths (29). The modeling exercise was repeated with the Bax helix 1 rotated into an antiparallel orientation with the Bcl-XL helix 1 to fit with another set of disulfide cross-linking data. The parallel and antiparallel models were used as starting models in the Rosetta-based program FlexPepDock (30,31) to optimize the position of the peptide and the conformation of the side chains in and near the interface. Two hundred trial structures were generated and scored for each model. The 10 trial structures with the most favorable scores were inspected with PyMOL. The parallel model and antiparallel model with the best agreement with the disulfide cross-linking data were selected for further analysis.
The overall structural organization model for the Bcl-XL⅐Bax heterodimer bound to the membrane was constructed from a composite of the BH1-3 groove⅐BH3 and parallel helix 1⅐helix 1 interface models described above. Helix 9 of Bcl-XL was built from residues 206 -230 using the program PHYRE2 (32). Helices 4, 5, 6, 7, 8, and 9 of Bax were extracted from the Bax monomer structure (PDB entry 1F16). Using PyMOL, these helices were added to the composite structure such that helix 9 of Bcl-XL and helices 5, 6, and 9 of Bax were located under a virtual plane representing the cytosol-MOM boundary, whereas helices 4, 7, and 8 of Bax were located above the plane with the composite structure. Further, the residues that were buried in the membrane, as indicated by the IASD labeling results, were placed under the plane, whereas the residues that were exposed to the aqueous milieu were placed above the plane. The positions of these helices relative to each other and to the composite structure were adjusted such that the distances between the indicated residues were in the range expected for the residues that could be cross-linked by BMH.

RESULTS
Experimental Design-The interface of Bcl-XL⅐Bax heterodimer that was formed in membranes was mapped using site-specific disulfide cross-linking and photocross-linking. The former provided structural information about the interface at a higher resolution than the latter, because a disulfide bond can form only when the C 1 ␤ -C 2 ␤ distance of the two cysteines is ϳ3-5 Å and the dihedral angle (C 1 ) about the disulfide bond is close to Ϯ90°(33). However, disulfide crosslinking occurs on a time scale of seconds to minutes (34,35) and may capture both the most abundant Bcl-XL⅐Bax complex that is in the most stable conformation and the rare complexes that are in less stable conformations and reached during the reaction time due to thermal fluctuations. In contrast, the photocross-linking occurs within nanoseconds at a very low efficiency (36), thereby providing a "snapshot" of only the most stable and abundant complex while not sampling the less stable and rare complexes.
The topology of Bcl-XL⅐Bax heterodimer in membranes was determined by labeling the membrane-bound proteins with IASD, a sulfhydryl-reactive and membrane-impermeant reagent with two negative charges. A cysteine in the protein that is exposed to the aqueous milieu will be labeled. On the other hand, a cysteine that is buried in a hydrophobic environment (e.g. a lipid bilayer, a protein core, or a protein complex interface) will not be labeled, unless a detergent, such as CHAPS, or a denaturant, such as urea, is added to disrupt the bilayer or denature the protein or protein complex, respectively (8).
Two complementary in vitro membrane systems were used. First, the synthetic liposomal membrane consisting of the MOM-characteristic lipids but not proteins (7) simplified the data interpretation. Second, the isolated mitochondria provided a native environment for protein interactions for verification of the results from the liposome system.
To ensure that the structural information obtained from the above assays was from functional Bcl-XL⅐Bax complexes in membranes, only Bcl-XL and Bax mutants that were active in the MOMP assay were used in the cross-linking and labeling experiments. Moreover, mutations that could disrupt the heterodimer formation were generated, and their effects on inhibition of Bax-mediated MOMP by Bcl-XL were determined to define the structure-function relationship.
The Bcl-XL and Bax Mutants Used in the Interface-and Topology-mapping Experiments are Functional-In order to map the interface and topology of the Bcl-XL⅐Bax complex formed in membranes, we made Bcl-XL and Bax mutants with a single cysteine or lysine at specific locations ( Fig. 1) (9). The mutant proteins were synthesized in vitro using a coupled transcription/translation system, and their MOMP activities were measured in an ELISA-based cytochrome c release assay (24). Each Bax mutant was activated by either purified tBid protein or a Bax peptide consisting of the BH3 region (BH3 peptide) and incubated with the mitochondria that lack both Bax and Bak (Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria) either in the absence or presence of a Bcl-XL mutant. The resulting cytochrome c release was quantified. Like wild-type Bax, each single-cysteine Bax mutant released cytochrome c in a tBid-dependent ( Fig.  2A) or BH3 peptide-dependent (data not shown) manner. Like wild-type Bcl-XL, each single-cysteine Bcl-XL mutant inhibited the cytochrome c release by the single cysteine Bax mutant that was paired with the Bcl-XL mutant in the disulfide cross-linking experiments ( Fig. 2A). The additional single-cysteine Bax and Bcl-XL mutants used in IASD labeling and BMH crosslinking experiments were also active ( Fig. 2B). All of the singlelysine Bcl-XL mutants used in the photocross-linking experiments had activities similar to that of the wild-type protein and the lysine-null (K0) mutant (Fig. 2C). All of the single-lysine Bax mutants used in the photocross-linking experiments were active in the MOMP assay ( Fig. 2C), in Bax and Bak double knock-out mouse embryonic fibroblast cells (bax Ϫ/Ϫ /bak Ϫ/Ϫ MEFs) (Fig. 2D), in HEK293 cells (9), or in both cells.
In Membranes, Bcl-XL and Bax Dimerize via the Canonical BH1-3 Groove⅐BH3 Helix Interface after Activation-We performed disulfide cross-linking experiments to map the heterodimer interface using the active single-cysteine Bcl-XL and Bax proteins synthesized in vitro. We first did the experiments with the MOM liposomal membranes to avoid complications from the other mitochondrial proteins. We also used the Bax BH3 peptide instead of tBid to activate the soluble Bcl-XL and Bax proteins and generate the membrane-bound proteins, because tBid can bind to Bcl-XL and Bax in the membranes (5, 6) and hence may interfere with the Bcl-XL⅐Bax interaction. Like tBid, the BH3 peptide could activate the Bcl-XL and Bax proteins (data not shown) but did not bind to the membranes, as shown previously (7). As shown in Fig. 3A, a disulfide-linked product with an apparent molecular mass (M r ) close to that predicted for the Bcl-XL⅐Bax heterodimer was detected in the membranes when the indicated pairs of single-cysteine Bcl-XL and Bax mutants were targeted to the membranes. Particularly, the product (indicated by an arrow in lane 2 or 10 on the nonreducing gel) was formed when a Bcl-XL protein with the cysteine in the BH1-3 groove, V126C or L194C, was paired with a Bax protein with the cysteine in or near the BH3 region, L59C or M74C, respectively.
The following four results demonstrated that the arrow-indicated product was a disulfide-linked heterodimer of the sin- The native cysteines (underlined) were changed to alanine to create the cysteine-null (C0) mutants. Single-cysteine Bax and Bcl-XL mutants were created from the respective cysteine-null mutants by individually replacing the residues in boldface type with cysteine. Arrowheads, Met 74 of Bax and Gly 138 and Arg 139 of Bcl-XL that were changed to glutamate, alanine, and aspartate in Bax M74E and Bcl-XL G138A and R139D mutants, respectively. C, single-lysine Bcl-XL mutants. Bcl-XL sequence is shown with BH motifs and helices indicated as in B. The native lysines (underlined) were changed to arginine to create the lysine-null (K0) mutant. Single-lysine Bcl-XL mutants were created from the lysine-null mutant by individually replacing the residues in boldface type with lysine.
gle-cysteine Bcl-XL and Bax after they were activated by the BH3 peptide and targeted to the membranes. (i) The product appeared on the non-reducing gel and disappeared on the reducing gel (Fig. 3A, lanes 2 and 10) and was not formed when the cysteine-null (C0) Bcl-XL and Bax mutants were used (Nonreducing; lane 4) and was greatly reduced when NEM was added to block the sulfhydryl moiety before oxidation by CuPhe (Non-reducing; lane 1 versus lane 2 and lane 9 versus lane 10). Therefore, it is a disulfide-linked product formed by the singlecysteine Bcl-XL and Bax. (ii) The product was formed when both proteins were present in the membranes but not when one protein was omitted (Fig. 3A, Non-reducing, lane 2 versus lanes 6 and 8 and lane 10 versus lanes 12 and 14). Therefore, it is not a Bcl-XL or Bax homodimer. (iii) When one of the proteins was labeled by [ 35 S]Met but the other was not, the product was still detectable, but the radiation intensity was less than that when both proteins were labeled (Fig. 3B, arrow-indicated bands in lane 2 versus lanes 4 and 6 and lane 8 versus lanes 10 and 12). These data suggest that the heterodimers detected in lanes 4 and 10 were formed by the radioactive Bcl-XL (visible in the phosphorimages, indicated by filled circles) and the nonradioactive Bax (invisible in the phosphorimages), and the heterodimers detected in lanes 6 and 12 were formed by the radioactive Bax (visible, indicated by open circles) and the nonradioactive Bcl-XL (invisible). In contrast, the heterodimers detected in lanes 2 and 8 were formed by the radioactive Bcl-XL and Bax that were both visible in the phosphorimages. (iv) When the BH3 peptide was omitted, although a substantial amount of Bcl-XL was bound FIGURE 2. Activities of Bax and Bcl-XL mutants in Bax ؊/؊ /Bak ؊/؊ mitochondria and mouse embryonic fibroblasts. A-C, MOMP activity in the Bax Ϫ/Ϫ / Bak Ϫ/Ϫ mitochondria. Wild-type (WT) or the indicated mutant Bax proteins were synthesized in vitro, activated by tBid protein, and targeted to the Bax Ϫ/Ϫ / Bak Ϫ/Ϫ mitochondria in the absence or presence of WT or the indicated mutant Bcl-XL proteins that were synthesized in vitro. Cytochrome c release from the mitochondria was measured using ELISA. Data shown are average fractions of cytochrome c release from two to three independent experiments with the ranges indicated by error bars. D, the apoptotic activity of WT, lysine-null, and single-lysine mutant Bax in the bax Ϫ/Ϫ /bak Ϫ/Ϫ MEFs. The MEF cells were infected with retrovirus that co-expressed Bax and GFP from the same message using the IRES sequence between the two coding regions and treated with etoposide of the indicated concentrations. The cells were examined for apoptosis by annexin V staining. Both panels show the fraction of adherent annexin V-positive cells in the GFP-positive (infected and expressing Bax; black bar) and GFP-negative (uninfected and not expressing Bax; white bar) populations. The type of Bax protein expressed by the virus that was used to infect the cells is indicated below the plot. GFP, control virus that expressed only GFP; WT, K0, K21, etc., virus that expressed GFP and wild-type, lysine-null, or single-lysine Bax, respectively. Data shown are the averages from eight independent replicates. For each replicate, a minimum of 356 cells/well (average ϭ 1084, maximum ϭ 2613) or 309 cells/well (average ϭ 849, maximum ϭ 1442) was analyzed for the 2 or 0.5 M etoposide-treated cells, respectively. Error bars, S.D. The resulting proteoliposomes were isolated and oxidized by CuPhe for 30 min. NEM and EDTA were then added to stop the oxidation. For the 0 min controls, NEM and EDTA were added prior to the addition of CuPhe. The resulting samples were analyzed by non-reducing and reducing SDS-PAGE and phosphorimaging. B, the in vitro synthesized radioactive (R) and/or non-radioactive (N) Bax and Bcl-XL mutants were activated by the BH3 peptide and targeted to the MOM liposomes. The resulting proteoliposomes were processed and analyzed as described in A. C, the in vitro synthesized radioactive Bax and Bcl-XL mutant proteins were incubated in the absence or presence of the MOM liposomes and/or the BH3 peptide. The resulting proteoliposomes were processed and analyzed as described in A. D and E, in vitro synthesized radioactive Bax and Bcl-XL mutants were activated by tBid and targeted to the MOM liposomes (D) or activated by the BH3 peptide and targeted to the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria (E). The resulting membranes were isolated, oxidized, and analyzed as described in A. F, the in vitro synthesized radioactive (R) and/or non-radioactive (N) Bax and Bcl-XL mutant proteins were activated by the BH3 peptide and targeted to the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria. The resulting mitochondria were isolated, oxidized, and analyzed as described in A. In all panels, protein standards are indicated on the side of phosphorimages with their molecular masses (M r ). Arrows, disulfide-linked Bax⅐Bcl-XL heterodimers; open or filled circles, Bax or Bcl-XL monomers, respectively.
to the membranes, the amount of Bax bound to the membranes was negligible, and the disulfide-linked heterodimer was undetectable in the membranes (Fig. 3C, lanes 6 and 14). In the absence of the membranes, virtually no Bcl-XL or Bax monomer was recovered from the membrane fraction in the sucrose gradient; nor was the disulfide-linked heterodimer (lanes 4 and 12). Therefore, the heterodimer detected in Fig.  3A was formed by a disulfide linkage of the indicated cysteines in the Bcl-XL and Bax proteins that were activated by the BH3 peptide and targeted to the membranes.
To verify whether the heterodimer interface detected in the liposomes after activation by the BH3 peptide is also formed in a more biological setting, we used tBid to activate the Bcl-XL and Bax proteins. Similar disulfide-linked heterodimers were formed in this system (Fig. 3D). Further, we targeted both proteins that were activated by the BH3 peptide to the Bax Ϫ/Ϫ / Bak Ϫ/Ϫ mitochondria and detected similar disulfide-linked heterodimers (Fig. 3, E and F). Therefore, the groove⅐BH3 interface is formed not only in the liposomal membranes but also in the mitochondrial membranes after activation by either the BH3 peptide or the BH3 protein.
We also conducted photocross-linking experiments using Bcl-XL and Bax mutants with a single photoreactive lysine analog located at specific positions. After activation by the BH3 peptide and targeting to the MOM liposomes, the in vitro synthesized [ 35 S]Met-labeled Bcl-XL or Bax protein with the photoreactive probe, ANB, attached to the ⑀-amino group of the lysine located in or near the BH1-3 groove or the BH3 region cross-linked with the purified His 6 -tagged Bax or Bcl-XL protein, respectively (Fig. 4, A and B). These data are consistent with those from disulfide cross-linking experiments and suggest the existence of the groove⅐BH3 interface in the Bcl-XL⅐Bax heterodimer that is formed in the membranes.
The above cross-linking data support a structural model in which the BH1-3 groove⅐BH3 helix interface, as seen previously by crystallography (17), exists in the Bcl-XL⅐Bax heterodimer formed in the MOM after both proteins are activated by the BH3 protein or peptide (Fig. 5). Clearly shown in the model, Val 126 and Leu 194 in the BH1-3 groove of Bcl-XL are in close proximity with Leu 59 and Met 74 in the BH3 helix of Bax, respectively (Fig. 5A). These hydrophobic residues interact to contribute a significant portion of the binding energy that holds the complex together. For example, Leu 59 and Met 74 in the BH3 helix of Bax bind to the first and the fifth hydrophobic pocket in the BH1-3 groove of Bcl-XL, respectively (17). Val 126 in the groove of Bcl-XL has a van der Waals interaction with not only Leu 59 but also Leu 63 , which are the first and the second conserved hydrophobic residues in the BH3 region of Bax, respectively. The side chain of Leu 194 in the groove of Bcl-XL also has favorable van der Waals interactions with the aliphatic portion of the side chain of Arg 78 in the BH3 helix. When these Bcl-XL/ Bax residue pairs are replaced by the cysteine pairs (V126C/ L59C and L194C/M74C), the model predicts that the S 1 ␥ -S 2 ␥ distance between the paired cysteines and the C 1 dihedral angle are close to the range for disulfide bond formation ( Fig. 5B) (33). This prediction is supported by the disulfide cross-linking data (Fig. 3). The model also predicts that the C 1 ␤ -C 2 ␤ distance between Cys 62 , a native cysteine located in the BH3 helix of Bax, and the Val 126 located in the groove of the Bcl-XL mutant is 9.8 Å (Fig. 5A), which is too great a distance to result in a disulfide bond when the Val 126 was changed to cysteine (V126C). Similar predictions were also made for the C 1 ␤ -C 2 ␤ distances between the other two mismatched cysteine pairs in the Bax⅐Bcl-XL complex: L59C/L194C and M74C/V126C (data not shown). These predictions were consistent with the observation that no disulfide-linked dimer was detected between the corresponding mismatched Bax/Bcl-XL mutant pairs (data not shown). It was also clear that the residues in the BH3 helix of Bax and the groove of Bcl-XL, which generated photocross-linked heterodimers when replaced by ANB-lysine (Fig. 4), were located in or near the BH3⅐groove interface in the model (Fig. 5C).
The Helix 1 Regions of Bcl-XL and Bax Interact with Each Other in a Parallel Manner, Forming a Novel Interface in the Heterodimer-To determine if regions of Bcl-XL and Bax other than the BH1-3 regions are also involved in the heterodimerization, we used photocross-linking to scan more Bcl-XL and Bax mutants with a single ANB-lysine positioned in the other regions. The photoreactive probes located in helix 1 of both Bcl-XL and Bax generated the heterodimerspecific photoadducts in liposomes (Fig. 6), thereby indicating involvement of helix 1 in the heterodimerization. To determine if the helix 1 regions from both proteins interact with each other, we did disulfide cross-linking experiments with the following single-cysteine Bcl-XL/Bax mutant pairs: E7C/M20C, S18C/F30C, and S23C/R34C. After these mutants were activated by the BH3 peptide and targeted to the MOM liposomes (Fig. 7A) or the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria (Fig. 7B), multiple disulfide-linked products were detected, particularly in the mitochondria. Among these products, the ones marked with arrows are the Bcl-XL⅐Bax heterodimers because they were generated only when both proteins were targeted to the membranes and because their molecular masses were close to that of the heterodimer (Fig. 7, A and B, lanes 6, 12, and 18). As expected, the heterodimer-specific bands were still detectable when one of the two proteins was not labeled with [ 35 S]Met, although the intensities of the bands were weaker than those when both proteins were labeled in most cases (Fig. 7C, arrowindicated bands). When the BH3 peptide was omitted, although a substantial amount of Bcl-XL was bound to the liposomal membranes, the amount of Bax bound to the membranes was negligible, and the disulfide-linked heterodimers were undetectable in the membranes (Fig. 7D, lanes 6, 14, and 22). In the absence of the membranes, virtually no Bcl-XL or Bax monomer was recovered from the membrane fraction in the sucrose gradient; nor was the disulfide-linked heterodimer (Fig. 7D,  lanes 4, 12, and 20). Therefore, the disulfide-linked heterodimers detected in Fig. 7, A and B, were formed by the Bcl-XL and Bax mutants with the indicated cysteines in their helix 1 after they were activated by the BH3 peptide and targeted to the membranes.
When some single-cysteine Bcl-XL and Bax mutants were individually targeted to the membranes, disulfide-linked products were generated. Each of the products has a molecular mass close to that of the corresponding homodimer (Fig. 7, A and B,  indicated by filled or open triangles, respectively), suggesting that these products were disulfide-linked homodimers of Bcl-XL or Bax.
Additional disulfide-linked products were detected when certain single-cysteine Bcl-XL or Bax mutant was targeted to the mitochondria (Fig. 7, B and C, indicated by stars and  squares). These products were probably heterodimers of the respective single-cysteine mutant and mitochondrial proteins, because treatment of the mitochondria with NEM before targeting the single-cysteine mutant to the mitochondria and oxidation by CuPhe blocked the formation of these products (Fig.  7B, lane 20 versus lane 18; data not shown).
Using the disulfide cross-linking data from the single-cysteine Bcl-XL/Bax mutant pairs, E7C/M20C, S18C/F30C, and S23C/R34C, we built a molecular model for the helix 1⅐helix 1 interface in the Bcl-XL⅐Bax complex (Fig. 8A). We used the FlexPepDock program (30) to dock the helix 1 peptide from the NMR structure of Bax monomer (PDB entry 1F16) against the helix 1 in Bcl-XL protein from the crystal structure of the Bcl-XL protein⅐Bax BH3 peptide complex (PDB entry 3PL7). The two helix 1 regions, each with the three cysteine substitutions, were aligned in parallel and positioned such that the C 1 ␤ -C 2 ␤ distance and the C 1  output models were ranked by their Rosetta energy scores (not shown). One of the top 10 models is shown in Fig. 8A (left). The distribution of the scores was broad when plotted against the root mean square deviation with the starting model. In other words, the bottom of the docking funnel was flat. The top 10 models differed in their scores by only 1-2 Rosetta energy units. In the selected model, the S 1 ␥ -S 2 ␥ distances and the C 1 ␤ -S 1 ␥ -S 2 ␥ -C 2 ␤ dihedral angles in the three cysteine pairs are near the ranges that allow disulfide formation (33).
We then replaced the three cysteine pairs in the structure shown in the left panel of Fig. 8A with the corresponding wildtype residues and used the resulting structure as a starting model in the FlexPepDock program to generate another 200 output models for the wild-type helix 1⅐helix 1 heterodimer interface. The right panel of Fig. 8A depicts one of the top 10 models, which predicts how Bcl-XL and Bax interact via their helix 1 when the two helices are parallel to each other. The model also indicates how the two helices form a previously unreported dimer interface that has the three wild-type Bcl-XL-Bax residue pairs, Glu 7 -Met 20 , Ser 18 -Phe 30 , and Ser 23 -Arg 34 , in proximity to each other. The model predicted some stabilizing interactions, including a hydrogen bond between the side chain N ⑀ of Bax Arg 34 and the backbone carbonyl oxygen of Bcl-XL Gln 19 , and complete or partial burial of the hydrophobic side chains of Bax Leu 26 , Leu 27 , and Phe 30 and Ile 31 . These favorable interactions are countered by the burial of the negatively charged side chain of Bcl-XL Glu 7 . This model was also consistent with the photocross-linking data because the residues that could form photoadducts, when replaced with ANBlysine (Fig. 6), were also in or near the dimer interface (Fig. 8B).
The Helix 1 Regions of Bcl-XL and Bax Can Also Interact in an Antiparallel Fashion to Form an Alternative Interface in the Heterodimer-To determine whether the helix 1⅐helix 1 heterodimer interface can form only in the parallel orientation, we  (Fig. 3), presented in stick form, and their ␤-carbon atoms (C 1 ␤ and C 2 ␤ ) linked by dashed lines with the distances in Å indicated. Cys 62 , a native cysteine in the BH3 region of Bax, is also presented in stick form with its C ␤ linked to the C ␤ of Bcl-XL Val 126 and the distance in Å indicated. B, the two Bcl-XL/Bax residue pairs, Val 126 /Leu 59 and Leu 194 /Met 74 , shown in A are replaced by two cysteine pairs, which are presented in stick form with their ␥-sulfur atoms (S 1 ␥ and S 2 ␥ ) linked by virtual bonds and the distances in Å indicated. The C 1 ␤ -S 1 ␥ -S 2 ␥ -C 2 ␤ dihedral angles about the V126C/L59C and L194C/M74C disulfide bonds are Ϫ52.2 and ϩ22.0°, respectively. C, the crystal structure of the Bcl-XL protein⅐Bax BH3 peptide complex is shown with the residues that generated the heterodimer-specific photoadducts when replaced by ANB-lysine (Fig. 4) presented in stick form. In all panels, the BH1, BH2, BH3, and BH4 regions of Bcl-XL protein are colored blue, cyan, red, and orange, respectively, and the BH3 region of the Bax peptide is colored green. did disulfide cross-linking experiments using mismatched single-cysteine mutant pairs that were activated by the BH3 peptide and targeted to the MOM liposomes. Surprisingly, Bcl-XL E7C or S23C mutant with the cysteine located near the N or C terminus of helix 1 formed a disulfide-linked dimer with Bax R34C or M20C with the cysteine located near the C or N terminus of helix 1, respectively (Fig. 9A). These disulfide crosslinking data support an "antiparallel dimer model" in which the two helix 1 regions from Bcl-XL and Bax are aligned antiparallel to each other (Fig. 9B). The model predicted stabilizing contributions from complete or partial burial of the hydrophobic side chains of Bcl-XL Val 10 and Trp 24 and Bax Ile 19 , Leu 27 , Phe 30 , and Ile 31 . The Rosetta energy score (not shown) for the antiparallel model (Fig. 9B) was very close to that of the parallel model (Fig.  8A), suggesting that the helix 1⅐helix 1 dimer interface equally sampled the two arrangements of the Bax peptide. Further, the Rosetta energy scores calculated from the helix 1⅐helix 1 interface models were similar to that from the BH1-3 groove⅐BH3 helix interface model ( Fig. 5A; data not shown), implying similar contributions from these interactions to the overall stability of the Bcl-XL⅐Bax complex. These Rosetta energy scores were also similar to those reported previously for the groove⅐BH3 interfaces formed by Bcl-XL and BH3 peptides from various BH3-only proteins (37), suggesting similar affinities for the binding of Bcl-XL to the Bax peptides and the BH3 peptides. On the other hand, the flexible binding interaction at the helix 1⅐helix 1 interface (Figs. 8A and 9B) contrasted with the rigid binding interaction at the groove⅐BH3 helix interface (Fig. 5A). We thus investigated their relationship and contribution to the function of the Bcl-XL⅐Bax complex with the mutational analyses described below.
The Helix 1⅐Helix 1 Interaction Is Largely Dependent on the BH1-3 Groove⅐BH3 Helix Interaction, and Both Interactions Are Important for Bcl-XL to Inhibit Bax-mediated MOMP-To determine whether the novel helix 1⅐helix 1 interface can form independently of the canonical BH1-3 groove⅐BH3 helix interface, we tested several mutations that would disrupt the groove⅐BH3 interface and monitored their effect on both interfaces. According to the groove⅐BH3 interface model (Fig. 5A), Gly 138 in the groove of Bcl-XL has a favorable van der Waals interaction with Asp 71 in the BH3 helix of Bax. Replacement of the Gly 138 with alanine (G138A) would not only eliminate the van der Waals interaction but also generate a severe steric collision between the ␤-carbon of Ala 138 and the backbone carbonyl oxygen of Gly 67 in the BH3 helix of Bax (Fig. 10A, left). The next residue in the groove of Bcl-XL, Arg 139 , uses its side chain to form two hydrogen bonds as part of a strong salt bridge with the side chain of Asp 68 in the BH3 helix of Bax. Replacement of the Arg 139 with aspartate (R139D) would introduce a strong electrostatic repulsion with the Asp 68 of Bax and eliminate the strong salt bridge (Fig. 10A, left). On the Bax side, the side chain of Met 74 in the BH3 helix projects into the fifth hydrophobic pocket in the groove of Bcl-XL and forms favorable van der Waals interactions with the side chain of Tyr 195 in Bcl-XL. Replacement of the Met 74 with glutamate (M74E) would eliminate the hydrophobic interaction and generate a collision with the side chain of Tyr 195 (Fig. 10A, right). As predicted by the model, each mutation abolished the disulfide cross-linking of Bcl-XL with Bax in mitochondria via the V126C/L59C pair in the groove⅐BH3 interface (Fig. 10B, lane 2  versus lanes 4 and 6).
We then monitored the effects of the mutations in the groove⅐BH3 interface on the helix 1⅐helix 1 interface formation. Each of the mutations inhibited the disulfide cross-linking of Bcl-XL and Bax via the S23C/R34C pair in the parallel helix 1⅐helix 1 interface to a different extent (Fig. 10C). Thus, the M74E mutation in Bax BH3 helix completely abolished the cross-linking (lane 6 versus lane 8), whereas the R139D mutation in the Bcl-XL groove inhibited the cross-linking more than the G138A mutation ( lane 12 versus lanes 14 and 16). The three mutations also inhibited the heterodisulfide cross-linking via the E7C/R34C and S23C/M20C pairs in the antiparallel helix 1⅐helix 1 interface to similar extents as those via the cysteine pair in the parallel interface (Fig. 10D, lane 4 versus lane 6 and  lane 12 versus lanes 14 and 16). Therefore, the formation of the helix 1⅐helix 1 interface, no matter whether in the parallel or the antiparallel conformation, is at least partially dependent on the formation of the groove⅐BH3 interface.
We did the MOMP assay to test whether the physical disruption of the interfaces by the mutations impacted the inhibition of Bax-mediated MOMP by Bcl-XL. As shown in Fig. 10E, Bcl-XL with V126C in the BH1-3 groove greatly inhibited the MOMP activity of Bax with L59C in the BH3 helix, whereas Bcl-XL with S23C or E7C in helix 1 greatly inhibited the activity of Bax with R34C in helix 1 (columns 1-3 and 9 -12). However, the single-cysteine Bcl-XL mutants barely inhibited the MOMP activity of the corresponding single-cysteine Bax mutants that also have the M74E mutation (columns 4 -6 and 13-16). Along with the great reduction in Bcl-XL activity against the Baxmediated MOMP, the M74E mutation in Bax also greatly impaired the formation of both the groove⅐BH3 helix interface and the helix 1⅐helix 1 interface, either parallel or antiparallel, as shown by the disulfide cross-linking data (Fig. 10, B (lane 2  6)). Similarly, when the R139D mutation was introduced into Bcl-XL V126C and S23C, the resulting Bcl-XL mutants partially lost the inhibitory activity on Bax L59C-and R34C-mediated or M20C-mediated MOMP, respectively (Fig. 10E, column 3 versus column 7, column 11 versus column 17, or column 21 versus column 22). Consistent with the partial loss of Bcl-XL activity against the Bax-mediated MOMP, the R139D mutation in Bcl-XL blocked the helix 1⅐helix 1 interaction to a lesser extent compared with the M74E mutation in Bax (Fig. 10, C The resulting membrane-bound proteins were processed and analyzed as described in the legend to Fig. 3. C, the in vitro synthesized radioactive (R) and/or non-radioactive (N) single-cysteine Bcl-XL and Bax proteins were activated by the BH3 peptide and targeted to the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria. The resulting mitochondria-bound proteins were processed and analyzed as in B. D, the in vitro synthesized radioactive single-cysteine Bcl-XL and Bax proteins were incubated in the absence or presence of the MOM liposomes and/or the BH3 peptide. The resulting proteoliposomes were processed and analyzed as in A. In all panels, the labels are as described in the legend to Fig. 1. In addition, filled or open triangles indicate the disulfide-linked Bcl-XL or Bax homodimers, and stars or squares indicate the disulfide-linked Bcl-XL or Bax⅐mitochondrial protein complexes, respectively. NEM-Mito in B indicates that the mitochondria used in the reaction were pretreated with NEM to block the sulfhydryl moieties in the mitochondrial proteins, preventing their cross-linking with the sulfhydryl moieties in the Bcl-XL and Bax proteins. A, left, to build the initial model, the helix 1 extracted from the Bax monomer structure (PDB entry 1F16) was manually positioned onto the helix 1 of the Bcl-XL protein in the Bcl-XL protein⅐Bax BH3 peptide complex structure (PDB entry 3PL7), such that the three cysteine pairs, E7C/M20C, S18C/F30C, and S23C/R34C, which resulted in disulfide-linked Bcl-XL⅐Bax heterodimers in Fig. 7, are in a geometry suitable for disulfide linkage. The resulting complex structure was input into FlexPepDock program, and one of the top 10 output models is shown with the cysteine pairs presented in stick form and their ␥-sulfur atoms (S 1 ␥ and S 2 ␥ ) linked by virtual bonds and the distances in Å indicated. The C 1 ␤ -S 1 ␥ -S 2 ␥ -C 2 ␤ dihedral angles about the disulfide bond for the three cysteine pairs are ϩ99.2°, ϩ114.5°, and Ϫ178.6°, respectively. Right, to generate the final model, the cysteines in the left panel were changed back to the corresponding wild-type residues. The resulting complex structure was the starting model in an automated peptide docking experiment with the FlexPepDock program. One of the top 10 output models from the docking experiment is shown with the respective wild-type residue pairs presented in stick form and their ␤-carbon atoms (C 1 ␤ and C 2 ␤ ) linked by dashed lines with the distances in Å indicated. B, a model for the parallel helix 1⅐helix 1 interface in the Bcl-XL⅐Bax heterodimer was built based on the model presented in the right panel of A and the photocross-linking data shown in Fig. 6. The residues that generated the heterodimer-specific photoadducts when replaced by ANB-lysine are presented in stick form. In all models, the BH1-4 regions of Bcl-XL protein are colored as in Fig. 5, and the helix 1 of Bax is colored magenta. For simplicity, the BH3 helix of Bax was omitted from the models.  12 versus lane 16)), although the two mutations blocked the groove⅐BH3 helix interaction equally (Fig. 10B, lane 2 versus lanes 4 and 8). Interestingly, the effect of the G138A mutation in Bcl-XL on its inhibition of Bax-mediated MOMP depended on the cysteine location in the mutants. Whereas the G138A mutation caused a partial loss in the Bcl-XL V126C inhibition of Bax L59C (Fig. 10E, column 3 versus column 8), it did not affect the Bcl-XL S23C inhibition of Bax R34C or M20C (Fig. 10E, column 11 versus column 18 or  column 21 versus column 23). The differential effects of the G138A mutation on the anti-Bax activity of Bcl-XL in the MOMP could be explained by the differential effects of the mutation on the physical interaction of Bcl-XL with Bax. Thus, as indicated by the disulfide cross-linking data, the G138A mutation largely disrupted the groove⅐BH3 helix interaction between Bcl-XL V126C and Bax L59C (Fig. 10B, lane 2 versus lane 6) but only slightly inhibited the helix 1⅐helix 1 interaction between Bcl-XL S23C and Bax R34C or M20C (Fig. 10, C (lane 12 versus lane 14) and D ( lane  12 versus lane 14)). Therefore, the MOMP activity data from these Bcl-XL and Bax mutants are directly correlated with the physical interaction data, supporting a conclusion that the physical interactions in both interfaces are important for Bcl-XL to inhibit Bax-mediated MOMP.
The BH1-3 Groove of Bcl-XL Directly Competes with the BH1-3 Groove of Bax for Binding to the BH3 Helix of Bax to Inhibit Bax Homo-oligomerization in the MOM-The effects of the mutations in the BH1-3 groove of Bcl-XL on its interaction with Bax and inhibition of Bax-mediated MOMP suggest a critical role for the groove in the function of Bcl-XL. Conceivably, FIGURE 9. Disulfide cross-linking of Bax and Bcl-XL proteins with a single cysteine in their helix 1 that supports the antiparallel helix 1⅐helix 1 dimer model. A, the in vitro synthesized [ 35 S]Met-labeled single-cysteine Bcl-XL and Bax proteins were activated by the BH3 peptide and targeted to the MOM liposomes. The resulting membrane-bound proteins were processed and analyzed as described in Fig. 7. B, the antiparallel helix 1⅐helix 1 dimer model was based on the disulfide cross-linking data from two single-cysteine Bcl-XL/Bax mutant pairs, E7C/R34C and S23C/M20C, shown in A. It was generated by the FlexPepDock program and presented similarly as described in the legend to Fig. 8A. Left, the two cysteine pairs are presented in stick form with their ␥-sulfur atoms (S 1 ␥ and S 2 ␥ ) linked by virtual bonds and the distances in Å indicated. The C 1 ␤ -S 1 ␥ -S 2 ␥ -C 2 ␤ dihedral angles about the disulfide bond for the two cysteine pairs are ϩ146.7°and Ϫ74.0°, respectively. Right, the two respective wild-type residue pairs are presented in stick form with their ␤-carbon atoms (C 1 ␤ and C 2 ␤ ) linked by dashed lines and the distances in Å indicated. The BH1-4 regions of Bcl-XL protein and the helix 1 of Bax are colored as in Fig. 8. For simplicity, the BH3 helix of Bax was omitted from the models.  C and D). The mitochondria-bound proteins were processed and analyzed as described in the legend to Fig. 3. E, the indicated Bax proteins were synthesized in vitro, and their BH3 peptide-dependent cytochrome c release activities in the Bax Ϫ/Ϫ /Bak Ϫ/Ϫ mitochondria were assayed in the absence or presence of the indicated in vitro synthesized Bcl-XL proteins as described in the legend to Fig. 2. Data shown are average fractions of cytochrome c release from 2-4 independent experiments with the ranges indicated by error bars. the BH1-3 groove in Bcl-XL would compete with a similar groove in Bax that has been suggested to bind the BH3 helix of Bax nucleating Bax oligomerization in the MOM (13). To test this scenario directly, we monitored the effect of Bcl-XL on Bax homodisulfide cross-linking via T56C in the BH3 helix and R94C in the BH1-3 groove that was previously used to capture a Bax homodimer formed via the BH3 helix⅐groove interface (12,13). The single-cysteine Bax mutants formed a disulfidelinked homodimer after they were activated by the BH3 peptide and targeted to the MOM liposomes (Fig. 11A, lane 2, open  triangle), consistent with the previous finding from the mitochondria isolated from the apoptotic cells that expressed the same Bax mutants (12). The addition of purified recombinant Bcl-XL protein to the reaction inhibited the disulfide crosslinking of the Bax mutants (Fig. 11A, lane 4 versus lane 2). To demonstrate that the inhibition is directly caused by the competition between the BH1-3 grooves of Bcl-XL and Bax for the same BH3 helix of Bax, we added Bcl-XL E129C mutant with a single cysteine in the BH1-3 groove to the Bax T56C with a single cysteine in the BH3 helix. As predicted by the groove⅐BH3 helix interface model for Bcl-XL⅐Bax heterodimer (Fig. 5), Bcl-XL E129C formed a disulfide-linked heterodimer with Bax T56C after they were activated by the BH3 peptide and targeted to the MOM liposomes (Fig. 11B, lane 2, arrow), suggesting that the similar BH1-3 grooves in Bcl-XL and Bax can interact with the same residue in the BH3 helix of Bax, resulting in a direct competition. As expected, the R139D mutation in the groove of Bcl-XL E129C largely inhibited the heterodisulfide linkage with Bax T56C (Fig. 11B, lane 6 versus lane 2). The R139D mutation also partially abolished the inhibition of Bax T56C-mediated MOMP by Bcl-XL E129C (Fig. 11D, columns  1-3 versus column 5). Altogether, these experiments demonstrated that the inhibition of Bax MOMP activity by Bcl-XL is directly mediated by the competition between the Bax homointeraction and the heterointeraction with Bcl-XL in the MOM via the similar BH3 helix⅐BH1-3 groove interfaces.
ABT-737 Dissociates the Membrane-bound Bcl-XL⅐Bax Heterodimer and Restores Bax Homo-oligomerization and MOMP Activity-ABT-737, a BH3 mimic compound, binds to the BH1-3 groove of Bcl-XL and inhibits its interaction with BH3only proteins (22, 38 -40). Because Bcl-XL binds to Bax not only through the BH1-3 groove⅐BH3 interface but also the helix 1⅐helix 1 interface, we tried to determine whether ABT-737 could fully inhibit the Bcl-XL interaction with Bax in membranes. We first tested the effect of ABT-737 on the formation of the two interfaces. When ABT-737 was included in the reaction with Bcl-XL E129C, Bax T56C, the BH3 peptide, and the MOM liposomes, it dramatically blocked the disulfide crosslinking of the two proteins in the membranes (Fig. 11B, lane 2  versus lane 4), suggesting that ABT-737 outcompetes the BH3 helix of Bax for interaction with the groove of Bcl-XL. As predicted from the dependence of helix 1⅐helix 1 interface formation on groove⅐BH3 interface formation, the compound also partially inhibited the disulfide cross-linking of Bcl-XL S23C and Bax R34C with the cysteines located in the helix 1⅐helix 1 interface (Fig. 11C, lane 2 versus lane 4). Moreover, the compound almost fully reversed the inhibition of the homodisulfide cross-linking between Bax T56C and Bax R94C by recombinant Bcl-XL (Fig. 11A, lanes 4 versus 6), indicating that ABT-737 has freed the BH3 helix of Bax from the groove of Bcl-XL, which in turn binds to the BH1-3 groove of other Bax. Consistent with the ABT-induced reformation of the Bax homocomplex in the presence of Bcl-XL, the addition of ABT-737 almost completely reversed the Bcl-XL inhibition on the MOMP by BH3 peptideactivated Bax (Fig. 11D, column 3 versus column 4 and column 8  versus column 9).
Heterodimerization of Bcl-XL and Bax Does Not Alter Their Membrane Topology-To determine the topology of the Bcl-XL⅐Bax heterodimer bound to the liposomal membrane after activation by the BH3 peptide, we used the compartment-specific IASD labeling. We first did the experiments using Bcl-XL and Bax mutants with a single cysteine located in the interfacial regions. As shown in Fig. 12A, in the presence of excess purified recombinant Bax protein, the in vitro synthesized Bcl-XL S18C or S23C with the cysteine in helix 1 or V126C with the cysteine in the BH1-3 groove was labeled by IASD without using the detergent CHAPS to solubilize the membrane or the denaturant urea to destabilize the proteins or their complexes. The addition of CHAPS, urea, or both did not increase the labeling significantly. Likewise, in the presence of excess purified recombinant Bcl-XL protein, the CHAPS-and urea-independent IASD labeling was observed for the in vitro synthesized Bax F30C or R34C with the cysteine in helix 1 or L59C with the cysteine in the BH3 helix (Fig. 11B). Because most of the in vitro synthesized Bcl-XL or Bax should be in complex with the purified Bax or Bcl-XL in the presence of excess purified Bax or Bcl-XL, respectively, the IASD labeling data suggest that these interfacial residues in the Bcl-XL⅐Bax heterodimer are located outside of the membrane and are accessible from the aqueous milieu. In contrast, the IASD labeling of Bcl-XL L194C with the cysteine in the groove near the C terminus of helix 8 (Fig. 12A) or Bax M74C with the cysteine near the C terminus of the BH3 helix (Fig. 12B) was increased by the addition of CHAPS, urea, or both, indicating that these residues are buried at least partially in the membrane, in the protein, or in the protein complex.
We next determined the effect of Bcl-XL on the location of helices 5, 6, and 9 of Bax. We previously found that these three helices are embedded in the MOM in the activated mitochondrial Bax (8). As shown in Fig. 12C (top), in the presence of excess purified Bcl-XL, most of the in vitro synthesized, BH3 peptide-activated, and MOM liposome-bound Bax V121C, T140C, and S184C with the cysteine located in helices 5, 6, and 9, respectively, were not labeled by IASD unless CHAPS was added to the reaction either by itself or together with urea. These results suggest that these three residues, and these three helices by inference, were buried in the membrane at least partially. Similar IASD labeling profiles were observed for the three Bax mutants when the Bcl-XL was absent (Fig. 12C, top), suggesting that the activated Bax, which was not in complex with Bcl-XL, had a membrane topology similar to that of the inactivated Bax, which was in complex with Bcl-XL. The fractions of IASD-labeled Bax in the absence or presence of Bcl-XL with or without CHAPS from 2-5 independent experiments were quantified and shown in Fig. 12C (bottom). These quantitative data indicate that Bcl-XL binding to Bax did not significantly alter the multispanning topology of the activated Bax in the membrane.
We also determined the effect of Bax on the location of helices 5, 6, and 9 of Bcl-XL, the three putative membrane-spanning helices, based on the structural homology with Bax and with the bacterial pore-forming toxins (41,42). In the presence and absence of excess purified Bax, most of the Bcl-XL T216C and V224C with the cysteine in helix 9 were not labeled by IASD in the absence of CHAPS and urea, and the addition of either CHAPS or urea increased the labeling significantly (Fig. 12D,  top). These results suggest that these residues, and helix 9 by inference, were buried partially in the membrane and partially in the protein or the protein complex. In contrast, most of the Bcl-XL E153C and S164C with the cysteine in helices 5 and 6, respectively, were labeled by IASD in the absence of CHAPS and urea, and the addition of CHAPS, urea, or both did not alter the labeling significantly. These results suggest that these two residues, and helices 5 and 6 by inference, were mostly exposed to an aqueous milieu. The average fractions of IASD-labeled Bcl-XL in the absence or presence of Bax with or without CHAPS from three to four independent experiments (Fig. 12D, bottom) indicated that Bax binding to Bcl-XL did not change the tail-anchored topology of Bcl-XL in the membrane. Based on all of the IASD labeling data, we concluded that (i) the presence of the other protein did not significantly alter the topology of Bax and Bcl-XL in the MOM liposomal membrane; (ii) Bax in the heterodimer was probably in the multispanning conformation, whereas Bcl-XL was probably in the tail-anchored conformation; and (iii) the two interfaces in the heterodimer were largely located outside of the membrane.
In contrast to Bcl-XL, two bands were detected in the IEF gel for the in vitro synthesized Bax proteins, even in the absence of IASD labeling (Fig. 12, B and C, lane 1 in all panels). The two bands were also observed with the cysteine-null (C0) Bax mutant in the absence of IASD labeling (Fig. 12E). When a positively or negatively charged residue in Bax C0 was changed to an uncharged cysteine, the bands shifted accordingly in the IEF gel relative to the bands of Bax C0 (Fig. 12E). We suspected that a post-translational modification specific to the wheat germ-based in vitro translation system altered the isoelectric point of the Bax proteins, causing the differential migrations in the IEF gel, because the Bax C0 protein synthesized in the reticulocyte lysate-based system migrated as a single band (Fig.  12E). Nevertheless, the modified and unmodified Bax proteins were equally accessible to IASD labeling under most conditions (Fig. 12, B and C).
The Overall Structural Organization of Bcl-XL⅐Bax Heterodimer in the Membrane-Because both Bcl-XL and Bax have one or more helices embedded in the membrane, it is conceivable that these helices may interact to form an additional heterodimer interface inside the membrane. We used various cross-linking methods to test this possibility. The photocrosslinking data from photoreactive probes located in these membrane-embedded helices indicate the proximity of these helices to the other protein in the heterodimer (Fig. 13). However, the disulfide cross-linking experiments with cysteine placed throughout these helices did not capture any heterodimer (data not shown), suggesting that the distances between these helices were too long to be bridged by a disulfide bond. In support of this notion, chemical cross-linking with BMH, a membranepermeant and sulfhydryl-reactive bifunctional reagent, generated a heterodimer between Bcl-XL T219C with the cysteine in helix 9 and Bax K123C, R134C, or L181C with the cysteine in helix 5, 6, or 9, respectively (Fig. 14). Because the two sulfhydryl-reactive maleimides in BMH were spaced ϳ13 Å apart, and the reactive nitrene in the photoreactive probe was ϳ12 Å away from the C ␣ of the lysine, both photocross-linking and chemical cross-linking data were consistent with a model in which helix 9 of Bcl-XL is embedded together with helices 5, 6, and 9 of Bax in the membrane, without being close enough to form an additional heterodimer interface.
We also used chemical cross-linking to assess the distance between the cytosolic part and the membrane-embedded part of the Bcl-XL⅐Bax heterodimer. As shown in Fig. 14, Bcl-XL T216C or T219C with the cysteine in helix 9 was cross-linked to Bax Cys 62 , M74C, or M79C with the cysteine in the BH3 helix by BMH, whereas Bcl-XL L194C with the cysteine in helix 8 was cross-linked to Bax R94C with the cysteine in helix 4. Thus, the distances between these paired cysteines were ϳ13 Å. These experimentally determined distances, together with the information about the interface and the membrane topology, were used to build a schematic model for the overall structural organization of the Bcl-XL⅐Bax heterodimer that was partially embedded in the membrane (Fig. 15).

DISCUSSION
The BH1-3 Groove⅐BH3 Helix Interaction in the Membranebound Bcl-XL⅐Bax Heterodimer and the Functional Consequence-Accumulating mutagenesis and structure data indicate that a Bcl-XL BH1-3 groove⅐Bax BH3 helix interface is the exclusive site of interaction between the two proteins and therefore responsible for inhibition of Bax by Bcl-XL. However, it was still unclear whether this interaction existed in the membrane-bound complex and how this interaction integrated into the inhibitory function of Bcl-XL on Bax-mediated MOMP. We used sitespecific cross-linking approaches to systematically map the interface of the Bcl-XL⅐Bax heterodimer formed at the natural and artificial MOM. Collectively, our data support a direct physical interaction between the hydrophobic BH1-3 groove of Bcl-XL and the amphipathic BH3 helix of Bax for the membrane-bound proteins, with an interface similar to that revealed by crystallography for the soluble Bcl-XL protein and Bax BH3 peptide (17). These data included the following: (i) formation of disulfide bonds between cysteines in the groove of Bcl-XL and those in the BH3 helix of Bax; (ii) formation of heterodimer-specific photoadducts via ANB-labeled lysines in the groove and the BH3 helix; and (iii) disruption of the groove⅐BH3 interaction by the mutations in the groove and the BH3 helix and by  ) were synthesized in the wheat germ-based in vitro system. The resulting radioactive Bcl-XL or Bax protein, either alone or together with the purified non-radioactive recombinant Bax (rBax) or Bcl-XL (rBcl-XL) protein, respectively, were activated by the BH3 peptide and targeted to the MOM liposomes. The resulting proteoliposomes were isolated and treated with IASD in the absence or presence of CHAPS, urea, or both. After 30 min, the labeling reactions were stopped by ␤-mercaptoethanol. For the 0 min controls, the samples were pretreated with ␤-mercaptoethanol before the addition of IASD. The resulting radioactive proteins were resolved using IEF and detected by phosphorimaging. Circles and triangles, unlabeled and IASD-labeled Bcl-XL proteins, respectively; square and angle brackets, unlabeled and IASD-labeled Bax, respectively. The phosphorimaging data for IASD labeling of radioactive Bcl-XL and Bax mutants in the top panels of C and D and the similar data from 1-4 independent replicates were quantified. In the corresponding bottom panels, the average fractions of IASD labeling for the mutants under the specified conditions are presented as bar graphs of the specified patterns with the ranges indicated by error bars. E, the indicated [ 35 S]Met-labeled Bax mutants were synthesized in the wheat germ (WG)-or rabbit reticulocyte lysate (RRL)-based in vitro translation system and analyzed with IEF and phosphorimaging. The K119C or D142C mutant was constructed from the C0 mutant by changing Lys 119 or Asp 142 to cysteine, respectively. the groove-binding compound ABT-737. Interestingly, previous studies showed that two of the mutations, G138A in Bcl-XL and M74E in Bax, which inhibited the heterodimerization in membranes in our study, abolished Bcl-XL inhibition of Bax in cells (17,43). In addition, ectopic expression of Bax in cells enhances sensitivity to ABT-737 when Bcl-XL is present (44). Consistent with these findings, these mutations and ABT-737 reversed the Bcl-XL repression of Bax-mediated MOMP in our cell-free system, suggesting that the BH1-3 groove⅐BH3 helix interaction in the membrane-bound Bcl-XL⅐Bax heterodimer inhibits Bax.
It was suggested that the BH3 motif of Bax is exposed upon activation, resulting in Bax autoactivation and homo-oligomerization (5, 9, 10, 12, 13, 24, 45). The results presented here show that the BH3 motif of Bax is sequestered in the BH1-3 groove of Bcl-XL and that this heterointeraction inhibits Bax homointeraction in membranes via its BH3⅐BH1-3 groove interface. Therefore, the binding of Bcl-XL to Bax could potentially neutralize two activities mediated by the BH3 motif of active Bax in membranes: (i) the recruitment of cytosolic Bax to the membranes by preventing Bax autoactivation, facilitating retrotranslocation of Bax from mitochondria to cytosol (46), or both and (ii) the homo-oligomerization of membrane-bound Bax. Both of these activities are essential for Bax-mediated MOMP.
The Helix 1⅐Helix 1 Interaction in the Membrane-bound Bcl-XL⅐Bax Heterodimer and the Functional Implication-Although it was well documented that the BH3 motifs of proapo- ptotic members of Bcl-2 family and the BH1-3 groove of antiapoptotic members dictate their binding affinities and specificities in solution (47)(48)(49), it was speculated that regions other than BH1-3 in Bax and Bcl-XL may contribute to their hetero-and homointeractions and functions. First, Bax homo-oligomerization is mediated not only by the BH1-3 regions but also by helices 1 and 6 and the loop connecting helices 1 and 2 (9,10,12). Second, Bax activation by BH3only proteins and autoactivation by previously activated Bax are mediated by a BH3-binding triggering pocket composed of helices 1 and 6 and the loop connecting helices 1 and 2 (7,45,50,51). Third, the helix 1 of Bcl-XL, which overlaps with the BH4 region, is essential for binding to soluble Bax, thereby preventing Bax from inserting into membranes and perhaps retrotranslocating Bax from membranes to cytosol and is hence required for Bcl-XL to inhibit Bax-mediated apoptosis (5,52). Fourth, the 6A7 epitope of Bax, which overlaps with helix 1, is not accessible to the 6A7 antibody in the Bax sequestered by Bcl-XL in cells (39,53).
Here, we demonstrated that the helix 1 regions in membrane-bound Bax and Bcl-XL directly interacted with each other and contributed to the functionality of the membranebound heterodimer. In contrast to the rigid groove⅐BH3 interface, the helix 1⅐helix 1 interface is flexible. The groove and the BH3 helix are formed by well organized structural elements that are ideal for forming a rigid binding interface. In contrast, the helix 1 in both proteins is connected to the rest of the protein by a long and unstructured loop that provides the freedom for the helices to sample different binding interfaces to achieve a flexible interaction. It will be interesting to explore whether and how the multiple BH and non-BH regions of Bcl-XL and Bax are involved in the interactions with other family members at membranes, because these interactions are expected to alter the functional Bcl-XL⅐Bax interaction according to our embedded together model (3,54). The assays established here will be valuable tools for these explorations.
Beside disulfide-linking to each other, Bcl-XL and Bax with a single cysteine in the BH4/helix 1 region could be disulfidelinked to a few unknown mitochondrial proteins (Fig. 7), suggesting that this region of Bcl-XL and Bax might exert activities other than regulation of MOMP. Bcl-XL was shown to regulate FIGURE 14. Chemical cross-linking of single-cysteine Bcl-XL and Bax proteins. The indicated in vitro synthesized [ 35 S]Met-labeled single-cysteine Bcl-XL and Bax proteins were activated by the BH3 peptide and targeted to the MOM liposomes. The resulting proteoliposomes were isolated and subjected to BMH cross-linking. The resulting radioactive proteins and their adducts were analyzed by reducing SDS-PAGE and phosphorimaging. Arrows, BMH-cross-linked heterodimers. Other labels are the same as those in Fig. 3.
FIGURE 15. A model for the overall structural organization of the Bcl-XL⅐Bax heterodimer in membranes. The cytosolic part of the heterodimer model was assembled by merging the BH1-3 groove⅐BH3 helix interface model (Fig. 5A) with the parallel helix 1⅐helix 1 interface model (Fig. 8A, right). The MOMembedded part was assembled with the ␣-helix 9 (␣9) of Bcl-XL and the ␣-helices 5, 6, and 9 (␣5, ␣6, and ␣9) of Bax. The distances between these membraneembedded helices were set according to the BMH cross-linking data (Fig. 14). The distances between the helix ␣8 or ␣9 of Bcl-XL and the BH3 helix (␣BH3) or helix ␣4 of Bax, respectively, were further adjusted based on the BMH cross-linking data. Dashed lines link the residues that were cross-linked by BMH when they were mutated to cysteines. The residues shown as spheres are embedded in the MOM, because their cysteine substitutions could not be substantially labeled by IASD unless CHAPS was added (Fig. 12). The residues shown as sticks are exposed to the cytosol, because their cysteine substitutions could be labeled by IASD in the absence of CHAPS and urea. mitochondrial membrane potential by modulating the activity of voltage-dependent anion channel via an interaction that requires the BH4/helix 1 region (55). The soluble form of Bax, which was unable to promote MOMP, positively regulated mitochondrial fusion through interaction with mitofusin-2 (56). On the other side, the MOM-bound Bax, even when sequestered by Bcl-XL and unable to induce MOMP, promoted mitochondrial fission (39,57). Interaction between Bcl-XL and Bax may regulate their interactions with these and other mitochondrial proteins and the corresponding activities, some of which are intimately connected with apoptosis (58,59).
Does the Structural Difference between the Bax Homocomplex and the Bax⅐Bcl-XL Heterocomplex Explain Their Functional Difference?-Despite the striking similarity in their monomeric structure, it is puzzling how Bcl-XL interacts with Bax and thereby halts the homo-oligomerization of Bax. It is also unclear why Bcl-XL cannot form large oligomeric pores in membranes but Bax can. Now it becomes even more perplexing that Bcl-XL and Bax interact at membranes via two interfaces, the canonical groove⅐BH3 helix interface and the novel helix 1⅐helix 1 interface, which potentially can mediate the formation of a higher order hetero-oligomer. However, the disruptive mutations in the former interface also partially disrupt the latter one, suggesting that the formation of the latter interface partially depends on the formation of the former one. One explanation of this result is that both interfaces are used to form a more stable heterodimer, as shown in our overall structural model (Fig. 15). Alternatively, the binding mediated by one interface is too weak to extend the heterodimer that is formed by the other interface to a higher order oligomer. In addition, Bax homo-oligomers are probably formed by the monomers that are deeply embedded into membranes (8). In contrast, our topology data for the Bcl-XL⅐Bax heterocomplex (Fig. 12) suggest that although the embedding of the Bax in the heterocomplex in the membrane is similar to that of the Bax in the homo-complex, the embedding of Bcl-XL seems to be shallower compared with that of Bax. Therefore, the different conformations of Bcl-XL and Bax in the membrane may be another factor that distinguishes the activities of the two proteins in the membrane.
Are Bcl-XL and Bcl-2 Different?-Besides the subcellular localization, the differences between Bcl-2 and Bcl-XL, two antiapoptotic proteins with similar soluble domain structures, were implied in many ways. First, Bcl-2 and Bcl-XL have different binding spectrums to BH3-only proteins as well as to Bax and Bak (39,48). Second, Bcl-XL is more effective than Bcl-2 in inhibition of apoptosis in some cell types (60). Third, our ISAD labeling data suggest that the Bcl-XL in complex with Bax may be in the same tail-anchored conformation that it adopts when targeted to membranes in the absence of Bax. In contrast, the full antiapoptotic function of Bcl-2, including inhibition of Bax, requires a changed membrane-bound conformation (18,20). Fourth, we found that Bcl-2 can be converted to a Bax-like molecule through interaction with Bim and tBid, whereas a similar conversion of Bcl-XL was not detected at least in liposomes (61).
Conclusions-Our study revealed two interfaces in the membrane-bound Bcl-XL⅐Bax complex. The first one is rigid with the BH3 helix of Bax sequestered in the BH1-3 groove of Bcl-XL, whereas the second one is flexible with the two helices 1 interacting either in parallel or antiparallel. Formation of the second interface largely depends on the first one, yet both contribute to the overall stability of the heterodimer that inhibits homo-oligomerization of Bax and subsequent MOMP. Therefore, the rigid BH3⅐groove interface positions the two helices 1 such that they form a flexible interface, which contributes significantly to energetic stabilization of the membrane-bound Bcl-XL⅐Bax heterodimer that is pivotal to apoptosis regulation.