Bcl-2 Homodimerization Involves Two Distinct Binding Surfaces, a Topographic Arrangement That Provides an Effective Mechanism for Bcl-2 to Capture Activated Bax*

The homo- and heterodimerization of Bcl-2 family proteins is important for transduction and integration of apoptotic signals and control of the permeability of mitochondria and endoplasmic reticulum membranes. Here we mapped the interface of the Bcl-2 homodimer in a cell-free system using site-specific photocross-linking. Bcl-2 homodimer-specific photoadducts were detected from 11 of 17 sites studied. When modeled into the structure of Bcl-2 core, the interface is composed of two distinct surfaces: an acceptor surface that includes the hydrophobic groove made by helices 2 and 8 and the loop connecting helices 4 and 5 and a donor surface that is made by helices 1-4 and the loop connecting helices 2 and 3. The two binding surfaces are on separate faces of the three-dimensional structure, explaining the formation of Bcl-2 homodimers, homo-oligomers, and Bcl-2/Bax hetero-oligomers. We show that in vitro the Bcl-2 dimer can still interact with activated Bax as a larger oligomer. However, formation of a Bax/Bcl-2 heterodimer is favored, since this interaction inhibits Bcl-2 homodimerization. Our data support a simple model mechanism by which Bcl-2 interacts with activated Bax during apoptosis in an effective manner to neutralize the proapoptotic activity of Bax.

Bcl-2 family proteins are key regulators of apoptosis. These proteins share sequence homology in Bcl-2 homology (BH) 1 domains and function to promote or prevent apoptosis. Antiapoptotic proteins such as Bcl-2 and Bcl-x L show homology in four BH domains (BH1 to -4). Proapoptotic proteins can be grouped into "multidomain" and "BH3-only" subfamilies. Mul-tidomain proapoptotic proteins such as Bax and Bak display homology in BH domains 1-3, whereas BH3-only proteins such as Bid and Bim are similar structurally to multidomain family members, but sequence similarity is limited to only the BH3 domain. The current model for how Bcl-2 family proteins regulate apoptosis involves three sequential processes: (i) BH3only proteins are activated by various death signals; (ii) the active BH3-only proteins then either activate multidomain proapoptotic proteins or inhibit antiapoptotic proteins or both; and (iii) unless inhibited by antiapoptotic Bcl-2 proteins, activated multidomain proapoptotic proteins form oligomers in the mitochondrial outer membrane that release proapoptotic proteins such as cytochrome c and Smac/DIABLO from the mitochondrial intermembrane space. The released proteins trigger activation of the caspases and nucleases that eventually dismantle the cell (1)(2)(3)(4).
The interaction between Bcl-2 family proteins is obviously important for coupling apoptotic signals to the alteration of mitochondrial membrane permeability. Three-dimensional structures of monomeric antiapoptotic proteins such as Bcl-x L , Bcl-2, Bcl-w, and CED-9 as well as proapoptotic proteins such as Bax and Bid have been determined (5)(6)(7)(8)(9)(10)(11)(12). A hydrophobic surface groove common to the structures of all of the antiapoptotic proteins is encircled by five helices (helices 2-5 and 8) and connecting loops. The equivalent hydrophobic groove in the multidomain proapoptotic protein Bax is occluded by its Cterminal hydrophobic helix (helix 9). Although the overall structure is similar, the BH3-only protein Bid does not contain this hydrophobic groove. Further, a complementary hydrophobic surface that would interact with the hydrophobic groove is not obvious in the structures of monomeric Bcl-2 family proteins, making modeling of a dimeric complex difficult (13). However, one possible interface for the heterodimer between anti-and proapoptotic proteins has been deduced from structures of Bcl-x L in complex with a BH3 domain peptide from a proapoptotic protein such as Bak, Bad, or Bim (14 -16). In all of these complex structures, the hydrophobic groove of Bcl-x L binds the peptide as an ␣-helix.
Modeling a heterodimer between an anti-and a proapoptotic protein based on these complex structures provided explanations for some early mutagenesis data. For example, replacement of glycine 145 with alanine (G145A) in the BH1 domain of Bcl-2 and equivalent G138 mutations in Bcl-x L both diminish heterodimerization with Bax and antiapoptotic function (17)(18)(19)(20)(21). In both cases, the glycine replaced is located in the bottom of the hydrophobic groove that has been proposed to constitute one binding interface. Conversely, some mutations in the BH3 domain of Bax, Bad, Bid, and Bim abolish the binding of these proteins to Bcl-2 or Bcl-x L as well as their proapoptotic activity (22)(23)(24)(25)(26)(27)(28)(29). These mutations usually alter the hydrophobic face of the BH3 helix predicted to make critical contacts with the hydrophobic groove of Bcl-x L based on the peptide-bound Bcl-x L structures. Thus, a model for heterodimerization between proand antiapoptotic proteins exists that satisfactorily accounts for the existing structural and functional data.
Much less is known about homodimerization of Bcl-2 family proteins. Most single point mutations created in several Bcl-2 family proteins have failed to disrupt homodimer formation (17,18). Only Jeong et al. (21) reported recently that G138A mutation in Bcl-x L reduced the homodimer formation. More dramatic deletion mutagenesis has been used to disrupt Bcl-2 homodimerization (30,31), but the principle caveat of this approach is that the overall structure of the protein is probably being altered, making molecular interpretation of the mutagenesis results difficult. It is surprising that the structure of the homodimer remains unknown, since homodimers have been reported as the major form of intracellular Bcl-2 family proteins for more than 10 years (e.g. see Refs. 17 and 21). To determine the mechanisms by which Bcl-2 family proteins interact and function under normal as well as apoptotic conditions, it is essential to determine how both homo-and heterodimerization are regulated. As an essential step in understanding these interactions, we have mapped the binding surfaces involved for two functionally well characterized molecules, Bcl-2 and Bax. To identify binding surfaces, we performed cross-linking experiments using both purified and in vitro synthesized proteins, each having a cross-linker attached to a single site. Our physical map for the Bcl-2 homodimer interface confirms a known binding surface, reveals an additional binding surface, and suggests a mechanism by which homodimers can form higher oligomers as well as how homodimers can dissociate to form heterodimers with activated Bax.

EXPERIMENTAL PROCEDURES
Plasmids and Proteins-To construct Bcl-2 mutants, we used a plasmid that contains the coding region of full-length Bcl-2 protein in the vector pSPUTK (32). The lysine-null Bcl-2 mutant (K0) was generated by mutating all four lysine codons in the Bcl-2 coding region to codons for arginine. Other Bcl-2 mutants with a single lysine codon at a particular position were generated by mutating the corresponding codon of the K0 mutant to the lysine codon and are designated here as K followed by a number that indicates the codon position in the sequence. For example, the mutant Bcl-2 K108 has a single lysine codon at position 108 of the Bcl-2 coding sequence. The last 22 codons of the bcl-2 gene were then replaced with a single asparagine codon to create Bcl-2⌬TM mutants with a single lysine at the desired location. These mutants are named similarly; thus, the C-terminal deletion of Bcl-2 Lys 108 is designated Bcl-2⌬TM K108.
All mutagenesis was done using appropriate primers and an overlapping PCR-based method (33). The sequence of all plasmids was verified by DNA sequencing. For transfection of MCF-7 cells, the coding region of Bcl-2 K0 or other single lysine mutants was shuttled from the pSPUTK-based plasmid to the vector pRc/CMV (Invitrogen). Details of all plasmid preparations will be provided upon request.
Prevention of Apoptosis by Bcl-2 K0 in MCF-7 Cells-MCF-7 cell culture, transfection, adriamycin treatment, and focal adhesion kinase (FAK) and poly(ADP-ribose) polymerase (PARP) cleavage assays were conducted as described (36,37). Briefly, MCF-7 cells were grown in ␣-minimal essential medium supplemented with 10% fetal bovine serum. The pRc/CMV vector, or a pRc/CMV plasmid encoding either wild type Bcl-2 (WT Bcl-2) or Bcl-2 K0 was used to transfect MCF-7 cells, and clones that stably express Bcl-2 proteins were selected. The expression of endogenous Bcl-2 in the MCF-7 cells was reduced by growing the cells without estrogen (␣-minimal essential medium without phenol red, supplemented with charcoal-filtered fetal bovine serum) for 6 days prior to adriamycin treatment. After the addition of 10 M adriamycin, the cells were incubated for 48 h and then harvested for FAK and PARP cleavage assays. 10 million cells were lysed in 1.5 ml of ice-cold lysis buffer (37) by passing through a 26-gauge needle four times. Cellular debris and nuclei were removed by centrifugation at 13,000 rpm for 10 min in a microcentrifuge. Aliquots of the supernatant corresponding to equal cell numbers were analyzed by SDS-PAGE followed by immunoblotting using antibodies specific for Bcl-2 (36), FAK (Pharmingen), or PARP (Biomol).
Preparation of ANB-labeled Bcl-2 Proteins-The N ⑀ -(5-azido-2-nitrobenzoyl)-Lys-tRNA Lys (⑀ANB-Lys-tRNA Lys ) was prepared as described (38,39). RNAs encoding the Bcl-2 mutants were prepared from the corresponding pSPUTK-based plasmids using SP6 RNA polymerase as described (40). Translations of these RNAs were carried out in a rabbit reticulocyte lysate-based in vitro translation system with the ⑀ANB-Lys-tRNA Lys as described (41), except that the reactions contained 110 mM KOAc and were conducted for 50 min at 26°C.
Dimerization, Photocross-linking, and Characterization of Photoadducts-For photocross-linking of 5-azido-2-nitrobenzol (ANB)-labeled Bcl-2⌬TM and 6H-Bcl-2⌬TM, purified 6H-Bcl-2⌬TM was added to 10 l of in vitro translated ANB-Bcl-2⌬TM to a final concentration of 2.5 M. The sample was shielded from light and incubated at 4°C for 1 h with occasional mixing to allow the proteins to interact. The sample was then photolyzed for 10 min at 0°C as before (42). A 2-l aliquot was removed and analyzed directly by SDS-PAGE (as the total). To the remaining sample, 200 l of Ni 2ϩ -nitrilotriacetic acid-agarose (Qiagen; 3% (v/v) suspension in buffer A: 20 mM HEPES (pH 7.5), 100 mM KOAc, 5 mM Mg(OAc) 2 , 10% (v/v) glycerol, 0.5% (v/v) Triton X-100, and 5 mM imidazole) were added and mixed at 4°C for 1.5 h to allow binding of 6H-Bcl-2⌬TM and its adducts to Ni 2ϩ -beads. The beads were then washed three times with 400 l of buffer A and one time with 400 l of phosphate buffer (140 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , and 1.8 mM KH 2 PO 4 (pH 7.4)). The proteins bound to Ni 2ϩ -beads (as the Ni 2ϩbound) were eluted at 65°C for 30 min with a reducing SDS-sample buffer as before (40). The samples were loaded onto a 15% gel for SDS-PAGE. The radioactivity in dried gels was detected using an Amersham Biosciences PSI PhosphorImager.
For photocross-linking of ANB-labeled Bcl-2⌬TM mutants to Bax, purified Bax and Triton X-100 were added to 10 l of in vitro translated ANB-Bcl-2⌬TM to final concentrations of 0.5 M and 0.25% (v/v), respectively. After binding and photolysis as described above, Bax was immunoprecipitated from the samples using Bax-specific antiserum as before (36). The immunoprecipitates were analyzed by SDS-PAGE and phosphorimaging as described above.

RESULTS
Experimental Design-The Bcl-2 homodimer interface was mapped using site-specific photocross-linking. RNAs encoding a mutant Bcl-2 containing a lysine codon at a single specific location (Fig. 1A) were translated in vitro in the presence of ⑀ANB-Lys-tRNA Lys and [ 35 S]Met (to facilitate product detection). The ⑀ANB-Lys-tRNA Lys is a functional aminoacyl-tRNA analog that recognizes the lysine codon in RNA and incorporates a lysine analog with a photocross-linker covalently attached to the lysine side chain into the polypeptide being synthesized in vitro. In this way, the 35 S-labeled Bcl-2 proteins, each with a photocross-linker attached to a specific location, were produced.
The ANB-and 35 S-labeled Bcl-2 was then incubated with nonradioactive His-tagged recombinant Bcl-2 to allow dimerization, and upon exposing the sample to UV light, a nitrene was generated from the ANB moiety (43). The nitrene is a powerful electrophile that rapidly reacts with any nearby het-eroatoms possessing nonbonding electron pairs (S, O, N, etc.), double bonds, or even C-H bonds, including such heteroatoms or bonds from the bound His-tagged Bcl-2. The presence of each Bcl-2 in a covalently linked dimer photoadduct was shown by its affinity to Ni 2ϩ -chelating beads and by the detection of 35 S in the photoadduct. Since the ANB probe is located at a single specific site, the detection of a His-tagged Bcl-2-specific photoadduct demonstrates that the ANB-labeled site in the Bcl-2 synthesized in vitro is located at or adjacent to the Bcl-2 dimerization interface. Therefore, sites of contact or close approach can be mapped for the two proteins by altering the location of the photoreactive lysine. The precision of this mapping approach will be limited by the long flexible lysine side chain that links the ANB probe to the protein surface site. The distance between the nitrene and the C ␣ of the lysine is roughly 12 Å if the entire side chain assumes a fully extended, rigid conformation. However, most photocross-linking reactions will occur at a shorter distance because of the randomness of probe motion and photochemical reaction. Therefore, the maximum uncertainty for the location of a site in the interface map will be much less than 12 Å.
Unlike soluble Bcl-2 family proteins such as Bcl-w, the hydrophobic tail of Bcl-2 expressed in cells is not found in the hydrophobic binding groove on the protein surface. Instead, the protein is constitutively anchored to mitochondrial, ER, and nuclear membranes with the C-terminal hydrophobic helix inserted into the lipid bilayer and the rest of the protein exposed to the cytoplasm (29). Thus, the cytosolic domain of Bcl-2 is most likely involved in homodimerization. Therefore, as a first step toward determining the native Bcl-2 homodimer interface, we examined the interaction of the cytosolic domains of Bcl-2 using the Bcl-2⌬TM proteins described above.
Since a single Lys residue must be positioned at a specific site in Bcl-2 to achieve specific ANB-labeling, we substituted all four Lys residues in wild type Bcl-2 with Arg, generating a Lys-null Bcl-2 protein (Fig. 1A, Bcl-2 K0). To ensure that the substitution did not inactivate the protein, the antiapoptotic activity of Bcl-2 K0 was compared with that of Bcl-2 WT after constructing stable MCF-7 cell lines expressing similar amounts of each of the proteins. MCF-7 cells were used for these studies, because the endogenous Bcl-2 expression can be reduced by growing the cells in estrogen-free medium, thereby sensitizing the cells to apoptosis and allowing measurement of the activity of the exogenous Bcl-2 mutants specifically (compare expression of the endogenous protein (neo) with the exogenous proteins (WT and K0) in Fig. 1B). Both Bcl-2 and Bcl-2 K0 inhibited the cleavage of the well characterized caspase 3/7 substrates FAK and PARP equally, confirming that Bcl-2 K0 retains close to wild type activity in MCF-7 cells. Consistent with these observations, both proteins also inhibited cytochrome c release from mitochondria (data not shown). Therefore, the Lys-null Bcl-2 is fully functional in MCF-7 cells and hence can be used as the template for constructing the single Lys Bcl-2 mutants.
The selection of the other positions in Bcl-2 that were suitable for incorporating a photocross-linker to map the interface was based on the result of a modeling experiment using the Bcl-2 core structure (6). The incorporation of an ANB-Lys at these positions would have minimal or no effect on the rest of Bcl-2 structure. The only exception is that an ANB-Lys at position 96 might clash with residues from the C terminus. But since these C-terminal residues in the Bcl-2 structure are in a flexible region, the ANB-Lys at position 96 may be accommodated without having a dramatic effect on the overall structure of Bcl-2. Unlike lysine, the side chain of ANB-lysine is neutral and therefore does not leave a charge in the hydrophobic groove FIG. 1. Sequence and function of Bcl-2 mutants. A, the Bcl-2 sequence is shown with BH domain sequences highlighted by dashed lines above and helices by arrows below. All four lysines (underlined) were changed to arginines to create Bcl-2 K0. The last 22 residues in parentheses were deleted in all Bcl-2⌬TM mutants. The residues replaced individually by ANB-Lys are in boldface type and underlined. The Gly 145 is in boldface italic type. B, adriamycin-treated MCF-7 cells that stably express exogenous Bcl-2 WT and Bcl-2 K0 as well as transfect with the empty vector (neo) were lysed. Aliquots of lysate from equal numbers of cells were analyzed by SDS-PAGE followed by immunoblotting using antibodies specific for FAK, PARP, or Bcl-2, as indicated below each blot. The caspase cleavage product of FAK or PARP is indicated by an arrowhead beside the relevant bands. The relative molecular masses (M r ) of standards are indicated to the left of each panel.
when positioned within the groove. However, the ANB moiety is relatively polar and may reduce the hydrophobicity of the groove somewhat. Nevertheless, in most locations, it appears that the ANB-lysine does not inhibit Bcl-2 dimer formation (see Fig. 2) (data not shown). Furthermore, previous mutagenesis experiments have failed to identify any single point mutations that disrupt Bcl-2 homodimers, suggesting that the homodimer interface may be extensive enough that many single site alterations can be tolerated individually. Consistent with the prediction from our modeling study and previous mutagenesis data, most of the single lysine Bcl-2 mutants that we generated retained as least some anti-apoptotic activity when assayed in the MCF-7 cells (data not shown). Since these mutants containing a charged lysine residue instead of the neutral ANB-lysine continue to inhibit apoptosis, the reduction in hydrophobicity due to the lysine does not interfere with Bcl-2 interactions important for function.
Mapping the Bcl-2 Homodimer Interface by Site-specific Photocross-linking-Several previous studies suggest that the helix 2 of Bcl-2 family proteins containing the BH3 sequence is likely to be involved in homo-and heterodimerization (6,14,44). To examine the possibility that this region is in the homodimer interface Bcl-2⌬TM K108, a mutant with a single ANB-Lys replacing the last residue of helix 2 (Tyr 108 ) was used for photocross-linking to purified His-tagged Bcl-2⌬TM. Two specific photoadducts were detected between the proteins ( Fig.  2A, lane 8, arrowheads). In control experiments ( Fig. 2A), neither product was formed when UV irradiation (lane 6), ⑀ANB-Lys-tRNA Lys (lane 9), RNA encoding Bcl-2⌬TM K108 (lane 1), or His-tagged Bcl-2⌬TM (lane 7) was omitted. Both adducts were preferentially bound to Ni 2ϩ -chelating beads compared with other products seen in the total photocross-linking reaction (data not shown). The formation of both photoadducts was via the ANB probe attached to Lys 108 , since neither was detected when RNA encoding the Lys-null Bcl-2⌬TM K0 was used ( Fig. 2A, lanes 2-5). The relative molecular mass (M r ) of both photoadducts is close to 50 kDa, the M r predicted for the Bcl-2⌬TM homodimer. The difference in the apparent M r between the two photoadducts most likely resulted from the ANB probe at Lys 108 of in vitro synthesized Bcl-2⌬TM reacting with two different sites in the His-tagged Bcl-2⌬TM, since precedents of this phenomenon have been reported previously (45,46). Among the other radioactive proteins detected, the most abundant one was the in vitro synthesized Bcl-2⌬TM protein ( Fig.  2A, filled circle). The other one was a translation product from a reticulocyte mRNA, since it also appeared when ⑀ANB-Lys-tRNA Lys , UV light, or both Bcl-2⌬TM K108 and K0 RNAs were omitted ( Fig. 2A, arrow).  1, 6, 7, and 9, respectively. The adducts were also absent when RNA encoding Bcl-2⌬TM K0 was used (lanes 2-5). Background detected in some or all of the control reactions included in vitro synthesized Bcl-2⌬TM and an unidentified reticulocyte protein, indicated by a filled circle and arrow, respectively. The M r of standards is indicated between the two panels. B and C, photocross-linking of His-tagged Bcl-2⌬TM proteins to in vitro synthesized Bcl-2⌬TM mutant proteins, each with a single ANB probe attached to a Lys residue within the acceptor surface or the donor surface, respectively (defined in Fig. 3). The symbols used to indicate Bcl-2⌬TM photoadducts, background proteins and M r of standards are defined in A.
Three lines of evidence from this experiment suggest that adduct formation is due to position 108 being in the interface of a Bcl-2 homodimer. (i) The ANB probe at Lys 108 of one Bcl-2 molecule was able to react specifically with the other Bcl-2 molecule in solutions containing ϳ2.5 M ( Fig. 2A) to ϳ0.25 M (data not shown) total Bcl-2. (ii) Cross-linking between the two Bcl-2 molecules was observed in reactions containing a high concentration of reticulocyte lysate proteins, including ϳ150 M globin. (iii) The cross-linking occurred via the ANB-derived reactive nitrene which has a very short lifetime (nanoseconds) and covalently links to the side chain of Lys 108 .
To map the hydrophobic groove of Bcl-2 previously predicted to be involved in Bcl-2 dimerization (parts of helices 2, 3, 4, and 8 as well as the loops between helices 2 and 3 and helices 4 and 5 including the BH1-3 domain sequences), individual mutants were generated with the ANB probe at each of the residues 96, 98, 99, 100, 102, 103, 106, 107, 139, and 202. These proteins were used to perform photocross-linking experiments with Histagged Bcl-2⌬TM protein as above. Although a complete set of controls was included in each experiment, for clarity only reactions with and without ⑀ANB-Lys-tRNA Lys are presented in Fig. 2B. Specific photoadducts were detected when the probe was located at position 99, 100, 103, 107, 139, or 202 but not for other positions (Fig. 2B). These results provide direct physical evidence that the hydrophobic groove of Bcl-2 is involved in Bcl-2 homodimerization. They also demonstrate that the outcome of the photocross-linking experiments is very sensitive to the location of the ANB probe, as expected.
Additional mutants were assayed to determine if other parts of the Bcl-2 surface including the BH4 domain sequence or helix 1 are involved in homodimerization. We found that positions 17 or 22, 109, 118, and 129 are near the dimer interface, since ANB at these positions generated specific photoadducts containing the two Bcl-2⌬TM proteins, whereas ANB at positions 105 and 164 did not result in significant adduct formation compared with the controls (Fig. 2C) (data not shown). Thus, the homodimerization interface is not limited to the hydrophobic groove.
Modeling the Bcl-2 Homodimer Interface-The photocrosslinking data in Fig. 2 were used to model which residues would be in the Bcl-2 homodimer interface, based on the NMR structure of the Bcl-2 core (6). Mapping these residues on the NMR structure suggests that there are two distinct binding surfaces in the homodimer that we refer to as acceptor and donor surfaces (Fig. 3, A and B, respectively). As shown in Fig. 3A, the acceptor surface is defined by photoadducts obtained when the cross-linker was positioned at the residues that are colored red. This surface overlaps extensively with the hydrophobic groove, including the BH1-3 domain sequences that was previously implicated in Bcl-2 family protein dimerization. The glycine at position 145 in the middle of the acceptor surface abolishes hetero-but not homodimerization when mutated to alanine (Fig. 3A, green). The donor surface is oriented ϳ90°to the acceptor surface, and the positions resulting in photoadducts are colored blue. This surface begins on one side of helix 2, a result consistent with the assignment of this helix as the correspondent to the ligand helix that binds to the hydrophobic acceptor groove on Bcl-x L (14 -16). However, our map suggests that the donor surface extends beyond helix 2 into helices 1, 3, and 4, overlapping the BH3 and -4 domain sequences. Unlike previous models, the donor surface we identified is exposed in the NMR structure of Bcl-2 core and does not require a structural rearrangement proposed previously to expose helix 2 such that it can form the entire binding surface. Position 145 is not part of the donor surface identified here.
Bcl-2 Can Form Oligomers-One prediction of our model is that Bcl-2 may form homo-oligomers, since each Bcl-2 protein has more than one binding surface, and utilization of each need not be mutually exclusive. Consistent with this prediction, gel filtration chromatography demonstrated that although Histagged Bcl-2⌬TM proteins eluted from a Superdex 200 column mainly as monomers, dimers and oligomers were also evident (Fig. 4). The actual ratio of dimer and oligomer to monomer in the sample is probably higher than that shown by the chromatogram (Fig. 4A), since some dimers and oligomers could dissociate into monomers during chromatography. The elution profile was not changed when the Bcl-2⌬TM proteins were incubated in a solution containing reticulocyte lysate and ATP (i.e. the same solution used for in vitro translation except for the radioactive amino acid and aminoacyl tRNA) prior to gel filtration (data not shown), indicating that at least a fraction of Bcl-2 remains in complexes even in the presence of chaperones and ATP.
Activated Bax Interacts with Both Bcl-2 Monomers and Dimers-Taken together, the results above suggest that Bcl-2 proteins can form both dimers and oligomers via homotypic interactions that use two distinct binding surfaces. Given these interactions, how would a proapoptotic Bcl-2 family member, such as Bax, interact with Bcl-2? To determine which form(s) of Bcl-2 (monomer, dimer, or both) can be bound by Bax, we used in vitro synthesized Bcl-2⌬TM that was ANB-labeled at Lys 107 and purified recombinant Bax protein to perform additional photocross-linking experiments. The Bax protein was a fulllength protein expressed in bacteria and purified under conditions that kept the protein in a native conformation similar to that of Bax in healthy cells (35). In healthy cells, Bax proteins reside in the cytosol and do not dimerize with each other or with other Bcl-2 family members. Upon induction of apoptosis, Bax proteins are switched to an active conformation and translocate to mitochondrial and ER membranes, where they form homo-oligomers or heterodimerize with other Bcl-2 family members. Active full-length Bax is likely to interact with the cytosolic domain of membrane-bound Bcl-2 to form a heterodimer, since the C-terminal tail of Bcl-2 is buried in the lipid bilayer and is therefore not available for the initial interaction.
As shown in Fig. 5A, photoadducts between the Bcl-2⌬TM protein and Bax protein were detected. The size of the adducts is most consistent with the formation of a heterodimer. Furthermore, the cross-linked proteins were recognized by the antibody specific for Bax (Fig. 5A, lane 4, arrowheads) and were formed only when Bax protein, UV light, and ⑀ANB-Lys-tRNA Lys were all present (Fig. 5A, lane 4 versus lanes 1-3) and Bcl-2⌬TM RNA was used to program the in vitro translation reaction (data not shown). Triton X-100 (0.25%) was required for the formation of the cross-linked Bcl-2/Bax heterodimer (Fig. 5A, lane 4 versus lane 5), consistent with previous findings indicating that certain nonionic detergents can switch Bax to a dimerization-competent conformation (47).
Thus, activated Bax can interact directly with Bcl-2 monomers to form Bax/Bcl-2 heterodimers. Therefore, it is important to determine whether or not the activated Bax protein can also interact with Bcl-2 homodimers, since most of the Bcl-2 proteins isolated from mammalian cells appear to be in homodimer or oligomer state (48,49). To determine whether Bax can interact with the Bcl-2 homodimer directly, we used the thiospecific chemical cross-linker BMH to cross-link two Histagged Bcl-2⌬TM proteins, each containing a single Cys at position 168. BMH will covalently link two Bcl-2⌬TM molecules together if the two Cys residues are about 16 Å away, a

FIG. 4. Oligomerization of Bcl-2. A, oligomerization of purified
His-tagged Bcl-2⌬TM protein examined by gel filtration chromatography. The A 280 was monitored for each eluted fraction and plotted. The arrow indicates a fraction that was not collected for analysis. The void volume for the column and elution positions for protein standards are indicated at the top of the plot by V 0 and M r , respectively. The predicted M r for the Bcl-2 monomer, dimer, trimer, and tetramer are about 25, 50, 75, and 100 kDa, respectively. B, the proteins in fractions 5-11 (shown in A) examined by SDS-PAGE and silver staining. His-tagged Bcl-2⌬TM protein was detected in all fractions. 15-Fold less sample was loaded for fraction 11 than for other fractions.

FIG. 5. Interaction of Bax with both Bcl-2 monomer and dimer.
A, photocross-linking of in vitro synthesized Bcl-2⌬TM protein with a single ANB attached at Lys 107 to Bax protein detected by immunoprecipitation using antibodies specific for Bax. The photoadducts were detected in the reaction shown in lane 4 (indicated by arrowheads) but not in control reactions in which ⑀ANB-Lys-tRNA Lys , UV irradiation, Bax protein, or Triton X-100 was omitted, shown in lane 1, 2, 3, or 5, respectively. The in vitro synthesized Bcl-2⌬TM protein was indicated by a filled circle. B, BMH cross-linking of Bcl-2⌬TM proteins with single Cys at position 168 detected by SDS-PAGE and Coomassie staining. The BMH cross-linked Bcl-2⌬TM dimer and monomer are indicated by an arrowhead and a filled circle, respectively. C, photocross-linking of BMH cross-linked His-tagged Bcl-2⌬TM proteins to in vitro synthesized, ANB-labeled Bax protein. The three major photoadducts detected in Ni 2ϩ -bound fractions after SDS-PAGE by phosphorimaging were identified as follows: adducts of Bax with Bcl-2⌬TM monomer (bracket), BMH cross-linked Bcl-2⌬TM dimer (arrow), and BMH cross-linked Bcl-2⌬TM/reticulocyte protein (oval). These adducts were not detected in the control reaction in which the BMH cross-linked Bcl-2⌬TM protein, UV irradiation, or ⑀ANB-Lys-tRNA Lys was omitted, shown in lane 1, 2, or 3, respectively. The in vitro synthesized Bax protein is indicated by a filled circle.
distance possible for these residues in a Bcl-2 homodimer formed via the two different binding surfaces we identified above. We detected BMH-cross-linked Bcl-2⌬TM dimer only in the presence of BMH, not the solvent control (Ϫ BMH) (Fig. 5B,  lanes 3 and 4, arrowhead). When a Cys-null Bcl-2⌬TM protein (C0) was used in parallel experiments, no cross-linked Bcl-2⌬TM proteins were detected, verifying that the BMH crosslinking reaction is specific (Fig. 5B, lanes 1 and 2).
To test whether Bax can cross-link to the Bcl-2 homodimers, we incubated the sample from the BMH cross-linking reaction with in vitro synthesized Bax protein with ANB probes attached to the endogenous Lys residues. Based on the migration position, one of the photoadducts detected in the Ni 2ϩ -bound fraction was due to cross-linking between ANB-labeled Bax and the BMH-linked 6H-Bcl-2⌬TM dimer (Fig. 5C, lane 4 , arrow). Photoadducts between Bax and Bcl-2⌬TM monomers were also detected (Fig. 5C, lane 4, bracket). These bands are not Bax homodimers because Bax homodimers do not bind to the Ni 2ϩ resin (data not shown), similar to the Bax monomers (Fig. 5C, lane 1, filled circle). The photoadduct indicated by an oval is probably a cross-linking product of Bax oligomers with 6H-Bcl-2⌬TM. It is also possible that the complex includes one or more proteins from the reticulocyte lysate. Together, these results demonstrate that Bcl-2 monomers, dimers, and possibly larger oligomers bind activated Bax, suggesting that any of these may neutralize Bax apoptotic activity.
Bax Binds to the Acceptor Face of Bcl-2-Our data support a model in which a Bcl-2 homodimer is formed by interaction of the acceptor surface of one molecule with the donor surface of another molecule (Fig. 3). In principle, Bcl-2/Bax heterodimer interface can be mapped in the same way. However, of the various ANB-Bcl-2⌬TM mutants that we have tested, only ANB-Bcl-2⌬TM K107 cross-linked to recombinant Bax ( Fig.  5A) (data not shown). Since position 107 is in the acceptor surface of Bcl-2 (Fig. 3A), this result strongly suggests that Bax binds the acceptor surface of Bcl-2. Consistent with Bax binding to only the acceptor surface, a Bcl-2 mutation (G145A) known to abrogate the interaction between Bcl-2 and Bax (17) is located in the middle of the acceptor surface (Fig. 3A). To verify that an interaction between Bcl-2 G145A and Bax is not detected by photocross-linking, we assayed cross-linking between in vitro synthesized Bcl-2⌬TM G145A with ANB attached to Lys 107 and Bax. As shown in Fig. 6A, cross-linking was eliminated by the mutation, substantiating the notion that the acceptor surface in Bcl-2 serves as the binding site for Bax. In addition, since the G145A mutation did not affect the crosslinking of two Bcl-2⌬TM proteins via ANB-Lys 107 (Fig. 6B,  lanes 1-3), even when the mutation was created in both Bcl-2⌬TM proteins (Fig. 6B, lanes 4 -7), the mutation did not abolish homodimerization. Consistent with our finding, Bcl-2 G145A was shown to retain homodimerization and certain anti-apoptotic functions in other systems (17,20). Therefore, the Bcl-2 homodimer interface must be different than that of Bcl-2/Bax heterodimer.
Inhibition of Bcl-2 Homodimerization by Activated Bax-The fact that Bax can bind to a Bcl-2 homodimer that is held together with a covalent cross-link demonstrates that the homodimer contains an accessible binding site for Bax, as predicted from our model. However, if binding of Bax to Bcl-2 impacts Bcl-2 homodimerization, as predicted by some models  7. Inhibition of Bcl-2 homodimerization by Bax. A, photocross-linking of in vitro synthesized Bcl-2⌬TM protein with a single ANB attached to Lys 107 to His-tagged Bcl-2⌬TM protein at various Bcl-2⌬TM/Bax ratios. The Bcl-2⌬TM homodimer photoadduct was detected in the Ni 2ϩ -bound fractions as shown in Fig. 2. The amount of the photoadduct was reduced as the molar ratio of His-tagged Bcl-2⌬TM to Bax in the reaction (indicated at the top of the gel) was reduced. The arrowhead and filled circle indicate the cross-linked Bcl-2⌬TM dimer and monomer, respectively. B, photocross-linking of in vitro synthesized Bcl-2⌬TM protein with a single ANB attached to Lys 107 to Bax protein at various Bcl-2⌬TM/Bax ratios. Aliquots of the samples used for A were examined for the Bcl-2⌬TM/Bax photocross-linking by immunoprecipitation using antibodies specific for Bax. The arrow indicates the photoadduct of the Bcl-2⌬TM and Bax, whose amount was increased when the molar ratio of Bcl-2⌬TM to Bax was reduced. (50), such an effect would not be revealed by this assay. To determine whether or not Bax binding to Bcl-2 inhibits Bcl-2 homodimerization, photocross-linking experiments were performed with in vitro synthesized, ANB-labeled Bcl-2⌬TM K107 and His-tagged Bcl-2⌬TM in the presence of increasing amounts of detergent-activated Bax. As shown in Fig. 7A, lanes 1-4, the cross-linking of two Bcl-2⌬TM proteins was inhibited by the Bax in a concentration-dependent manner. Moreover, the amount of photoadduct between Bax and Bcl-2⌬TM correlated positively with the relative amount of Bax added to the reaction (Fig. 7B, lanes 1-4). Taken together with the data presented in Fig. 5, this result suggests that Bax/Bcl-2 interaction reduces the amount of Bcl-2 homodimer by either binding to the Bcl-2 monomer or binding to and disrupting the Bcl-2 dimer or both.

DISCUSSION
Our photocross-linking studies reveal that the interface of the Bcl-2 homodimer is much more extensive than predicted from structural studies using peptides. The binding interface can be divided into acceptor and donor surfaces and includes at least part of all four conserved BH domain sequences. The acceptor surface, consisting of helix 2, the loop between helix 4 and 5, and helix 8, includes the BH1-3 domain sequences. The donor surface that consists of helices 1-4, including the BH3 and -4 domain sequences, is oriented ϳ90°from the acceptor surface. Because of the geometry of the two surfaces, binding at one surface does not occlude the other. Therefore, one Bcl-2 molecule can bind either Bax or Bcl-2 via its acceptor surface and simultaneously another Bcl-2 monomer via its donor surface. This aspect of our model explains the formation of both homo-and hetero-oligomers. Moreover, our competition data suggest that binding of Bax to Bcl-2 may initially take place on homodimeric Bcl-2 molecules but may reduce the affinity of the homodimer, leading it to dissociate releasing another monomer. If the monomeric Bcl-2 has a higher affinity for Bax than the homodimer, the Bcl-2 monomers generated in this way may effectively sequester more activated Bax proteins.
Implications of the Two-binding Surface Model for Bcl-2 Dimerization-Bcl-2 homodimerization has been studied using a yeast two-hybrid system and various Bcl-2 deletion mutants (30,31). Some mutant Bcl-2 proteins were still able to interact although they contained very large deletions. For example, a Bcl-2 mutant with the BH4 domain deleted interacted with Bcl-2 mutants with either the BH1 or BH2 domain deleted. This result is consistent with our two-binding surface model, because the BH4 deletion mutant still has an intact acceptor surface, and deletion of either BH1 or BH2 is not predicted to affect the donor surface of the other Bcl-2. In contrast, if both Bcl-2 mutants lacked either intact acceptor or donor surfaces they did not interact in the two-hybrid assay. For example, deletion of the BH1 domain sequence from both bait and target proteins analyzed in yeast abolished the interaction, since the deletion destroyed the same surface from both proteins. Similarly, deletion of the BH1 sequence from one protein and the BH2 sequence from the other also abolished the interaction between the two proteins, because both deleted sequences belong to the acceptor surface. This result also confirms our prediction that Bcl-2 homodimers consist of donor-acceptor pairs.
The binding affinity (EC 50 ) measured in semisolid binding assays for the Bcl-2 and Bcl-x L homodimers (66.2 and 109.7 nM, respectively) (51) is much lower than the K d measured in fluorescence equilibrium binding assays for Bcl-x L BH3 peptide binding to Bcl-x L (325 M) (14). These quantitative data strongly suggest that the dimer interface is much more extensive than that between the BH3 peptide and the Bcl-x L hydrophobic groove, revealed by the NMR structures (14 -16).
The Role of the Acceptor and Donor Surfaces in Protein Binding by Bcl-2-Our data suggest that the Bcl-2 acceptor surface binds either Bcl-2 or Bax, whereas the donor surface binds only another Bcl-2. The NMR structure of Bcl-x L in complex with BH3 peptides from different BH3-only proteins implies that the acceptor surface of Bcl-2 may also bind BH3-only proteins. Thus, the hydrophobic groove in the acceptor surface may be able to accommodate a variety of ligands.
Bcl-2 family proteins have been reported to bind to a variety of proteins, including scaffold proteins Apaf-1, CED-4, and Bag-1, mitochondrial proteins VDAC and FKBP38 (an inherent calcineurin inhibitor), kinase Raf-1 (an effector of H-Ras), phosphatase calcineurin, GTPases R-Ras and H-Ras, prenylated Rab acceptor PRA1, and transcription regulators p53 (and p53binding protein p53-BP2) and Nur77/TR3 (for a review, see Refs. 52 and 53) (54 -61). The detailed interface of most of these interactions has not yet been elucidated, and some of these interactions may be the result of low specificity interactions with the acceptor surface of Bcl-2. However, consistent with our data showing that the BH4 domain is part of the interface for Bcl-2 homotypic interaction, there is evidence that Bcl-2 BH4 domain sequences are involved in most of these Bcl-2 heterotypic interactions. Our model suggests that such interactions can occur with Bcl-2 homodimers, since there is always one of each binding surface that is unused in the dimer. Thus, our model clearly explains how, although the vast majority of the Bcl-2 in the cell is homodimeric, the same sequences identified as important for homodimerization can engage in a variety of other protein-protein interactions.