Bax Forms an Oligomer via Separate, Yet Interdependent, Surfaces*

Interactions of Bcl-2 family proteins regulate permeability of the mitochondrial outer membrane and apoptosis. In particular, Bax forms an oligomer that permeabilizes the membrane. To map the interface of the Bax oligomer we used Triton X-100 as a membrane surrogate and performed site-specific photocross-linking. Bax-specific adducts were formed through photo-reactive probes at multiple sites that can be grouped into two surfaces. The first surface overlaps with the BH1–3 groove formed by Bcl-2 Homology motif 1, 2, and 3; the second surface is a rear pocket located on the opposite side of the protein from the BH1–3 groove. Further cross-linking experiments using Bax BH3 peptides and mutants demonstrated that the two surfaces interact with their counterparts in neighboring proteins to form two separated interfaces and that interaction at the BH1–3 groove primes the rear pocket for further interaction. Therefore, Bax oligomerization proceeds through a series of interactions that occur at separate, yet allosterically, coupled interfaces.

Interactions of Bcl-2 family proteins regulate permeability of the mitochondrial outer membrane and apoptosis. In particular, Bax forms an oligomer that permeabilizes the membrane. To map the interface of the Bax oligomer we used Triton X-100 as a membrane surrogate and performed site-specific photocross-linking. Bax-specific adducts were formed through photo-reactive probes at multiple sites that can be grouped into two surfaces. The first surface overlaps with the BH1-3 groove formed by Bcl-2 Homology motif 1, 2, and 3; the second surface is a rear pocket located on the opposite side of the protein from the BH1-3 groove. Further cross-linking experiments using Bax BH3 peptides and mutants demonstrated that the two surfaces interact with their counterparts in neighboring proteins to form two separated interfaces and that interaction at the BH1-3 groove primes the rear pocket for further interaction. Therefore, Bax oligomerization proceeds through a series of interactions that occur at separate, yet allosterically, coupled interfaces.
Mitochondrial outer membrane permeabilization (MOMP) 4 is the commitment step of apoptosis regulated by the proteins of Bcl-2 family (1-3). These proteins contain at least one of the four Bcl-2 homology (BH) motifs. Bax and Bak are required for MOMP and contain BH1, -2, and -3 motifs, although a BH4 motif was recently found in these proteins based on structural comparison with antiapoptotic family members (4). Bcl-2, Bclx L , and other antiapoptotic members contain all four BH motifs and function to inhibit Bax/Bak-mediated MOMP. The other members of the Bcl-2 family, such as Bad and Bim, only contain a BH3 motif. Another common feature within the Bcl-2 family is a hydrophobic sequence located at the C terminus of some members. Structures of Bax and Bak have been revealed (5,6). Both structures share a similar fold that contains a hydrophobic groove formed by BH1-3 regions (termed as BH1-3 groove hereafter). The structure of Bak, in which the BH1-3 groove is empty, was determined by using the soluble domain without the C-terminal hydrophobic sequence. In contrast, the BH1-3 groove in the structure of full-length Bax, as shown in Fig. 1A, is occupied by the C-terminal hydrophobic ␣-helix (␣9). Unlike Bak that is constitutively anchored in the MOM by the C-terminal hydrophobic sequence, Bax is a monomeric protein in the cytosol and integrates into the MOM only after activation by BH3-only proteins or other factors (7)(8)(9)(10)(11). Active Bax and Bak in the MOM form oligomers that are required for MOMP (9,(12)(13). To understand membrane permeabilization by Bax and Bak, it is important to characterize the interface between the proteins in the oligomers.
Following a treatment by truncated Bid (tBid) Bak oligomerizes in the MOM via two interfaces that were recently revealed (14,15). The first interface is formed by reciprocal binding of the BH3 region to the BH1-3 groove of two Bak molecules (Fig.  1B, front:front interface); the second one involves two interacting ␣6 helices located on the opposite side of the protein from the first interface (Fig. 1B, rear:rear interface). In the absence of tBid or other apoptotic stimuli, Bak is integrated into the MOM via the C-terminal hydrophobic sequence with the N-terminal domain located in the cytosolic side of the MOM. It appears that, in the absence of an apoptotic signal, most Bak remains monomeric in the MOM, but some fraction can be bound by VDAC2 (voltage-dependent anion channel 2) or by an antiapoptotic family member such as Bcl-x L that binds to the MOM after activation (16 -18). The binding with Bcl-x L is likely mediated by an interaction between the BH3 region of Bak and the BH1-3 groove of Bcl-x L (19). During apoptosis Bad or another BH3-only protein may activate Bak by displacing Bak from the BH1-3 groove of Bcl-x L without directly binding to Bak (16,20,21). The freed Bak can then oligomerize via the two interfaces described above.
Although Bax oligomerization can also be induced by tBid, the mechanism is likely different. Monomeric Bax is equilibrated between cytosolic soluble and peripheral MOM-bound states in healthy cells. Upon apoptotic stimulation tBid interacts with Bax exposing the targeting signals at both N-and C-terminal regions that localize Bax to mitochondria (11,13,(22)(23)(24)(25)(26). Once at mitochondria Bax embeds its ␣9 as well as ␣5 and ␣6 helices into the MOM, which follows by oligomerization (27). Therefore, Bax activation may always require a direct interaction with BH3-only protein or some other activator protein, whereas such interaction may not always be required for Bak activation. This implies that the Bax oligomer may contain a BH3-only protein that may mask one of the two interacting surfaces similar to those revealed in Bak oligomer thereby inhibiting oligomerization. Thus, should Bax oligomerize in the same way as proposed for Bak, the BH3-only protein that activates Bax must be displaced (Fig. 1B). Alternatively, Bax may stay in the complex with the activating BH3-only protein but use different surfaces to oligomerize. There is also evidence in some systems that oligomerization of Bak may require a direct interaction with a BH3-only protein (28 -30).
The possibility that one of the potential surfaces for Bax oligomerization interacts with a BH3-only protein is supported by a NMR structure in which a stabilized helical BH3 peptide from Bim binds Bax through a pocket that consists of ␣1, the loop connecting ␣1 and ␣2, and ␣6 (Fig. 1A, colored in dark gray) whose counterpart in Bak forms the rear:rear interface in the oligomer (31). Additional evidence for binding of a BH3-only protein to the rear pocket comes from a recent observation that a hemagglutinin-tagged ␣1 of Bax interacts with several BH3only proteins, including Bim (28). However, this study also suggests that the BH1-3 groove may serve as the second binding pocket for the BH3-only protein because coprecipitation of tBid and Bax was abolished by a mutation in the pocket. Thus, should the activated Bax interact at both front and rear surfaces to form the two interface oligomer (Fig. 1B), the BH3-only activator must leave Bax prior to oligomerization. However, the one interface oligomer model in which the BH3-only protein remains bound to Bax oligomer cannot be ruled out, because binding of BH3-only protein to the rear pocket of Bax may induce a conformational change in Bax that would release ␣9 from the BH1-3 groove at the front side exposing the BH3 motif, which then could interact with the rear pocket of and activate the next Bax in the same way as the Bim BH3 peptide. In fact a Bax BH3 peptide can activate Bax, and the activated Bax can then activate another Bax through a mechanism termed auto-activation (32), a notion that was coined previously to describe a similar activation mechanism for Bak (33). Therefore, both auto-activation and oligomerization of Bax could be mediated by only one type of interface formed between the front and rear surfaces (Fig. 1C). The BH3-only protein that initially activates Bax may remain bound to the rear pocket of the first Bax in the oligomer or may be displaced only when the oligomer circulates.
To determine which model best describes Bax oligomerization, we employed a site-specific photocross-linking approach with single photoreactive probes located in both BH1-3 groove and rear pocket. Prior to cross-linking Triton X-100 micelles were added to induce Bax oligomerization similar to that seen when a BH3-only protein is added to Bax and the MOM. Unlike the conclusion from a recent study that Bax exists as monomer in the micelle (34), our data show that Bax forms an oligomer that contains two types of interface similar to those proposed for Bak oligomer. Because the rear pocket of Bax binds BH3only protein during activation, our results indicate that the BH3-only protein must depart from the pocket before the activated Bax can oligomerize. We suggest that the interaction between two activated Bax proteins at the BH1-3 groove may be the trigger for the departure of the BH3-only protein. Moreover, the interaction mediated by the BH1-3 regions at the front:front interface primes the rear pocket to form the rear: rear interface. Therefore, our study reveals a mechanism by which Bax can form an oligomer through two interdependent surfaces.

EXPERIMENTAL PROCEDURES
Plasmids, Proteins, and Peptides-To construct plasmids of Bax mutants for in vitro transcription and translation, we FIGURE 1. Two competing models of Bax oligomerization. A, structure of Bax monomer. The structure was drawn based on the coordinates 1F16 in the Protein Data Bank using PyMOL (DeLano Scientific). All ␣-helices and the loop between ␣1 and ␣2 are indicated. The front surface that includes BH1-3 regions is colored in medium gray. The rear surface formed by ␣1, the loop between ␣1 and ␣2, and ␣6 are colored in dark gray. The schematic underneath represents the Bax structure with front and rear surfaces. B, two-interface model. Bax oligomerizes through two different kinds of interface. One is formed by interaction of the front surfaces of neighboring molecules, and other by interaction of the rear surfaces. No BH3-only protein remains on Bax oligomer after it activates Bax. C, one-interface model. Bax oligomer contains only one type of interface that is formed by interaction of the front surface of a molecule and the rear surface of its neighbor. BH3-only protein that activates the first Bax in the oligomer may remain on its rear surface. used a plasmid that contains the coding region of full-length human Bax inserted between NcoI and EcoRI sites in the vector pSPUTK (Stratagene). Lys-null Bax mutant (K0) was generated by mutating all Lys codons in the Bax coding region to Arg codon. Bax mutants with a single Lys codon at a particular position were generated by mutating the corresponding codon of the K0 mutant to Lys codon and are designated here as "K" followed by a number that indicates the codon position. For example, the mutant Bax K73 has a single Lys codon at position 73 of the sequence. The codon for Gly 67 in Bax K79 and K134 was changed to Arg codon to create the G67R mutants.
To construct plasmids of Bax mutants for testing the proapoptotic activity in cells, we used pBabe-MN-Bax-IRES-GFP plasmid (27). The NcoI site at 1226 bp of the plasmid was mutated to make the NcoI site at 8 bp unique. The NcoI-StuI fragment of the Bax coding region in the mutated plasmid was replaced by that from the pSPUTK-Bax K0 plasmid to generate pBabe-MN-Bax K0-IRES-GFP plasmid. Then, the NcoI-BamHI fragment of pBabe-MN-Bax K0-IRES-GFP plasmid was replaced by that from pSPUTK-Bax single K mutant plasmid to generate the corresponding pBabe-MN-Bax single K mutant-IRES-GFP plasmid.
The plasmid for expression a His 6 -tagged Bax in Escherichia coli was modified from the pTYB1-Bax plasmid (5) by inserting a coding sequence for MH 6 GENLYFQGT prior to the Bax coding sequence. The codon for Gly 67 of Bax in the plasmid was changed to Arg codon to create the G67R mutant. All of the Bax mutant plasmids were generated by using appropriate primers and the overlapping PCR-based method (35) or the QuikChange method (Stratagene), and verified by DNA sequencing.
Proapoptotic Activity of Bax Mutants in Cells-HEK293 cells were transfected with a plasmid encoding the eco receptor to generate a cell line that can be infected with murine retrovirus. Replication-incompetent murine retroviruses that contain the pBabe-MN-Bax mutant-IRES-GFP plasmid described above and hence coexpress the Bax mutant and green fluorescent protein (GFP) were prepared using Phoenix cells as a packaging line, and the amount of virus was determined using realtime PCR. This is necessary as the more pro-apoptotic a Bax mutant is the less virus the packaging cells make before they die. Furthermore, viruses expressing a pro-apoptotic protein do not form plaques. Equal numbers of virus were used to infect the HEK293 human embryonic kidney cells for 48 h (infection efficiency was ϳ20%). One of duplicate sets of the cells was stained with Annexin V, whereas the other set was treated with tumor necrosis factor ␣ (TNF␣) and cycloheximide for 6 h and then stained with Annexin V. The fraction of Annexin V-positive cells was determined for green cells (infected) and non-green cells (uninfected). This process underestimates the total amount of cell death, because only adherent cells are counted and at a single time point (not cumulative). Furthermore, cells that were not scored as green may express small amounts of Bax (and GFP) and therefore have increased Annexin V staining compared with cells infected with virus that expressed only GFP. Nevertheless, the assay can clearly reveal the proapoptotic activity of the Bax mutants relative to the wild-type Bax that was assayed in parallel.
Gel-filtration Chromatography to Assay Bax Oligomerization Induced by Triton X-100-Purified His 6 -tagged Bax (2.2 M) and/or Bcl-2⌬TM (2.2, 4.4, or 6.6 M) or Bcl-2⌬TM-G145A (6.6 M) protein was incubated in 25 l of buffer A containing 0.25% (v/v) Triton X-100 at 4°C for 1 h. The sample was adjusted to 300 l with buffer B (300 mM NaCl, 0.2 mM dithiothreitol, 2% (w/v) CHAPS, and 20 mM HEPES, pH 7.5) and then mixed with 50 l of Ni 2ϩ -nitrilotriacetic acid-agarose (50% (v/v) suspension in buffer B). After rocking at 4°C for 2 h the Ni 2ϩ -resin-bound proteins were washed twice with 1 ml of buffer B to remove Triton X-100, judged by reduction of the A 280 to a baseline level (39). Then the proteins were eluted with 120 l of buffer B containing 250 mM imidazole. The eluted proteins (100 l) were injected into Superdex 200 HR10/30 column in AKTA-FPLC (Amersham Biosciences). The chromatography was developed in buffer B at a flow rate of 0.4 ml/min, and fractions of 500 l were collected. The fractions were precipitated with Cl 3 CCOOH and analyzed by 15% SDS-PAGE, immunoblotting with a Bax-or Bcl-2-specific antibody, and the corresponding proteins were detected with enhanced chemiluminescence.
For the reaction containing purified His 6 -Bax protein and in vitro translated [ 35 S]Met-Bax protein, 2.2 M His 6 -Bax and 0.25% Triton X-100 were added to 100 l of [ 35 S]Met-Bax, and the sample (ϳ140 l) was incubated at 4°C for 1 h. The sample was then adjusted to 350 l with buffer B, and mixed with 100 l of the Ni 2ϩ -resin suspension. The sample was processed as described above. Then one-fourth of the Cl 3 CCOOH-precipitated sample was analyzed by the SDS-PAGE and immunoblotting to detect His 6 -Bax, and three-fourths of the sample was analyzed by the SDS-PAGE and phosphorimaging to detect [ 35 S]Met-Bax. For the His 6 -Bax and [ 35 S]Met-Bax incubated in the absence of Triton X-100, the sample was processed in parallel as the proteins incubated in the presence of Triton X-100, except that the steps to remove Triton X-100 were omitted.

RESULTS
Bax Oligomerization Induced by Detergent-To determine the binding site on Bax for oligomerization using site-specific photocross-linking as described (36), we need to prepare one polypeptide containing a photoreactive cross-linking probe and the other a unique tag to permit isolation of the photoad-duct. This method requires one Bax protein to be synthesized by in vitro translation to incorporate the photoreactive probe at a specific site, and [ 35 S]Met as a tracer. The binding partner is a purified recombinant Bax protein containing a His 6 tag at the N terminus that can be used to enrich the radioactive photoadduct using Ni 2ϩ -chelating resin. Ideally oligomerization of these Bax proteins should be studied in the MOM. However, the yield of oligomers was not enough for the crosslinking study. We thus chose to study the Bax oligomer formed in detergent micelle as a membrane surrogate.
Non-ionic detergent, Triton X-100, was known to induce Bax conformational change and homooligomerization in cell lysate (8,40). The detergent treatment also activated recombinant Bax to permeabilize liposomes and mitochondria (41). To determine whether the in vitro synthesized [ 35 S]Met-Bax and the purified His 6 -Bax behave like the Bax in cell lysate that only forms oligomer in the presence of detergent, these proteins were first incubated in the absence of detergent. The resulting sample was treated with a zwitterionic detergent CHAPS at the concentration of 2% that is Ͼ0.6%, the critical micelle concentration, and subjected to gelfiltration chromatography. CHAPS was included in these steps to make the sample composition similar to the samples incubated in the presence of Triton X-100 (see below), and CHAPS does not change the oligomeric state of Bax (8,40,42). The eluted fractions were analyzed by SDS-PAGE and either phosphorimaging to detect the [ 35 S]Met-Bax or immunoblotting with a Bax-specific antibody to detect the His 6 -Bax. As shown in Fig. 2A, both Bax proteins were eluted mainly before a protein marker with relative molecular mass of 25 kDa. Because the molecular mass predicted for each Bax protein (ϳ21 kDa) in a micelle with 4 -14 CHAPS molecules (average size of a CHAPS micelle, ϳ6 kDa) is ϳ27 kDa (43), these results suggest that both proteins are monomeric in the absence of Triton X-100.
Next, the [ 35 S]Met-Bax and His 6 -Bax were coincubated in the presence of Triton X-100 at the concentration of 0.25% that is Ͼ0.05%, the critical micelle concentration. The resulting sample was bound to Ni 2ϩ -resin to enrich the oligomer formed by the His 6 -Bax and [ 35 S]Met-Bax. The bound fractions were washed extensively with 2% CHAPS to remove Triton X-100. This is necessary because a Triton X-100 micelle contains 100 -155 Triton X-100 molecules on average with molecular mass ϳ80 kDa (43), equivalent to a Bax tetramer, thereby preventing the use of gel-filtration chromatography to analyze Bax oligomerization. The bound fraction with Triton X-100 exchanged by CHAPS was then eluted by imidazole and subjected to the gel-filtration chromatography. Fractions eluted from the column were analyzed as described above. As shown in Fig. 2A, both Bax proteins were eluted as oligomers in CHAPS micelles, mainly trimer (ϳ69 kDa) but large oligomers up to 200 kDa were also visible. Therefore, the gel-filtration data demonstrated that the interaction between the in vitro synthesized Bax and the recombinant Bax can be induced by Triton X-100 treatment resulting in a large extent of oligomerization similar to endogenous Bax.
Bax Oligomerization Inhibited by Bcl-2-It has been well established that Bcl-2 inhibits Bax oligomerization in biological systems (44). To determine whether Bcl-2 can inhibit Bax oligomerization in Triton X-100, the His 6 -Bcl-2⌬TM was incubated with the His 6 -Bax protein in 0.25% Triton X-100. Triton X-100 in the sample was exchanged for CHAPS prior to the gel-filtration chromatography as described above. Fractions from the gel-filtration chromatography were analyzed by SDS-PAGE and immunoblotting with an antibody specific to Bax or Bcl-2. As shown in the left panel of Fig. 2B, Bax protein shifted from the trimeric and large oligomeric factions to the dimeric and monomeric fractions in the presence of the Bcl-2 (Fig. 2B,  right panel). Furthermore, more Bax proteins were seen in the dimeric and monomeric fractions as the concentration of Bcl-2 was increased. Therefore, Bcl-2 inhibits Triton X-100-induced Bax oligomerization in a dose-dependent manner. Of note, the Bcl-2 signal in the immunoblots does not correlate with the amount of Bcl-2 used in the experiments. Due to the limited linear range of immunoblotting it was necessary to load smaller amounts of the fractions from experiments with more Bcl-2 to ensure that the resulting blots remained in the linear range for comparison across the fractions. The amount of protein lost during the gel filtration is small and the same among the experiments as reflected by the similarity in total intensity for the blots shown.
To determine whether the inhibitory effect of Bcl-2 on Bax oligomerization is dependent on the interaction between Bcl-2 and Bax, we used His 6 -Bcl-2⌬TM-G145A, a Bcl-2 mutant protein that does not interact with Bax (36,45). As shown in the bottom panel of Fig. 2B, the mutant Bcl-2 does not inhibit Bax oligomerization, even at the highest Bcl-2 to Bax ratio, demonstrating that the Bcl-2 inhibition of Bax oligomerization requires the Bcl-2-Bax interaction. This result also indicates that the generation of monomeric Bax in the presence of Bcl-2 is due to the Bcl-2-Bax interaction. However, the resulting Bcl-2-Bax complex seems not stable, because Bax was eluted mainly as monomers in the presence of the highest concentration of Bcl-2. This may be due to CHAPS in the buffers, which has previously been shown to dissociate tBid-Bax complex and therefore may also dissociate the Bcl-2-Bax complex albeit less efficiently (18). Consistent with this possibility Bcl-2 was also eluted mainly as monomer (ϳ25 kDa) in CHAPS micelles (ϳ6 kDa) (Fig. 2B, right panel). Nevertheless, because the G145A mutation also abolishes the interaction of Bcl-2 with Bax and anti-Bax activity in cells (44,45), our results suggest a biological relevance for the interaction between these proteins in Triton X-100.
Strategy to Capture Bax Homodimer by Site-specific Photocross-linking-To determine if the detergent-induced Bax homo-interaction can be captured by photocross-linking, we translated RNA encoding Bax in vitro in the presence of [ 35 S]Met and a modified lysyl-tRNA with an ANB moiety covalently attached to the ⑀-amino group of the lysine. The ⑀ANB-Lys-tRNA Lys is a functional aminoacyl-tRNA analog that recognizes the Lys codon in Bax RNA and incorporates the photoreactive Lys analog into the Bax protein (46). The ANBlabeled Bax protein was then incubated with the purified His 6 -Bax protein in the presence of Triton X-100 to induce their interaction. Excess His 6 -Bax was added to prevent the ANB-Bax from homo-interaction thereby increasing the yield of photoadduct between the ANB-and His 6 -labeled proteins. Upon exposing the sample to UV light, a nitrene was generated from the ANB moiety, which is a powerful electrophile that rapidly reacts with any nearby heteroatoms possessing non-bonding electron pairs, double bond, or even C-H bond, including such heteroatoms or bonds in the bound His 6 -Bax. The photolysis thus generated a photoadduct of covalently linked [ 35 S]Met-Bax and His 6 -Bax that was then enriched by Ni 2ϩ -resin, separated from other proteins by SDS-PAGE, and detected by phosphor-imaging.
Human Bax contains nine lysines and hence has the potential to be labeled by ANB at multiple sites. If one or more of these sites are close to or within the surface of [ 35 S]Met-Bax that contacts His 6 -Bax, photoadducts can be formed between the two proteins. However, the precise locations of the photoreactive probe would be difficult to determine, and residues other than the lysines that are located in the binding surface would not be seen. To use the photocross-linking approach to map the binding surface, we generated a series of Bax mutants, each with a single Lys located at a specific location where the photoreactive probe will be incorporated. Each single Lys Bax mutant has a Lys substituting for a specific residue in a Lys-null Bax (Bax K0) that was constructed by replacing all of the lysines in wildtype Bax (Bax WT) with Arg (Fig. 3A).
The following criteria were employed to design the single Lys Bax mutants. 1) Only surface residues were replaced by Lys to avoid disrupting the protein core. 2) Conserved substitutions were preferred, wherever suitable, to minimize the impact on the protein surface. Of note, some of the lysines in Bax WT were kept individually to create several single Lys mutants. 3) Representative Bax mutants with a single Lys in some of the structurally important motifs were assayed in HEK293 cells for proapoptotic activity alongside Bax WT and Bax K0.
In the assay GFP was coexpressed with the Bax proteins from an internal ribosome entry site (IRES) in the same messenger RNA to identify cells expressing exogenous Bax. Annexin V labeling and GFP expression were measured by flow cytometry and used to compare the extent of apoptosis in retrovirus infected (GFP-positive) and non-infected (GFP-negative) cells. The histograms in Fig. 3B show the fraction of non-infected cells (white bars) compared with the fraction of infected cells (black bars) that stained with Annexin V in the exact same well of the microtiter plate. Assume that Bax expression correlates with GFP expression from the IRES, the results clearly show that expression of Bax increased apoptosis in the cells. Furthermore, the increase of apoptosis was due to expression of exogenous Bax, because less apoptosis was detected in the cells that were either not infected or infected with the control retrovirus that expressed only GFP. This single time point assay does not measure cumulative cell death and underestimates the difference between "uninfected" and infected cells, because the threshold GFP expression used to separate the populations was arbitrary, and it is likely that some cells expressing low amounts of Bax and GFP were scored as uninfected. Even with these caveats the bias introduced should result in a similar underestimation of the proapoptotic activity for WT-Bax and the mutants. Importantly, there was a similar increase in Annexin V staining in the cells expressing WT Bax compared with the cells expressing the Bax mutants (Fig. 3B, upper panel). Thus expression of both WT Bax and the Bax mutants induced a similar extent of apoptosis in otherwise untreated cells. The cell death in these populations is mediated by the response of cells to expression of the exogenous Bax and the stress of being grown in culture. When these cells were treated with TNF␣, there was an increase in Annexin V labeling that was further exacerbated by expressing either WT Bax or the mutant Bax (Fig. 3B, lower panel). These data suggest that the expression of Bax and the Bax mutants facilitated TNF␣-induced apoptosis similarly. By considering all the data together we conclude that, when expressed in cells, the Bax mutants are approximately as proapoptotic as WT Bax.
Bax Homodimerization Can Be Detected by Site-specific Photocross-linking-To determine whether the homodimerization of Bax could be detected using site-specific photocrosslinking we used Bax K73, a functional mutant with a single Lys replacing Asn 73 , the last residue of the BH3 region (Fig. 3). The BH3 region is well known to play an important role in Bax homo-interaction (20,(47)(48)(49). The Bax mutant was synthesized by in vitro translation to incorporate [ 35 S]Met and an ANB-Lys at position 73. The resulting protein was incubated with His 6 -Bax in the presence of Triton X-100. After photolysis, a product with an apparent molecular mass close to that predicted for the Bax dimer was detected (Fig. 4A, lane 2, indicated  by arrow). The product is a photoadduct, because it did not form in the control experiment when ⑀ANB-Lys-tRNA Lys or UV irradiation was omitted (Fig. 4A, lanes 1 and 3). The product contains both [ 35 S]Met-and His 6 -Bax because it is radioactive and bound to Ni 2ϩ -resin, and was not formed in the absence of His 6 -Bax (Fig. 4A, lane 4). Therefore, the product is the photoadduct formed by [ 35 4A, lanes 1-3) is due to its binding to His 6 -Bax. The same explanation applies to other data shown in the other lanes of Figs. 4 and 5, because they were also obtained from the Ni 2ϩ -bound fraction, but not to the data shown below in Fig. 7C that were from the total photocross-linking sample.
As expected when RNA encoding the Bax K0 control was translated, ANB-Lys was not incorporated into the Bax protein, because there is no Lys codon in the RNA. Although the [ 35 S]Met-Bax K0 protein still bound to the His 6 -tagged Bax, and was therefore recovered from the Ni 2ϩ -bound fraction (Fig. 4A,  lanes 5-7, indicated by the open circle), no photoadduct was generated, as expected. The product indicated by the closed circle is not a photoadduct, because it was also formed in the absence of the ANB and UV irradiation. Therefore, the photoadduct formed by [ 35 S]Met-Bax K73 and His 6 -tagged Bax was via the ANB probe attached to Lys 73 . The photoadduct formation is due to position 73 being close to the interface of Bax homodimer, because the ANB probe at Lys 73 of one Bax molecule was able to react specifically with micromolar concentrations of the His 6 -tagged Bax molecule in the presence of high concentrations of reticulocyte proteins from the in vitro translation system, including ϳ150 mM globin. Moreover, the cross- linking occurred via the Lys 73 -ANB-derived reactive nitrene that has a relatively short lifetime of nanoseconds.
To check whether the site-specific photocross-linking captured a biologically relevant Bax dimer complex, we added His 6 -Bcl-2⌬TM or its G145A mutant to the reaction. As shown in  4). In contrast, very little Bax/Bcl-2 photoadduct was formed when the His 6 -Bcl-2⌬TM-G145A was used (indicated by the arrowhead in lane 5), and as expected the mutant did not inhibit the formation of Bax homo-photoadduct (compare the arrow-indicated bands in lanes 2 and 6). These findings are consistent with the competitive nature of Bax homoand hetero-interaction with Bcl-2 observed in biological systems, which relies on the interaction between the two proteins and is prevented by the G145A mutation (45,47,50). Further evidence that the G145A mutation inhibits the Bcl-2 interaction with Bax can be found by comparing the circle-indicated [ 35 S]Met-Bax monomer band in lane 5 with other lanes in Fig. 5B. Clearly there is almost no [ 35 S]Met-Bax detected in lane 5, which indicates that the radioactive protein did not bind to the His 6 -tagged G145A Bcl-2 mutant. These data also indicate that the Bax synthesized by in vitro translation behaved similarly to Bax expressed in cells as it bound to WT Bcl-2 but not the G145A mutant. As mentioned above, Bax synthesized in vitro also bound His 6tagged Bax (Fig. 4A), which is another indication that the protein is active. Finally a mutation that inhibited the function of Bax in cells also inhibited the binding between these Bax proteins (see Fig. 8), consistent with the notion that our system reconstitutes authentic binding interactions between the proteins.
Interface of Bax Homo-oligomer Revealed by Site-specific Photocross-linking-Because photocross-linking of Bax with an ANB-Lys located at position 73 with His 6 -Bax indicates the proximity of this position to an interface of the two molecules in the complex, other sites of contact or close approach for the molecules in the complex can be similarly mapped by altering the location of the photoreactive Lys. We therefore prepared 23 additional Bax mutants, each with a single ANB-Lys located at a specific location, and performed the photocross-linking reaction. The results shown in Fig. 5 indicate that the following residues are close to the interface of the neighboring proteins in the Bax complex: 24 in helix 1, 37, and 47 in the loop between helices 1 and 2 (termed loop 1), 68 and 69 in helix 2, 73 in loop 2 (data shown in Fig. 4A), 79 in helix 3, 82 in loop 3, and 134 and 145 in helix 6. In addition, residue 189 and/or 190 near the C terminus is close to the interface as indicated by the data from Bax K189,190.  2 and 3), ⑀ANB-Lys-tRNA Lys (lanes 1 and 7), or UV irradiation (data not shown) was omitted. The schematic underneath illustrates that His 6 -tagged Bax, and His 6 -tagged Bcl-2⌬TM compete for interaction with [ 35 S]Met-and ANB-labeled Bax. Other symbols are as described in panel A.
A map for the interface of Bax oligomer was constructed based on the photocross-linking results (Fig. 6). A striking feature of this map is that most of the interface sites are located in or near the rear pocket on Bax (originally identified as the binding site for a stabilized Bim BH3 helix (31), see Fig. 1A), and in the helix 2, which shares a strong sequence and structural homology with the Bim BH3 helix (5,31,51). For example, residues 24, 47, 134, and 145 are located in the rear pocket, and residues 68, 69, and 73 are in the BH3 region that is part of the front pocket. Thus, this interface map seems in line with the one interface model in which a Bax oligomer is formed via a series of front:rear interfaces (Fig. 1C). Other sites involved in photocross-linking can be included in the interface, if it can be extended to include residue 79 in helix 3 and 82 in the loop 3 located near the front pocket, and 37 in the loop 1 located near the rear pocket. The expansion of the interface is conceivable considering that the Bax:Bax interface may have more contacts than the Bax:BH3 peptide interface. However, the photocrosslinking data also can be explained by the two-interface model that includes the front:front and rear:rear interfaces (Fig. 1B).
Photocross-linking of Bax BH3 Peptide with Bax Protein Supports the Two-interface Model for Bax Oligomerization-To test which interface model is correct for the Bax oligomer, we used a peptide that contains both helices 2 and 3 (H2-H3) of Bax that could interact with either the front or the rear pocket depending on which model is correct. The peptide was crosslinked to [ 35 S]Met-Bax with ANB labeled at a single Lys either in the front or rear pocket (Fig. 7). A radioactive photoadduct that has a slightly higher molecular mass than [ 35 S]Met-Bax was detected when the photoreactive probe was located in the front pocket (K73 and K79) but not the rear pocket (K24, K37, K47, K134, and K147) (Fig. 7A, lane 2, indicated by arrow, Fig.  7B; data not shown for K73). This photoadduct is most likely formed by the peptide and the [ 35 S]Met-Bax, because it was not detected in the absence of the peptide (Fig. 7A, lane 4). Moreover, the photoadduct formed in the presence of a short peptide that only contains helix 2 (H2) had a lower apparent molecular mass than the one formed in the presence of the long H2-H3 peptide (Fig. 7A, lane 6, indicated by arrow). The yield of the photoadduct containing the short peptide is lower than the one containing the long peptide, indicating that the short peptide has lower affinity for Bax protein than the long one. This is consistent with our previous finding that the long peptide can activate Bax to permeabilize membranes more efficiently than the short one (32). Interestingly, the peptide with a Gly 67 to Arg mutation that did not activate Bax in membranes also did not cross-link to the Bax with the photoreactive probe labeled at the front pocket (Fig. 7A, lanes 5 and 7), supporting that the peptide-protein interaction detected in the photocross-linking reaction is functionally relevant.
Since the H2-H3 peptide is part of the front pocket of Bax protein, cross-linking of it to the sites located in the front pocket of Bax protein supports the two interface model in which one of the interfaces is formed by the front pockets from two adjacent Bax proteins (Fig. 1B). In contrast, the cross-linking experiments did not reveal the front:rear interaction predicted by the one interface model (Fig. 1C).
Interaction at the Front Pocket of Bax Alters the Interaction at Its Rear Pocket-In the cross-linking reaction containing [ 35 S]Met-Bax with the photoreactive probe labeled at K134 or K145, there was a photoadduct formed in the presence of the mutant H2-H3 peptide (Fig. 7B, lane 8 or 10, respectively, indicated by the arrowhead). This adduct was most likely resulted from cross-linking of [ 35 S]Met-Bax with a reticulocyte protein in the in vitro translation system. It was noted that the adduct was not formed in the presence of the wild-type H2-H3 peptide (Fig. 7B, lanes 7 and 9). Because the wild-type peptide interacts with [ 35 S]Met-Bax, but the mutant does not (Fig. 7A), only the authentic peptide interaction with Bax inhibits the Bax interaction with the reticulocyte protein. The cross-linking of the reticulocyte protein with Bax was only detected when the photoreactive probe is labeled at K134 and K145, but not other lysines. Therefore, the reticulocyte protein most likely interacts with the rear pocket where the K134 and K145 are located.
Because the peptide interacts with Bax at the front pocket, the interaction of the peptide with Bax at the front pocket inhibits the interaction of the reticulocyte protein with Bax at the rear pocket. The interaction at the front pocket thus likely alters the conformation of the rear pocket releasing the reticulocyte protein.
Intriguingly the cross-linking of [ 35 S]Met-Bax with the reticulocyte protein was also inhibited in the presence of His 6 -Bax (Fig. 7C, compare lanes 1 with 3, and 4 with 6, indicated by arrowhead), suggesting that binding of His 6 -Bax to the front pocket of [ 35 S]Met-Bax may release the reticulocyte protein from the rear pocket, thereby exposing the rear pocket for binding of other His 6 -Bax resulting in the [ 35 S]Met-Bax:His 6 -Bax photoadduct (Fig. 7C,  lanes 1 and 4, indicated by arrow). Taken together these data suggest that the interaction at the front pocket primes the rear pocket for interaction.
To examine this hypothesis in a different way we introduced a G67R mutation into the front pocket of Bax. This mutation, when introduced into both [ 35 S]Met-and His 6 -Bax proteins, greatly inhibited the interaction between the two proteins, because much less [ 35 S]Met-Bax coprecipitated with Ni 2ϩ -bound His 6 -Bax-G67R than His 6 -Bax (Fig. 8A, (Fig. 8A, top image, lane 8, indicated by  arrow). When compared with the cross-linking observed between the two proteins without the G67R mutation (Fig. 5, K79, lane 2) the cross-linking was not inhibited when the mutation was only in one of the proteins (e.g. [ 35 S]Met/ANB-Bax-K79-G67R) (Fig. 8A, top image, lane 9, indicated by arrow; data not shown for cross-linking of [ 35 S]Met/ANB-Bax K79 to His 6 -Bax-G67R). These results can be interpreted by a model in which the front:front interface is symmetric. As illustrated in Fig. 8B, the G67R mutation in the front pocket of one binding partner cannot disrupt the entire interface and hence the interaction (middle diagram), whereas the mutation in both binding partners abolishes the interaction (bottom diagram).
Surprisingly the G67R mutation in the front pocket also greatly reduced the cross-linking through the probe attached to FIGURE 6. Potential interface of Bax oligomerization. Based on the site-specific photocross-linking data shown above, the surfaces of Bax that may form the oligomer interface were modeled on the previously determined Bax structure (5). The residues that resulted in Bax dimer-specific photoadduct when replaced by ANB-Lys and hence were close to or within the interface are in red, whereas the residues that did not result in the photoadduct and hence were likely far from the interface are in blue. The radii of these residues are increased to 6 Å to reflect the uncertainty of the photocross-linking-based mapping technique (36). The position of these residues in the Bax sequence is indicated by number. The view on the left and right shows the front pocket and rear pocket, respectively. The schematic underneath each view illustrates that the Bax protein with ANB-Lys at either the front or rear pocket interacts with the His 6 -tagged Bax.
K134 in the rear pocket (Fig. 8A, top image, compare the arrowhead-indicated bands in lanes 2 and 3), and this inhibitory effect was only observed when the mutation was introduced into both Bax proteins in the cross-linking reaction. Therefore, when the front-pocket mutation disrupted the front:front interface, it also disrupted the rear:rear interface (illustrated in Fig. 8C, bottom diagram), suggesting that binding at the rear pocket is dependent on the binding at the front pocket. The binding at the front pocket may induce a conformational change at the rear pocket (Fig. 8C, top diagram), as suggested by the fact that binding of Bax peptide or protein to the front pocket inhibited binding of the reticulocyte protein to the rear pocket (Fig. 7). We therefore concluded that binding of two Bax proteins through the front pockets primes the rear pockets for binding of other Bax proteins (illustrated in the top diagram of Fig. 8C, the two proteins on the left as well as the two on the right bind through the front pocket first, then the two resulting dimers bind through the conformation changed rear pocket to form a tetramer).

DISCUSSION
We have used site-specific photocross-linking to systematically map the interface of Bax oligomer. By combining the results from this and previous studies (22,25,28,(31)(32), it is Implications for Bax Activation and Oligomerization-The salient feature of our model is that each protein in the oligomer contributes two surfaces to interact with neighboring proteins (Fig. 9). Although a single exposed surface is sufficient to form dimers, both surfaces need  Fig. 5. C, schematic illustration for how the conformation and interaction at the rear pocket was altered by interaction at the front pocket, and how the conformation and interaction were affected by the front pocket mutation. FIGURE 9. A stepwise model for Bax oligomerization. Bax oligomerization may occur through the following steps. 1, upon activation, Bax changes conformation to expose the front pocket; 2, the conformation-altered Bax dimerizes through interaction at the front pocket; 3, the interaction at the front pocket alters the conformation at the rear pocket; and 4, two Bax dimers with the altered rear pocket bind and form a tetramer that can grow into larger oligomers by adding more similar Bax dimers at the rear pocket ends. Neighboring Bax molecules in the dimer and tetramer are shaded differently for clarity, although they have the same conformation and interaction. to be exposed before oligomerization can take place. Soluble Bax monomers are most likely in a conformation with the BH1-3 groove occluded by ␣9 but the rear pocket exposed (the substrate before step 1 in Fig. 9) (5). A BH3-only protein may bind the rear pocket of Bax using their BH3 motif as suggested previously (22,28,31). The binding may change Bax conformation releasing ␣9 from the BH1-3 groove such that the hydrophobic ␣9 can insert into membrane (Fig. 9, step 1). Dislodging of ␣9 would expose and alter the BH1-3 groove to form the front pocket that in turn could bind the front pocket of another Bax, which has been similarly activated by a BH3-only protein (step 2). Alternatively or additionally, the first activated Bax may use its exposed BH3 region to bind the rear pocket of a soluble Bax monomer. This binding would activate the Bax monomer in the same way as the binding of BH3-only protein does. In this way the initial activation of Bax by BH3-only protein can be amplified through Bax auto-activation as we observed before (32). No matter how Bax proteins are activated, some of them will have the front pocket exposed. These pockets can engage with each other to form the first Bax:Bax interface via the front:front interaction demonstrated in the current cross-linking study generating a dimer (Fig. 9, step 2). The formation of the front:front interface may alter the rear pocket of the two Bax proteins thereby releasing the activator, either a BH3-only or a Bax protein, from the pocket (step 3), a possibility that is consistent with the data that interaction of Bax BH3 peptide at the front pocket of a Bax protein releases the reticulocyte protein from the rear pocket of the Bax protein (Fig. 7). The interaction via the re-exposed but altered rear pocket of two Bax dimers generates a Bax tetramer (step 4). The tetramer can propagate into a larger oligomer by interaction with other dimers that are generated through the steps 1-3. Interestingly Bak, a structural homolog of Bax, was shown to form oligomers with two similar interfaces in recent studies (14,15). These studies also show that the front:front interface is symmetric and formation of the rear:rear interface depends on the interaction at the front:front interface. Therefore, despite their difference in terms of intracellular location and interaction prior to activation, Bax and Bak may form similar oligomers via similar mechanisms after activation, consistent with their functional redundancy.
We noted that G67R mutation was recently suggested to inhibit the tBid-induced conformational change that is required for Bax oligomerization (28). This could provide an alternative interpretation of our cross-linking data with the mutant. However, our interpretation is supported by the fact that the G67R Bax mutant still cross-linked with a wild-type Bax, therefore, the mutant should be in an oligomerization competent conformation. The difference between our and their observations is most likely that the mutant protein can still be activated by detergent. In fact even tBid can partially activate Bax resulting in oligomer formation as detected in their gel-filtration assay, suggesting that the mutant retains some potential to be activated and form oligomers.
The Bax oligomer may grow while each protein in the oligomer is in the tail-anchoring conformation with the ␣9 helix inserted into the membrane and the rest of the protein retained on the membrane surface similar to the locale suggested for Bak (52). The interaction between the ␣9 sequences may occur at this stage as suggested by our cross-linking data from the photoreactive probe attached to the C-terminal end of ␣9 (Fig. 5,  K189,190). The tail-anchored oligomer may switch to the multispanning conformation with additional sequences, such as ␣5 and ␣6, inserted into the membrane. This conformational switch may occur after the oligomer reaches a certain size, and may lead to permeabilization of the membrane. However, such a mechanism completely disagrees with our previous study in which monomers of activated Bax with ␣5, ␣6, and ␣9 embedded in the membrane were found in the absence of oligomerization (27). Therefore, the conformational change caused by binding of BH3-only protein to the rear pocket of soluble or peripheral membrane-bound Bax may induce a transition from the cytoplasmic to the multispanning state. The BH3 region in the multispanning Bax may be exposed on the membrane surface to engage other soluble/peripheral Bax molecules inducing the same conformational transition. In this alternative model, the Bax oligomer grows as each monomer switches to the multispanning conformation. The membrane permeabilization may start when the multispanning Bax oligomer grows to certain size. The extent of permeabilization or the pore size may increase thereafter as more multispanning monomers are added to the oligomer. Because there is no evidence for a maximal Bax pore size, there may be no constraints on the eventual size of the oligomer. The cross-linking approaches that we and others have used will help us to map the interface for the poreforming oligomer (14,36).
Implications for How Bcl-2 Inhibits Bax Activation and Oligomerization-Bcl-2 can bind to BH3-only proteins preventing them from activating Bax. Bcl-2 can also sequester the activated Bax proteins inhibiting their conformational change and oligomerization (1,18). Thus, Bcl-2 may bind to the exposed BH3 region in the activated Bax to prevent it from activating other Bax molecules through auto-activation (32) and from oligomerizing with other activated Bax (44). In addition, it is also possible that Bcl-2 may bind to the rear pocket of the Bax dimer to prevent it from forming the tetramer. Our preliminary cross-linking data suggest that both types of Bcl-2-Bax interaction exist. Further experiments are underway to determine the extent of overlap between the Bax:Bcl-2 and Bax: Bax interfaces.
Triton X-100 Induces Bax Oligomerization Accompanied by Conformational Change-We used Triton X-100 to induce Bax oligomerization that likely occurs after or concomitantly with the conformational change induced by the detergent. Triton X-100 exposes an N-terminal epitope between residues 13 and 19 that is recognized by 6A7 monoclonal antibody (8,53). This epitope is buried in the soluble Bax in healthy cells (8) and becomes exposed upon apoptotic stimuli when the Bax is bound to the mitochondrial membrane (26,28,54). In solution the 6A7 epitope can be reversibly exposed by liposomes (37). Addition of tBid to the Bax/liposome mixture makes the epitope exposure irreversible, which accompanies Bax oligomerization in the liposomal membrane. Interestingly the 6A7 epitope is buried in the hetero-complex formed between Bax and Bcl-2 or Bcl-x L in Triton X-100 (8,40). Because this epitope overlaps with the helix 1, which is part of the rear pocket of Bax (see Fig. 1A), Bcl-2 or Bcl-xL may bind to this pocket and mask the epitope. Together these observations suggest that the conformational change of Bax induced by Triton X-100 shares some characteristics with the conformational change induced by apoptotic stimuli.
Numerous studies have shown that Bax oligomerizes in various detergents, including Triton-X-100, similarly to the membraneembedded Bax (18,40,42,44). Bax oligomers formed in detergents or membranes can be detected by gel-filtration chromatography and chemical cross-linking. It is therefore surprising that Bax was found as a monomer in the micelles of several detergents, including Triton X-100, in a recent study (34). This study used a fluorescent Bax mutant lacking ␣9 and fluorescence correlation spectroscopy and fluorescence-intensity distribution analysis. The results from our photocross-linking study are not in accord with the above conclusion. In particular, photoadducts of Bax dimer were detected in Triton X-100 micelles when the photoreactive probe was located at multiple sites on the Bax protein, demonstrating that the Bax proteins in the micelle are in close proximity, which strongly suggests a direct molecular interaction. The reason for the difference in conclusions is likely that we used full-length Bax protein, whereas Ivashyna et al. used truncated protein without the C-terminal ␣9. Consistent with this possibility we have shown that the C terminus is involved in the Bax:Bax interaction by crosslinking through a C-terminal photoreactive probe (Fig. 5,  K189,190).
In summary, our site-specific photocross-linking study provides the first comprehensive interface map of the Bax oligomer that is formed in detergent micelles. Our system mimics apoptotic stimuli to activate Bax monomers and native membranes to let the activated Bax form oligomer. Our data support a twointerface model in which Bax oligomerizes through two separated interfaces, and formation of one interface depends on formation of the other interface. Furthermore, Bax activation and oligomerization likely progress through a sequence of steps, each of which may be targeted for therapeutic gains to treat apoptosis-related diseases.