Identification of an activation site in Bak and mitochondrial Bax triggered by antibodies

During apoptosis, Bak and Bax are activated by BH3-only proteins binding to the α2–α5 hydrophobic groove; Bax is also activated via a rear pocket. Here we report that antibodies can directly activate Bak and mitochondrial Bax by binding to the α1–α2 loop. A monoclonal antibody (clone 7D10) binds close to α1 in non-activated Bak to induce conformational change, oligomerization, and cytochrome c release. Anti-FLAG antibodies also activate Bak containing a FLAG epitope close to α1. An antibody (clone 3C10) to the Bax α1–α2 loop activates mitochondrial Bax, but blocks translocation of cytosolic Bax. Tethers within Bak show that 7D10 binding directly extricates α1; a structural model of the 7D10 Fab bound to Bak reveals the formation of a cavity under α1. Our identification of the α1–α2 loop as an activation site in Bak paves the way to develop intrabodies or small molecules that directly and selectively regulate these proteins.

T he commitment of cells to apoptotic cell death is determined by interactions between members of the Bcl-2 protein family on the mitochondrial outer membrane (MOM) 1,2 . Members of this family contain one to four Bcl-2 homology (BH) domains, and are divided into three sub-classes: prosurvival members that contain the BH1-BH4 domains; pro-apoptotic BH3-only members; and pro-apoptotic Bak and Bax that also contain the BH1-BH4 domains.
A key step in apoptosis is the loss of MOM integrity, which requires Bak and Bax activation followed by their structural conversion into pore-forming oligomers [2][3][4] . Both Bak and Bax contain nine a-helices, including a C-terminal transmembrane domain (a9), a buried BH3 domain (a2), as well as a hydrophobic surface groove (a2-a5) that can engage in interactions with other members of the Bcl-2 family. Whereas Bak is inherently mitochondrial, Bax is largely cytosolic with its a9-helix partly sequestered in the a2-a5 groove 5 until Bax accumulates on the MOM following an apoptotic stimulus 6,7 .
Bak and Bax activation (that is, unfolding) are triggered when BH3-only proteins (for example, Bid or Bim) bind transiently to the a2-a5 groove [8][9][10][11] . In Bax, but not Bak, access of the activator to the a2-a5 groove requires initial binding to a second site (rear pocket) between a1 and a6 to displace a9 (refs [12][13][14]. Bak may also be activated at sites other than the a2-a5 groove, as several proteins reported to directly activate Bak appear to lack a BH3-like motif [15][16][17][18] . Binding of BH3-only proteins to the Bak and Bax a2-a5 groove initiates unfolding of a2 followed by dissociation of both a1 and the a6-a8 latch 8,9,19 . The unfolded proteins collapse onto the mitochondrial surface and dimerize via a reciprocal BH3:groove interaction to nucleate the oligomers thought to permeabilize the MOM 5,20-25 .
Here we report the proximal a1-a2 loop as a second activation site in Bak and in mitochondrial Bax. This site can be targeted by antibodies to induce the same Bak and Bax homo-oligomerization and pore formation as that induced by BH3-only proteins. A structural model of the 7D10 Fab bound to Bak supports biochemical evidence that antibody binding to the a1-a2 loop acts by directly dissociating a1.

Results
An antibody to Bak triggers mitochondrial permeabilization. While using antibodies to characterize Bak conformational changes triggered by tBid, we found that an anti-Bak antibody, clone 7D10, could trigger cytochrome c release from mitochondria expressing human Bak (hBak, Fig. 1a). During the incubation, Bak had become activated as shown by sensitivity to limited proteolysis ( Fig. 1b; Supplementary Fig. 1a-c), and had oligomerized as shown by disulfide-linked dimers induced by addition of the oxidant copper phenanthroline (CuPhe, Fig. 1c). Two alternative antibodies, 8F8 and anti-FLAG, that bound Bak N-terminal to a1, failed to activate Bak and FLAG-Bak, respectively (Fig. 1a-c; Supplementary Fig. 2a,b). These data demonstrate that an antibody can trigger Bak activation, oligomerization and mitochondrial cytochrome c release, and that the epitope recognized by 7D10 may be an important site for activating Bak.
The 7D10 antibody binds to the a1-a2 loop of human Bak. 7D10 is a rat monoclonal antibody raised against human BakDC25 (ref. 22). By peptide array we had defined 51 GVAAP 55 at the start of the a1-a2 loop (Fig. 1d) as the minimal set of residues required for 7D10 binding, with G51 and P55 as particularly important residues within this sequence 19 . We then tested whether substituting each residue in this region (with cysteine) decreased 7D10 binding and activation. When tested by immunoprecipitation, G51 and P55 were confirmed to be important for 7D10 binding (Fig. 1e) and for 7D10-induced cytochrome c release (Fig. 1f). Curiously, 7D10 could immunoprecipitate BakP55C only after its activation by tBid (Fig. 1e), suggesting that cysteine at position 55 is permissive for binding of 7D10, but that proline at that position helps to orientate the epitope in non-activated Bak. The substitutions did not affect tBid-induced cytochrome c release (Fig. 1f), or Bak-mediated apoptosis in cells ( Supplementary Fig. 3a), although we cannot exclude the possibility that variants may have slightly more or less function than wild-type Bak.
To test if 51 GVAAP 55 was sufficient for Bak activation by 7D10, four sets of substitutions were made in mouse Bak (mBak, Fig. 1g), which we had found was not activated by 7D10 (Fig. 1a). The substitutions did not affect Bak-mediated apoptosis in cells ( Supplementary Fig. 3b). In mitochondria experiments, the GVAAP and AEGVAAP sequences were not sufficient for 7D10 to immunoprecipitate non-activated mBak or induce cytochrome c release, whereas the GAAAPAD sequence was sufficient (Fig. 1h,i). The importance of the aspartate residue was confirmed by the single substitution (N55D) allowing mBak binding and activation by 7D10 (Fig. 1h,i), and by the reverse substitution (D57N) in hBak inhibiting activation by 7D10 ( Supplementary Fig. 4a-c). Thus, as discussed for the proline residue, this aspartate residue may help 'present' other epitope residues in non-activated hBak. In addition, both the hBak P55C and D57C variants were susceptible to proteinase K even before their activation ( Supplementary Fig. 5a), suggesting that P55 and D57 structure the a1-a2 loop in a way that allows recognition by 7D10 but not proteinase K. Further, D57 is not in contact with the antibody, as shown by the ability of D57C to disulfide bond (D57C:D57C linkage) between Bak molecules after 7D10 has bound and triggered Bak activation ( Supplementary  Fig. 5b,c). Similarly, the A49C:A49C linkage after 7D10 activation indicates that the other end of the epitope is delimited by A49 ( Supplementary Fig. 5b,c). Thus, the functional 7D10 epitope in non-activated Bak is the 51 GVAAPAD 57 sequence, with P55 and D57 orientating other epitope residues (for example, G51) that directly contact the antibody.
Antibody-triggered activation requires epitope close to a1. To investigate the mechanism of antibody-triggered Bak activation and in particular the relevance of the epitope being close to a1, the FLAG epitope was placed in four positions in the a1-a2 loop (Fig. 2a). Each variant retained pro-apoptotic function in cells ( Supplementary Fig. 3c), and each variant could be immunoprecipitated with anti-FLAG antibody (Fig. 2b). Notably, antibody-triggered cytochrome c release occurred only if the FLAG epitope was N-terminal in the a1-a2 loop (n-loop replacement or insertion; Fig. 2c). Moreover, when the 7D10 epitope was re-positioned 11 residues away from a1, 7D10 could bind but could no longer activate Bak (n-loop insert, Fig. 2b,c). As a1 dissociation is an important step in Bak activation by tBid 19 , the anti-FLAG and 7D10 antibodies may act by binding close to a1 and directly destabilizing its many contacts with the remainder of Bak.
7D10 binding to Bak does not directly target a2. To test which Bak conformational changes mediated 7D10-induced activation of Bak, we used disulfide tethers to restrain the various structural elements in Bak (Fig. 2d). Each tether was efficient and did not hinder 7D10 binding to Bak ( Supplementary Fig. 6a,b). Three tethers (A28C:L163C, Y41C:A79C and V142C:T148C) previously shown to block Bak activation by tBid 8,19,23 also blocked activation initiated by 7D10 (Fig. 2e), indicating that although tBid and 7D10 bind to different sites on Bak (Fig. 1d), most of the subsequent unfolding events (illustrated in Supplementary  Fig. 6c) are equivalent. A tether of the a1-a2 loop to the a5-a6 loop (E59C:T148C), similar to a tether (L45C:M137C) shown to block Bax activation 12 , also blocked Bak activation by both tBid and 7D10 (Fig. 2e). This tether may prevent core/latch separation, although the tether is to the hinge region rather than to the latch (a6-a8) itself. In contrast, because the a1-a2 loop lies  across a1, tethering the loop would greatly hinder a1 dissociation. The final tether, from the middle of the a1-a2 loop to a1, was designed to prevent the effect of antibody binding to the N-terminal end of the loop from being transduced to the C-terminal end of the loop and to a2 (H43C:M60C; Fig. 2d). In spite of this tether, 7D10 still activated Bak (Fig. 2e), further indicating that 7D10 binding directly impacts a1, and not a2.
The 7D10 Fab efficiently activates Bak. To test whether the bivalency or bulk (B150 kDa) of the 7D10 antibody might contribute to Bak activation, the 7D10 antibody fragment (Fab, B50 kDa) was purified ( Supplementary Fig. 7a,b). When Fab was incubated with mitochondria, Bak underwent activation and oligomerization ( Supplementary Fig. 7c) and released cytochrome c with similar molar efficiency as the full antibody ( Supplementary Fig. 7d). Thus, the bivalency and bulk of 7D10 are not essential for its activation of Bak.
The 7D10 Fab alters oocyte mitochondrial morphology.
The 7D10 Fab was then tested in cells by microinjection into immature human oocytes. The 7D10 Fab, but not a control Fab, induced mitochondrial aggregation that progressed over 16 h (Supplementary Fig. 8a-c). The morphology was consistent with mitochondrial fusion, supported by a decrease in mitochondrial particle number and increase in size ( Supplementary Fig. 8f,g). We also observed an apparent deformation of the plasma membrane creating a highly contoured surface compared with controls ( Supplementary Fig. 8d,e). This is consistent with early stages of apoptosis seen in oocytes under other conditions 26 . Importantly, changes in mitochondrial morphology were not apparent when the 7D10 Fab was injected into mouse oocytes ( Supplementary  Fig. 9a-e), consistent with an on target effect of the Fab on human Bak. Notably, the human oocytes appear not to have undergone MOMP, as TMRM was retained in the aggregated mitochondria ( Supplementary Fig. 8a-c). Further, mitochondrial fission rather than fusion most often accompanies MOMP 27,28 , although Bak and Bax may also contribute to mitochondrial fusion [29][30][31] .
Mitochondrial Bax can be activated by antibody to a1-a2 loop. Due to the structural and functional similarities of Bak and Bax, we anticipated that Bax would also be activated by an antibody to its a1-a2 loop. This was first tested in a mitochondrial form of Bax (S184L), that like Bak, adopts a non-activated conformation in which a9 forms a transmembrane domain [32][33][34] , and that converts to the same activated conformation as Bax in response to tBid 25 . Because of its mitochondria location, BaxS184L does not require initial binding of tBid to the rear pocket to extrude a9 (refs 12-14). When the 7D10 epitope was positioned in the a1-a2 loop of mitochondrial Bax (BaxS184L-GVAAPAD; Fig. 3a), the 7D10 (anti-Bak) antibody could bind (Fig. 3b) and release cytochrome c (Fig. 3c). In addition, a novel anti-Bax antibody, clone 3C10, that recognized a peptide encompassing residues 31-45 at the start of Bax a1-a2 loop (Fig. 3d) was able to bind mitochondrial Bax (Fig. 3e) and induce dimerization and cytochrome c release (Fig. 3f).
As Bax is normally cytosolic, we tested whether 3C10 could also activate WT Bax that was present in cytosolic extracts or that was generated as recombinant protein. Unexpectedly, although 3C10 could bind WT Bax from both sources (Fig. 3g), it failed to trigger Bax translocation to the mitochondria or cytochrome c release (Fig. 3h). Notably, 3C10 also prevented tBid-triggered Bax translocation (Fig. 3i) and cytochrome c release (Fig. 3j). A similar blockade effect was very recently reported for a Fab that bound residues of a1, a6 and the a1-a2 loop 35 .
Structural model shows 7D10 binding causes cavities under a1. The mechanism of antibody-mediated Bak activation was further investigated by generating a molecular model of the 7D10 Fab bound to Bak (Fig. 4). We note that Bak is normally anchored in the outer membrane by a9 forming a transmembrane domain 33 , a feature that may affect Bak activation. However, membranes were not incorporated in the MD simulations as the structure of membrane-bound Bak is not available. Models of Bak obtained from MD simulations were docked with homology-based models of 7D10: several models of both Bak and 7D10, representing different possible loop conformations of each, were cross-docked against one another in an attempt to account for flexibility in the uncomplexed proteins. The final model of the complex after molecular dynamics simulation ( Fig. 4a) was also assessed for the effect of antibody binding on the structure of Bak (Fig. 4b,c).
The final model (Fig. 4a) has a large interaction surface (1,411 Å 2 ) and high shape (0.670) complementarity. Moreover, the interaction surface reflects our biochemical identification of the 7D10 epitope ( Fig. 1d-i) and the availability of cysteine residues at either end of the epitope for linkage in 7D10-activated Bak oligomers (A49C and D57C; Supplementary Fig. 5b-d).
Notably, the 7D10 epitope is flanked by CDR2 and CDR3 of the heavy chain. A much smaller contact forms between the light chain and the C terminus of a3; this interaction is dominated by weak van der Waals interactions, which may help explain why 7D10 can also bind and activate BaxS184L-GVAAPAD (Fig. 3b,c). Insertion of CDR3 in the groove underneath the a1/loop region may contribute to activation, as both 7D10 and the M2 anti-FLAG antibody failed to activate Bak if the related epitope was positioned B11 residues away from a1 ( Fig. 2a-c). antibody induces cytochrome c release. Membrane fractions from Bak À / À Bax À / À MEFs, those cells expressing human Bak (hBak), or Bax À / À MEFs, were incubated with tBid or with the 7D10 or 8F8 antibodies. Supernatant (Sup) and pellet (Mito) fractions were assessed for cytochrome c release. (b) 7D10 triggers Bak conformational change as indicated by susceptibility to proteinase K. Incubations from a were treated with proteinase K and immunoblotted for Bak. Note that 7D10 binding at the loop masks a cleavage site (lane 4, Supplementary Fig. 1a), and that uncleaved Bak and light chain co-migrate. (c) 7D10 triggers Bak oligomerization. Incubations from a were treated with oxidant (CuPhe) to induce disulfide bond formation and immunoblotted for Bak. M, monomer; M x, intramolecular linked monomers; D, intermolecular linked dimers. (d) The 7D10 trigger site in Bak is distinct from the canonical BH3-only trigger site. Cartoon representation of BakDN19DC25 (2IMT, white) highlighting the a1-a2 loop (blue), and a3 and a4 of the hydrophobic groove (green). (e) Mutation of Bak G51 or P55 inhibits binding by 7D10. Membrane fractions from Bak À / À Bax À / À MEFs expressing the indicated hBak variants were incubated with or without tBid followed by immunoprecipitation with 7D10 and immunoblotting for Bak. IP, immunoprecipitated; UB, unbound; #, light chain. (f) Mutation of Bak G51 or P55 prevent Bak activation and cytochrome c release by 7D10. Membrane fractions from e were incubated with tBid or 7D10 and assessed for cytochrome c release. (g) Substitutions in mouse Bak to generate the 7D10 epitope. (h) The 51 GVAAPAD 57 sequence allows binding of 7D10 to mouse Bak. Membrane fractions from Bax À / À MEFs (mBak cells), or Bak À / À Bax À / À MEFs expressing hBak or the indicated mBak variants, were incubated with or without tBid followed by immunoprecipitation with 7D10 and immunoblotting for Bak. (i) The 51 GVAAPAD 57 sequence allows 7D10 to activate mouse Bak and release cytochrome c. Membrane fractions from h were incubated with tBid or 7D10 and assessed for cytochrome c release. In a-c,e,f,h and i data are representative of three independent experiments. Molecular dynamics simulations of non-activated Bak indicated significant flexibility specifically in the region containing the 7D10 epitope and a3 (Fig. 4b, left), consistent with the high B-factors observed in this region in X-ray structures of Bak (2IMT, 2JCN, 2YV6) 36,37 . Binding of 7D10 to the epitope resulted in reduced flexibility in both regions (Fig. 4b, right), consistent with 7D10 binding to a preferred conformation. Insertion of the CDR3 loop appears to pry a1 from a2, evidenced by the creation of cavities between these two helices (Fig. 4c,  right). In contrast, in unbound Bak only a small cavity is observed between a5 and the a3-a4 loop (Fig. 4c, left), again consistent with the high B-factors in X-ray structures. This weakening of the association of a1 likely represents the initial destabilizing event leading to a1 dissociation from the remainder of Bak (see diagram in Supplementary Fig. 10). Following this event, exposure of the N terminus allows high-affinity interaction of the a1-a2 loop with the heavy chain to persist without hindering formation of BH3:groove dimers (see diagram in Supplementary Fig. 10).

Discussion
We report here that certain antibodies can bind and activate Bak and mitochondrial Bax to induce cytochrome c release. Anti-Bak antibodies required the epitope to be at the start of the a1-a2 loop, and acted by directly dissociating the a1-helix. As antibodies have not been reported to activate either Bak or Bax, nor has the a1-a2 loop region been identified as a trigger site, our findings provide novel opportunities to develop therapeutics that directly activate Bak, and perhaps Bax.
The activation site we have identified at the start of the a1-a2 loop is remote from two sites targeted by BH3-only proteins. The a2-a5 hydrophobic groove in Bak and mitochondrial Bax is engaged by BH3-only proteins to release a1 and unlatch a6-a8 (refs [8][9][10][11]19). A second site at the a1-a6 rear pocket in Bax (but not Bak) is initially engaged by BH3-only proteins to release a9 from the groove 12,13 . The section of the a1-a2 loop bound by three activating antibodies, 7D10, anti-FLAG and 3C10, lies between these two sites.
An intriguing finding was the ability of the 3C10 antibody to trigger cytochrome c release via mitochondrial but not cytosolic Bax, whereas tBid released cytochrome c via both forms. A partial explanation derived from the ability of 3C10 to block Bax translocation triggered by tBid. Indeed, a recent study reported that Bax translocation triggered by tBid can also be blocked by a synthetic Fab that binds residues in a1, a6 and the a1-a2 loop 35   Membrane fractions from Bak À / À Bax À / À MEFs expressing the indicated double cysteine variants were pre-incubated with CuPhe to induce a tether (T; Supplementary Fig. 6a), and then incubated with tBid or 7D10 and assessed for cytochrome c release. Note that the H43C:M60C tether lies between the 7D10 epitope and a2, but does not block 7D10 activation of Bak. In b,c and e data are representative of three independent experiments. conformational changes in the N-terminal surface 35 . As the rear pocket does not seem to be involved in antibody-triggered activation of mitochondrial Bax (or Bak), 3C10 may be unable to activate cytosolic Bax because it suppresses release of a9 from the Bax groove, or because it more directly prevents Bax from associating with the mitochondrial outer membrane.
While BH3-only proteins do not bind the a1-a2 loop, several non-Bcl-2 proteins reported to directly activate Bak or Bax, may do so. These include Bif-1, ab amyloid oligomers, protein disulfide isomerases and p53 (refs 15,18,38-40), although p53 has recently been reported to bind to the Bax a6-a9 region 41 . As these proteins lack welldefined BH3 domains, they may bind to the a1-a2 loop, or to a region nearby, to dislodge the a1 helix. It is also possible that proteins may bind to this region to inhibit Bak and Bax activation, as proposed for the Bcl-2 BH4 fragment 42 , and act in a similar way to tethering the a1-a2 loop within Bak (Fig. 2d,e) or within Bax 12 .
Bak conformational changes induced by antibody (see diagram in Supplementary Fig. 10) are similar to those induced by BH3-only proteins (see diagram in Supplementary Fig. 6c), as both agents induced similar Bak conformational change and oligomerization, and cytochrome c release initiated by either agent was blocked by each of four tethers within Bak. However, the initial unfolding events triggered by both agents were different, consistent with distinct binding sites on Bak. tBid binding first initiates movement of a2 ( Supplementary  Fig. 6c) 8,19 , consistent with binding to the a2-a5 groove. In contrast, 7D10 binding initially triggers a1 dissociation, as 7D10 and FLAG epitopes allowed activation only when placed proximal to a1, and tethering the loop to a1 did not block 7D10-triggered Bak activation. Moreover, molecular dynamics of a 7D10:Bak structural model caused extrication of a1 from a2 to form cavities between the two helices.  (*, non-specific band) (g) 3C10 binds WT Bax. Lysates of Bak À / À Bax À / À MEFs expressing Bax (Exp.) or of Bak À / À Bax À / À membranes reconstituted with recombinant Bax (100 nM, Rec.) were immunoprecipitated with 3C10 and immunoblotted for Bax. (h) 3C10 fails to trigger WT Bax translocation to mitochondria and release of cytochrome c. Permeabilized Bak À / À Bax À / À MEFs expressing hBax (Exp.) or Bak À / À Bax À / À MEF membranes reconstituted with recombinant Bax (50 nM, Rec.) were incubated with tBid or 3C10. Supernatant (Sup) and pellet (Mito) fractions were immunoblotted for Bax, VDAC or cytochrome c. (#, light chain). (i,j) 3C10 blocks Bax translocation and integration and cytochrome c release triggered by tBid. Bak À / À mouse liver mitochondria were incubated as in h, but an additional sample also contained both 3C10 and tBid. Samples were assessed for Bax membrane integration (i), or for cytochrome c release (j) as in (h). In b,c and e-j data are representative of three independent experiments.
As either Bak or Bax is essential for mitochondrial apoptosis, developing agents that directly trigger their activation is of great interest 43,44 . Developing anti-cancer agents that target the Bak loop may complement BH3-mimetics that target the Bcl-2 prosurvival proteins 1,45 . While BH3-mimetics release endogenous BH3-only proteins that activate both Bak and Bax 46-48 , the 7D10 antibody, and other activators that might directly dislodge a1, are likely to specifically bind Bak due to the lack of sequence homology in the loop regions. This specificity also implies that this class of activators cannot be sequestered by prosurvival Bcl-2 proteins such as Bcl-x L and Mcl-1, and could bypass resistance caused by those proteins.
It remains to be demonstrated that the 7D10 Fab-induced changes in oocyte mitochondrial morphology are due to Bak conformational change. There are very few studies investigating apoptosis in oocytes at the immature fully grown germinal vesicle stage we have used. It is well known that early oocytes in primordial follicles undergo a TAP63-mediated apoptosis in response to DNA damage 49 but downregulation of TAP63 on the initiation of oocyte growth is thought to make fully grown oocytes, such as those used here, more refractory to apoptosis 50 . Some evidence of oocyte fragmentation to form apoptotic bodies has been reported when mature ovulated oocytes are allowed to age in vitro but the outward signs of apoptosis remain controversial at these mature stages 51,52 . The limited availability of healthy human oocytes for these studies is a practical limitation on our ability to formally prove the association of mitochondrial morphology with Bak conformational change.
In conclusion, we have identified the proximal a1-a2 loop as a second activation site in Bak and in mitochondrial Bax. This site can be targeted by antibodies to directly dissociate a1, and as a result induce the same Bak oligomerization and pore formation that is induced by BH3-only proteins. This site may allow the development of novel therapeutics such as intrabodies based on 7D10, and possibly small molecules, that pry a1 from a2. Unlike the BH3-mimetics, such agents would specifically activate either Bak or Bax, and, as they would not be sequestered by prosurvival Bcl-2 proteins, they may bypass resistance due to those proteins. Conversely, agents that bind the a1/loop region and prevent a1 dissociation may provide novel anti-apoptotic agents.

Methods
Cloning and expression of Bak and Bax variants. Bak and Bax mutants were generated by site-directed mutagenesis and overlap extension PCR using Phusion High-Fidelity Master Mix (New England BioLabs). The two BaxS184L variants also contain two cysteine substitutions in the BH3 domain (S55C) and groove (R94C) to monitor BH3:groove interaction after Bax activation and oligomerization 53 . Primer sequences are listed in Supplementary Table 1. PCR products were digested with EcoRI and XhoI, and cloned into the pMX-IRES-GFP retroviral expression vector, and the introduced mutations confirmed by DNA sequencing. The variants were then stably expressed in SV40-immortalized Bak À / À Bax À / À mouse embryonic fibroblasts (MEFs), and were a gift from Professor David Huang. MEFs were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum, 0.1 mM L-asparagine and 55 mM 2-mercaptoethanol. All cultures were maintained at 37°C in a humidified 10% CO 2 incubator. Bak activation assessed by limited proteolysis. Bak conformational change was assessed by limited proteolysis 55 . Briefly, 50 ml of membrane fractions resuspended in MELB with 4 mg ml À 1 pepstatin A (Sigma) were pre-chilled to 0°C then incubated with 30 mg ml À 1 proteinase K (Roche) or trypsin (Sigma) on ice for 20 min, or 0.06 U ml À 1 enterokinase (Merck Millipore) at room temperature for 2 h. The reactions were stopped by addition of 1 mM phenylmethylsulfonyl fluoride (PMSF) and immunoblotted with antibodies to the Bak BH3 domain (4B5).
Disulfide bond formation in Bak. Bak conformational change and oligomerization was assessed by inducing disulfide bonds within or between Bak molecules. Membrane fractions were exposed to the redox catalyst copper (II)(1,10-phenanthroline) 3 (CuPhe) diluted 100-fold from a stock containing 30 mM CuSO 4 and 100 mM 1,10-phenoanthroline in 4:1 water:ethanol 56 . Reactions were incubated for 30 min on ice and then re-suspended in SDS sample buffer containing 20 mM EDTA before analysis by non-reducing SDS-PAGE and immunoblotting. Where indicated, CuPhe was also used to 'tether' non-activated Bak by forming disulfide bonds between cysteine residues 19 . Briefly, fractions were incubated with CuPhe (1 mM) for 5 min on ice followed by quenching with 20 mM N-ethylmaleimide (NEM) for 10 min at room temperature.
Immunoprecipitation. Briefly, whole cell or membrane fractions from 3-5 Â 10 6 cells were solubilized with 1% w/v digitonin in lysis buffer (20 mM Tris, 135 mM NaCl, 1.5 mM MgCl 2 , 1 mM EGTA, 10% glycerol, pH 7.4) for 1 h on ice. Lysates were centrifuged at 13,000 g for 5 min and pre-cleared with 50 ml Protein G sepharose (Amersham Biosciences). Pre-cleared supernatant was then incubated with antibody (4 mg) and Protein G sepharose. Unbound proteins were collected and the resin washed with lysis buffer containing up to 0.1% w/v digitonin. Immunoprecipitated proteins (IP) were eluted by boiling in sample buffer, and together with unbound (UB) and total lysates (input), were analysed by immunoblotting. Several rat monoclonal antibodies generated in-house were used for immunoprecipitation, including clones 8F8 and 7D10 that recognize Bak 22,23 , and clone 3C10 that recognizes Bax (Fig. 3d). Other antibodies include mouse monoclonal anti-FLAG (M2; Sigma) and several anti-Bak antibodies described in Supplementary Fig. 2. Bax membrane integration. To assess Bax membrane integration, membrane fractions were subjected to carbonate extraction. Mitochondria isolated from Bak À / À mouse liver were incubated for 2 h at 37°C with 50 nM recombinant Bax, and with tBid and 3C10 as indicated. The incubations were centrifuged at 13,000 g for 5 min to obtain cytosolic and pellet fractions. The pellet was re-suspended in 0.1 M Na 2 CO 3 (pH 11.5) and incubated on ice for 20 min. pH was neutralized with an equal volume of 0.1 M HCl and the sample incubated at 22°C for 5 min. Samples were then supplemented with nuclease buffer from a 10 Â stock (400 mM Tris, 100 mM MgSO 4 , 10 mM CaCl 2 ) and 1 U DNase, before being centrifuged at 13,000 g for 10 min, and supernatant (peripherally attached) and pellet (membrane inserted) fractions immunoblotted for Bax.

SDS
Generation and purification of 7D10 Fab. The 7D10 antibody was cleaved with papain to generate the Fab fragment. Papain (Sigma) at 1 mg ml À 1 was solubilized and activated in 10 mM cysteine and 20 mM EDTA in PBS for 10 min on ice. Following 10-fold dilution in PBS, papain was added to the 7D10 antibody at a ratio of 1:20 and incubated at 37°C followed by inactivation with 30 mM iodoacetamide (Sigma). Papain-cleaved antibody was dialysed overnight in Buffer A (10 mM acetic acid, pH 4.5) and applied to a Mono S column equilibrated in Buffer A. A linear 20 ml gradient with Buffer B (10 mM acetic acid, 500 mM NaCl, pH 4.5) was used to separate Fab and Fc in 0.5 ml fractions. Image analysis. Accurately scaled single slice oocyte images at 3 h post injection were converted using ImageJ to 8-bit then a threshold was applied accordingly to each image. An ROI around each oocyte was then used to analyse the number and size of particles.
Molecular dynamics and protein-protein docking. Computational docking of Bak and 7D10 was performed using structures of 7D10 obtained from the RosettaAntibody3 online server 58 , and structures of Bak obtained from molecular dynamics (MD) simulations. The sequence of 7D10 is available from International Patent Application No. PCT/AU2015/000290. Flexibility (beyond just side-chain movement) was incorporated in the docking by considering multiple structures of both Bak and 7D10. Molecular dynamics simulations used the GROMACS (v5.0.4) package of programs 59 with the OPLS-aa force field 60 . Ionizable residues were assumed to be in their charged state. Each molecule was solvated in a water box extending 10 Å beyond all atoms; sodium and chloride ions were added to neutralize the system and provide a final ionic strength of 0.1 M. Protein and water (with ions) were coupled separately to a thermal bath at 300 K using velocity rescaling 61 applied with a coupling time of 0.1 ps, whereas the pressure was coupled to an isotropic barostat using a time constant of 1 ps and compressibility of 4.5 Â 10 5 per bar. All simulations were performed with a single non-bonded cutoff of 10°and applying a neighbour-list update frequency of 10 steps (20 fs). The particle-mesh Ewald method was used to account for long-range electrostatics 62 (applying a grid width of 1.2 Å and a fourth-order spline interpolation). Bond lengths were constrained using the LINCS algorithm 63 . All simulations consisted of an initial minimization of water molecules, followed by 100 ps of MD with the protein restrained. Following positional restraints, all restraints on the protein were removed and MD continued for a further 200 ns. Coordinates were archived throughout the simulation at 10 ns intervals-these structures were used in subsequent docking calculations. The models of the 7D10 antibody Fv region included the original homology model and 10 top-scoring models from H3 loop refinement.
The ZDOCK program 64 was used to dock 21 structures of Bak to 11 structures of the 7D10 Fv with 6°rotational sampling of the ligand. Docking to 7D10 was limited to the CDR loops (including three residues on either side of each loop). The Bak a8-helix was blocked from participating in the interface since its proximity to the membrane should prevent antibody binding, thus providing a weak constraint on the likely orientation of Bak relative to the membrane. Subsequent re-scoring of the top-scoring 2,000 results from each of the 231 ZDOCK calculations was conducted using the ZRANK (v2) program 65 . The top-scoring 2,000 complexes following ZRANK rescoring, inclusive of all 462,000 complexes, were refined using the RDOCK program 66 . The top-scoring 20 complexes from refinement using RDOCK were manually examined for consistency with experimental data. Shape complementarity of the top-scoring 20 RDOCK refined structures was conducted using the Sc program 67 .
MD simulations were performed on both the unbound Bak and the final Bak-7D10 complex obtained from the RDOCK analysis; these simulations were each run for 100 ns. Cavity analysis was performed using the trj_cavity tool 68 .
Data availability. The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information files.