Assembly of Bak homodimers into higher order homooligomers in the mitochondrial apoptotic pore

In mitochondrial apoptosis, Bak is activated by death signals to form pores of unknown structure on the mitochondrial outer membrane via homooligomerization. Cytochrome c and other apoptotic factors are released from the intermembrane space through these pores, initiating downstream apoptosis events. Using chemical crosslinking and double electron electron resonance (DEER)-derived distance measurements between specific structural elements in Bak, here we clarify how the Bak pore is assembled. We propose that previously described BH3-in-groove homodimers (BGH) are juxtaposed via the ‘α3/α5’ interface, in which the C-termini of helices α3 and α5 are in close proximity between two neighboring Bak homodimers. This interface is observed concomitantly with the well-known ‘α6:α6’ interface. We also mapped the contacts between Bak homodimers and the lipid bilayer based on EPR spectroscopy topology studies. Our results suggest a model for the lipidic Bak pore, whereby the mitochondrial targeting C-terminal helix does not change topology to accommodate the lining of the pore lumen by BGH.


The Bak homodimers oligomerize via 'α3/α5 interface' as well as 'α6:α6 interface' in mitochondria.
To determine how Bak homodimers oligomerize in the mitochondrial outer membrane, we mapped the proximity of amino acid residues in the Bak oligomeric pore using disulfide cross-linking 35 (Fig. 2a). Stable expression of full length Bak mutants containing single, double and triple cysteine substitutions at strategic positions was performed in Bax −/− Bak −/− mouse embryonic fibroblasts (MEFs) (see Methods). These Bak mutant proteins targeted to the mitochondria normally, as evidenced by the Western blot analysis (Fig. 2b). The parent cysteine-less Bak (lane 1, Fig. 2b) and the cysteine substitution mutants (lanes 2-11, Fig. 2b) were expressed in varying quantities relative to the wild-type Bak (lane 12, Fig. 2b) (from the lowest ~80% for 162C to the highest ~130% for 111C). These mutant proteins were active in apoptotic pore formation in the mitochondrial outer membrane, as evidenced by the efficient release of cytochrome c from the mitochondria (Fig. 2c-e). When the Bak proteins were activated by p7/p15 Bid, approximately 80-90 percent of the cytochrome c molecules were released from the mitochondria except for mutant 111C (Fig. 2c,d). In the absence of p7/p15 Bid, less than 20 percent of the cytochrome c was released in all the cases (Fig. 2c,e). These data indicated that the cysteine substitution Bak mutant proteins expressed in the MEF mitochondria were mostly intact in their structure and apoptotic function.
In the mouse BGH structure, the α -carbon atom (C α ) of residue 69 on helix α 2 in one α 2-α 5 polypeptide chain is in close proximity to the C α of reside 111 on helix α 4 in the other paired polypeptide (spheres in purple and cyan, respectively, Fig. 2a). The shortest distance between the β -carbon atoms of the cysteines introduced at these two locations is 4.6 Å in the BGHs of the GFP-Bak tetramer and the thiols of these residues can be in closer proximity (Fig. 1d). Thus, upon oxidation by copper(II)(1,10-phenanthroline) 3 reagent, two disulfide bonds will be formed between the cysteine residues (i.e., for 69C′ /111C and 69C/111C′ ) due to the symmetric nature of BGH (Fig. 2a). This will result in a Bak dimer with a shifted mobility in the denaturing polyacrylamide gel electrophoresis (PAGE) as previously shown in human Bak by Dewson et al. 24 . This was indeed the case (lane 8, Fig. 2f). The mobility shift was observed only when Bak was activated by p7/p15 Bid (lanes 7 and 8, Fig. 2f), proving the proximity of the cysteines only in the activated Bak but not in the inactive Bak. In contrast, the disulfide bond was not formed significantly in Bak mutant proteins containing only one cysteine at residue 69 or 111 (lanes 1 and 2, and 3 and 4, respectively, Fig. 2f), regardless of Bak activation by p7/p15 Bid. This further supports that the gel shift in lane 8 was due to the disulfide formation between cysteines at residues 69 and 111, which can be reduced under a reducing condition (Supplementary Information Figure S1c). Collectively, these results confirm that the BGH structure was formed in mitochondrial membrane by mouse Bak when it was activated by p7/p15 Bid, which is consistent with our previous in vitro data 27 and with Dewson et al. 24 .
When an additional cysteine residue such as 143C (the penultimate C-terminal residue of α 5 helix) was present in Bak 69C/111C mutant (i.e., in Bak 69C/111C/143C), large oligomers of even numbered Bak proteins were formed upon oxidation after activation with p7/p15 Bid (lane 10, Fig. 2f; also see Supplementary Information Figure S1c). This was not observed in the absence of Bak activation (lane 9, Fig. 2f), indicating that 143C was brought to the oligomerization interface only when Bak was activated. Consistent with this, a dimer was formed in Bak 143C mutant in a p7/p15 Bid-dependent manner (lanes 5 and 6, Fig. 2f). These results showed that 143C  29 , respectively, using Pymol 59 . The rootmean-square deviation (RMSD) values for the color-coded polypeptide backbone chains were calculated using Pymol 59 .
was brought close to each other between BGHs in the Bak oligomeric pore, but not within a single BGH since the two 143C residues in a Bak BGH (i.e., 143C and 143C′ in Fig. 2a) are separated too far away for disulfide formation (~50 Å between C α atoms). In conclusion, the above results showed that the C-termini of α 5 helices around residue 143 in the BGHs were near the Bak oligomerization interface. Likewise, even-numbered high order oligomers were formed in a p7/p15 Bid-dependent manner only in Bak 69C/111C/96C but not in Bak 96C ( Fig. 2g; lanes 5 and 6, and lanes 1 and 2, respectively). The large distance between two 96C residues in a BGH (~45 Å between C α atoms) also precludes the possibility of disulfide bond formation within a BGH. Thus, these results indicated that residue 96C, i.e., the C-termini of α 3 helices, were juxtaposed in the oligomeric Bak between neighboring BGHs. Finally, similar results were also observed in Bak 162C and Bak 69C/111C/162C (lanes 3, 4, 7 and 8), indicating that residue 162C, the penultimate C-terminal residue of helix α 6, was also at the oligomerization interface in Bak pore, consistent with the formerly known 'α 6:α 6 interface' 23 .
Additionally, the monomers of Bak 86C were cross-linked upon activation by p7/p15 Bid, consistent with the proximity of the two symmetry-related 86C residues in the BGH structure (86C and 86C′ ; ~10 Å or ~13 Å between C β or C α atoms, respectively) (Fig. 2h). This also indicated that the BGH structure was preserved in the Bak oligomeric pores in membrane.
The above results collectively showed that, in addition to α 6 helices, the C-termini of helices α 3 and α 5 were juxtaposed between BGHs in oligomeric Bak (Fig. 2a) in apoptotic mitochondria, thus demonstrating the existence of the 'α 3/α 5 interface. ' This is consistent with our earlier in vitro results obtained with recombinant mutant Bak proteins in liposomes 27 . EPR data further support the existence of 'α3/α5 interface' in oligomeric Bak pore. Using site-directed spin labeling (SDSL) method (Fig. 3a), the inter-spin distances in the range of 15-80 Å can be measured by the double electron electron resonance (DEER) method 36 . There would be multiple spin-spin interactions between BGHs as well as within a BGH if spin-labeled Bak monomers formed oligomeric Bak pores (Fig. 3b,c). This was indeed the case in the Bak oligomeric pores formed with Bak/84R1, a Bak monomer spin labeled at residue 84 (Fig. 3d,e; also see Supplementary Information Figure S2). Clearly, three well-resolved peaks were observed in the probability vs. distance function obtained from the X-band DEER data using DeerAnalysis2013 program 37 (Fig. 3d). Due to the short phase memory time of the electronic spins of the nitroxide labels in X-band experiment, the evolution time was limited and thus the accuracy of the longer distance was compromised. The Q-band DEER was thus used to overcome this. As shown in Fig. 3e where i runs over multiple observations of the same intensity, and j runs over all crystallographic unique intensities. R factor = Σ ||F obs | − |F calc ||/Σ |F obs |. R free was calculated with 5% of the reflections selected. and the distances over a longer range could be measured, which also revealed three distinct probability peaks when analyzed by DeerAnalysis2013 program (right panel) (also see Supplementary Information Figure S3c, 4th row). The probability distribution peak at ~27 Å (with the width of ~5 Å at half-height) was assigned to the intra-dimer spin pair and the other two peaks at ~33 Å and ~49 Å were assigned to the inter-dimer spin pairs (Fig. 3f). This was based on the control experiments shown in Supplementary Information Figure S3. When 84R1 residues were present in the BGHs of His-GFP-Bak tetramer in solution (Supplementary Information Figure S3a), only a single peak at ~27 Å was observed in the short distance range (Supplementary Information Figure S3c, 2nd and 3rd rows), indicating that this peak corresponds to the intra-BGH 84R1-84R1' distance, consistent with modeling of the spin label rotamers in a BGH (see Supplementary Information Figure S3d,e). If the structure of 84R1 in BGH is known, a 2-dimensional modeling of the two neighboring BGHs can be done by triangulation with the three distances determined above (Supplementary Information Figure S2d). The tetrameric GFP-Bak spin labeled at residue 84C could not be crystallized. We thus attempted to model the rotamers of 84R1 in silico, based on its low solvent accessibility (Supplementary Information Figure S2b Figure S3d). When the amino acid side chains around 84R1s were rearranged, (c) When a singly spin labeled Bak forms an oligomeric pore via homodimer formation, various intra-and inter-dimer spin-spin distances (d 1 , etc.) can be determined by the DEER method. (d,e) X-(d) and Q-band (e) DEER data for oligomeric Bak formed with sBak-Δ C-His spin labeled at residue 84 (Bak/84R1) were obtained (red traces, left panel; black dotted lines are theoretical fit) and analyzed with DeerAnalysis2013 37 , resulting in the indicated most probable distances (Å) (width at half-height in parenthesis). (f) Rotamers of 84R1, 84R1 a and 84R1 b , on the polypeptide chains A and B, in the hydrophobic pockets of BGH and their dihedral angles X 1 -X 5 . The error range of the dihedral angles is ~20°. The carbons in 84R1 a and 84R1 b are in green and cyan, respectively. The white, red, blue and yellow spheres represent hydrogens, oxygens, nitrogens and sulfurs, respectively. The images were generated using Pymol 59  hydrophobic pockets could be created on the surface of BGH (see Supplementary Information Figure S3e for details). Into these, rotamers of 84R1 could be placed in such a way that their N-O moieties are sequestered from the protein surface and the rotation of the nitroxide rings is inhibited (Fig. 3f, left panel). Considering the X 1 and X 2 dihedral angles of these rotamers, they correspond to the {t,m} rotamers 39 . The inter-spin distance (between the nitrogen atoms of the NO groups of 84R1 and 84R1') in the BGH was ~24 Å. This is close to the measured distance, 26.6 (5.2) Å, within the range of the probability peak (Fig. 3e). When the 84R1-84R1' inter-spin vector associated with the BGH was superimposed to the calculated inter-spin vectors from the triangulated points of R1s (Supplementary Information Figure S2d), the C-termini of α 3 and α 5 helices, specifically, residues 96Cs and 143Cs, could be brought close to each other surprisingly (with the C β -C β distance of 6.5 Å and 14 Å, respectively) ( Fig. 3g), consistent with the current cross-linking data (Fig. 2) and with the short inter-spin distances between 96R1s and 143R1s, respectively, as reported earlier 27 .
The BGHs and helix α6 are adsorbed to the membrane surface at shallow depths. To better define how Bak homodimers interact with the membrane, we measured the membrane immersion depths of selected residues in BGH and helix α 6 using SDSL (Fig. 4a, S5 and S6).
Residues Y106 and F117 are located on helix α 4 on the hydrophobic surface of BGH (Fig. 4b). Their corresponding R1 sidechains were located at the depths of ~17 Å and ~12 Å from the membrane surface, respectively (Fig. 4a,b). The immersion depths of α 5 residues on the hydrophobic surface such as 130R1, 138R1, 141R1 and 144R1 were ~7 Å, ~9 Å, ~18 Å, ~11 Å, respectively, indicating that these residues were also located in the hydrocarbon region of the lipid bilayer (Fig. 4a,b). Residue 125R1 located in the first helical turn of α 5 helix also had a depth of ~7 Å, indicating that the N-terminus of α 5 was also in contact with the membrane. These results collectively showed that the hydrophobic surface of BGH is in contact with the membrane.
In the α 5-6 loop (residues 145-148), residue 147R1 was water-exposed while others such as 145R1 and 148R1 were buried at ~15 Å depth from the membrane surface (Fig. 4c), indicating that the loop was partially exposed to water ( Fig. 4d; also see Supplementary Information Figure S6b legend).
Residues on helix α 6 were also interacting with the membrane as summarized in Fig. 4a,c,d. The immersion depths of R1s in helix α 6 (residues 149-163) had an oscillating pattern as a function of residue locations (Fig. 4a,c). The depths could be best-fitted with an α -helix inserted into the membrane with its helical axis tilted toward the N-terminus at an angle of ~30° (Fig. 4a,d, and S6c). The direction of the greatest depth was very close to the radial line for residue 153R1 in the helical wheel diagram for α 6 helix (Fig. 4c, dotted vertical arrow). In this state, residues 149R1, 151R1, 152R1, 153R1 and 157R1 were located in the acyl chain region of the lipid bilayer (white region in Fig. 4d) while residues 155R1 and 159R1 were in the head group region (gray region in Fig. 4d) (Also see Fig. 4a). Residues 150R1, 161R1 and 162R1 had immobile EPR lineshapes (Supplementary Information Figure S5d) with small accessibility parameters for oxygen and (Supplementary Information Figure S6), indicating that these residues are either in protein interior or in tertiary contact (Supplementary Information Figure S6b).
Noting that helix α 6 was a surface-adsorbed helix and that the α 5-6 loop was also partially exposed to water, the pattern in the depths of α 5 residues could be best explained by a surface-adsorbed helix with its axis tilted toward the C-terminus (Fig. 4d,e). These results, along with the depths of residues in α 4 helix (106R1 and 117R1), showed that the BGH and the α 6 helices in the Bak homodimer were adsorbed to the membrane surface, consistent with Westphal et al. 30 .

Discussion
The X-ray crystal structure of the mouse BGHs determined here has the characteristic binding pocket formed by helices α 2-α 5, to which the BH3 domain in the extended α 2-α 3 helix of its symmetry-related partner is bound 25,29 (Fig. 1d-f ). As a result, a raft-like structure of two-layers of α -helices is formed as is also seen in other BGHs formed by human Bak and Bax (Fig. 1e,f). The surface formed by a pair of α 4-α 5 helices is hydrophobic and curved, suited for interaction with the membrane (Figs 1e and 4). The chemical cross-linking data from Bak 86C and Bak 69C/111C in apoptotic mitochondria (Fig. 2) were consistent with the BGH structure determined here (Fig. 1). The EPR spectra of spin-labeled residues attached to various locations of the BGH were very similar whether they were present in the tetrameric GFP-Bak in solution or in oligomeric Bak in membrane (Supplementary Information Figure S4f). Also, the distance between 84R1s within a BGH domain remained essentially the same in the above two states (Supplementary Information Figure S3c). All these strongly suggest that the BGH structure in the oligomeric Bak pore in the membrane is very similar to the X-ray crystal structure of BGH observed in solution state, consistent with our previous report 27 .
In the GFP-Bak tetramer, the two BGH units form a partly open hydrophobic pocket in which the hydrophobic surfaces are sequestered away from the surface and thus not readily available for interaction with the membrane (Fig.1d). Furthermore, between the two BGHs, the C-terminal residues of the two closer α 3 helices are separated at a large distance (~40 Å) unlike what was observed in the membrane (Fig. 2). Thus, the 'α 3/α 5 interface' was implicated neither in the GFP-Bak tetramer nor in the crystal contacts (Supplementary Information Figure S1b).
The immersion depths of the R1s in oligomeric Bak indicated that the BGH and α 6 helices are adsorbed to the membrane surface at shallow depths (Fig. 4), consistent with others 30 . In our BGH structure, the two central α 5 helices in the BGH form an angle of approximately 15 (± 1) degrees relative to a hypothetical horizontal plane that is set parallel to the α 2-α 3 helices (Fig. 4e). Assuming that BGH is immersed flat in the membrane, the helical tilt of α 5 would be approximately 15 (± 1) degrees relative to the membrane surface. The membrane-immersion depths of 130R1, 138R1, 141R1 and 144R1 in α 5 helix appear to be consistent with this assumption (Fig. 4d,e). Note that the immersion depth of a R1 side chain depends not only on the position Scientific RepoRts | 6:30763 | DOI: 10.1038/srep30763 of its α -carbon (C α ) but also on the side chain's direction relative to the membrane normal vector. For this reason, pairs of residues such as 130R1 and 138R1, 106R1 on α 4 and 141R1 on α 5 had similar depths despite the differences in the depths of the C α atoms (Fig. 4b). The chemical cross-linking results clearly demonstrated the proximity of the C-termini of α 3 and α 5 helices between neighboring homodimers in the Bak oligomeric pore formed in the mitochondrial outer membrane (Fig. 2a,f,g), confirming our in vitro study 27 and its biological relevance. Very recently, similar results were also observed in oligomeric Bax 28 , indicating that this 'α 3/α 5 interface' is common both in Bak and Bax oligomeric pores. The DEER results also support the existence of this interface (Fig. 3g).
Recently, Westphal et al. proposed a model of lipidic pore formed by apoptotic Bak oligomers 30 . In this model, Bak BGHs and α 9 helices were assumed to remain on the flat region of the membrane while the helical hairpin, formed by α 6 and α 7-α 8 extended helices, was hypothesized to line the central lumen of the lipidic pore in a transmembrane orientation, reaching well beyond the core of the membrane. However, our molecular modeling indicated that the α 6-α 8 helical hairpin with the extended length of ~30 Å, if it existed, is too short to reach beyond the midpoint of a lipidic pore when it is adsorbed to the surface of a lipidic pore formed in a ~45-50 Å-thick lipid bilayer. Furthermore, if the hypothesized α 6-α 8 helical hairpin existed on the surface of the lipidic pore lumen, parallel arrangement of the hairpins within the pore lumen would make it difficult for α 6 helices to make direct contacts between them, contrary to the cross-linking result with Bak/162C (Fig. 2g) and the short inter-spin distance between 162R1-162R1, ' which is ~5-12 Å 27 .
Based on the nitroxide inter-spin distances in Bax, Bleicken et al. 32 proposed an alternative model of Bax lipidic pore, where the Bax homodimers 'clamp' the toroidal surface of the lipidic pore as mentioned in the Introduction. They assumed that the transmembrane orientation of helix α 9 alternates in the membrane. However, it was suggested that α 9 helices are associated in a parallel transmembrane (TM) orientation in Bax apoptotic pores 28,40 . Iyer et al. also suggested that the 'α 9:α 9 interface' in Bak pore is formed by parallel association of α 9 helices in a transmembrane orientation 31 . Thus, it's difficult, if not impossible, to envision that the TM helix of Bax or Bak will switch its orientation during pore formation. Zhang et al. recently suggested that Bax α 9 helices line the large lipidic pores formed by Bax 28 . In case of Bak, a TM sequence was not essential in pore formation 33 and its direct contribution to the pore structure was not supported experimentally 31 . Now, a more detailed working model of the Bak lipidic pore, built on our previous one 27 , is proposed to resolve the above issues (Fig. 5a). Here, the TM α 9 helices are hypothesized to interact with each other, forming the 'α 9:α 9 interface' within the flat region of the membrane around the lipidic pore as suggested by others 28,31,40 . The extended α 6-α 8 helices, while being adsorbed to the lipidic pore lumen, are hypothesized to be in a pseudo-parallel orientation 27 , tethering the BGHs to α 9 helices. This forms the 'α 6:α 6 interfaces' between the neighboring Bak homodimers, consistent with our and others' results 23,27,41 . In this model, the topological arrangements of α 5-α 6 helical hairpins on the lipidic pore surface can also be explained as schematically shown in Fig. 5b. When a BGH is adsorbed within the curved surface of the lipidic pore as depicted in Fig. 5a, the α 5-α 6 helical segment is inserted into the membrane with their helical axes tilted toward the C-and N-terminus, respectively (Fig. 5b), consistent with the experimental results (Fig. 4d). Note that the two possible arrangements of the α 5-α 6 helical hairpins in Fig. 5b correspond to the two conformations of each monomer in the BGH adsorbed to the lipidic pore (Fig. 5a). Also note that the C-termini of α 3 and α 5 helices can be brought to a close proximity between BGHs of any neighboring Bak homodimers, which are presumably on the curved surface of the lipidic pore, forming the 'α 3/α 5 interface' (Fig. 5a). Currently, the exact location of BGHs within the lipidic pore is not known 18,19,21 . The dimension of a BGH is approximately 40 Å from top to bottom when oriented as depicted in Fig. 5a, which can be placed within the lipidic pore. The topology of Bax pores or arcs recently determined by the atomic force microscopy (AFM) 18 showed a protrusion of about 40 (± 10) Å at the rim relative to the membrane surface. Earlier, Epand et al. reported a protrusion of 24 Å at the pore rim in an AFM image of Bax pore with the diameter of ~20 nm 42 . These suggest that BGHs and/or other domains (e.g., helices α 1 or α 6-α 9) of Bax or Bak might cover both the flat region and the toroidal surface of the lipidic pore. Finally, noting that a simple 2-dimensional triangulation with the measured intra-and inter-dimer distances could explain the proximity of residues in the 'α 3/α 5 interface' (Fig. 3), the diameters of the lipidic pores formed by Bak in liposomes in this study should be very large, which is consistent with other's reports [18][19][20][21]42,43 .
In conclusion, we have determined the mouse Bak BGH structure, which allowed more accurate modeling of the DEER distances observed within and between the BGHs. The curved hydrophobic surface of the BGH was immersed in the membrane at a shallow depth. The BGHs were shown to oligomerize via the 'α 3/α 5 interface' in mitochondria. These findings led us to propose a probable assembly of the Bak homodimers in the mitochondrial apoptotic pore. This sheds important insights into the action mechanism of Bak or Bax in mitochondrial apoptosis pathway. However, the function of the novel 'α 3/α 5 interface' in Bak oligomerization and pore formation is unknown and it requires further investigation. Methods X-ray crystallography. Protein expression and purification. A DNA fragment for α 2-α 5 of mouse BAK (residues 66-144) was subcloned into the pET-28a-GFP A206N vector (designated as pYEGFP_A206N _BAK_ H2-H5_pET28a) 25,44 . The resulting vector expresses an amino (N)-terminally hexahistidine tagged green fluorescent protein (with A206N mutation) fused to mouse Bak α 2-α 5 helices (designated as His-GFP-Bak) in E. coli BL21 (DE3) (see Fig. 1a). The N-terminal sequence of His-tag and the thrombin cleavage sequence is MGSSHHHHHHSSGLVPR/GSH in a single letter amino acid code (thrombin recognition sequence and cleavage site indicated by an underline and '/' , respectively). Cells were grown to the OD 600nm of ~1.2 at 37 °C in Super Broth and were induced to express the fusion protein with 1mM IPTG for 15 hrs at 18 °C. Cells were centrifuged at 4,700 g and the cell pellets were resuspended in TBS (20 mM Tris pH 8.0, 150 mM NaCl) buffer containing 10 mM imidazole. The cells were treated with lysozyme, freeze-thawed and lysed by sonication. The protein was purified with Ni-NTA metal affinity chromatography, followed by N-terminal his tag removal by thrombin cleavage. The tag-less GFP-Bak was purified by another round of Ni-NTA metal affinity chromatography in TBS Scientific RepoRts | 6:30763 | DOI: 10.1038/srep30763 buffer. The protein was concentrated to a final concentration of ~20 mg/ml using Amicon Ultra concentrator with the molecular weight cutoff of 50 kDa (Millipore).
Crystallization of GFP-Bak, data collection and structure determination. Purified GFP-Bak was crystallized by hanging-drop vapor diffusion method adapting Czabotar et al. 25 as described in the Supplementary Information. Diffraction data were collected at the GM/CA-CAT, Advanced Photon Sources in Argonne National Laboratory and the structure was determined by molecular replacement as described in detail in the Supplementary Information (also see Table 1  Preparation of MEF cell lines expressing mutant Bak protein. Retroviral plasmids harboring genes for single, double, or triple cysteine substitution BAK mutant proteins were prepared with the template plasmid pMSCV-mBAK-IRES-GFP 46 using the QuikChange Site-directed mutagenesis kit (Stratagene) according to the manufacturer's instructions. Retroviral preparations were made by cotransfecting the ecotropic 293 cells with pMSCV-mBAK-IRES-GFP vector, pECO and pSG5-BCLx using Lipofectamine 2000 (Invitrogen). After 72 hrs of transfection, the cell culture medium containing the virus was collected and cleared by centrifugation at 3,000 g for 10 min at room temperature. The viral supernatant was then used to infect the MEF cells derived from bax −/− bak −/− double knockout mice 47 . MEF cells stably expressing the Bak proteins were selected by serial passages (minimum 3) in the presence of puromycin (2 μ g/ml) in the above cell culture medium for 7 days on cell culture flasks (Genesee, San Diego, CA).

Isolation of mitochondria.
For each Bak mutant, puromycin-selected cells above were expanded and plated onto 4 culture dishes (15 cm in diameter, Genesee, CA) in the selection medium. The cells were harvested by scraping and the mitochondria were isolated at 4 °C from these cells using a mitochondria isolation kit (Thermo Scientific) according to the manufacturer's instructions. Cells, resuspended in the resuspension buffer in the kit, were disrupted by 10 passages through a 21 G syringe needle. Heavy membrane fractions were removed by two consecutive centrifugations at 700 g for 10 min at 4 °C. Mitochondrial fractions were pelleted by centrifuging the resulting supernatant at 12,000 g for 15 min. The resulting pellets were gently resuspended in a trehalose buffer (300 mM trehalose, 10 mM KCl, 1 mM EGTA, 10 mM HEPES, pH 7.4) to a final protein concentration of ~2 mg/ml. The protein concentration was determined using Pierce ™ BCA Protein Assay Kit (Thermo scientific).
Cytochrome c release assay. Mitochondria (60 μ g in protein quantity) were spun down at 12,000 g for 10 min at 4 °C. They were resuspended in 100 μ l of the cytochrome c release assay buffer (20 mM HEPES/KOH pH 7.5, 100 mM sucrose, 80 mM KCl, 1 mM ATP, 80 μ M ADP, 5 mM Na Succinate, 1 mM DL-dithiothreitol (DTT)) in the presence of 0 or 100 nM p7/p15 Bid, and further incubated for 30 min at 30 °C. A volume of 50 μ l of the reaction mixture was set aside on ice for the cross-linking experiments below. Cytochrome c released into the medium was collected by centrifuging the remaining samples at 12,000 g for 10 min at 4 °C. The resulting pellet was resupended in the assay buffer (50 μ l). A volume of 10 μ l of 6x SDS sample buffer (0.375 M Tris pH 6.8, 12% (w/v) SDS, 60% (v/v) glycerol, 0.6 M DTT, 0.06% (w/v) bromophenol blue) was mixed with 50 μ l of the resulting supernatant and resuspended mitochondrial samples. One sixth of each paired sample was subjected to SDS-PAGE under a reducing condition, followed by immunoblotting. The primary and the secondary antibodies used were mouse monoclonal anti-cytochrome c antibody (Santa Cruz, Cat. # sc-13156)/Anti-rabbit IgG (Perkin Elmer, Cat. # NEF812001EA). The percentage of released cytochrome c was determined by measuring the intensities of the Western blotting images using ImageJ software.
Liposome preparation. Large unilamellar vesicles (LUVs) mimicking the lipid composition of mitochondrial contact sites were made as described (See Supplementary Information). LUVs encapsulating fluorescein isothiocyanate-dextran 10 (FITC-dextran, 10 kDa, Invitrogen) were prepared with the same lipid composition and stored in the presence of 18% (v/v) glycerol as described 33 .
Preparation of oligomeric Bak in membrane. Oligomeric Bak samples were prepared using the above LUVs in the presence of the activator protein p7/p15Bid with a mixture of the spin-labeled sBak-Δ C-His proteins and the unlabeled soluble Bak molecule (sBak/C154S-Δ C-His) at a ratio of 3:4 (for depth measurement) or 7:0 (for DEER experiment) as described 33  EPR spectroscopy. X-band continuous wave (CW) EPR experiments were carried out as follows. CW EPR spectra of the singly spin-labeled sBak-Δ C-His proteins (in 18% (v/v) glycerol) in solution or in membrane-inserted oligomeric BAK samples, were obtained on a Bruker EleXsys 580 spectrometer using a Bruker High Sensitivity resonator or a loop gap resonator (JAGMAR, Krakow, Poland) 50 at 2-mW incident microwave power using a field modulation of 1.0-1.5 Gauss at 100 kHz at room temperature. Power saturation method was used to measure the accessibility parameters of air oxygen and NiEDDA (Nickel(II) ethylenediaminediacetate) (i.e., ∏ (O 2 ) and ∏ (NiEDDA) at 5 mM or 50 mM). The accessibility parameter of a R1 residue to a collision reagent is a quantity that is proportional to the collision frequency between the spin label and the collision reagent (e.g., molecular air oxygen or Ni(II)ethylenediaminediacetate (NiEDDA)), which can be used to map the topological locations of proteins 51 . Samples in a volume of 3 μ ls were placed in a gas-permeable TPX capillary (Molecular Specialties, Inc., Milwaukee, WI) and the power saturation data were obtained by recording the central lines of the EPR spectra of the samples in the window of 15 Gauss over 0.4-100 milliwatts microwave incident power successively in the absence or presence of air oxygen or NiEDDA (5 mM or 50 mM) using the loop gap resonator 50 as described 52,53 . Nitrogen gas (Nitrogen HP 99.995%, Specialty Gases of America, Inc., Toledo, OH) was used to flush the samples. Power saturation data were analyzed to calculate the P 1/2 values using the R program (version 2.12.0) 54 as described 48 . The accessibility parameters, ∏ , were calculated as defined 52,53 ; ∏ (x) = {(P 1/2 (x)− P 1/2°) / Δ H pp }/{P 1/2 (DPPH)/Δ H pp (DPPH)}, where x = O 2 or 5 mM (or 50 mM) NiEDDA; and P 1/2° is the P 1/2 value without any collision reagent under nitrogen gas; P 1/2 (DPPH) is the P 1/2 value of the standard sample of crystalline 2,2-diphenyl-1-picrylhydrazyl (DPPH) in KCl; Δ H pp and Δ H pp (DPPH) are the peak-to-peak line widths of the sample's and the DPPH's EPR spectra, respectively. For depth measurement, the Φ value, which is the natural log of the ratio of ∏ (O 2 ) to ∏ (50 mM NiEDDA) (i.e., log e [∏ (O 2 )/∏ (50 mM NiEDDA)]) was determined for each R1 residue. The Φ value was converted to the membrane immersion depth using a Φ -depth calibration curve as reported 33 . The depth standards used were PC tempo, N-tempoylpalmitamide, 5-doxyl-PC, 7-doxyl PC, and 10-doxyl PC, for which the immersion depths were − 5.0, 0.0, 8.1, 10.5 and 14.0 Å, respectively 55 . NiEDDA was synthesized as described 53 .
DEER experiments were done using the 4-pulse DEER sequence 56 as described 27 . The X-band DEER experiments were carried out with an in-house Bruker EleXsys 580 spectrometer as described 27 . The Q-band DEER spectroscopy was carried out at the National Biomedical EPR Center, Milwaukee on a Q-band Bruker ELEXSYS 580 equipped with an EN5107D2 resonator and a 10 W amplifier at 80 K using a four-pulse sequence. Q-band DEER measurements were done after exchanging the sample buffer with deuterated buffers. First, the deuterated 20 mM Tris, 150 mM NaCl, pH 8 (TBS) buffer was prepared as follows; A quantity of 8.6 milligrams of Tris-HCl (FisherScientific), 5.5 milligrams of Tris base (FisherScientific), and 43.8 milligrams of NaCl (Sigma-Aldrich) were dissolved in a final volume of 5 ml deuterated water (D 2 O)(100%, Sigma-Aldrich). Deuterated buffer A was made by dissolving 23.8 milligrams of HEPES (Sigma-Aldrich), 55.9 milligrams of KCl (Sigma-Aldrich) into a final volume of 5 ml D 2 O and the pH was adjusted to 6.6, which is equal to pD 7.0 57 . Spin labeled His-GFP-Bak proteins prepared in TBS were buffer-exchanged in the above deuterated TBS by repeating two cycles of 10-fold dilution and centrifugal concentration in a concentrator (MWCO of 50 kDa). Oligomeric Bak samples prepared in membrane as described above in buffer A were resuspended in 100 μ ls of the deuterated buffer A. These were centrifuged at ~110,000 × g for 30 min at room temperature. The heavy buffer layer was removed by using a glass capillary. Finally, thus prepared buffer exchanged samples were mixed with deuterated glycerol (Sigma-Aldrich) to a final concentration of 18% (v/v) for cryoprotection, typically in 13 μ l. Samples were contained in fire-sealed quartz capillaries (1.1 mm × 1.6 mm; VitroCom) and flash frozen in a dry ice and acetone mixture and loaded onto the spectrometer for DEER experiments. DEER data were analyzed with DeerAnalysis 37 or DEFit program 58 .