Characterization of an alternative BAK-binding site for BH3 peptides

Many cellular stresses are transduced into apoptotic signals through modification or up-regulation of the BH3-only subfamily of BCL2 proteins. Through direct or indirect mechanisms, these proteins activate BAK and BAX to permeabilize the mitochondrial outer membrane. While the BH3-only proteins BIM, PUMA, and tBID have been confirmed to directly activate BAK through its canonical BH3 binding groove, whether the BH3-only proteins BMF, HRK or BIK can directly activate BAK is less clear. Here we show that BMF and HRK bind and directly activate BAK. Through NMR studies, site-directed mutagenesis, and advanced molecular dynamics simulations, we also find that BAK activation by BMF and possibly HRK involves a previously unrecognized binding groove formed by BAK α4, α6, and α7 helices. Alterations in this groove decrease the ability of BMF and HRK to bind BAK, permeabilize membranes and induce apoptosis, suggesting a potential role for this BH3-binding site in BAK activation. Mitochondrial apoptosis is controlled by BCL2 family proteins, and the BH3-only proteins often act as sensors that transmit apoptotic signals. Here the authors show how the BH3-only proteins BMF and HRK can directly activate the BCL2 protein BAK and interact with BAK through an alternative binding groove.

BH3-only proteins sense various cellular stress signals and transform them into death signals 4,5 . For example, endoplasmic reticulum (ER) stress causes BIM upregulation and subsequent cell death 6 ; BMF transforms signals arising during cellular detachment or actin depolymerization to apoptosis 7 ; and HRK converts growth factor withdrawal signals to cell death in neurons 8 . Another BH3-only protein, BIK, also transduces signals initiated by various anticancer agents, including DNA crosslinking agents 9 , lenalidomide 10 , and anti-BCR (B-cell receptor) therapy 11 , into cell death signals. Accordingly, elucidating the action of these BH3-only proteins is important for understanding how various treatments induce cell death.
BH3-only proteins can induce mitochondrial apoptosis in two different ways: (1) By displacing activated BAK/BAX from antiapoptotic BCL2 family members 12,13 and (2) by directly interacting with and activating BAK/BAX [14][15][16] . Depending on their ability to directly activate BAK/BAX or not, BH3-only proteins are divided into direct activators, which can directly bind BAK and/or BAX to induce their conformational change, oligomerization, and activation 15,17,18 , and sensitizers 15,19 , which inhibit antiapoptotic BCL2 family proteins to release the direct activators or activated BAK/BAX 20,21 .
Experiments examining which BH3-only proteins are direct activators and which are sensitizers have yielded divergent conclusions. BIM and tBID were originally reported to be the only activators 15,19 . PUMA was subsequently suggested to be a direct activator [22][23][24] , albeit with lower potency than BIM 25 . Whether BIK, BMF and HRK are direct activators is less clear. The original experiments identified these three BH3-only proteins as sensitizers 15,19 , but later studies suggest these proteins can also activate BAK or BAX, again with lower potency 26 . Moreover, a recent study using chimeras in which the BID BH3 domain is replaced by BH3 peptides from other BH3-only proteins also indicates that BMF, HRK and BIK can activate BAK/BAX 27 , but this conclusion disagrees with results in a yeast model system that questions whether BIK and BMF act as direct activators 28 .
A number of studies have found that BIM, PUMA, and tBID interact with the canonical BH3-binding groove formed by BAK α3, α4, and α5 17,25,29,30 . However, the interaction surfaces of BMF and HRK on BAK have not been reported. In this study, we evaluate the abilities of BMF, HRK, and BIK to directly interact with BAK and induce BAK-mediated membrane permeabilization. Moreover, using nuclear magnetic resonance (NMR) spectroscopy, molecular dynamics (MD) simulations and site-directed mutagenesis, we map the sites where BMF and HRK interact with BAK. Interestingly, these studies suggest that activation of BAK by BMF and possibly HRK involves binding to a previously unrecognized heterodimerization groove on BAK.

Results
Direct binding of BMF and HRK BH3 peptides to BAK. We have previously shown BIM, NOXA, tBID, and PUMA can directly bind and activate BAK 17,25 . Because all known direct activator BH3-only proteins interact with BAK through their BH3 domains 15,30 , we first asked whether synthetic BMF, HRK, and BIK BH3 peptides (Fig. 1a) can interact with BAKΔTM. BIM and PUMA BH3 peptides served as positive controls, whereas BAD BH3 peptide (the prototypic sensitizer) served as a negative control 25 .
Surface plasmon resonance (SPR) indicated that the BMF BH3 domain binds to BAKΔTM with a dissociation constant (K D ) of <1.0 μM (Figs. 1b, f and Supplementary Fig. 1a), which is smaller than the K D of BIM BH3 with BAKΔTM (Figs. 1d, f and Supplementary 1c). The HRK BH3 peptide binds to BAKΔTM with K D in the low micromolar range (Figs. 1c, f and Supplementary Fig. 1b), which is similar to the affinity of BIM and PUMA BH3 peptides for BAKΔTM (Figs. 1d, f and Supplementary Fig. 1c, d) 26,[29][30][31] . In contrast, BAD BH3 domain showed a much lower affinity (Fig. 1f). Direct comparison of the SPR curves (Fig. 1g) showed that BMF BH3 exhibited the fastest association with BAKΔTM and slowest dissociation, resulting in the smallest K D .
Surprisingly, SPR did not detect interactions between the BIK BH3 peptide and BAKΔTM (Fig. 1e-f). Circular dichroism spectroscopy revealed similar α-helicity of the BH3 peptides ( Supplementary Fig. 2a, b), ruling out the possibility that the differences in K D s reflected differences in ability of these peptides to form secondary structures.
To further evaluate the ability of these peptides to activate BAK, we studied cytochrome c release using mitochondria from Bax −/− mouse embryonic fibroblasts (MEFs) (Supplementary Fig. 3a). Treatment with the BCL2/BCLx L inhibitor navitoclax, the MCL1 inhibitor S63845, or combination of the two failed to induce cytochrome c release, suggesting that Bak is not autoactivated 21 in these mitochondria (Figs. 2c, d and Supplementary Fig. 3b). Consistent with these results, BH3 peptide from the sensitizer protein BAD also did not induce cytochrome c release even at 10 μM (Fig. 2e). In contrast, BIM BH3 peptide, which reportedly has the strongest ability to activate Bak 27 , induced near maximal cytochrome c release at 0.2 μM; and BID BH3 peptide-induced cytochrome c release at 2 μM (Fig. 2e). BMF BH3 peptide-induced cytochrome c release at concentrations similar to the BID BH3, whereas the HRK and BIK peptides only induced cytochrome c release at 10 μM, the highest concentration tested.
Because cytochrome c release from Bax −/− mitochondria did not exactly match the liposome permeabilization results shown in Figs. 2a, b, we hypothesized that BH3 peptide-induced cytochrome c release from mitochondria of Bax −/− MEFs might reflect a combination of direct Bak activation and antiapoptotic paralog neutralization. To test this idea, S63845 or navitoclax was applied to the mitochondria along with BIK BH3 or HRK BH3. In the presence of S63845, HRK BH3-induced cytochrome c release at a concentration 10 times lower than BIK BH3 (Fig. 2f), in agreement with the binding and the liposome permeabilization results. In contrast, navitoclax did not enhance either BIK or HRK BH3induced cytochrome c release ( Supplementary Fig. 3c), suggesting that Bak was predominantly bound to Mcl1 after activation.
We also examined cytochrome c release by incubating mitochondria from Bak −/− Bax −/− MEFs with BH3 peptides plus purified human BAKΔTM. Under these conditions, where more recombinant BAK was present, BIM BH3 also induced the most cytochrome c release, BMF-and HRK-induced release of an intermediate amount, and BIK BH3 barely induced cytochrome c release (Fig. 2g). Collectively, these results indicate that under cell-free conditions BIM BH3 peptide is the strongest BAK activator tested, BMF BH3 and HRK BH3 are weaker, and BIK BH3 is barely able to activate BAK.
BMF and HRK induce direct BAK activation in cells. To assess Bak activation by full-length BMF and HRK proteins, we measured induction of apoptosis by BMF and HRK. In initial experiments, wild type (WT) MEFs were transfected with EGFPtagged BH3-only proteins ( Supplementary Fig. 4a), incubated for 24 h, and stained with Annexin V to assess phosphatidylserine externalization in EGFP + (successfully transfected) cells. Typically, >60% cells were EGFP + 24 h after transfection. In the WT MEFs, EGFP-tagged BIM, PUMA, BMF, and HRK all induced apoptosis in >75% of EGFP + cells, suggesting potent proapoptotic action of these BH3-only proteins in situ. In contrast, BAD, which is not an activator of either BAK or BAX, only induced apoptosis in~25% of WT MEFs. Because of poor BIK expression in the MEFs ( Supplementary Fig. 4a), results with this protein were inconclusive. Upon transfection into Bak −/− Bax −/− MEFs, none of these proteins induced apoptosis ( Supplementary  Fig. 4b), confirming that cell death induced by these BH3-only proteins depends on Bak or Bax.
In Jurkat cells, which express~8 times more BAK than BAX 33   b-e SPR (relative units (RU)) observed when immobilized BAKΔTM was exposed to a series of concentrations of BMF BH3 b, HRK BH3 c, BIM BH3 d, or BIK BH3 e. f Approximate dissociation constant ranges of BAKΔTM interacting with the indicated BH3 peptides estimated from three independent SPR assays using steady-state analysis ( Supplementary Fig. 1). Because artefacts were observed when high concentration of certain peptides were applied, the determination of BAK:BH3 peptide affinities did not meet the criterion required for steady-state analysis that top concentrations should be at least threefold above the reported K D . g Direct comparison of BAKΔTM interacting with the indicated BH3 peptides at 1600 nM. K D dissociation constant. Source data are provided as a source data file.  Fig. 3a, b). Moreover, EGFP-BIK also induced apoptosis in 85% of EGFP + cells, perhaps reflecting the fact that BAK is autoactivated in Jurkat cells 21 and can be displaced from BCLx L and MCL1 by BIK. This apoptosis was accompanied by BAK activation, as detected by an antibody that recognizes an active BAK conformation ( Fig. 3c) 17,25 and abolished by BAK knockout (Fig. 3d), consistent with the important role of BAK in apoptosis induction in these cells.
To further explore BMF-and HRK-induced BAK activation without the confounding effects of known endogenous BH3 activators and BAK autoactivation, we generated BIM −/− PUMA −/− BID −/− NOXA −/− BAK −/− BAX −/− Jurkat cells using CRISPR-Cas9 technology (Fig. 3e) and stably transfected these HexaKO cells with a plasmid containing BAK behind a doxycycline inducible promoter (Fig. 3f). When BAK was induced by 10 ng/mL doxycycline for 16 h, the BAK concentration was much lower than endogenous BAK levels in WT Jurkat cells (Fig. 3f), and no autoactivated BAK was detectable in BCL2, BCLX L , and MCL1 immunoprecipitates (Fig. 3g) 21 . If BMF or HRK were simply inducing apoptosis by displacing activated BAK from antiapoptotic BCL2 family members, there should be no BMF-or HRK-induced apoptosis in this cell line (because there is no pre-activated BAK to displace). However, both BMF and HRK-induced apoptosis in 45-55% of successfully transfected cells in both clones (Fig. 3h, i), suggesting that BMF and HRK can directly activate BAK in this cell line. In contrast, BIK could only induce apoptosis in~20% of successfully transfected HexaKO cells, again suggesting a much weaker BAK activating ability.
Collectively, results in Fig. 3 indicate that BMF and HRK are sufficient to induce direct BAK activation and BAK-mediated apoptosis in different cell lines.
A possible second binding site for BH3 peptides on BAK. Earlier studies identified the canonical BH3-binding groove of BAK as the site for interaction with the BH3 domains of BIM, PUMA, and tBID 17,25,29,30 . To address whether BMF and HRK bind to the same site, an NMR-based chemical shift perturbation assay was used. In brief, 15 N-labeled BAK 15-186 was purified as described in the Methods. After the assignment of the peaks on 1 H, 15 Table 1). Because of protein precipitation during incubation with BIM BH3 peptide, we could not obtain high-quality 1 H, 15 N-HSQC perturbation data. Accordingly, PUMA BH3 was used as a positive control. As shown in Fig. 4a and Supplementary Fig. 6b, PUMA BH3 induced two types of chemical shift changes on BAK 1 H, 15 N-HSQC spectrum. One group of residues, including Y89, D90, S91, F93, T95, L97, H99, K113, I114, L118, G126, A130, L132, G133, and F134, exhibited ligand-induced decreases in peak intensity or signal disappearance. Another group of residues, including Q94, Q98, F157, L163, H165, and C166, exhibited focal, doseresponsive chemical shift changes. When mapped onto the structure of BAK (PDB ID: 2IMS 34 , Fig. 5a and Supplementary   ), these residues were mostly located in the canonical BH3binding groove, in agreement with previous studies 25,29,30 . Interestingly, at the same concentration of BMF or HRK BH3 domain, changes for a group of BAK residues (Y89, D90, S91, F93, T95, and L97 on helix α3 as well as G126, A130, L132, G133, and F134 on helix α5) were absent or much lower than with PUMA. Instead, addition of either the BMF or HRK BH3 peptide-induced prominent chemical shift changes that involved residues M96, Q98, and H99 on BAK helix α3; K113, I114, L118 on α4; F157, F161, M162, L163, H165, and C166 on α6; and A168 on helix α7 ( Fig. 4b and Supplementary Fig. 6a, c, d). Most of the residues were situated within an α4/α6/α7 groove on BAK (Fig. 5b, c), raising the possibility that the BH3 domains of BMF and HRK might bind to the α4/α6/α7 groove of BAK, although some of the residues in the canonical groove (Q94, L97, Y108, L118, G126, L131, and F134) also displayed limited but detectable reduction in peak intensity after addition of BMK or HRK peptides (see Supplementary Fig. 6c-d and Discussion).
To further investigate whether the BMF and HRK peptides preferentially bind the α4/α6/α7 groove or canonical BH3binding groove, we conducted comprehensive, microsecond, isobaric-isothermal MD simulations of BAK with the BIM, BMF, and HRK BH3 peptides docked at either groove. These simulations used FF12MC, an AMBER protein forcefield with improved effectiveness in simulating localized disorders of folded proteins that has accurately simulated experimental folding times for fast-folding proteins 35,36 . Using the model protein CLN025 37 as a control to validate our conformational stability analysis method, we showed that the most populated conformers at 300 K and 277 K are nearly identical to the experimentally determined    CLN025 conformer as indicated by the alpha carbon root mean square deviation of 1.7 Å between the simulated and experimental conformers, and that the CLN025 population increases as the temperature decreases (65% at 300 K and 71% at 277 K as listed in Supplementary Table 2), indicating the correlation of the population with the relative stability of CLN025. Using this method, we obtained the relative populations (Supplementary  Table 2) of the most populated BAK complexes with the three BH3 peptides at either groove ( Supplementary Fig. 7). These populations (given in parentheses hereafter) suggest that: (1) BIM BH3 binds with a strong preference to the canonical α3/α4/α5 groove (75%) over the noncanonical α4/α6/α7 groove (35%), in agreement with earlier studies 17,29,30 ; (2) BMF BH3 binds both grooves with a possible preference for the noncanonical groove (65% for the canonical and 69% for the noncanonical); and (3) HRK BH3 also binds both grooves (74% for the canonical and 53% for the noncanonical). Despite its higher population in the canonical groove, HRK is thought to also bind the noncanonical groove based on its high population at the noncanonical groove (53%) relative to BIM at the same site (35%). This involvement of both grooves for BMF and possibly HRK is in agreement with the chemical shift perturbation data ( Supplementary Fig. 6c, d).
In addition, the simulations revealed that BAK Y108 and H165 moved out of the α4/α6/α7 groove, exposing a deep hydrophobic cavity to accommodate the conserved L137 of BMF or L37 of HRK, whereas F161 moved up slightly and remained in the groove to (Fig. 5d, e). The simulations also showed that BAK F161 interacted with BMF L137 or HRK L37, whereas BAK H164 and H165 formed salt bridges with the conserved D142 of BMF or D42 of HRK (Fig. 5f, g). A hydrophobic surface analysis indicated that the α4/α6/α7 groove induced by the two peptides was primarily hydrophobic (Fig. 5h). These computational studies are consistent with the NMR results showing prominent chemical shift perturbations along the α4/α6/α7 groove upon binding of BAK to BMF or HRK BH3.
Effects of BAK and BH3 mutations. To further assess whether binding of BMF and possibly HRK BH3s involves the α4/α6/α7 groove, SPR was used to assay interactions of BIM, BMF and HRK BH3 peptides with WT or mutant BAKΔTM. Two mutations that physically block the canonical BH3-binding groove (BAK G126S and BAK F93E) diminished binding of BIM BH3 domain (Fig. 6a left and b), in agreement with previous results 17,25,38 , but did not appreciably affect binding of BMF and HRK BH3s (Fig. 6a middle and right, Fig. 6b). Conversely, the helix α6 mutations BAK F161A and H164A/H165A diminished binding of the BMF and HRK BH3 domains (Fig. 6c, d) but did not significantly change binding of BIM peptide. These SPR results show that physical blockage of the canonical binding groove affects BIM BH3 binding much more than BMF or HRK binding, consistent with the NMR and MD simulation results suggesting that BMF and HRK can bind at two sites.
We also examined the effect of the BAK F161A mutation using MD simulations. Because α4 separates the α4/α6/α7 and α3/α4/α5 grooves (Fig. 5a-c), the F161A mutation simultaneously contracted the BAK noncanonical groove and expanded the canonical BH3-binding pocket. These changes modestly reduced the population of the BIM BH3 peptide at the canonical groove from 75 to 50% (Supplementary Table 2). However, the populations for BMF binding at the noncanonical and canonical grooves were markedly reduced to 32% and 22%, respectively, in the F161A mutant from 69 and 65% for WT BAK (Supplementary Table 2). Likewise, those of HRK at the noncanonical and canonical grooves of the F161A mutant were also markedly decreased to 22% and 32% from 53 and 74% for WT BAK, respectively (Supplementary Table 2). These simulations suggest that the F161A mutation inhibits binding of BMF and HRK BH3 peptides at both grooves while sparing the binding of BIM at the canonical groove.
To further study these interactions, we also mutated three conserved hydrophobic residues in the BH3 domain of BMF or HRK to glutamate (Fig. 6e). These mutations diminished binding of the BH3 peptides to BAKΔTM (Fig. 6f-g), consistent with the MD simulations, suggesting that L137 is involved in the binding at the noncanonical groove and all three hydrophobic residues are involved in binding at the canonical groove 17,25 . Taken altogether, the results shown in Figs. 4-6 suggest that the BAK binding to the BH3 domains of BMF and possibly HRK involves, at least in part, the α4/α6/α7 groove.
BAK activation is also impacted by α6 helix mutations. We also examined the impact of the F161A mutation on BAKmediated liposome permeabilization. As shown in Fig. 7a, this mutation markedly decreased BAK-mediated liposome permeabilization initiated by BMF or HRK BH3 domains, but did not affect liposome permeabilization initiated by BIM or PUMA.
In further studies, we reconstituted Bak −/− Bax −/− DKO MEFs with WT BAK or BAK mutants and examined the cytotoxicity of transiently expressed BH3-only proteins fused to EGFP (Fig. 7b-d). Apoptosis induction by EGFP-BIM ( Supplementary  Fig. 8a) or EGFP-PUMA (Fig. 7c) was not inhibited in cells expressing BAK F161A but was inhibited by the BAK G126S mutation even in the presence of the reciprocal N86G mutation (Fig. 7d) that allows BAK oligomerization 17,38 . In contrast, cytotoxicity of EGFP-BMF or EGFP-HRK was markedly diminished by the BAK F161A mutation ( Supplementary Fig. 8a and Fig. 7c), reflecting inability of EGFP-BMF or HRK to induce activation of BAK F161A (Fig. 7e), but was relatively unaffected by the BAK G126S/N86G mutation ( Fig. 7d and Supplementary  Fig. 8b).
To rule out the possibility that decreased ability of BMF and HRK to induce BAK F161A activation results from altered binding of BAK F161A to antiapoptotic BCL2 proteins, we compared the binding of BAK WT and F161A to antiapoptotic BCL2 proteins using pull-down assays. As shown in Fig. 7f, the F161A mutation did not change binding of BAK to BCLX L , BCL2, or MCL1.  Fig. 6 BMF and HRK bind the BAK α4/α6/α7 groove. a SPR (relative units (RU)) as a function of time observed when similar levels (about 4000 RU) of WT BAKΔTM or BAKΔTM with mutations in the canonical BH3-binding groove were immobilized and exposed to 1600 nM BIM BH3 (left), BMF BH3 (middle), or HRK BH3 (right). b Maximum RU values obtained from binding isotherms using a series of concentrations of the corresponding peptides in a, and p = 0.007, 0.008 for BIM BH3; 1.0, 0.48 for BMF BH3; and 0.06, 0.26 for HRK BH3 when maximum RU values of binding to BAK F93E or G126S, respectively, were compared with WT BAK. c SPR as a function of time observed when similar levels (~4000 RU) of WT BAKΔTM or the indicated BAKΔTM α4/α6/α7 groove mutant were immobilized and exposed to 1600 nM BIM BH3 (left), BMF BH3 (middle), or HRK BH3 (right). d Maximum RU values obtained after exposure of the indicated BAK species to a series of concentrations of the corresponding peptides in c. p = 0.52, 0.32 for BIM BH3; 0.004, 0.001 for BMF BH3; 0.0003, 0.001 for HRK BH3 when maximum RU values of binding to BAK F161A, or 164/165 A, respectively, were compared to WT BAK. e Sequences of 3E BMF and HRK BH3 peptides bearing alterations in the indicated hydrophobic residues. f SPR observed when immobilized BAKΔTM was exposed to 1600 nM BMF BH3 or its mutant (left) and HRK BH3 or its mutant (right). g Maximum RU values obtained after exposure of BAKΔTM to a series of concentrations of the peptides in f. p = 0.001 and 0.001 for BMF and HRK BH3, respectively, when maximum RU values observed with the 3E mutants were compared with corresponding WT BH3 peptides. Error bars in b, d, g: mean ± S.D. of three independent experiments. ns, p > 0.05; **, p < 0.01; ***p < 0.001; two-tailed paired t test for maximum RU values of mutants vs WT. Source data are provided as a source data file.
To assess BMF-initiated BAK activation under endogenous conditions, we examined detachment-induced apoptosis, a process that is diminished in BMF-depleted cells ( Supplementary  Fig. 8c) 7 . When this BMF-dependent process was subsequently assessed in Bak −/− Bax −/− DKO MEFs reconstituted with BAK WT or BAK F161A, significantly less anoikis was induced in cells reconstituted with BAK F161A (Fig. 7g), strengthening the view that endogenous BMF might, at least in part, bind the alternative site to trigger BAK-mediated apoptosis.
Collectively, the results in Fig. 7 suggest that canonical BH3binding groove can be irreversibly occluded by mutations without affecting certain BH3-only proteins such as BMF, but simultaneous disruption of binding to the canonical BH3-binding pocket and α4/α6/α7 groove impairs BAK activation and cell death induced by these proteins.

Discussion
In the present study, we characterized the interaction of the BH3only proteins BMF, HRK, and BIK with the mitochondrial permeabilizer BAK. Our experiments not only demonstrated direct interactions between BMF or HRK BH3 and BAK, but also provided evidence that a second site that might play a role in BH3-binding. This second site has potential implications for future efforts to modulate BAK activity.
Our SPR studies indicated that the BAK displays a higher affinity for BMF BH3 than for BIM and PUMA BH3s, two wellestablished BAK activators. Moreover, BAK also bound the HRK and PUMA BH3s with comparable affinities (Fig. 1). Although BAK showed lower affinity for the BIM and PUMA BH3s than in our previous studies 17,25 , the differences might reflect immobilization of BAK rather than BH3 peptides and a different analysis method (Supplementary Fig. 1). Our further experiments showed that the BMF and HRK BH3 domains can induce BAK-mediated liposome permeabilization (Fig. 2a-b) and mitochondrial cytochrome c release (Fig. 2e, g), in agreement with previous studies concluding that BMF and HRK can directly activate BAK 26,27 . Although BIK has also been reported to be a direct BAK activator, in our study interaction of the BIK BH3 domain with BAK was   .0002 for PUMA, BMF, and HRK, respectively, when the percentages of Annexin V + cells reconstituted with BAK F161A were compared with those reconstituted with WT BAK c, and p = 0.10, 0.0009, 0.001 for BIM, BMF, and HRK, respectively, when the percentages of Annexin V + cells of reconstituted with BAK G126S/N86G were compared with those reconstituted with WT BAK d. In addition, cell lysates were prepared for immunoprecipitation with BAK Ab-1 antibody and blotting for BAK e. f After S-tagged WT or F161A BAK was transfected into MEFs for 24 h, cell lysates were subjected to pull-down with S-protein agarose and blotting with the indicated antibodies. g Anoikis was induced for 48 h in Bak −/− Bax −/− DKO MEFs reconstituted with WT BAK or BAK F161A. The percentage of cells with sub-G1 DNA was assayed (p = 0.0006). Right panels in c and d, whole-cell lysates subjected to immunoblotting. Error bars: mean ± S.D. of three independent experiments. ns, p > 0.05; **p < 0.01; ***p < 0.001, paired two-tailed t test for comparison of dextran release or cell death induced by mutants vs WT. Source data are provided as a source data file. undetectable using SPR (Fig. 1e). Nonetheless, BIK BH3 peptide at a high concentration (10 µM) induced cytochrome c release from mitochondria of Bax −/− MEFs (Fig. 2e) even though the BAD BH3 peptide, the BCL2/BCLx L inhibitor navitoclax and/or the MCL1 inhibitor S63845 did not (Fig. 2c-e and Supplementary  Fig. 3b), suggesting low but detectable ability of BIK BH3 to activate BAK.
In further experiments, the ability of EGFP-BMF, EGFP-HRK, and EGFP-BIK to induce apoptosis in transfected cells was examined. Previous studies have suggested BAK could be activated through BIM-, PUMA-, and BID-induced direct activation or concentration-dependent autoactivation 15,17,21 . To exclude the influence of these possible mechanisms, we generated HexaKO Jurkat cells, which lack BAK, BAX, BIM, PUMA, BID, and NOXA, introduced an inducible BAK expression construct, and then studied BMF-and HRK-induced apoptosis. Under conditions where BAK is not autoactivated (Fig. 3g) and also in the absence of the well-defined activators BIM, PUMA, BID, and NOXA (Fig. 3e), full-length BMF or HRK can induce BAKmediated apoptosis (Fig. 3h-i). Moreover, EGFP-tagged fulllength BIK also induced more apoptosis than the sensitizer BH3only protein BAD (Fig. 3h-i), again suggesting that BIK is a weak BAK activator. We note, however, that expression of EGFP-BIK is lower than other BH3-only proteins fused to EGFP, perhaps because of a short BIK half-life in MEFs and Jurkat cells.
Activation of BAK through binding to the canonical BH3binding groove has been previously demonstrated by NMR 30 , xray crystallography 29 , and site-directed mutagenesis assays 17,25 . In these studies, BH3-only proteins have been shown to interact with the canonical BH3-binding groove of BAK to initiate a BAK conformational change that triggers BAK oligomerization and activation 29,30,38 . Interestingly, the present results suggest that the BMF and HRK BH3 peptides interact, at least in part, with an alternative binding site on BAK. Evidence for this alternative binding site comes from NMR spectroscopy, MD simulations, and site-directed mutagenesis.
To further evaluate this possibility, we performed extensive MD simulations of BH3 binding to either the canonical groove or the α4/α6/α7 groove. For each BH3 peptide binding to each site, we obtained 20,000 conformations at 100-ps intervals of the MD simulations and calculated the population of the most populated conformer of each BAK with BH3 binding at either groove. BIM, a BH3 peptide that is well established to bind the canonical groove 17,18,29,41 had a dominant population of 75% at the canonical groove but only 35% at the α4/α6/α7 groove, suggesting the lower affinity of BIM for the alternative groove. In contrast, BMF and HRK BH3 peptides bound to both grooves with populations of 53-74% for both grooves (Supplementary Table 2). These results suggest that (1) BIM binds almost exclusively to the canonical groove, (2) BMF binds to both grooves with a possible slight preference for the noncanonical groove, and (3) HRK binds to both grooves with a preference for the canonical groove.
Further studies examined two types of mutations on binding: those that interfere with binding at only the canonical BH3binding pocket (G126S, F93E) and those that interfere with binding at both sites by expanding the canonical site and contracting the possible alternative site (F161A and H164A/H165A). Mutations that occlude the canonical BH3-binding pocket 38 abolished the binding of BIM BH3 but not BMF or HRK BH3 to BAK (Fig. 6a-d). In marked contrast, mutations that simultaneously expand the canonical site and narrow the alternative site diminished binding of BMF and HRK BH3s but marginally affected the binding of BIM BH3. These results appear to indicate BMF and HRK BH3s are able to bind to a site on BAK even when the canonical binding groove is occluded 38 . Binding of BMF and HRK to both sites in the wild-type protein and the alternative binding site in the G126S or F93E mutant would certainly be compatible with these results. Consistent with these results, Bak −/− Bax −/− MEFs expressing BAK G126S/N86G were less sensitive to BIM-induced apoptosis than cells expressing WT BAK, but the BAK G126S/N86G mutation had much less effect on BMF-or HRK-induced apoptosis (Fig. 7d). In contrast, the F161A mutation, which affects both grooves, diminished BMFand HRK-induced apoptosis and the BMF-dependent process of anoikis ( Fig. 7g and Supplementary Fig. 8b).
The most straightforward explanation for the NMR, MD simulations, and mutational data is that certain BH3 peptides such as BMF and, to a lesser extent, HRK are able to bind and activate BAK through a previously unrecognized peptide-binding groove comprised of BAK α4, α6, and α7 helices, especially when the canonical BH3-binding groove is irreversibly occluded by mutations. Nonetheless, because the interaction of BMF or HRK BH3 with the α4/α6/α7 groove has not been biochemically confirmed in intact cells, e.g., by cross-linking studies in situ 16 , this model remains one of several potential explanations for the results. In particular, the NMR data do not exclude the possibility that the BH3 peptides can bind at the canonical groove and allosterically perturb residues located in the noncanonical groove. However, if BMF or HRK BH3 peptide were binding exclusively to the canonical groove, it is difficult to explain why so many chemical shift perturbations and disappearances seen with PUMA peptide binding at that groove are markedly diminished or absent. Although one potential solution to this conundrum could conceivably be an alternative binding mode in which the BHK and HRK peptides bind in a different orientation across the canonical BH3-binding groove, any model postulating this different binding mode at the canonical groove must take into account (i) the lack of abundance of any conformer in this alternative orientation in the MD simulations and (ii) the lack of impact of the canonical groove-perturbing G126S and F93E mutations on BMF and HRK binding and action. Nonetheless, further studies such as in situ cross-linking 16 will be required to definitively distinguish between alternative explanations for the results.
The identification of the BAK α4/α6/α7 groove as an alternative BH3-binding site has several possible implications. In particular, this site might serve as a potential target for the development of direct BAK activators to treat disorders where apoptosis is impaired. Conversely, the alternative binding site might serve as a potential receptor for one end of Protac molecules designed for diseases with excessive BAK-induced apoptosis. Either possibility would further validate the alternative binding site and simultaneously provide new insights into BAK function.
Protein expression and purification. Plasmids encoding BAKΔTM (GenBank BC004431, residues 1-186) in pET29b(+) was kindly provided by Qian Liu and Kalle Gehring (McGill University, Montreal, Canada) 34 . Plasmids encoding BAK 15-186 and the corresponding BAK mutants were generated using site-directed mutagenesis. All plasmids were subjected to automated sequencing to verify the described alterations and confirm that no additional mutations were present.
Plasmids were transformed into E. coli BL21. After these, cells were grown to an optical density of 0.8, isopropyl βd-1-thiogalactopyranoside was added to a final concentration of 1 mM and incubation was continued for 24 h with shaking at 16°C. Bacteria were then washed and sonicated intermittently on ice in TS buffer (10 mM Tris-HCl, 150 mM NaCl, pH 7.4, and 1 mM phenylmethylsulfonyl fluoride). His 6 -tagged proteins were then applied to Ni 2+ -NTA-agarose columns, which were washed with 20 volumes of TS buffer and 10 volumes TS buffer containing 40 mM imidazole, followed by elution with TS buffer containing 200 mM imidazole.
BAKΔTM or mutants were immobilized on a CM5 chip using a Biacore T200 biosensor with a blank channel as a negative control. Binding assays were performed at 25°C using Biacore buffer containing the indicated BH3 peptides injected at 30 µl per min for 1 min using a Biacore T200 Control Software. Proteins were allowed to dissociate during perfusion with Biacore buffer at 30 µl per min for 10 min and then desorbed with 2 M MgCl 2 . All binding affinities were derived by Biacore T200 Evaluation Software using Steady State analysis (Biacore, Uppsala, Sweden).
Liposome permeabilization assay. Release of F-d10 from large unilamellar vesicles (LUVs) was monitored by fluorescence dequenching 25 using a fluorimetric plate reader (Thermo Scientific) with SoftMax Pro software. After addition of purified His 6 -BAKΔTM and/or the BH3 peptides to LUVs (final lipid concentration 10 µg ml −1 ), 96 well plates were incubated at 37°C and recorded (excitation 485 nm, emission 538 nm) every 10 sec. FITC-labeled dextran release was calculated using the equation ((F sample − F blank )/(F Triton − F blank ) × 100%), where F sample , F blank , and F Triton are the fluorescence recorded for the reagent-, buffer-, and 1% Triton-treated LUVs, respectively.
Immunoblotting. Cells were solubilized in SDS sample buffer containing 4 M urea, 2% (w/v) SDS, 62.5 mM Tris-HCl (pH 6.8), 1 mM EDTA and 5% (v/v) 2-mercaptoethanol, and sonicated 40 times. Aliquots containing 50 μg of protein were then separated on SDS-PAGE, transferred to nitrocellulose membrane, and probed by antibodies overnight at 4°C with indicated dilutions, followed by incubation with horseradish peroxidase-labeled secondary antibody at room temperature for 1 h. Membranes were developed using ECL.
NMR sample preparation and assignments. For NMR samples, bacteria were grown in M9 medium. Uniformly labeled BAK 15-186 was produced using M9 medium containing 0.5 g L −1 99% 15 N-ammonium sulfate and 2.5 g L −1 99% 13 C-glucose as the sole nitrogen and carbon sources, respectively. After Ni 2+ -NTA column purification, the labeled protein was further purified using Superdex-75 size exclusion chromatography (GE) and dialyzed against 20 mM NaH 2 PO 4 (pH 6.7) containing 100 mM NaCl, 1 mM DTT, and 2 mM EDTA.
All NMR spectra were recorded at 303 K on a Bruker DMX850 spectrometer with Topspin software. To assign backbone residues of BAK 15-186, a set of standard 2D and 3D spectra was recorded, including 1 H-15 N-HSQC (heteronuclear single-quantum correlation), HNCACB, CACB(CO)NH, HNCA, HN(CA)CO. All NMR data were processed with NMRPipe 44 and analyzed with Sparky 3 software 45 .
NMR chemical shift perturbation assay. In order to define the residues in BAK responsible for the binding of PUMA, BMF, and HRK BH3 peptides, 0.5 mM 15 Nlabeled BAK 15-186 was titrated with unlabeled PUMA, BMF, and HRK in different molar ratios. A series of HSQC spectra was then acquired at 303 K on the Bruker DMX850 spectrometer. A plot of chemical shift perturbations (CSPs) across all residues was generated for each BH3 peptide. The threshold value was defined as average CSPs value plus one standard deviation. CSPs for each residues were calculated using the formula CSPs = [(ΔδNH 2 + ΔδN 2 /25)/2] 1/2 . The average CSPs values of PUMA, BMF, and HRK peptide were 0.01, 0.02, and 0.01, respectively.
Log phase Jurkat cells or MEFs growing in antibiotic free medium were transiently transfected with the indicated plasmid using a BTX 830 square wave electroporator delivering a single pulse (10 msec) at 240 mV or 280 mV for Jurkat or MEFs, respectively. Cells were incubated for 24 h, stained with APC-coupled annexin V, subjected to flow microfluorimetry, and analyzed using CytExpert software 47 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data supporting the findings of this study are available in the manuscript and supplementary files or are available from the corresponding authors upon request. Source data are provided with this paper.