Introduction

Proteins of the BCL-2 family are the major regulators of ‘intrinsic’ apoptosis, the programmed cell death induced by nearly all intracellular stress signals [1, 2]. The family comprises three interacting subgroups: the pro-apoptotic BCL-2 homology 3 (BH3)-only proteins (e.g. BIM and BID), which signal for apoptosis; the critical effectors BAX and BAK, which permeabilize the mitochondrial outer membrane (MOM); and the pro-survival members (e.g. BCL-2 and MCL-1), which act by constraining members of both pro-apoptotic factions. Once activated, BAX and BAK can form homo-oligomers that perforate the MOM and thereby release cytochrome c and other proteins that trigger cellular demolition by the caspase proteases.

BAX comprises a bundle of nine alpha helices (α1-α9) [3]. Residues in helices α2-α5 form a hydrophobic groove on the surface, which represents the canonical interaction site with BH3 domains of other pro-apoptotic BCL-2 family members [4]. In contrast to its close relative BAK, which is mainly located on the MOM, inactive BAX is primarily cytosolic and, in a proportion of the BAX molecules, the α9 helix containing its transmembrane (TM) domain is bound within its canonical groove [2, 3, 5]. In the steady state, both BAX and BAK may still continually shuttle between the cytosol and the MOM, albeit at very different rates [6,7,8].

Complex interactions between BCL-2 family members tightly control the activation of BAX and BAK [9]. The BH3-only proteins can provoke apoptosis by targeting and neutralising the pro-survival proteins, thereby releasing activated BAX and BAK. In addition, certain BH3-only proteins (e.g. BID and BIM) can also bind transiently to the effectors and directly activate them [2, 10]. For BAX, two BH3 binding sites with distinct roles seem to co-operate in controlling its conversion from cytosolic inert monomers into the pore-forming homo-oligomers. BH3 engagement of an incompletely characterised α1/α6 ‘trigger site’ may prepare BAX for full activation by aiding α9 release and thereby increase MOM translocation and integration, facilitating subsequent BH3 binding to the canonical groove for full BAX activation [11,12,13,14]. BH3 engagement of the groove then induces unfolding of BAX into a ‘core’ (helices α2-a5) and ‘latch’ (α6-a8), leading to the core “symmetric dimers” [15] that nucleate the homo-oligomers [4, 16]. An alternative BH3 binding site involving helix α6 may also aid full activation of BAK [17]. Other mechanisms of BAX/BAK activation, such as lipid-mediated spontaneous activation [18, 19] and autoactivation [20], have also recently been reported. An emerging issue for BAX and BAK oligomerization is how these proteins interact with the lipid bilayer. Notably, for BAK oligomers, lipids were recently shown to help bridge adjacent dimers [21]. However, the lipids observed connected grooves within adjacent dimers formed by their α4 and α5 helices. Flexible hydrophilic loop regions such as α1–α2 have not been implicated.

Several studies have implicated the α1−α2 loop in BAX activation. The BAX NMR structure revealed that the central region of the α1−α2 loop was more rigid than the rest of the loop and appeared to associate with a portion of α6 [3]. A role for the loop as a ‘molecular switch’ in the early steps of BAX activation was first proposed by Gavathiotis et al. [12, 14]. Cross-linking studies showed that the loop’s central region could juxtapose with helix α6 [12] and that linking the loop to α6 blocked BAX activation [11, 12], indicating that efficient BAX activation requires flexibility of the loop. Intriguingly, antibodies with epitopes in the loop can prevent activation of cytosolic BAX but trigger activation of mitochondrial BAX [22,23,24]. Furthermore, we and others have shown that the conformation of the BAX α1−α2 loop affects BAX activity and that allosteric crosstalk between the loop and helix α9 can influence exposure of the α9 TM domain and thereby inhibit BAX MOM association and integration [11, 22].

To date, however, no BAX mutations supporting a role for the α1−α2 loop have been reported and the specific loop regions/residues important for BAX function remain elusive. Therefore, we have sought to identify BAX α1−α2 loop mutants and to explore the underlying mechanisms. We have used three different mutagenic approaches: replacements with loop segments from other BCL-2 family members, systematic alanine scanning and analysis of loop missense mutations reported in tumours. Our findings suggest that the N-terminal and C-terminal halves of the BAX α1−α2 loop have distinct roles in BAX activation, and that the first nine residues of the loop are important for efficient BAX activation.

Materials and methods

Recombinant proteins

Full-length BAX variants were generated by PCR site-directed mutagenesis and produced as described [3, 11]. Briefly, BAX constructs cloned into pTYB were expressed in E. coli ER2566 and proteins purified by chitin affinity chromatography and gel filtration (Superdex75). Single-cysteine BAX variants were generated in constructs containing C62S and C126S. All other BAX variants retained the native cysteines. Recombinant truncated BID (tBID) was produced as described [25].

Mitochondrial cytochrome c release

Mouse Liver Mitochondria (MLM) were prepared from C57BL/6 Bak/ mice as described [26]. Purified MLM was supplemented with recombinant full-length BAX and tBID (see above), and cytochrome c release assessed by SDS-PAGE and immunoblotting as reported [11]. The percentage of cytochrome c released was determined by densitometry with Image Lab 6.0 (Biorad, USA) software.

Assessment of BAX melting temperature

The melting temperature (Tm) of BAX variants was determined as described [11].

PEG-maleimide labelling

Single-cysteine BAX variants (50 nM) were labelled with PEG-maleimide by treatment with 0.2 mM methoxy polyethylene glycol (PEG)-maleimide (5 kDa, PLS-234, Creative PEGworks) on ice for 1 min, then quenched with 20 mM N-ethylmaleimide (Sigma).

Cell culture and retroviral infection

Bak‒/‒Bax‒/‒ MEFs (C57BL/6), immortalised by transfection with DNA encoding SV40-large T-antigen, were passaged in Dulbecco’s Modified Eagles Medium supplemented with 10% foetal calf serum, 55 µM 2-mercaptoethanol and 250 µM asparagine at 37 °C and 10% CO2. BAX constructs were cloned into the retroviral expression vector pMSCV-IRES-GFP and transduced into MEFs using packaging cell line HEK293T cells. All cell lines were routinely checked for Mycoplasma using the MycoAlert™ mycoplasma detection kit (Lonza).

Cell death assays

Etoposide-treated cells were stained by adding 200 nM TMRE (tetramethylrhodamine, ethyl ester) (Sigma), a cell-permeating dye that accumulates in active mitochondria of live cells, to the culture medium for 20 min at 37 °C. Cells were then harvested, re-suspended in phosphate-buffered saline (PBS) and cell death (%TRME-negative) assessed by flow cytometry.

Carbonate extraction

MEFs were pre-treated (1 h) with 25 µM Q-VD (MP Biomedicals Astralasia) and then incubated with 5 µM etoposide. After 8 h, cells were harvested and permeabilized with 0.025% w/v digitonin in MELB buffer (100 mM KCl, 2.5 mM MgCl2, 100 mM sucrose, 20 mM HEPES/KOH pH 7.5) for 10 min on ice; then cytosol and mitochondria-enriched heavy membrane were separated by centrifugation. Isolated MLM samples were simply fractionated by centrifugation. Membrane fractions were then re-suspended in sodium carbonate (0.1 M, pH 11.5) and incubated on ice for 20 min before addition of an equal volume of 0.1 M HCL. After treating the samples with DNase I (5 Units/50 µl), the supernatant fraction containing peripheral proteins and pellet fraction containing membrane-integrated proteins were separated by centrifugation. Cytosol, peripheral and integrated fractions were then run on SDS-PAGE and immunoblotted for BAX.

Limited proteolysis

BAX conformational changes were assessed by limited proteolysis using proteinase K (PK). In brief, 50 µl of recombinant protein (50 nM) was treated with 30 μg/ml PK (Roche) on ice for the indicated times. The reaction was stopped by adding 1 mM phenylmethylsulfonyl fluoride (Sigma). Samples were run on SDS-PAGE and immunoblotted with BAX clone 3 antibody (BD).

SDS-PAGE and Immunoblotting

For SDS-PAGE, lysates of whole cells or cellular fractions diluted with an equal volume of 2× reducing SDS sample buffer (1.2% SDS, 30% glycerol, 5% 2-mercaptoethanol, 0.02% bromphenol blue, 150 mM Tris/HCl, pH 6.8) were electrophoresed through NuPAGE Bis-Tris gels (Invitrogen) and transferred to PVDF membrane. Membranes were blocked in 5% w/v non-fat milk in TBST (20 mM Tris, 150 mM NaCl, 0.1 % Tween 20, pH 7.4) prior to immunoblotting. Primary antibodies used were rabbit polyclonal anti-BAX NT (Millipore; Cat# 06-499, RRID:AB_310143), mouse monoclonal anti-BAX clone 3, which recognises BAX aa55-178 (BD Biosciences; Cat# 610982, RRID:AB_398295), monoclonal anti-cytochrome c (BD Biosciences; Cat# 556433, RRID:AB_396417). Secondary antibodies were (HRP)-conjugated anti-rabbit and anti-mouse Ig (Southern Biotech; Cat# 4010-05 and # 1010-05). Proteins were visualised by Luminata Forte Western HRP substrate (Millipore; Cat# WBLUF0500).

Intracellular flow cytometry

Etoposide-treated cells were harvested and washed in PBS prior to permeabilization with 0.025% w/v digitonin in MELB (100 mM KCl, 2.5 mM MgCl2, 100 mM sucrose, 20 mM HEPES/KOH pH 7.5) for 10 min on ice. Cells were then stained with 0.5 µg/ml of conformation-specific mouse monoclonal anti-BAX 6A7 or clone 3 (BD Biosciences; Cat# 556467 and Cat# 610982) in MELB buffer for 30 min on ice. After several washes in MELB, cells were incubated with mouse-specific PE-conjugated secondary antibody (Southern Biotech Cat# 1030-09) in MELB buffer for 30 min on ice in the dark. Cells were then washed, resuspended in PBS and analysed by flow cytometry.

Quantification and statistical analysis

All experiments were repeated in at least three biological replicates. The details of experiments, including statistical tests used, number of experiments, and dispersion and precision measures, are stated in the figure legends. All statistical measurements were done in GraphPad Prism version 7.0d for Mac (GraphPad Software). Two-tailed unpaired t tests were used to determine statically significant differences between two groups and two-way ANOVA (using Dunnett’s multiple comparison test) to compare multiple groups. Western blot images were quantitated by densitometry with ImageLab software 6.0.0 (Bio-Rad).

Results

The N-terminal half of the α1−α2 loop influences BAX activation

We sought to determine whether the BAX α1−α2 loop is simply a spacer without sequence-specific function or whether certain segments are important for BAX activity. We first made chimeric loops by replacing the N-terminal or C-terminal half of the loop with cognate loop regions from the related proteins BAK (pro-apoptotic) and BCLW (pro-survival) (Fig. 1A and B), which have α1−α2 loops very similar in length to that in BAX (Fig. S1A). As a sensitive assay of BAX function, we supplemented mouse liver mitochondria (MLM) from Bak‒/‒ mice, which naturally lack BAX [27], with the recombinant BAX chimeric proteins and increased levels of tBID and monitored cytochrome c release. The chimeras replacing the C-terminal half of the BAX α1−α2 loop, termed BAX-BAK and BAX-BCLW, retained WT activity (Fig. 1C, Fig. S1B).

Fig. 1: Replacing the N-terminal half of the BAX α1−α2 loop reduces BAX activity.
figure 1

A BAX loop swap chimeras with the first or second half of the BAX α1−α2 loop (bold black) replaced with the corresponding regions of the α1−α2 loop of either BAK (red) or BCLW (green). The residues changed from WT BAX are underlined. B BAX structure (PDB: 1F16) showing the N-terminal (blue) and C-terminal (purple) region of the BAX α1–α2 loop. C Impact of the loop substitutions on BAX-mediated cytochrome c release. Bak‒/‒ MLM were supplemented with 10 nM recombinant BAX and increasing concentrations of tBID for 1 h, at 37 °C. Supernatant and pellet samples were then immunoblotted for cytochrome c (see representative immunoblots in Fig. S1B) and percent cytochrome c release determined densitometrically. Black lines are WT and red the mutant. Data are means ± SEM of at least three independent experiments. Significant differences from WT BAX were measured by two-tailed unpaired t tests (*P < 0.05, **P < 0.01, ***P < 0.001).

Interestingly, however, the BAX variants having the N-terminal half of the α1−α2 loop replaced with the cognate sequence from either BAK or BCLW (termed BAK-BAX and BCLW-BAX) responded significantly less well (Fig. 1C, Fig. S1B). The replacement of the first nine BAX residues with that of BAK, which changed six residues (Fig. 1A), increased the IC50 about five-fold, whereas the more divergent BCLW substitution, which changed eight residues, produced a more than 20-fold increase in the IC50 (Fig. 1C, Fig. S1B). These results demonstrate that the BAX α1−α2 loop is not merely a random spacer sequence and highlight a potential important role of its N-terminal portion for efficient BAX function.

Alanine mutagenesis of the two ends of the BAX α1−α2 loop affects BAX activity oppositely

Since the loop swap mutants implicated the α1–α2 loop in BAX function, to identify critical loop residues systematically, we then used alanine scanning mutagenesis, an unbiased approach that has allowed us to identify new BAX loss-of-function (LOF) mutants [11]. The scan generated non-overlapping triple-alanine mutants along the BAX α1–α2 loop and, where alanine was already present, included a valine substitution (Fig. 2A). Although most of these mutants retained normal function (Fig. 2B, Fig. S2), replacing the first or the last three residues of the loop significantly affected the ability of BAX to induce cytochrome c release. Interestingly, whereas N-terminal variant LV1 reduced BAX activity about two-fold (p < 0.01), the C-terminal LV8 enhanced its activity to a similar extent (p < 0.01) (Fig. 2B; Fig. S2). These findings further support a role for the α1-α2 loop in BAX activation and suggest that the N- and C-terminal ends of the loop may have opposing roles in the activation.

Fig. 2: Triple-alanine mutagenesis reveals distinct functions for the N-terminal and C-terminal end of the loop.
figure 2

A BAX triple-alanine mutagenesis scan of the α1−α2 loop. B Ability of the triple-alanine mutants to affect tBID-induced cytochrome c release was analysed as in Fig. 1C. Representative immunoblots are shown in Fig. S2. Black lines are WT and red the mutant. Data are means ± SEM of at least three independent experiments. Significant differences from WT BAX were measured by two-tailed unpaired t-tests (**P < 0.01).

Tumour mutations in the BAX α1−α2 loop reduce BAX activity

Next, we focused on mutations in the BAX α1−α2 loop found in tumour cells and their impact on BAX function. Intriguingly, collating somatic BAX missense mutations described in cancer cell lines or tumour samples [28] or listed on the COSMIC database (https://cancer.sanger.ac.uk/cosmic) [29] revealed a ‘hotspot’ of mutations in the N-terminal half of the loop, where five of its nine residues have suffered one or two mutations (Fig. S3A). Since the triple-alanine mutagenesis indicated that some mutations in this region might only modestly impact BAX function (Fig. 2), we decided first to generate BAX variants harbouring a combination of α1−α2 loop mutations found in different tumours rather than making single-residue mutations (Fig. 3A). Notably, all three such loop variants (denoted ‘Tumour All’, ‘Tumour 1 and 2’) showed significantly less cytochrome c release at low tBID concentrations (Fig. 3B, Fig. S3B). The increase in IC50 ranged from two-fold for ‘Tumour 1’ and three-fold for ‘Tumour 2’ to five-fold for ‘Tumour All’, which has five residues mutated. Thus, tumours harbouring missense mutations in this loop region may have a survival advantage during cytotoxic stress.

Fig. 3: BAX variants harbouring mutations found in tumours attenuate BAX activity.
figure 3

A BAX variants with mutations in the α1−α2 loop found in tumours (see Fig. S3A). B Their activity was analysed as in Fig. 1C. Representative immunoblots are shown in Fig. S3B. Black lines are WT and red the mutant. Data are means ± SEM of at least three independent experiments. Significant differences from WT BAX were measured by two-tailed unpaired t-tests (*P < 0.05, ***P < 0.001, ****P < 0.0001).

We next sought to identify specific N-terminal α1-α2 loop residues required for its function.

To do so, we generated single-residue BAX variants harbouring some of the missense mutations found in tumours that induce more drastic changes (e.g. charge swaps) and tested their impact on BAX activity. However, none of the three single- residue α1-α2 loop mutants tested (E41K, A42S and E44Q) differed significantly from WT BAX in activity (Fig. S3C, D), emphasising the importance of distinctive regions of the α1-α2 loop rather than specific single residues.

The LOF BAX α1−α2 loop variants are impaired in MOM translocation and/or integration

The BAX α1−α2 loop has been implicated in the early steps of BAX activation [11, 12, 14, 22]. To assess whether the LOF variants identified here have a defect in MOM translocation and/or integration in response to a death stimulus, we tested their stable integration into the lipid bilayer by carbonate extraction. Indeed, at low tBID concentrations, all loop LOF variants had less BAX integrated into the MOM. Whereas tBID increased MOM integration of WT BAX about three-fold, it evoked little increase in the chimeric BAX α1−α2 loop mutants BAK-BAX and BCLW-BAX (Fig. 4A), the triple-alanine mutant LV1 (Fig. 4B) or the BAX tumour loop variants (Fig. 4C). In contrast, the second-half loop swaps BAX-BAK and BAX-BCLW showed relatively normal BAX translocation and integration (Fig. S4A), as did BAX LV8 (Fig. S4B), despite its ability to increase cytochrome c release (Fig. 2B). Thus, the diminished BAX function in the N-terminal variants appears to be due to reduced translocation to the MOM and/or integration within it.

Fig. 4: Mutations in the first half of the BAX α1−α2 loop reduce BAX integration into the MOM upon a death stimulus.
figure 4

BAX variants A BAK-BAX and BCLW-BAX, B LV1 and C Tumour All, Tumour 1 and Tumour 2 were tested for MOM translocation and integration upon tBID treatment. Bak-/- MLM supplemented with 10 nM recombinant BAX were treated with graded tBID concentrations for 1 h at 37 °C and BAX membrane integration determined by carbonate extraction of the membrane fraction. The supernatant (S/N) and integrated fractions were immunoblotted for BAX. For each variant, the fold increase of integrated BAX in the tBID-treated sample over the untreated control was determined by densitometric analysis.

Mutations in the BAX α1−α2 loop allosterically reduce exposure of the α9 TM domain

We then explored the molecular mechanism responsible for the defective MOM integration in the LOF α1−α2 loop variants. We first compared their thermal stability to BAX WT to exclude the possibility that changed protein stability impaired BAX activation. However, none of the N-terminal LOF loop variants (i.e. BAK-BAX, BCLW-BAX, LV1 or the tumour mutants) differed significantly in melting temperature (Tm) from BAX WT (Fig. S5A), ruling out major conformational changes. In contrast, most of the C-terminal loop variants (BAX-BAK, BAX-BCLW, LV5, LV6 and LV8) have a significantly lower Tm (Fig. S5A), suggesting that this loop segment is important for overall BAX stability.

As recent studies from our lab and others have suggested that an allosteric interaction between the BAX α1−α2 loop and the distant α9 TM may control exposure of the membrane anchor and thereby affect BAX integration into the MOM [11, 22], we tested whether the BAX loop LOF variants display changes in α9 conformation that could explain their defective MOM integration. Robin et al. showed that the extent of BAX α9 exposure can be assessed by the susceptibility of certain BAX α9 single-cysteine substitutes, such as V180C (Fig. 5A), to labelling by the cysteine-reactive PEG-maleimide [22]. Therefore, we generated a panel of BAX α1−α2 loop variants harbouring a single cysteine substitution in α9 (V180C) and performed PEG-maleimide labelling. Notably, all the N-terminal loop variants with reduced function (LV1, BAK-BAX, BCLW-BAX and Tumour All) showed significantly less PEG-labelling of V180C than did BAX WT (Fig. 5B). The reduction ranged from about 35% to 85% (p < 0.05 to < 0.0001). These data strongly indicate that these mutations inhibit BAX translocation and cytochrome c release by increased sequestration of α9 in the BAX canonical groove. In contrast, PEG labelling of V180C was not significantly inhibited by the C-terminal variant LV8 nor by the C-terminal loop swap variant BAX-BCLW (Fig. S5B), although BAX-BCLW did have some reduction in PEG labelling.

Fig. 5: Reduced exposure of helix α9 in BAX variants with mutations in the first half of the loop.
figure 5

A At the top, BAX (PDB: 1F16) showing the cysteine replacement V180C (red) in α9 (blue). Below, cartoon showing inactive cytosolic BAX existing as conformers with α9 held in the canonical groove and with α9 exposed. A death signal (e.g. tBID) shifts the balance to a conformer with an altered α1-α2 loop and the α9 TM exposed to induce MOM targeting. B PEG-maleimide labelling of BAX α9. The mutations LV1, BAK-BAX, BCLW-BAX and Tumour All were introduced into BAX having a single cysteine in helix α9 (V180C) and the recombinant proteins incubated with the cysteine-reactive PEG-maleimide (5 kDa) for 1 min on ice. Labelling of BAX was then assessed by immunoblotting (upper panel: representative experiment). Data in lower panel are means ± SEM of four independent experiments. Statistical analysis of differences from WT BAX was determined by one-way ANOVA, using Dunnett’s multiple comparison test (*: P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). # The cysteine in the BCLW part of the α1−α2 loop was replaced by a serine. C Conformation of BAX helix α9 in α1−α2 loop variants assessed by limited PK proteolysis. In inactive WT BAX, the major cleavage is after Phe176 in α9, whereas the α1−α2 loop cleavage site (mainly after Met38) is protected22,30,39. The BAX variants (50 nM recombinant protein) were treated with PK (5, 10 or 15 min on ice), run on SDS PAGE and immunoblotted for BAX. The immunoblots are representative of three independent experiments.

Another well-established test for BAX conformational changes, including α9 exposure, is limited Proteinase K (PK) treatment [11, 22, 30]. As expected, cytosolic WT BAX was mainly cleaved in α9, almost completely after 10 min of PK exposure, whereas its α1−α2 loop exhibited little cleavage (Fig. 5C). The cleavage pattern of LOF loop variants LV1, BAK-BAX and Tumour All resembled that of WT BAX, but α9 cleavage was markedly retarded (Fig. 5C), consistent with reduced α9 exposure. Loop variant BCLW-BAX, however, showed more cleavage in both the α1−α2 loop and α9 (Fig. 5C). This is not entirely surprising given that BCLW-BAX changes eight BAX loop residues (Fig. 1A) and had the greatest Tm drop among LOF variants (Fig. S5A). The C-terminal loop variant LV8 also showed a PK cleavage pattern very distinct from WT BAX, suggestive of structural rearrangements and destabilization (Fig. S5C). Surprisingly, the C-terminal loop swap BAX-BCLW, despite the lowest Tm of all variants (Fig. S5A), exhibited a relatively normal PK cleavage pattern (Fig. S5C).

Together, the above results provide strong evidence for crosstalk of the N-terminal half of the α1−α2 loop with the distant TM domain-containing helix α9 and suggest that their allosteric interaction plays an important role in fine-tuning BAX activation.

N-terminal BAX α1−α2 loop mutants also show reduced activity in cells

To explore if mutations in the LOF α1−α2 loop BAX variants identified using isolated mitochondria also reduced activity in a cellular environment, we stably expressed the loop variants LV1, BAK-BAX, BCLW-BAX and Tumour All in Bax/Bak double knockout (DKO) mouse embryonic fibroblasts (MEFs) and measured cell death after etoposide treatment. Indeed, all these loop variants showed significantly impaired BAX activity in the treated cells (Fig. 6A).

Fig. 6: BAX N-terminal α1−α2 loop variants also show reduced activity in cells.
figure 6

A Tests of activity of BAX α1–α2 loop mutants in cells. Bax/Bak DKO MEFs expressing the indicated BAX variants were treated, or not, with 5 µM etoposide for 24 h and mitochondrial membrane potential determined by TMRE staining and FACS analysis. Data are means ± SEM of four independent experiments. Statistical analysis of differences from WT BAX in etoposide-treated samples were analysed by ordinary one-way ANOVA, using Dunnett’s multiple comparison test (*P < 0.05, ***P < 0.001) or a paired RM (repeated measure) one-way ANOVA, using Dunnett’s multiple comparison test (#: P < 0.05). Upon etoposide treatment, the BAX N-terminal α1-α2 loop mutations impede exposure of the epitopes for B 6A7 and C clone 3. MEFs expressing the BAX variants were pre-incubated with 25 µM Q-VD.oph for 1 h at 37° and then treated with 5 µM etoposide. After 8 h, cells were permeabilized with 0.025% digitonin prior to staining with conformation-specific anti-BAX antibodies 6A7 or clone 3. After several washes, the samples were stained with a PE-conjugated secondary antibody and analysed by flow cytometry. The results are representative of at least two independent experiments. Mutations in α1-α2 reduce MOM translocation and integration of BAX. Bax/Bak DKO MEFs expressing the different BAX variants were pre-incubated with 25 µM Q-VD.oph for 1 h, treated with 5 µM etoposide for 8 h, subjected to carbonate extraction and fractions run on SDS-PAGE and immunoblotted for BAX. D shows representative blots. E Densitometric analysis of carbonate extracts (as in D). Data are means ± SEM of three independent experiments. Significant differences from WT BAX were revealed by two-way ANOVA, Dunnett’s multiple comparison test (*P < 0.05, ***P < 0.001).

As a control, we also tested the activity in cells of C-terminal loop variants LV8 and BAX-BCLW. In contrast to the mitochondrial assay (Figs. 1C, 2B), LV8 failed to show enhanced activity when expressed in cells, whereas BAX-BCLW did exhibit increased activity (Fig. S6A). Such inconsistencies are not altogether surprising, because the cellular etoposide response is much more complex than cytochrome c release from MLMs by BAX plus tBID. The BAX-BCLW results indicate that mutations in the C-terminal half of the α1−α2 loop can enhance BAX activity in cells, most likely because reduced BAX stability can facilitate the unfolding required for BAX activation.

Next, we explored whether the function of α1−α2 loop as an allosteric switch is required for BAX activity or instead simply fine-tunes BAX activity by affecting the kinetics of its activation. Time course experiments using etoposide-treated DKO MEFs expressing either BAX WT, the LOF variant LV1 or the gain-of-function variant BAX-BCLW suggest that the α1−α2 loop acts by influencing the activation kinetics. LV1 expressing cells showed significantly less cell death induction than BAX WT at 32 h of treatment, but not at 51 h (Fig. S6B). Conversely, cell death was induced significantly faster in cells expressing the C-terminal variant BAX-BCLW rather than WT BAX (Fig. S6B).

To confirm that BAX in DKO cells expressing an N-terminal BAX loop variant is in a less active state at an early time point (8-h treatment), we performed intracellular flow cytometry using the conformation-specific anti-BAX antibodies 6A7 and clone 3, which both recognise epitopes exposed only on activated BAX [31, 32]. As expected, untreated cells exposed neither epitope. However, in etoposide-treated cells, the LOF loop mutations strongly reduced exposure of the epitopes for 6A7 (Fig. 6B) and clone 3 (Fig. 6C). The cell population displaying 6A7 fell from 19.9% for WT BAX to between 2 and 9 % with different loop mutants (Fig. 6B), and clone 3 gave comparable results (Fig. 6C). These findings strongly re-enforce our conclusion that these BAX variants are much less active than their WT counterpart, because they are restrained in an inactive conformation. Conversely, the C-terminal variants LV8 and BAX-BCLW exhibited normal or slightly increased epitope exposure at this early time point (Fig. S6C).

Exposure of the 6 A7 epitope is thought to require BAX membrane association [33]. Hence, these results (Fig. 6B) support our conclusion that the BAX LOF loop variants have defects in MOM translocation and/or integration. We tested this directly by carbonate extraction on the DKO MEFs expressing the loop mutants, with or without etoposide treatment (Fig. 6D and E). The untreated cells exhibited no differences. However, as expected, every LOF loop variant was significantly less integrated than WT BAX (Fig. 6D and E). In contrast, with C-terminal loop variants LV8 and BAX-BCLW, MOM translocation/insertion (Fig. S6D) were relatively normal.

The evidence that cells expressing BAX variants harbouring mutations in the α1−α2 loop N-terminal region have retarded BAX activation (Fig. 6) confirms our results on isolated MLM (Figs. 14).

Discussion

As sketched in Fig. 7, this study demonstrates the important role of the BAX α1−α2 loop in the early steps of BAX activation. Different mutagenic approaches, including loop swaps, alanine scanning and investigation of tumour mutations, revealed that mutations in the N-terminal half of the α1−α2 loop (its first nine residues) lower BAX apoptotic activity (Figs. 13 and 6A), impede display of known activation markers (Fig. 6B, C) and retard its MOM association and integration (Figs. 4, 6D, 6E and 7). Molecular analysis revealed that these α1−α2 loop mutations act by allosterically reducing exposure of the distant BAX α9 TM domain (Figs. 5 and 7), thereby limiting its availability for MOM integration and favouring cell survival under cytotoxic stress (Figs. 6 and 7). Hence, we propose that the BAX α1−α2 loop is crucial for fine-tuning the early steps of BAX activation.

Fig. 7: A BAX activation model and the inferred impact of the N-terminal α1−α2 loop mutations on BAX activation.
figure 7

In the cytosol, inactive BAX conformers with α9 exposed or tucked into its canonical groove are in equilibrium. In response to a death stimulus (e.g. tBID), BH3 engagement of the α1/α6 ‘trigger’ site stabilises WT BAX in a conformation with α9 exposed, which favours increased BAX MOM association and integration. This facilitates BH3 binding to the BAX canonical groove, which elicits further conformational changes (including exposure of the BAX N-terminus and BH3 domain, and core-latch dissociation), finally leading to BAX BH3-in-groove dimerisation and oligomeric pore formation. Mutations in the N-terminal part of the α1–α2 loop allosterically reduce exposure of the distant BAX α9 TM domain and stabilise cytosolic BAX in a conformation less able to engage the MOM and to induce cytochrome c release upon a death stimulus.

Our findings rule out the possibility that the BAX α1−α2 loop is a random spacer sequence and show that it plays an important regulatory role in BAX activation (Fig. 7), as first proposed by Gavathiotis et al. [12, 14]. Whereas the native N-terminal loop sequence of nine residues proved important for optimal BAX function (Figs. 1B, 2B and 6A), modifying the C-terminal half of the loop did not impair BAX apoptotic function (Figs. 1, 2 and S6). Most C-terminal variants, however, had a de-stabilising effect on BAX, lowering its melting temperature by 5 to 11 degrees (Fig. S5), indicative of notable conformational changes. Thus, our results suggest that the C-terminal half of the loop may predominantly affect overall BAX stability, presumably through its interactions with other BAX regions. Reduced stability might well allow BAX to more readily assume an active conformation, explaining the modest but significant increase in activity for triple-alanine mutant LV8 in cytochrome c release assays (Fig. 2B) and for loop swap mutant BAX-BCLW in cell-based assays (Fig. S6A). However, LV8 significantly lowered the BAX melting temperature (Fig. S5A) but did not significantly increase cellular apoptotic activity (Fig. S6A). Perhaps certain C-terminal loop mutations have multiple effects on BAX activation that compensate one another.

It is not surprising that the N- and C-terminal halves of the α1−α2 loop seem to have distinct functions, because the BAX NMR structure [3] revealed a short helical segment between them, i.e. they resembled two separate loops.

All LOF loop variants identified in our mutagenesis screen (BAK-BAX, BCLW-BAX, LV1 and certain tumour mutants) reduced MOM translocation and/or integration upon a death stimulus, both with isolated mitochondria (Fig. 4) and when expressed in cells (Fig. 6D, E). Interestingly, Robin et al recently showed that antibodies with epitopes in the first half of the BAX α1−α2 loop can block BAX translocation and cytochrome c release by sequestering α9 in the canonical BAX groove [22]. Notably, our N-terminal α1−α2 loop mutants also stabilise BAX in an inactive conformation that reduces α9 exposure, because they reduced labelling of an installed α9 cysteine residue (Fig. 5B) and retarded proteinase K cleavage of α9 (Fig. 5C). Thus, these mutations produce a phenotype resembling such ‘blocking’ antibodies. Indeed, with the most effective such antibody (3C10), two of the crucial BAX residues in its epitope (R37 and M38) [22] are mutated to alanine in BAX LV1 (36GRM38), underlining this region’s importance. Hence, independent antibody blocking and mutagenesis studies converge to highlight the importance of the first nine residues in the loop for BAX activation. Also, our mutants strengthen the case that the α1−α2 loop is physiologically relevant to BAX regulation.

As suggested in the previous studies [11, 12, 22], our work confirmed allosteric crosstalk between the BAX α1−α2 loop and the distant α9 (Figs. 5 and 7). Thus, targeting the first half of the α1−α2 loop with an antibody or by mutagenesis increases sequestration of α9, which must be exposed for BAX activation upon a death stimulus. Interestingly, with a BAX variant that constitutively resides on the MOM (S184L), the loop antibodies instead activate BAX by unfolding BAX helix α1 [23]. Hence, the WT BAX α1−α2 loop appears to function as a ‘safety lock’, preventing unwarranted activation of cytosolic BAX.

Initiation of BAX activation upon BH3 binding to the proposed non-canonical BAX ‘trigger site’ on a1/a6 was suggested to require changing the α1−α2 loop from a ‘closed’ to an ‘open’ conformation [12, 14]. However, our studies on BAX variants with mutations in this ‘trigger site’ indicated that BH3 binding instead stabilises BAX with the α1−α2 loop in a ‘closed’ configuration [11], consistent with the findings here that mutations, especially in the first half of the loop, increase α9 sequestration (Fig. 5B) and reduce BAX integration into the MOM (Figs. 4 and 6E). However, because the PK cleavage site in the loop (mainly after M38) [22] is mutated in most of the LOF loop variants, conclusions about their loop conformation remain speculative. Our evidence that the N- and C-terminal parts of the loop have distinct roles (Figs. 1 and 2) might explain some of the conflicting observations made for the loop region.

Inactivating mutations in the BAX gene (typically frameshifts) can promote tumorigenesis and are poor prognostic factors in cancer [34]. Pertinently, the rare reported BAX missense mutations in tumour samples include a ‘hotspot’ in the first half of the α1−α2 loop but only a single mutation in the second half (Fig. S3A). This suggests that tumour cells expressing BAX N-terminal loop variants have a survival advantage, which might well contribute to malignant progression and therapy resistance. Indeed, although we were unable to identify single α1-α2 loop residues essential for BAX activity, BAX variants harbouring two to five such mutations identified in different tumours proved less active than WT BAX on isolated mitochondria (Fig. 3) and showed reduced killing in cells (Fig. 6A). Furthermore, like the other LOF α1−α2 mutants, these mutants impaired MOM integration upon a death stimulus (Figs. 4C, 6D, E) by reducing α9 exposure (Fig. 5). Our data may also help to explain the dearth of tumour mutations in the second half of the loop. Most mutations there significantly reduced BAX stability (Fig. S5A), which can facilitate the unfolding that activates BAX, and hence could antagonise tumour development, particularly if, like LV8 (Fig. 2) or chimera BAX-BCLW (Fig. S6), they can increase BAX activity.

Stabilising BAX in a conformation with the α9 helix exposed is crucial to allow efficient BAX integration into the MOM, where it becomes fully activated (Fig. 7). Interestingly, the proportion of BAX located on mitochondria correlates with sensitivity to death stimuli and therefore with therapeutic responsiveness [7, 8, 35]. Our data indicate that the α1−α2 loop helps set the crucial balance between cytosolic and MOM-integrated BAX by fine-tuning the early steps of BAX activation (Fig. 7) and therefore provides an appealing target for drugs that either directly promote BAX activation (for cancer) or inhibit it in pathologies with excessive apoptosis (e.g. for stroke or degenerative disorders) [36]. The discovery that antibodies against the BAX α1−α2 loop affect BAX activation [22, 23] also underscore this strategy’s promise. Pertinently, compounds targeting BAX may well have a therapeutic window [37, 38]. Hence, systematic mutagenesis and functional analysis of BAX, as in this study, are important to clarify not only how BAX is activated for death duty but also to reveal how best to target BAX therapeutically.