An Autoinhibited Dimeric Form of BAX Regulates the BAX Activation Pathway

SUMMARY

Pro-apoptotic BAX is a cell fate regulator playing an important role in cellular homeostasis and pathological cell death. BAX is predominantly localized in the cytosol where it has a quiescent monomer conformation. Following a pro-apoptotic trigger, cytosolic BAX is activated and translocates to the mitochondria to initiate mitochondrial dysfunction and apoptosis. Here, cellular, biochemical and structural data unexpectedly demonstrate that cytosolic BAX also has an inactive dimer conformation that regulates its activation. The full-length crystal structure of the inactive BAX dimer revealed an asymmetric interaction consistent with inhibition of the N-terminal conformational change of one protomer and the displacement of the C-terminal helix α9 of the second protomer. This autoinhibited BAX dimer dissociates to BAX monomers before BAX can be activated. Our data support a model whereby the degree of apoptosis induction is regulated by the conformation of cytosolic BAX and identify an unprecedented mechanism of cytosolic BAX inhibition.

Graphical Abstract

INTRODUCTION

Programmed cell death, or apoptosis, is a fundamental process that regulates the critical balance between cellular life and death in health and disease (Danial and Korsmeyer, 2004; Fuchs and Steller, 2011). The intrinsic pathway of apoptosis triggers mitochondrial outer membrane permeabilization (MOMP), which results in mitochondrial dysfunction, the release of soluble proteins from the mitochondrial intermembrane space to the cytosol and activation of caspases that are required for cellular demise (Tait and Green, 2010). The BCL-2 family proteins are the principal regulators of the intrinsic apoptotic pathway since they govern the commitment of the cells to MOMP. (Chipuk et al., 2010; Youle and Strasser, 2008). The BCL-2 family is divided into three subgroups of proteins: the pro-apoptotic effectors BAX and BAK that induce MOMP, the anti-apoptotic or survival proteins such as BCL-2 and MCL-1 that inhibit the proapoptotic proteins and block MOMP and the pro-apoptotic BH3-only proteins that directly activate BAX/BAK or inactivate anti-apoptotic BCL-2 proteins to enable or sensitize BAX/BAK activation (Chen et al., 2015; Shamas-Din et al., 2013).

Cells doubly deficient for BAX and BAK become resistant to various apoptotic stimuli, consistent with the notion that BAX and BAK control the gateway to MOMP and apoptosis (Wei et al., 2001). In contrast to BAK that resides at the mitochondrial outer membrane, BAX is predominantly localized in the cytosol in an inactive conformation (Edlich et al., 2011; Wolter et al., 1997). Upon cellular stress, cytosolic BAX is activated by the BH3-only molecules and translocates to the mitochondrial outer membrane to induce MOMP and release apoptogens (Desagher et al., 1999; Eskes et al., 2000; Goping et al., 1998; Nechushtan et al., 1999; Wei et al., 2001). The active conformation of BAX can be recognized specifically by the 6A7 antibody (Hsu and Youle, 1998), which detects the exposure of an N-terminal epitope (residues 12–24) upon conformational activation of BAX. NMR studies revealed that soluble BAX (Suzuki et al., 2000) has a globular bundle of nine α helices and is maintained in an inactive and 6A7-negative conformation. The C-terminal helix α9 is bound to the canonical BH3 groove in the inactive BAX structure while the exposure of α9 is required for mitochondrial anchoring and insertion of active BAX (Annis et al., 2005; Gavathiotis et al., 2010; Kim et al., 2009; Suzuki et al., 2000).

Structural studies using pro-apoptotic BH3 helices revealed an activation site at the N-terminal surface (α1/α6 trigger site) in the full-length BAX structure (Gavathiotis et al., 2008) and another activation site at the hydrophobic BH3 groove that is exposed in the α9-truncated BAX ΔC21 structure (Czabotar et al., 2013). BH3-triggered activation of full-length BAX is associated with the opening of the α1–α2 loop, the exposure of the 6A7 epitope and the allosteric conformational mobilization of the C-terminal helix α9 and α2 (BH3 domain) from the hydrophobic core of BAX (Gavathiotis et al., 2010; Gavathiotis et al., 2008; Kim et al., 2009). Likewise, BH3-induced activation of α9-truncated BAX through interaction with the hydrophobic BH3 groove resulted in the disengagement of the “latch” helices α6–α8 from the N-terminal surface (Czabotar et al., 2013). Models of BAX oligomerization at the membrane suggest that more dramatic conformational changes of the BAX structure are required for MOMP (Annis et al., 2005; Bleicken et al., 2014; Czabotar et al., 2013). These studies highlight that both activation mechanisms require conformational changes at the N-terminal surface and C-terminal α9 to elicit the active form of BAX that is accountable for MOMP induction.

While these studies provide insights into the mechanisms of BAX activation and the associated conformational changes, they also imply that additional mechanisms may regulate cytosolic BAX. Since the N-terminal activation site and the pocket of the C-terminal helix α9 are targets for BH3-mediated activation, an obvious question is whether there are additional interactions that stabilize and keep the cytosolic BAX conformation inactive. Interestingly, anti-apoptotic BCL-2 proteins were previously suggested to stabilize cytosolic, inactive BAX; however, no physical interaction has been biochemically detected in the absence of membrane or detergent, consistent with the mitochondrial localization of the anti-apoptotic BCL-2 proteins (Llambi et al., 2011; Shamas-Din et al., 2013).

Here, we present evidence that BAX can also exist in an inactive and dimeric conformation. We isolated this dimeric conformation and determined its crystal structure. The conformation of BAX in the crystal structure provides insights into a distinct mechanism that inhibits BAX from its conformational activation and mitochondrial translocation. We also showed that disruption of cytosolic BAX dimers releases BAX monomers that are primed for more potent activation and apoptosis induction. These findings define an autoinhibitory mechanism of BAX that can regulate apoptosis.

RESULTS

Cytosolic BAX adopts an inactive dimer conformation

Previous size-exclusion chromatography (SEC) analysis of cytosolic fractions suggested that BAX exists as a monomer in thymocytes (Hsu et al., 1998) but also cytosolic BAX appeared 20–30 kDa heavier than recombinant protein in established cell lines (Vogel et al., 2012). In contrast, SEC analysis of mitochondrial lysates, using detergent-containing buffer, has consistently showed that mitochondrial BAX is monomeric (Annis et al., 2005; Dlugosz et al., 2006; Kim et al., 2009). Interestingly, recombinant BAX purification by SEC shows predominantly BAX in a peak that corresponds to its monomeric form (centered on 15.5 ml fraction) and additionally in a second distinct peak that corresponds to dimeric BAX (centered on 14 ml fraction) based on the molecular weight (Figure S1A and 1A). To analyze cytosolic conformations of BAX, cytosolic fractions of several cell lines were separated from mitochondria and analyzed by SEC, which showed that endogenous cytosolic BAX was eluted in fractions that are consistent with a BAX monomer (centered on 15.5 ml fraction) but also a dimer (centered on 14 ml fraction) (Figure 1A). Specifically, BAX from the cytosol of mouse embryonic fibroblasts (MEFs) was consistently eluted as dimers. In contrast, BAX from the cytosol of OCI-AML3 leukemic cells was eluted in fractions consistent with BAX monomers, and both BAX monomers and dimers were identified from A375 melanoma cells (Fig. 1A). None of these cell lines have mutations in the BAX gene (Barretina et al., 2012). In Bax and Bak doubly deficient MEFs (DKO) that were stably reconstituted with human BAX at endogenous levels (DKO BAX), cytosolic BAX appeared to be consistent with a dimer (Figure 1A).

Figure 1. BAX forms an inactive dimer conformation.

A) Recombinant BAX (rec. BAX) and cytosolic extracts from WT MEF, OCI-AML3, A375, DKO MEF reconstituted with BAX, and BAX WT MEF treated with 1% Triton X-100 (cyt. BAX + 1% Triton) were analyzed by Superdex 200 (HR 10/30) gel filtration and anti-BAX immunoblot. The centers of the eluted monomer (M) and dimer (D) peaks are indicated. (B) Anti-apoptotic BCL-2, BCL-X_L, MCL-1 and BAX protein levels in MEFs determined by western blot analysis of separated cytosolic and mitochondrial fractions as confirmed by the anti β-tubulin and anti-VDAC immunoblots, respectively. (C) Size-exclusion chromatography (SEC) fractions of cytosolic extracts of MEFs containing dimeric BAX WT (fractions 13.5 and 14.5 ml) failed to be immunoprecipitated with the 6A7 antibody. When the fractions corresponding to dimers were incubated with 1% octylglucoside detergent, the 6A7 epitope was then exposed on BAX and BAX could be immunoprecipitated with the 6A7 antibody. (D) Dimerization of purified monomeric recombinant BAX at the indicated concentrations or prior to concentration (pre-concentration) was analyzed by Superdex 75 (HR 10/30) gel filtration. The peaks of monomer (M) and dimer (D) were eluted at ~11.8 ml and ~10.4 ml, respectively. Total amount of BAX was kept constant in different samples. (E) Representative dynamic light scattering data of BAX monomers and dimers taken from the SEC elution peaks in (D). (F) Dimerization of purified monomeric BAX 4M mutant at pre- and post-concentration to 10 mg/ml was analyzed by Superdex 75 (HR 10/30) gel filtration. (G) Quantification of BAX dimerization formed by BAX WT and BAX 4M based on SEC peaks at 10 mg/ml. (H) ANTS/DPX-encapsulated liposomes were incubated with SEC-isolated BAX monomers or dimers without or with BIM SAHB_A at the indicated concentrations. The release of entrapped fluorophore monitored with time is shown. (I) Liposome release assays as in (H) with each bar indicating the total liposomal release at 90 min. Data shown in (A-G) are representative of three independent experiments with similar results. Data in (H) and (I) are mean ± SD for assays performed in triplicate and two independent experiments. See also Figure S1.

Cytosolic and mitochondrial fractions of MEFs were subjected to western blot analysis and confirmed that anti-apoptotic proteins predominantly reside at the mitochondria (Figure 1B). Thus, heterodimerization with anti-apoptotic BCL-2 proteins is not supported by the significant amount of BAX in the cytosol compared to antiapoptotic BCl-2 proteins. Furthermore, the cytosolic BAX in MEFs was confirmed to adopt an inactive conformation as probed by the 6A7 antibody that only recognizes the active BAX conformation (Figure 1C). Since certain detergents activate BAX, we treated cytosolic BAX from MEFs with 1% Triton X-100 detergent (Hsu et al, 1998). Detergent-activated BAX was eluted predominantly at high molecular weight, presumably forming the BAX oligomers (Figure 1A). In addition, a portion of BAX was eluted as monomers, suggesting that detergent might induce dissociation of BAX dimers to monomers before the formation of BAX oligomers. As expected, detergent-activated BAX was recognized by the 6A7 antibody (Figure 1C). Taken together, these results demonstrate that cytosolic BAX can adopt a dimeric, inactive conformation in addition to the inactive monomer conformation.

BAX dimer in solution is inactive and resistant to activation

To elucidate the physiological role of the cytosolic BAX dimers, we investigated the molecular basis of BAX dimerization. Upon concentration, recombinant monomeric BAX formed a dimer as purified by SEC (Figure 1D). The monomeric and dimeric peaks were confirmed by dynamic light scattering to correspond to homogenous populations of BAX monomer (~21KD) and dimer (~42KD) respectively (Figure 1E). The amount of monomer to dimer transition is increased at pH higher than 6 and with increasing temperature (Figure S1B–S1D). Because of the heat-induced BAX oligomerization effect (Pagliari et al., 2005), more homogenous inactive dimers are generated at lower temperatures and kinetic rates (details in Experimental Procedures). Moreover, to prevent the formation of active dimers, we generated a BAX mutant (C62S, C126S, V121C, I136C), termed BAX 4M that was previously shown to lock BAX in the inactive conformation by a covalent disulfide bond (Czabotar et al., 2013). Consistently, the BAX 4M mutant under oxidizing conditions was able to undergo transition to the inactive BAX dimer with the same efficiency as the wild type BAX, excluding the likehood of an active swapped dimer structure previously identified for BAX (Figure 1F, 1G).

To investigate the functional activity of SEC-isolated, recombinant dimeric BAX, we used the established liposomal assays that probe the capacity of BAX to induce permeabilization of ANTS/DPX-encapsulated liposomes in the presence of the stapled BIM BH3 helix, BIM SAHB_A (Bird et al., 2014). As expected, the BAX monomer alone was not able to permeabilize liposomes but it induced membrane permeablization upon activation by the activator BIM SAHB_A, (Figure 1H, 1I). Similarly, the BAX dimer alone showed no capacity to induce membrane permeablization. In contrast, addition of BIM SAHB_A to dimeric BAX at the same doses demonstrated strikingly reduced capacity for membrane permeablization (Figure 1H, 1I). Similarly, tBID-induced BAX activation of dimeric BAX was signficantly impaired compared to monomeric BAX, altough, tBID is a more potent activator (Figure S1E). Notably, BIM SAHB_A increased more membrane association of BAX monomers than BAX dimers (Figure S1F), suggesting that the inability of BAX dimers to permeabilize liposomes is due to impaired translocation of BAX dimers to the membrane. Taken together, the data suggest that the BAX dimer adopts an inactive conformation that is resistant to BH3-mediated activation compared to monomeric BAX.

Crystal structure of the inactive BAX dimer

To gain further insights into the dimeric, inactive BAX, we performed structural studies. Crystallization studies using the wild type BAX (BAX WT) dimers were not successful, likely due to the partial transition into monomers as evidenced by SEC. Therefore, we screened several BAX mutants in order to generate a BAX dimer that persists over time. We identified such BAX mutants including BAX P168G that generated good diffraction crystals. To confirm that the mutation does not deviate the structure from that of BAX WT, we used NMR HSQC analysis of the monomer structures. Comparison of the ¹⁵N-¹H HSQC spectra between the BAX P168G mutant and BAX WT suggested that the structure of BAX is essentially the same in both proteins (Figure S2A–S2C). Notably, cross-peaks for the α8–α9 loop residues (F165-W170) were not observed in the BAX WT spectra that presumably were broadened beyond detection due to conformational motions of the loop. In contrast, single well-resolved peaks were observed for each α8–α9 loop residue in BAX P168G spectra, indicating a more rigid loop compared to BAX WT (Figure S2D). The α8–α9 loop was suggested as a hinge region controlling the movement of the c-terminal α9, and NMR data suggest that BAX P168G has less degree of freedom in this region.

A native X-ray data set from BAX P168G crystals was obtained to a resolution of 1.9 Å (Figure S2E, Table 1). The asymmetric unit contained two BAX molecules with excellent electron density maps in which all BAX residues could be traced, except from residues of the N-terminal unstructured region (residues 1–13) and four residues of the unstructured loop between helices α1 and α2 (residues 37–40) (Figure 2A). The BAX protomers within the BAX dimer structure resemble the inactive monomeric BAX structure determined by NMR (Suzuki et al., 2000), but they have noticeable differences in the orientation of helices and conformation of loops. Backbone r.m.s.d. over the ensemble of NMR structures is 2.4 Å ± 0.1 for heavy atoms present in both structures (Figure 2B). Most pronounced differences correspond to residues located in the C-terminal α9 and helices α1, α6 and α1–α2 loop. The two protomers in the BAX dimer structure have small differences in the structure (backbone r.m.s.d. is 0.34 Å).

Table 1.

Structural statistics of data collection and refinement

Data Collection
Protein	BAX P168G	BAX G67R
Beamline	BNL (X29)	APS (23IDD)
Space group	P21	P21
Molecule in asymmetric unit	2	2
Resolution (Å)	50.0-1.90 (1.93-1.90)	50.0-3.31 (3.31-3.25)
Cell Dimensions
a, b, c (Å)	63.65, 40.27, 65.30	40.58, 65.01, 65.59
α, β, γ (°)	90.0, 90.01, 90.0	90.0, 89.8, 90.0
No. unique observations	26198	4866
Completeness (%)	98.5 (100)	89.8 (78.3)
Rsym†	0.06 (0.65)	0.13 (0.18)
I/σ (I)	22.1 (2.3)	26.4 (13.4)
Redundancy	4.6 (4.9)	3.0 (2.7)
Refinement
Resolution (Å)	19.1–1.90	46.1–3.30
Reflections	24766	4484
Completeness (%)	98.6	90.1
Number of atoms	2754	2756
Solvent atoms	124	n/a
Rwork	0.20	0.22
Rfree	0.23	0.26
R.m.s. deviations
Bond lengths (Å)	0.008	0.014
Bond angles (°)	1.21	1.42
B-factors (Å2)
Overall	34.7	59.1
Main chain	32.5	58.5
Side chain	36.6	59.5
Water	40.6	n/a
Ramachandran plot (%) most favorable region additionally allowed region	98	90
	2	10

Numbers in parentheses are for the highest resolution shell

^†R_sym = Σ(I − ‹I>)/Σ ‹I>, where I is the intensity measurement for a given refraction and ‹I> is the average intensity for multiple measurements of this refraction.

Figure 2. Crystal structure of the inactive BAX dimer.

(A) Ribbon representation of the BAX dimer crystal structure. Helices (α) and loops (L) are depicted on the structures. Dimerization interaction interfaces are shown in pink and violet for each BAX protomer. Secondary structure cartoon representation of BAX, highlighting the location of the BCL-2 homology domains (BH) in light green, the transmembrane region (TM) in dark green and the dimerization interaction interfaces in pink and light blue, is shown above the crystal structure depiction. Dotted line represents the missing α1–α2 loop residues that have weak electron density. (B) Ribbon representation of each BAX protomer in view related to the representation in (A) at 90 degrees rotation around a vertical axis, showing the N-terminal and C-terminal dimerization interfaces. (C) Structural alignment of the solution structure of BAX (green) (PDB:1F16) and BAX protomer A (grey) within the BAX dimer. Views centered on helix α5 (top view) and the N-terminal activation site (right). See also Figure S2.

The dimerization interface formed by two protomers is extensive, burying almost 1900 Ų surface area (Figure 2A). Remarkably, the BAX dimer structure reveals a dimerization interface that includes the interaction of two structural surfaces critical for the activation of BAX (Figure 2A) (Czabotar et al., 2013; Gavathiotis et al., 2008; Kim et al., 2009): the N-terminal activation site from one BAX protomer and a C-terminal surface from the second BAX protomer that includes the C-terminal α9 helix. Importantly, this dimeric conformation and dimerization interface is independent of the P168G mutant. We also determined the same dimeric conformation and dimerization interface from the X-ray crystal structure of the BAX G67R dimer at resolution of 3.3 Å (Figure S3A–S3C, Table 1), although G67 is also located distantly from P168 residue and neither residues are part of the dimerization interface. Taken together, our structural analysis shows that BAX dimerizes in a distinct but inactive conformation in agreement with the biochemical and cellular data (Figure 1).

Determinants of inactive BAX dimerization

The asymmetric BAX dimer conformation reveals two unique interaction surfaces of BAX. The interaction surface of one BAX protomer includes the N-terminal BAX activation site (Gavathiotis et al., 2008) and a number of additional residues at its periphery. This N-terminal binding surface has a hydrophobic center from solvent-exposed hydrophobic residues of α1, α6 and the α1–α2 loop that are surrounded by charged residues, complementary to the C-terminal binding surface (Figure 3A). The interaction surface from the second BAX protomer involves a site at the C-terminal surface that includes the C-terminal helix α9. This C-terminal binding surface also has a hydrophobic center from solvent-exposed hydrophobic residues of α9 that are surrounded by polar and charged residues of helices α8, α7, α3, the α7–α8 loop, the α4–α5 loop and the α8–α9 loop (Figure 3A). The dimerization interface involves hydrophobic interactions, hydrogen bonds and salt bridges. Precisely, hydrophobic interacting residues of helix α9 (e.g. I175 and F176) from one protomer interact with helices α1 (M20 and A24), α6 (M137 and L141) and α1–α2 loop (L47 and V50) from the second protomer (Figure 3B, 3C). Several residues of the dimer interface form hydrogen bonds and charged interactions that contribute to the dimerization. Specifically, the following hydrogen bonds between pairs of the two BAX protomers are observed: residues R145 of α6 and Q171 of α9, residues E17 of α1 and M74 of α3, a salt bridge between residues K21 of α1 and E75 of α3 (Figure 3B, 3D), residues A46 of α1–α2 loop and Y164 of α8, residues E44 of α1–α2 loop and W107 of α5, and residues D48 of α1–α2 loop and N106 of α5 (Figure 3B, 3E). Thus, dimerization of BAX monomers is formed with highly complementary interactions from conserved residues of the BAX N-terminal and C-terminal interfaces (Figure S3D).

Figure 3. Determinants and mechanism of inactive BAX dimerization.

(A) Calculated vacuum electrostatics of the C-terminal and N-terminal dimerization interfaces from each BAX protomer showing the position of complementary hydrophobic, polar and charged residues. (B) Cartoon representation of the BAX dimer showing the interacting protomers in pink (C-terminal dimer interface) and violet (N-terminal dimer interface). The interacting residues are shown in sticks and are highlighted in three different regions of the dimerization interface: (C) hydrophobic core, (D) and (E) hydrogen bonds and salt bridges of the dimerization interface. See also Figure S3.

Structural insights suggest an auto-inhibition mechanism of BAX

It was previously shown that BH3 domain binding to the N-terminal activation site of BAX triggers the displacement of the α1–α2 loop into an open conformation; this conformational change was determined to be essential for BAX activation and coincided with the exposure of the 6A7 epitope (Gavathiotis et al., 2010) (Figure 4A). Structural comparison of the BAX:BIM BH3 interaction with the current BAX dimer structure, revealed that helix α9 from one protomer binds the N-terminal activation site of the other protomer in an orientation that preserves the α1–α2 loop in a closed and inactive conformation (Figure 4A). The BIM BH3 binding to the N-terminal activation site is mediated by hydrophobic and polar residues of α1 and α6, with complementary residues of the BIM BH3 helix (Figure 4D, 4E). In contrast, in the BAX dimer structure, α1 residues A24, L25, Q28 and Q32 interact instead with residues Q52 and V50 of the α1–α2 loop of the same BAX protomer (Figure 4B). This generates an alternative hydrophobic surface formed by the positioning of α1 and the α1–α2 loop, which facilitates the interaction with hydrophobic residues I175 and F176 of α9 from the adjacent BAX protomer (Figure 4B, 4C). This closed α1–α2 loop conformation is also stabilized by intermolecular hydrogen bonds mediated by residue D48 of the α1–α2 loop and the C-terminal interface of the adjacent BAX protomer (Figure 4B). Therefore, for one BAX protomer, these intermolecular interactions stabilize the α1–α2 loop in a closed conformation and prevent access to the N-terminal activation site (α1/α6). For the second BAX protomer, access to the C-terminal BH3 pocket is blocked and helix α9 is stabilized within its own hydrophobic pocket (Figure S4). These structural evidence provide a mechanism by which cytosolic BAX forms an autoinhibited dimer, which occludes key interactions required for the N-terminal conformational activation of BAX and prevents the displacement of helix α9 from its hydrophobic pocket that is required for mitochondrial translocation.

Figure 4. Structural insights suggest a BAX auto-inhibition mechanism.

(A) Structural comparison of the BAX:α9 interaction from the BAX dimer structure and BAX:BIM BH3 interaction (PDB ID: 2K7W). (B) Ribbon representation of the BAX:α9 interaction showing α1 residues, A24, L25, Q28 and Q32, interacting with residues Q52, V50 and D48 of the α1–α2 loop of the same BAX molecule (purple) and with the adjacent BAX molecule through α9 residues I175 and F176 and α4–α5 loop residues R109 and N106 (pink). (C) Surface representation of the BAX:α9 interaction showing residues of the N-terminal activation site as hydrophobic (yellow), positively charged (blue) and negatively charged (red) residues and BAX α9 helix hydrophobic residues in yellow sticks. (D) Ribbon representation of the BAX:BH3 interaction showing α1 residues, A24, L25, L27 forming interactions with conserved hydrophobic residues F159, I155 of BIM BH3. Residues Q28 and Q32 of BAX interact with residue N160 of BIM BH3. (E) Surface representation of the BAX:BH3 interaction shows that BIM BH3 helix (cyan) forms hydrophobic and electrostatic interactions with residues of α1 and α6. See also Figure S4.

The autoinhibited dimeric form of BAX dissociates to monomer and is then activated by a BH3 domain interaction

To confirm that the same dimerization interface is formed in the crystal structure and in solution, we used hydrogen-deuterium exchange mass spectrometry (HXMS) and probed the dimerization interface based on the accessibility of D₂O molecules to the BAX dimer surface. We analyzed the deuterium exchange of the dimeric and monomeric BAX (Figure 1D) and calculated the difference in % D₂O incorporation of the dimeric from monomeric BAX to highlight regions of BAX that are protected from solvent exposure upon formation of the BAX dimer (Figure S5). Consistent with the crystal structure, the HXMS analysis supported the dimerization interaction between the N-terminal interface and C-terminal interface. Mapping of the significantly protected areas on the structure of the dimer highlight solvent protection in helix α1, α6 and part of the α1–α2 loop (N-terminal interface) as well as α4–α5 loop, α7–α8 loop and part of α3, α4, α5 and α9 (C-terminal interface), while other regions of the BAX structure had little to no changes in the deuterium incorporation upon dimerization.

We next performed mutagenesis on residues of the N-terminal and C-terminal interface, using the BAX 4M construct to avoid BAX activation, and assessed dimerization by SEC (Figure 5A). Consistent with the crystal structure, mutations designed to disrupt the interactions between the N-terminal surface, including E17K, K21E and A24E, and the C-terminal interface E75K, T172K and F176A (Figure 3C, 3D), markedly decreased the ability of BAX to form a dimer. The P168G mutation showed enhanced dimerization, which is consistent with our structural studies. Of note, these mutations are unlikely to disrupt the swapped dimer structures found for other BCl-2 family proteins, based on the specific interaction interfaces involved in these swapped dimer structures (Dewson et al., 2012; Jeong et al., 2004; Lee et al., 2011; O'Neill et al., 2006). Moreover, we investigated the capacity of BAX WT and mutants to form the inactive cytosolic dimers in cells. We reconstituted DKO MEFs with BAX WT and select mutants at expression levels similar to that of WT MEFs (Figure S6A). P168G mutant formed an inactive cytosolic dimer as seen with BAX WT. In contrast, E75K, T172K, K21E and A24E mutants disrupted the cytosolic dimers and formed BAX monomers constistent with their ability to disrupt dimerization in vitro (Figure 5B).

Figure 5. The autoinhibited dimeric form of BAX dissociates to BAX monomers before its activation.

(A) Dimerization of purified monomeric recombinant BAX WT and mutants of the N-terminal interaction surface (purple bars) and the C-terminal interaction surface (pink bars) were analyzed by SEC and quantified by integration of areas under observed monomer and dimer peaks. Residues mutated are shown on the ribbon structures of BAX (B) Cytosolic fractions purified from DKO MEFs reconstituted with BAX WT or mutants were analyzed by SEC using Superdex 200 (HR 10/30). (C) Representation of the relationship between the inactive BAX dimers, BAX monomers and the BAX oligomers (D) Inactive BAX 4M dimers under oxidizing conditions dissociate into monomers in response to BIM SAHB_A. Reduction of the internally crosslinked BAX 4M mutant using 20 µΜ DTT followed by incubation with BIM SAHB can induce the oligomerizaton of the BAX 4M. (E) Quantification of % BAX monomers and % BAX oligomers based on SEC peaks shown in (D) and Figure S5C. Data are representative of three independent experiments with similar results. See also Figure S5.

Next, we hypothesized that the inactive BAX dimers should undergo reversible transition to monomeric BAX such that the BAX monomers could proceed to activation and further oligomerization in response to BH3-only proteins (Figure 5C). Therefore, we tested whether BIM SAHB_A binding to the N-terminal activation site can dissociate the BAX dimers. BAX 4M dimers were incubated with BIM SAHB_A in excess and subsequently analyzed by SEC. The inactive BAX dimers under oxidizing conditions could be competed by BIM SAHB_A and dissociated into monomeric BAX. Upon reduction of the internally crosslinked bond between V121C and I136C using DTT, BIM SAHB_A binding led to activation of the monomers, inducing BAX oligomerization (Figure 5D, 5E). Treatment of BAX 4M dimers with DTT alone was not sufficient to induce BAX oligomerization, indicating that BIM SAHB_A was required to induce BAX oligomerization (Figure 5E, S5C). We further assessed the kinetics of BIM SAHB_A-induced BAX oligomerization using either BAX monomers or BAX dimers, which revealed a delayed oligomerization from the BAX dimers than from the monomers (Figure S5D). Taken together, the HXMS analysis and the biochemical and cellular data further support the autoinhibited BAX dimer structure and suggest that dissociation of BAX dimers into monomers is required before for the binding of a BIM BH3 domain to activate BAX.

Inactive BAX dimers regulates BAX activation and BAX-mediated apoptosis

To investigate the physiological role of the BAX dimerization, we compared BAX WT and mutants that have different propensity to form dimeric and monomeric BAX, in their ability to promote apoptosis in cells (Figure 5A, 5B, S6A). Strikingly, retroviral-mediated transduction of dimerization impaired BAX mutants in DKO MEFs showed significantly increased activity in apoptosis induction compared to BAX WT that exist as cytosolic dimers (Figure 6A). To exclude the possibility that these dimerization impaired mutations may affect the intrinsic ability of BAX to be activated by BH3-only molecules, we isolated monomers of these BAX mutants using SEC and subsequently examined their ability to permeabilize liposomes upon activation (Figure S6B). Although these BAX mutants displayed varying activity in the liposome release assays, there was no significant increase in BH3-induced activation that would contribute to enhanced apoptosis observed in cells. On the other hand, transduction of BAX P168G that stabilizes cytosolic dimers induced less apoptosis than BAX WT (Figure 6A). Consistently, the P168G dimers were more resistant to BH3-induced activation compared to BAX WT dimers in the liposomal release assays (Figure S6C, 5A, 5B). Next, we compared BAX WT and mutant cells with propensity to form monomeric or dimeric BAX upon treatement with a typical apoptosis inducer such as staurosporine (STS). Upon STS treatment, BAX E75K monomer cells strinkingly underwent significantly more potent and faster apoptosis than the corresponding BAX WT dimer cells, resulting in more robust caspase activation (Figure 6B, 6C). Constistenly, the BAX P168G dimer cells showed slightly reduced potency and kinetics of caspase activation than BAX WT dimer cells. Further analysis of these cells after STS treatment by SEC of cytosolic and mitochondrial fractions showed that BAX E75K monomers, indeed underwent faster mitochondrial translocation and oligomerization as measured at 3 h (Figure 6D). However, the corresponding BAX WT dimer and BAX P168G dimer, consistent with their apoptosis induction, underwent slower mitochondrial translocation and oligomerization as cytosolic dimers were still present at 6h (Figure 6D). Taken together, the data indicate that when cytosolic BAX forms dimers, BAX activation and BAX-mediated apoptosis induction is hindered compared to the robust activation and apoptosis induction of the monomeric BAX.

Figure 6. Inactive BAX dimer regulates BAX activation and BAX-mediated apoptosis.

(A) Viability assays of DKO MEFs transiently transduced with retrovirus expressing human BAX WT or mutants as measured by annexin-V staining. P values <0.05 for BAX mutants compared to BAX WT. (B) Caspase 3/7 activation assays of DKO MEFs stably transduced with human BAX WT or mutants as measured by caspase 3/7 activation assays upon staurosporine (STS) treatment at 0.5 µΜ or 1 µΜ for 12h. (C) Caspase 3/7 activation assays of DKO MEFs stably transduced with human BAX WT or mutants upon 1 µΜ STS treatment at 3, 6 and 12h. (D) BAX WT and BAX P168G cells were treated with 1 µΜ STS for 6 h; BAX E75K cells were treated for 3 h due to faster cell death. Cytosolic and mitochondrial fractions were separated and analyzed by SEC. Data shown in A, B and C are mean ± SD from three independent experiments and in D are representative of three independent experiments with similar results. E) Autoinhibited dimeric BAX regulates the BAX activation pathway. Activation of the cytosolic BAX monomers (1) is initiated by BIM BH3 engagement of the α1/α6 trigger site (2), followed by discrete structural changes including α1–α2 loop displacement, 6A7 epitope exposure, BAX BH3 exposure, and α9 release (3). BAX is associated with the mitochondrial outer membrane that requires α9 release and exposure of the C-terminal BH3 pocket (4). Further BH3-mediated activation and BAX integration in the mitochondrial outer membrane into an undefined homo-oligomeric pore promotes release of mitochondrial apoptogenic factors such as cytochrome c (5). The autoinhibited dimeric BAX suppresses the initiation of structural changes in the cytosol by stabilizing the α1–α2 loop conformation, the 6A7 epitope and α9 in the inactive conformation (6). BH3-mediated interaction can disrupt the dimeric BAX to monomeric BAX before it induces the BAX activation pathway (7). See also Figure S6.

DISCUSSION

Pro-apopotic BAX is a critical BCL-2 famly member that, once deployed by BH3only proteins or other pro-apoptotic stimuli, becomes a conformationally active protein, which translocates to the mitochondrial outer membrane, oligomerizes and induces MOMP (Moldoveanu et al., 2014; (Walensky and Gavathiotis, 2011). Once BAX is activated and translocates to the mitochondria, it can be intercepted by anti-apoptotic BCL-2 proteins, which sequester the exposed BH3 domain of BAX (Chen et al., 2015; Czabotar et al., 2011; Dlugosz et al., 2006; Llambi et al., 2011) or additional surfaces required for conformational activation (Barclay et al., 2015; Ding et al., 2014; Ding et al., 2010).

Here, we identified an inactive dimer conformation of cytosolic BAX that its structure is consistent with the inhibition of either the N-terminal conformational change associated with the 6A7 epitope exposure of one protomer or the C-terminal conformational change associated with the mobilization of α9 and mitochondrial translocation for the other protomer. Regardless of the mechanism of BAX activation, both conformational changes at the N- and C-terminal surfaces are required for complete conformational activation of BAX leading to MOMP. Moreover, allosteric communication between the N- and C-terminal surfaces of the BAX structure was previously demonstrated (Gavathiotis et al., 2010). Therefore, our findings are consistent with a distinct regulatory step of cytosolic BAX that can hinder BAX activation and auto-translocation to the mitochondrial outer membrane (Figure 6E).

We found that the dimeric and monomeric conformations of BAX are in a reversible relationship, and that interaction with a BH3 domain can disrupt dimerization to release the BAX monomers, which subsequently can be activated and translocated to form the BAX oligomers. This is in agreement with previous studies that showed BAX translocation to the mitochondrial membrane as a monomer (Annis et al., 2005; Kim et al., 2009). We also found that increased amounts of a BH3 domain are required for the activation of the BAX dimer compared to the BAX monomer. The data suggest that the autoinhibited BAX conformation can regulate the degree of BAX activation and apoptosis induction by BH3-only proteins, serving as a safeguard mechanism to increase both the threshold for BAX activation and the amount of BAX activators required to commit the cell to apoptosis.

It is conceivable that formation of an inactive BAX dimer may support the function of BAX under homeostatic cell conditions. For example, previous studies proposed that cytosolic BAX exists in a dynamic equilibrium between the cytosol and mitochondria, supporting the deactivation and retrotranslocation of BAX from mitochondria back to the cytosol (Edlich et al., 2011; Schellenberg et al., 2013). Thus, the cytosolic BAX dimers may play a role in exchanging BAX monomers to the mitochondria and back and therefore regulate the response to pro-apoptotic stimuli. Consistent with this notion, leukemia cells with monomeric BAX showed a faster and more potent response to STS treatment, as evidenced by more complete BAX activation and oligomerization as well as caspase 3/7 activation, compared to MEFs that have dimeric BAX in the cytosol (Figure S6D, S6E). Along the same line, it is possible that cancer cells with more cytosolic BAX dimers will be more resistant to apoptotic insults including chemotherapeutic agents.

Our data reconcile previous works and show that cytosolic BAX monomers and dimers are feasible. We were unable to identify proteins bound to cytosolic BAX and the possibility of a cytosolic binding partner binding both the N-terminal and C-terminal surfaces of BAX is rather weak. However, the conditions that determine whether cytosolic BAX molecules will form dimers are likely cell-context dependent. The dissociation constant for the BAX dimer in vitro was estimated in the low µΜ range, 32 ± 10 µM (Figure S6F) which is consistent with the formation of the dimer in vivo based on the estimated average concentration of BAX levels in DU-145 cells, 3.16 ± 0.36 µM (Dussmann et al. 2010). Nevertheless, macromolecular crowding effect may change the local concentration of BAX and modulate the dimerization interaction in the cytosol (Zhou et al., 2008).

It is also conceivable that post translational modifications can affect BAX dimerization. For example, we found the BAX P168G dimers provide additional resistanse to BAX activation compared to wild type BAX both in vitro and in cells (Figure 5, 6 and S6C). Although P168G mutation does not affect the structure of BAX, our NMR data suggest that it reduces dynamics of the hinge region between α8–α9 loop (Figure S2). Previous reports described the P168 residue to be a target for Pin1 peptidyl-prolyl cis-trans isomerase, resulting in P168 cis-trans isomerization and inhibition of BAX (Shen et al., 2009). Also, mutations on P168 reduced the capacity of BAX to translocate and induce apoptosis (Schinzel et al., 2004). Our findings are consistent with these reports and suggest that BAX dimerization may be favored by the activity of Pin1 in specific cell-context. Interstingly, Pin1 overexpression is prevalent in human cancers (Lu & Hunter 2014), which may explain how certain cancer cells may promote cytosolic BAX dimerization. However, Pin1 can also function as a tumor suppressor, for example by promoting the apoptotic activity of p53.

In summary, we found that cytosolic BAX forms an autoinhibited dimer that regulates BAX activation. This mechanism of supressing BAX activation supports a model in which a cytosolic BAX conformation is selected for inhibiting or enabling apoptosis depending on the cellular context. The discovery of the cytosolic BAX dimers should help in elucidating mechanisms in homeostatic and pathological conditions of apoptosis and in therapeutic targeting approaches for BAX and anti-apoptotic BCL-2 proteins (Gavathiotis et al., 2012; Souers et al., 2013). Structural insights of the autoinhibited dimer offer important knowledge for the development of novel pharmacological modulators of BAX, which may engage either the N-terminal site or the C-terminal site to modulate dimerization and activation of BAX in diseases of deregulated apoptosis characterized by excessive cell death or survival.

EXPERIMENTAL PROCEDURES

Production of recombinant BAX

Human full-length BAX (1–192) wild type and mutants were generated in pTYB1 vector and expressed from BL21 (DE3) CodonPlus (DE3)-RIPL E. coli cells as described (Uchime et al. 2016) and detailed in the Supplemental Experimental Procedures.

Size-Exclusion Chromatography

Superdex 75 10/300 GL and 200 10/300 GL (GE Healthcare) columns were used for size exclusion chromatography of recombinant proteins and cell extracts. Recombinant protein was injected in columns equilibrated with a buffer containing 20 mM HEPES pH 7.2, 150 mM KCl, 1 mM DTT. 1mg of cytosolic or membrane extracts was applied to a Superdex 200 10/300 GL equilibrated in buffer containing 10 mM Tris, pH 7.5, 1 mM EGTA, 200 mM Sucrose and Complete Protease Inhibitors. Fractions of 500 µl were collected, and 30 µl of each fraction was analyzed by SDS-PAGE and immunoblotting. All operations were run at 4 °C. Gel filtration molecular weight markers (GE Healthcare) were also injected to the columns to obtain a standard curve for the estimation of the molecular weight of the proteins.

Crystallization

Homogeneous BAX protein in 20 mM HEPES pH 7.2, 150 mM KCl, 1 mM DTT buffer was concentrated to 10 mg/ml using a filtration unit (Milipore). Initial screening of crystallization conditions for BAX wild type and mutants were carried out by the sitting-drop vapor-diffusion method using 96-well Intelli plates (Hampton Research) at 293 K. Diffraction quality rod-shaped crystals were generated in 0.1 M Bis-Tris pH 6.5, 1.5 M ammonium sulfate via the hanging-drop vapor-diffusion method using 24-well VDX plates (Hampton Research). 1 µl of protein solution and reservoir solution were mixed and equilibrated against 1 ml of reservoir solution. Crystals were cryo-protected by soaking for 5 s in 20 µl of cryoprotectant solution containing 0.1 M Bis-Tris pH 6.5, 1.5 M ammonium sulfate and 25 % (v/v) glycerol, and flash-frozen in liquid nitrogen. Data collection and structure determination is described in the Supplemental Experimental Procedures.

BAX dimerization assays

Recombinant BAX monomer, wild type or mutants, was concentrated to 5–10 mg/ml in BAX buffer (10 mM HEPES pH 7, 150 mM KCl, 1 mM DTT) and incubated at various temperatures before diluting to a total volume of 250 µl and loaded onto a Superdex 75 HR 10/30 size-exclusion column (GE Healthcare). Dimerization was found to produce more consistently homogenous dimer at 20 °C compared to 37 °C albeit at a slower rate since at 37 °C some active dimer that proceeds to higher order oligomer is produced. Thus, to enable a consistent dimerization process and allow time for the active species to dimerize and oligomerize, we performed incubations at 20 °C incubating for at least 6 hours. For pH titration, the protein was concentrated to 5 mg/ml in 50 mM potassium phosphate buffer at pH 6, 7 or 8 and incubated at 20 °C before analysis using Superdex 75 HR 10/30 size exclusion column. BAX 4M (C62S, C126S, V121C, I136C) mutant was purified with 0.5 mM oxidised glutathione substituted into the SEC buffer. Samples were buffer exchanged into 10 mM HEPES pH 7.2 150 mM KCl, concentrated to 10 mg/ml and incubated at 20 °C before analysis by SEC using Superdex 75 HR 10/30 size exclusion column. Separation of monomeric and dimeric BAX was achieved using a flow rate of 0.5 ml/min at 4 °C. Protein standards (GE Healthcare) were used to calibrate the molecular mass of gel filtration peaks. Chromatogram traces are representative of several independent preparations of freshly SEC-purified monomeric BAX.

Liposomal permeabilization assays

Large unilamellar vesicles (LUVs) with lipid composition resembling the mitochondrial outer membrane were generated and entrapped with ANTS and DPX as described (Uchime et al. 2016) and detailed in the Supplemental Experimental Procedures.

Subcellular fractionation

To isolate cytosol and mitochondrial fractions, cells were lysed by Dounce homogenizer in lysis buffer LB containing 10 mM Tris, pH 7.5, 1 mM EGTA, 200 mM Sucrose and Complete Protease Inhibitors. The cell lysates were centrifuged at 700×g for 10 min to remove unlysed cells and nuclei. The supernatants were centrifuged at 12000×g for 10 min at 4°C and the resulting pellet was collected as the mitochondrial fraction. The membrane pellet was resuspended in LB + 1% CHAPS.

Cell culture, retroviral production and apoptosis assays

Reconstitution of BAX and BAX mutants into DKO cells was achieved by retroviral transduction of BAX–IRES–GFP, followed by MoFlo sorting for GFP-positive cells. Comparable expression of BAX WT and mutant proteins was confirmed by western analysis. For staurosporine treatment, MEFs were seeded in twelve-well clear-bottom culture plates for 16–18 hours and then treated with 1 µM staurosporine. For transient retroviral transduction of BAX or BAX mutant, DKO MEFs were infected with retrovirus expressing BAX or BAX mutant for 48 hours. Cell death was quantified by annexin-V (BioVision) staining. Flow cytometry was performed using a LSRFortessa (BD Biosciences) and data were analyzed using FACSDiva (BD Biosciences). The expression of BAX or BAX mutant was assessed by anti-BAX western blot. For caspase-3/7 activity assays, cells were seeded in 96-well plates overnight (1 × 10⁴ cells/well) in DMEM media at 37°C in a final volume of 100µL and then treated with staurosporine. Caspase-3/7 activity was assayed by addition of Caspase-3/7-Glo™ chemiluminescence reagent according to the manufacturer’s protocol (Promega) and luminescence measured using an Infinite F200 Pro microplate reader (Tecan). P values for statistical analyses were obtained using Student’s t test.