Minimalist Model Systems Reveal Similarities and Differences between Membrane Interaction Modes of MCL1 and BAK*

Background: BCL2 family protein interactions with and at mitochondrial membranes are poorly understood. Results: Fluorescence-based studies applied to minimalist model systems provide new insight into membrane activities of MCL1 and BAK under apoptotic-like conditions. Conclusion: Membrane interaction modes of MCL1 and BAK share particular features but also display important differences. Significance: BCL2 family protein function can be modulated at the mitochondrial membrane level through manifold mechanisms. Proteins belonging to the BCL2 family are key modulators of apoptosis that establish a complex network of interactions among themselves and with other cellular factors to regulate cell fate. It is well established that mitochondrial membranes are the main locus of action of all BCL2 family proteins, but it is difficult to obtain a precise view of how BCL2 family members operate at the native mitochondrial membrane environment during apoptosis. Here, we used minimalist model systems and multiple fluorescence-based techniques to examine selected membrane activities of MCL1 and BAK under apoptotic-like conditions. We show that three distinct apoptosis-related factors (i.e. the BCL2 homology 3 ligand cBID, the mitochondrion-specific lipid cardiolipin, and membrane geometrical curvature) all promote membrane association of BCL2-like structural folds belonging to both MCL1 and BAK. However, at the same time, the two proteins exhibited distinguishing features in their membrane association modes under apoptotic-like conditions. In addition, scanning fluorescence cross-correlation spectroscopy and FRET measurements revealed that the BCL2-like structural fold of MCL1, but not that of BAK, forms stable heterodimeric complexes with cBID in a manner adjustable by membrane cardiolipin content and curvature degree. Our results add significantly to a growing body of evidence indicating that the mitochondrial membrane environment plays a complex and active role in the mode of action of BCL2 family proteins.

known that under specific circumstances BCL2-type proteins can reverse their antiapoptotic phenotype to transform into BAX-type proapoptotic molecules (5). In addition, it remains debated whether all members within each BCL2 family subgroup share an identical mechanism of action.
From a structural point of view, BCL2-type and BAX-type proteins share a common all-␣ fold in solution that we will name the BCL2-like structural fold, comprising a primarily hydrophobic core helix surrounded by a bundle of six or seven amphipathic helices and their connecting loops. In addition, most multi-BH motif BCL2 family members contain a C-terminal segment dominated by hydrophobic residues, although they are not generally well conserved sequences. These C-terminal hydrophobic regions normally target multi-BH motif BCL2 family members to mitochondria and anchor them into the MOM; hence, they are regularly termed C-terminal membrane-anchoring (MA) or transmembrane domains (1,3,5). Nevertheless, evidence indicates that the C-terminal MA domain is not the sole region determining the targeting and anchoring into the MOM of BCL2 family proteins. As a prominent example, the BH3-only protein BID possessing a BCL2like structural fold but lacking a C-terminal MA domain efficiently translocates to and inserts into the MOM following apoptotic stimulation, with the mitochondrion-specific lipid CL playing a vital role in this process (6). In addition, it has long been recognized that BCL2-like structural folds of multi-BH motif BCL2 family members contain specific regions that can contribute to targeting and/or anchoring of these proteins into the MOM (7)(8)(9)(11)(12)(13).
BCL2 family proteins establish an intricate network of interactions among themselves and with many other cellular factors to regulate cell fate. Studies in membrane-free environments revealed a canonical protein-protein interaction mode between pairs of BCL2 family members (1). Here, an elongated hydrophobic groove present in BCL2-like structural folds of BCL2type and BAX-type proteins encompassing their BH1-BH3 motifs acts as the "acceptor site" for binding a "ligand" helix of a proapoptotic BCL2 family partner comprising its BH3 motif (1). Nevertheless, increasing evidence indicates that BCL2 family proteins can exhibit different interaction modes at the cytoplasm and at the MOM level (4). In addition, recent studies showed that many BCL2 family members are in a dynamic equilibrium between the cytosol and the MOM that depends upon the physiological status of the cell. Under non-apoptotic conditions, BAX, BAK, BCLXL, and MCL1 are constantly retrotranslocating from mitochondria into the cytosol; in response to apoptotic stress, retrotranslocation is stopped while all of these proteins accumulate at the MOM (13)(14)(15)(16)(17). How BCL2 family proteins retrotranslocate under non-apoptotic conditions is increasingly understood (14 -18), but the mechanisms leading to general BCL2 family protein accumulation at the MOM following a death stimulus are poorly understood. On top of this, it is also recognized that the mode of action of BCL2 family proteins can be modulated by compositional and structural features of the lipid bilayer portion of the MOM. However, we are only beginning to understand the precise influence of the MOM lipid bilayer environment on the mechanism of action of particular BCL2 family proteins (4, 19 -23).
Studying individual BCL2 family members or their mutual relationship at mitochondria in the course of apoptosis is a challenge because of the complex compositional nature of the MOM, which is further complicated by the fact that it can be spatially and dynamically connected to the mitochondrial inner membrane and to the endoplasmic reticulum, depending on the physiological status of the cell (22). During the last 2 decades, simple model systems that bypass cellular complexity and compositional diversity have provided powerful experimental means to obtain mechanistic insights into the apoptosis-regulatory mode of action of BCL2 family proteins (2, 6, 19 -27).
Here, we used minimalist model systems and a variety of fluorescence-based techniques to examine, side-by-side, membrane association and membrane level heterodimerization activities of BCL2-like structural folds belonging to MCL1 and BAK. Our results indicate that different apoptosis-related factors promote membrane binding of both proteins via mechanisms displaying common and distinguishing features. In addition, SFCCS and FRET analyses revealed that the BCL2-like structural fold of MCL1 forms a stable heterodimeric complex with cBID at the membrane level, whereas that of BAK does not. Furthermore, we report that MCL1⅐cBID heterodimerization is adjustable by membrane CL content and geometrical curvature. Altogether, our results support the notion that the function of BCL2 family members can be modulated at the mitochondrial membrane level through manifold mechanisms.
Protein Tryptophan (Trp) Fluorescence Measurements-Protein tryptophan fluorescence spectroscopy experiments were performed in an 8100 Aminco-Bowman luminescence spectrometer equipped with double-grating excitation and singlegrating emission monochromators (JobinYvon, Edison, NJ). The measurements were taken in 4 ϫ 4-mm quartz cuvettes. Trp fluorescence spectra were recorded by averaging 3-5 scans over a 300 -400 nm range at a scan rate of 1 nm/s, using an excitation wavelength of 295 nm. The slit widths for excitation and emission were kept at 4 nm. The contribution of buffer to sample fluorescence was subtracted as blank. Protein concentration was 300 nM.
Quasielastic Light Scattering Measurements-Quasielastic light scattering measurements were performed in a Malvern Zeta Sizer Nano ZS (Malvern Instruments, Malvern, UK). To analyze liposome size distribution, a helium-neon laser beam of 5 milliwatts was used, and the light scattered by the sample ( ϭ 633 nm) was detected with a photomultiplier placed perpendicular to the beam. Liposome size distribution was estimated based on particle mass (% volume) rather than scattering (% intensity).
Cytochrome c Release Assay-Mouse embryonic fibroblasts were kindly provided by Dr. Isabel Marzo (Universidad de Zaragoza, Zaragoza, Spain). Mouse embryonic fibroblasts were homogenized with the vertical passing of three rounds of 20, 15, and 10 strokes in a precooled 5-ml glass-Teflon Potter-Elvehjem homogenizer. Mitochondria-enriched fraction was obtained by differential centrifugation of the sample, using the following mitochondrial isolation buffer (MIB): 210 mM mannitol, 70 mM sucrose, 10 mM Hepes (pH 7.5), 1 mM EDTA, and protease inhibitors. Mitochondria were kept on ice and used within 2 h of preparation. Isolated mitochondria (1 mg protein/ ml) were incubated with cBID/cBID r (5 nM) alone or together with MCL1/MCL1 g (200 nM) for 30 min at 30 ºC in 125 mM KCl, 5 mM KH 2 PO 4 , 2 mM MgCl 2 , 1 mM DTT, and 10 mM HEPES-KOH, pH 7.2 (V final ϭ 50 l). Subsequently, reaction mixtures were centrifuged at 14,000 ϫ g for 10 min, and supernatant and pellet fractions were subjected to 15% SDS-PAGE and immunoblotting using anti-cytochrome c 7H8.2C-12 antibody.
Equilibrium Sucrose Gradient Centrifugation of Liposome/ Protein Mixtures-Proteins (250 nM) were first co-incubated with or without SUV/LUV (250 M) in KHE buffer for 30 min at 25°C. In the alkali extractability experiments, protein-liposome mixtures were then incubated in 100 mM Na 2 CO 3 (pH 11.5) for 30 min on ice. Samples were then adjusted to 1.4 M sucrose and loaded at the bottom of a discontinuous sucrose gradient containing a medium layer of 0.8 M sucrose and an upper layer of 0.5 M sucrose. Next, samples were centrifuged at 100,000 rpm for 3 h in a Beckman Optima TLX Benchtop ultracentrifuge using a TLA 120.2 rotor, followed by collection of four 250-l fractions. Samples were subsequently subjected to reducing SDS-PAGE on 15% gels, followed by on-gel fluorescence visualization using a Molecular Imager Versadoc (Bio-Rad). To identify liposome-containing fractions of the gradient, liposomes were prepared with 0.5 mol % L-␣-phosphatidylethanolamine-N-(lissaminerhodamine B sulfonyl). In all cases, at least 85% of L-␣-phosphatidylethanolamine-N-(lissaminerhodamine B sulfonyl) fluorescence was found at the uppermost two fractions of the gradient. Thus, we considered that the top two fractions correspond to the liposomecontaining fractions of the gradient, whereas the bottom two fractions correspond to the liposome-free fractions of the gradient.
Confocal Microscopy Fluorescence Cross-correlation Spectroscopy (FCCS) Analysis-For microscopy observations of protein recruitment to GUV, fluorescently labeled proteins and liposomes were incubated together for 2 h at 22°C in an observation chamber previously blocked with BSA to prevent attractive interactions between the vesicles and the glass substrate. Images were recorded in an inverted confocal fluorescence microscope (Nikon DECLIPSE C1, Nikon Inc., Melville, NY) with a total internal reflection fluorescence ϫ60 oil immersion objective. The excitation wavelengths used were 488, 561, and 635 nm, and emitted fluorescence was recorded using band pass filters of BP515, BP593, and a long pass filter of LP650, respectively. Fluorescence images were processed with ImageJ software using the plug-in "Radial profile," measuring integrated intensities along concentric circles centered at the middle of the GUV. Next, the measured intensity corresponding to the protein bound to the GUV membrane was normalized to the values obtained in solution.
FCCS measurements were performed at 22°C using a Con-foCor 3 module with attenuated excitation light from argon ion (488 nm) and helium-neon lasers (633 nm), and emitted fluorescence was recorded using a band pass filter of BP530 and a long pass filter of LP655, respectively. For solution FCCS experiments, we first calibrated the size of the focal volume using free Alexa488 and Alexa633 dyes. MCL1 g and cBID r were incubated in the observation chamber for 1 h before carrying out FCCS measurements. To obtain auto-and cross-correlation curves, raw fluorescence fluctuation data were fitted to a three-dimensional diffusion model with homemade software. For two-focus SFCCS measurements in GUV, photon arrival times were recorded with a Flex 02-01D/C hardware correlator, and the data were also analyzed with homemade software. Here, once the microscope had been aligned, the detection volume was repeatedly scanned perpendicular across the GUV equator in two parallel lines (the distance between the two lines, d, was measured by photobleaching on a film of dried fluorophores). From these two traces, the focal volume can be determined, avoiding the need for calibration. The measurement time was 300 s, and the bin time was 2 s. Fluorescence intensity data were arranged as a matrix such that every row corresponded to one line scan. The rows were aligned to correct for membrane movements by calculating the maximum of a running average over several hundred line scans and shifting it to the same column. The autocorrelation and spectral and spatial cross-correlation curves were computed from the intensity traces, and irregular curves resulting from instability and distortion were excluded from the analysis. We fitted the auto-and cross-correlation functions with a nonlinear least-squares global fitting algorithm (2Dimensions 2Focus 2Color). As a result of the anal-ysis of FCCS data, surface concentrations of single color particles (C r and C g ) and two-color particles (CC rg ) as well as their diffusion coefficients were determined. FCCS results were also corrected for fluorescence cross-talk and for protein labeling degrees (90% for MCL1 g , 80% BAK g , 70% for cBID r , and 100% for cBID r D95A). Complex % values were calculated with respect to the amount of the green particles (CC rg ϫ 100/C g ), being the percentage value between CC rg two-color particles (the green and red complexed molecules) and the total green particles (the free and the complexed ones).
FRET Measurements-FRET experiments were performed in an 8100 Aminco-Bowman luminescence spectrometer. Briefly, 100 nM NBD-MCL1/BAK (donor) and 400 nM Rho-cBID (acceptor) were incubated with and without liposomes (200 M lipid) for 30 min at room temperature in the dark. Fluorescence spectra were recorded by averaging 3-6 scans over a 510 -620 nm range at a scan rate of 1 nm/s, using an excitation wavelength of 465 nm for NBD. The slit widths for excitation and emission were 8 and 4 nm, respectively. We prepared four samples: B (blank, without donor or acceptor dyes), D (donor-containing), A (acceptor-containing), and DA (donor-and acceptor-containing). Subtraction of the B signal from that of D corrects for the significant light scattering signal and yields the net donor spectrum, whereas subtraction of the A signal from that of DA corrects for both scattering and any signal due to direct excitation of the acceptor, thereby yielding the net donor ϩ acceptor spectrum. FRET signals were estimated as described previously (28), by calculating ratios of net R DϩA (i.e. ratio of fluorescence of net donor ϩ acceptor spectrum signal at 580 nm and of fluorescence of donor ϩ acceptor spectrum signal at 540 nm) and net R D (i.e. ratio of fluorescence of net donor spectrum signal at 580 nm and of fluorescence of net donor spectrum signal at 540 nm).

Results
Design of Experimental Systems-Recombinant monocysteine versions of BCL2-like structural folds belonging to MCL1, BAK, and the apoptogenic form of BID (cBID) were site-specifically labeled with the green (g) Alexa488 fluorophore or the red (r) Alexa647 fluorophore to generate fluorescent MCL1 g , BAK g , and cBID r variants, respectively (Fig. 1A). We examined whether the mutagenesis and/or labeling procedures altered protein global structure or functionality. As shown in Fig. 1B, intrinsic fluorescence spectra of MCL1 g , BAK g , and cBID r were virtually indistinguishable from those of native protein counterparts. The three fluorescently labeled variants also retained the ability of their parent unlabeled proteins to modulate membrane permeability, as revealed by an assay measuring internalization of Rho-10-kDa dextrans into 30% CL GUV (Fig. 1C). In addition, we examined the ability of fluorescently labeled MCL1 and cBID variants to regulate mitochondrial cytochrome c release. As shown in Fig. 1D, cBID r released cytochrome c as effectively as cBID, whereas MCL1 g abolished this process akin to MCL1. Altogether, this set of experiments demonstrates that our fluorescently labeled BCL2 variants preserve basic structural and functional properties of their native protein counterparts.
As model membrane systems for our mechanistic studies, we used liposomes of variable sizes (SUV, LUV, and GUV) containing different molar percentages of CL (0% CL, 4% CL, 14% CL, 30% CL, and 100% CL). Due to their small size and high geometrical curvature (Fig. 1E), we consider SUV-type liposomes morphological surrogates of MOM-mitochondrial inner membrane contact sites (CS) and other curved lipid surfaces thought to be present at specialized regions of the MOM, such as mitochondrial membrane fission sites and MOM-endoplasmic reticulum junctions. Of note, evidence indicates that the number of MOM-mitochondrial inner membrane CS, mitochondrial membrane fission sites, and MOM-endoplasmic reticu-lum junctions increases during apoptosis (29 -31). By contrast, LUV-type and GUV-type liposomes display negligible membrane curvature at the protein scale due to their large size, thereby emulating the flat lipid surface thought to predominate at the MOM under non-apoptotic conditions. Regarding the lipid compositions chosen, we consider that (i) 0 -4% CL represents the average CL content of the MOM under normal conditions, (ii) 14 -30% CL reflects the CL content present at mitochondrial CS, and (iii) 100% CL emulates CL microdomains that may be formed during the apoptotic process at localized areas of the MOM (32). The latter are not unprecedented, because CL microdomains have been described in bacterial FIGURE 1. MCL1 g , BAK g , and cBID r maintain structural and functional properties of native unlabeled counterparts. A, three-dimensional structures of MCL1⌬N151⌬C23 (PDB code 1WSX), BAK⌬C21 (PDB code 2IMS), and BID (PDB code 1DDB) displaying as colored spheres the monocysteine residue where the Alexa-fluorophore is conjugated to generate MCL1 g , BAK g , and cBID r , variants used in this study. Scissors, caspase-8 cleavage site (Asp 59 2Gly 60 ) in the cBID structure. B, Trp fluorescence spectra of MCL1 g , BAK g , and cBID r variants and their wild-type, unlabeled counterparts (n ϭ 3). C, percentage of Rho-10-kDa dextran-permeabilized GUV in the presence or absence of the indicated BCL2 proteins. Mean values Ϯ S.D. (error bars) correspond to two independent experiments. D, effect of BCL2 proteins on the release of cytochrome c from mitochondria isolated from mouse embryonic fibroblasts. E, liposome size distribution analyzed by quasielastic light scattering for 30% CL SUV (red trace) and 30% CL LUV (black trace). IF, intensity of fluorescence; a.u., arbitrary units; WB, Western blot. membranes (33). Of note, although the role of CL during apoptosis is still unsettled, recent reports indicate that CL and/or its derivatives accumulate at the MOM early in the course of apoptotic cell death (34,35). It should also be taken into account that the BCL2-like structural folds of MCL1, BAK, and cBID are primarily oriented toward the cytosol-facing leaflet of the MOM, which should be enriched in CL relative to the intermembrane-facing leaflet of the MOM in accord with the asymmetrical interleaflet distribution of negatively charged lipids commonly found in biomembranes (32).
cBID, CL, and Curvature Promote Membrane Association of MCL1 g -In a first set of experiments, we evaluated whether MCL1 g ⅐membrane binding is affected by three different apoptosis-related factors: (i) the MCL1 ligand cBID, (ii) the mitochondria lipid trademark CL, and (iii) membrane geometrical curvature. To this aim, we incubated liposomes of variable sizes (SUV, LUV, and GUV) and CL contents with MCL1 g alone, cBID r alone, or MCL1 g plus cBID r .
To begin with, SUV-type or LUV-type liposomes were used as model membrane systems. After incubating the protein(s) with the vesicles, liposome-containing and liposome-free fractions were separated by equilibrium sucrose gradient ultracentrifugation, and the protein contents of each fraction were quantified by SDS-PAGE and fluorescence intensity analysis. On the one hand, in the absence of cBID r , we found minimal amounts of MCL1 g bound to MOM-like SUV or LUV containing 0 -4% CL (Fig. 2, A and B). In contrast, appreciable amounts of MCL1 g alone bound to mitochondrial CS-like SUV or LUV containing 14 -30% CL, whereas virtually all MCL1 g associated by itself with pure CL SUV or LUV (Fig. 2, A and B) (data not shown). Of note, much more MCL1 g bound on its own to SUV containing 14 -30% CL than to LUV containing 14 -30% CL (Fig. 2, A and B). Thus, high membrane geometrical curvature directly promotes MCL1 g association with liposomes containing CL levels present at mitochondrial CS. On the other hand, the addition of cBID r further stimulated MCL1 g binding to mitochondrial CS-like SUV and LUV (Fig. 2, A and B). cBID r also enhanced MCL1 g binding to MOM-like SUV but not to MOM-like LUV. Of note, the capacity of cBID r to stimulate MCL1 g binding to the liposomes correlated with the ability displayed by cBID r alone for binding to the vesicles.
We next analyzed MCL1 g binding to GUV-type liposomes containing increasing CL contents in the absence or presence of cBID r . Here, vesicles were incubated with the protein(s), and then the fluorescence intensity increase at the rim of the GUV membrane of multiple individual vesicles was visualized by confocal fluorescence microscopy and quantified. In qualitative agreement with results obtained with low curvature LUV-type liposomes, MCL1 g by itself only bound successfully to pure CL GUV. Furthermore, cBID r enhanced MCL1 g binding to mitochondrial CS-like GUV but not to MOM-like GUV, correlating with the intrinsic capacity of cBID r for binding the former but not the latter type of liposome, which also agrees with the behavior displayed by these proteins in the low curvature LUV system (Fig. 2, B-D). In summary, this set of results demonstrates that three distinct apoptosis-related factors (cBID, CL, and curvature) promote membrane association of the BCL2like structural fold of MCL1.

Different Membrane Association Modes of MCL1 g Revealed by Mutagenesis and Alkali Extractability-Solution-based
binding studies revealed a canonical protein-protein interaction mode between BH3-only proteins and BCL2-type or BAXtype proteins based on engagement of the BH3 domain of the proapoptotic ligand into an elongated hydrophobic groove present at the surface of the BCL2-type or BAX-type receptor (1,3). However, it is currently debated whether this canonical BH3-into-groove interaction mechanism is the exclusive mediator of functional interactions between different BCL2 family members at the MOM environment (4). To gain more insight into how cBID r stimulates MCL1 g membrane association in our reconstituted minimalist system, we evaluated the impact of cBIDD95A and MCL1 R244E point mutations that disrupt a critical intermolecular salt bridge according to the canonical BH3-into-groove interaction mechanism (1) (Fig. 3A).
Selective disruption of this protein-protein interaction mode through mutagenesis abolished cBID-mediated stimulation of MCL1 binding to 0% CL SUV (Fig. 3B), 30% CL LUV (Fig. 3C), and 30% CL GUV (Fig. 3D). By contrast, the cBID D95A and MCL1 R244E point mutations had minimal impact on MCL1 binding to 30% CL SUV (Fig. 3B), 100% CL LUV (Fig. 3C), and 100% CL GUV (Fig. 3D). These results together with those described in Fig. 2 indicate that MCL1 g can associate with liposomal membranes in two broadly distinct manners: (i) by engaging to membrane-bound cBID through the same canonical BH3-into-groove interaction mechanism described in solution binding studies or (ii) by directly binding to the liposomal lipid bilayer in a manner that depends on its CL content and geometrical curvature.
We also wished to discriminate whether MCL1 g is loosely bound to or firmly embedded into the liposomal membrane under different conditions. To this aim, we analyzed the alkali extractability of MCL1 g from SUV or LUV containing different amounts of CL in the presence or absence of cBID r (Fig. 3E). Upon alkali treatment, the majority of MCL1 g was extracted from mitochondrial CS-like LUV, whereas a substantially lower fraction of MCL1 was extracted from mitochondrial CS-like SUV, and minimal MCL1 g was extracted from pure CL vesicles (Fig. 3E). In all cases, cBID r had little effect in MCL1 g membrane extractability. These data suggest that CL and curvature, but not cBID, promote membrane insertion of MCL1 g .

MCL1 g and cBID r Form Stable Heterodimeric Complexes in CS-like GUV but Not in
Pure CL GUV-We next wished to analyze the ability of MCL1 to heterodimerize with cBID in solution and in the context of an intact lipid bilayer membrane environment. To this aim, we used FCCS and its variant SFFCS (36), which allowed us to obtain quantitative information of diffusion coefficients (D), concentrations, and complex formation between two spectrally different labeled species from the analysis of their autocorrelation and cross-correlation curves.
In a membrane-free environment, a fairly small cross-correlation curve for the MCL1 g ⅐cBID r complex was observed (Fig.  4A, blue line), with a low percentage of complex formation (ϳ7%) estimated at the highest protein concentrations tested (Fig. 4C). Nevertheless, when MCL1 g was incubated with the cBID r D95A mutant under the same conditions, the extent of cross-correlation decreased to background levels, suggesting that the weak MCL1 g ⅐cBID r complex formation occurred through a canonical BH3-into-groove interaction mechanism (Fig. 4).
Next, we evaluated MCL1 g ⅐cBID r heterodimerization at the level of the GUV membrane by SFCCS. To this aim, we first used mitochondrial CS-like GUV containing 30% CL, because we have previously shown that cBID r strongly stimulates MCL1 g binding to this type of liposome (Fig. 2D). Indeed, the good quality of MCL1 g and cBID r autocorrelation curves obtained in 30% CL GUV confirmed that MCL1 g and cBID r FIGURE 2. cBID r , CL, and curvature promote MCL1 g binding to liposomal membranes. A and B, SUV and LUV recruitment assay. Left-hand panels, Proteins were incubated for 30 min in the absence (SOL) or presence of SUV-or LUV-type liposomes containing different amounts of CL, followed by sucrose-gradient centrifugation to separate membrane-free fractions (Sol.) and membrane-containing fractions (Memb.) and analysis by SDS-PAGE and fluorescence detection for MCL1 g (green) and cBID r (red) bands. Right-hand panels, quantitation of binding of indicated proteins to SUV-or LUV-type liposomes containing different amounts of CL. Protein and lipid concentrations were 250 nM and 250 M, respectively. Data correspond to mean values Ϯ S.D. (error bars) for at least two independent experiments. C and D, GUV recruitment assay. C, indicated proteins were incubated with GUV containing different amounts of CL followed by analysis of samples by confocal fluorescence microscopy. Scale bar, 10 m. D, from the confocal fluorescence images, the ratio of maximum normalized integrated intensity values in membrane (IFmemb.) and solution (IFsol.) fluorescence obtained from radial profiles was measured for MCL1 g (100 nM) and cBID r (20 nM) in GUV of the indicated lipid compositions. In this box chart and raw data (dots) representation, the box represents the 96% confidence interval; inside the box, the media and median are represented by the small square and the line, respectively; and the errors correspond to 80% of the data. For each condition, at least 10 GUV were analyzed from 2-3 independent experiments. efficiently localize to the membrane of these vesicles (Fig. 5A,  left, green and red lines). Importantly, SFCCS analysis showed that MCL1 g and cBID r form stable complexes at the membrane of 30% CL GUV, reflected by the large positive amplitude of the cross-correlation curve (Fig. 5A, left, blue line). To try estimating the binding affinity between the two proteins at the membrane of these GUV, we analyzed MCL1 g ⅐cBID r complex formation at a wide range of MCL1 g and cBID r concentrations (Fig. 5A, three-dimensional plot at the right). Elevated values of complex formation were obtained at all protein concentrations, indicating that the affinity between membrane-localized MCL1 g and cBID r in CS-like GUV is so high that MCL1 g ⅐cBID r complex formation is always saturated. We also sought to determine whether MCL1 g ⅐cBID r complex formation in this type of liposome is reversible. To this aim, we added an excess of unlabeled cBID to 30% CL GUV that had been previously co-incubated with MCL1 g and cBID r (Fig. 5B). Adding an excess of unlabeled cBID drastically decreased the amplitude of the MCL1 g ⅐cBID r cross-correlation curve and diminished the percentage of MCL1 g ⅐cBID r complex forma-  2KBW). B and C, effect of MCL1 R244E and cBID r D95A mutations on protein binding to SUV-type or LUV-type vesicles containing different amounts of CL. cBID r D95A recruitment to liposomes was assessed as described in the legend to Fig. 2, A and B, whereas MCL1 R244E recruitment was assessed by immunoblotting and densitometric quantitation. D, effect of cBID r D95A on MCL1 g recruitment to GUV. MCL1 g and cBID r /cBID r D95A concentrations were 100 and 5 nM, respectively (30% CL GUV assays) and 2 and 1 nM, respectively (100% CL GUV assays). Other conditions were as described in the legend to Fig. 2. E, alkali extractability assay. Proteins preincubated with liposomes were treated with (ϩAlkali) or without (ϪAlkali) 100 mM Na 2 CO 3 , pH 11.5, and protein partition was assessed as described in Fig. 2. Error bars, S.E.

Comparison of Membrane Interaction Modes of MCL1 and BAK
tion to basal levels. From these experiments, we conclude that membrane-localized MCL1 g and cBID r form high affinity, stable, and reversible complexes in mitochondrial CSlike GUV.
Next, the same type of experiments were performed in 100% CL GUV. Here, also as expected, MCL1 g and cBID r efficiently localized to the GUV membrane, reflected by the good quality of MCL1 g and cBID r autocorrelation curves (Fig. 5C, left, green

. SFCCS analysis indicates that membrane-localized MCL1 g and cBID r form stable heterodimeric complexes in 30% CL GUV but not in 100% CL GUV.
A-D, two-focus SFCCS analysis of MCL1 g ⅐cBID r complex formation in CL-containing GUV. Different concentrations of MCL1 g and cBID r were incubated with GUV containing 30% CL in the absence (A) or presence of unlabeled cBID (B). In experiments with 100% CL GUV, MCL1 g was incubated with cBID r (C) or with cBID r D95A (D). Left-hand panels, raw data and fitted auto-and cross-correlation curves for the indicated protein and GUV combinations. In the right-hand panels, percentages of complex formation (Complex %) were represented on a three-dimensional plot as a function of individual protein concentrations (molecules/m 2 ). E, quantification of the effect elicited by unlabeled cBID or cBID r D95A on MCL1 g ⅐cBID r complex formation in GUV of the indicated lipid compositions. Data correspond to 3-6 independent experiments, with more than 40 GUV analyzed for each condition. F, diffusion coefficients (D, m 2 /s) of MCL1 g and cBID r at the different conditions analyzed. The box chart and raw data (dots) representations are as described in Fig. 2D. Data were obtained from 3-7 independent experiments, with n ϭ 61 for 30% CL GUV and n ϭ 63 for 100% CL GUV. In E and F, *, **, and ****, p ϭ 0.05-0.01, 0.01-0.001, and Ͻ0.0001, respectively. Error bars, S.D. and red lines). Surprisingly, in this case, the amplitude of the cross-correlation curve was negligible (Fig. 5C, left, blue line). Thus, despite MCL1 g and cBID r effectively localizing to the membrane of pure CL GUV, the two proteins do not stably heterodimerize therein. Analyzing cross-correlation percentages in 100% CL GUV treated with a wide range of protein concentrations revealed a measurable amount of MCL1 g ⅐cBID r complex formation at the highest protein concentrations tested (Fig. 5C, three-dimensional plot at the right). The MCL1 g ⅐cBID r complexes detected at such high protein concentrations were not an artifactual result, because the cross-correlation percentage returned to background levels when cBID was substituted by the MCL1 binding-defective cBID r D95A variant (Fig. 5, D and  E).
Finally, from the autocorrelation and cross-correlation curves, we estimated D and hydrodynamic radius (R H ) values for each protein. In solution, D values for cBID r , cBID r D95A , and MCL1 g were 92.5 Ϯ 8.7, 91.9 Ϯ 8.3, and 113.8 Ϯ 6.7 m 2 /s, respectively, with corresponding R H values of 2.37 Ϯ 0.2, 2.38 Ϯ 0.2, and 1.92 Ϯ 0.1 nm, respectively. These values are in accordance with molecular sizes determined for cBID and for MCL1 by NMR spectroscopy and x-ray crystallography (Fig. 1A). In 30% CL GUV, MCL1 g and cBID r displayed virtually identical D values around 5.7 m 2 /s, whereas MCL1 g ⅐cBID r complexes showed significantly lower D values (4.3 Ϯ 0.8 m 2 /s) (Fig. 5F). These D values also appear reasonable for membrane-associated proteins and protein complexes. Remarkably, MCL1 g displayed substantially lower D values in 100% CL GUV than in 30% CL GUV (Fig. 5F). These data suggest that in the presence of membrane-bound cBID r , MCL1 g adopts different conformations in mitochondrial CS-like GUV and in pure CL GUV, which is in accord with the results obtained in alkali extractability experiments using equivalent LUV-type liposomes (Fig. 3E). Nevertheless, the possibility cannot be excluded that the alterations in D could be due to changes in membrane viscosity induced by different CL contents.
Minimalist Systems Reveal Similarities and Differences between Membrane Interaction modes of BAK g and MCL1 g -Next, we examined the capacity of the BCL2-like structural fold of BAK for membrane association and for heterodimerization with cBID, using the same experimental strategy described above for MCL1.
First, we quantitatively analyzed the binding of BAK g to high curvature SUV and to low curvature GUV containing different amounts of CL, in the presence and absence of cBID. BAK g on its own bound quite efficiently to pure CL SUV/GUV, more modestly to mitochondrial CS-like SUV, and insignificantly to CS-like GUV and to MOM-like SUV/GUV (Fig. 6, A-C) (data not shown). Thus, high membrane CL content and curvature promote binding of BAK g on its own to liposomal membranes, which is in qualitative agreement with the behavior observed with MCL1 g . Furthermore, cBID r stimulated BAK g binding to MOM-like SUV and to CS-like SUV/GUV but not to MOMlike GUV, also matching the behavior observed with MCL1 g (compare Figs. 6 and 2).
Further experiments using heterodimerization-defective mutants of cBID and BAK supported the implication of a canonical BH3-into-groove interaction mechanism in cBID r -mediated stimulation of BAK g ⅐liposome binding, as observed with MCL1 g (Fig. 7, A-C). Nevertheless, the behaviors of BAK g and MCL1 g were not identical because the cBID D95A mutation inhibited to a lower degree BAK g ⅐liposome binding relative to MCL1 g ⅐liposome binding (compare Fig. 7, B and C, with Fig.  3, B-D). Alkali extractability experiments revealed another clear distinction between the membrane-interacting properties of BAK g and MCL1 g, because BAK g resisted alkaline extraction under all conditions examined (Fig. 7D).
BAK g and cBID r Do Not Form Stable Heterodimeric Complexes in CS-like GUV or in Pure CL GUV-Next, we analyzed BAK g and cBID r heterodimerization at the membrane level using SFCCS. As shown in Fig. 8, minimal BAK g ⅐cBID r heterodimerization was observed either in 30% CL GUV or in 100% CL GUV. Thus, despite the fact that cBID r efficiently recruits BAK g to the membrane of these two types of GUV, BAK g and cBID r do not form stable heterodimeric complexes therein.
We also estimated D values for BAK g associated with either 30% CL GUV or 100% CL GUV in the presence of cBID r . Similar D values were obtained for BAK g in both types of liposomes (Fig. 8D). These results together with those obtained in alkali extractability assays (Fig. 7D) suggest, but do not prove, that BAK g adopts similar conformations in mitochondrial CS-like membranes and in pure CL membranes, which is unlike the behavior observed with MCL1 g .
FRET-based Analysis of MCL1⅐cBID and BAK⅐cBID Heterodimerization at the Membrane Level-We performed FRET experiments to further evaluate the ability of BCL2-like structural folds of MCL1 and BAK to form stable heterodimeric complexes with cBID at the membrane level. To this aim, the same monocysteine MCL1/BAK (donor) and cBID (acceptor) variants used in the previous experiments were labeled either with NBD (donor) or with Rho (acceptor) fluorescent dyes. The fluorescenceemissionspectrumofNBDoverlapswiththeabsorbance spectrum of Rho, making these two fluorophores a convenient donor-acceptor pair with a relatively large R 0 of ϳ6 nm (37). We incubated donor and donor plus acceptor samples in solution or with SUV/LUV containing different CL amounts, followed by monitorization of the NBD fluorescence emission spectrum. FRET manifests as a decrease in donor emission (ϳ540 nm) and an increase in acceptor emission (ϳ580 nm) and can be quantified by calculating the ratio of donor plus acceptor fluorescence to donor fluorescence alone (28). A similar experimental strategy has been successfully used before to detect BAX⅐cBID heterodimerization (26).
Among all samples examined, the most prominent FRETbased interaction signal between MCL1-NBD and cBID-Rho corresponded to the mixture containing 30% CL LUV (Fig. 9, A  and B). This is consistent with the robust MCL1 g ⅐cBID r crosscorrelation signal observed in 30% CL GUV by SFCCS (Fig. 5A). By contrast, insignificant FRET was detected for the NBD-MCL1⅐Rho-cBID pair in the presence of 0% CL LUV (Fig. 9, A  and B), which is expected because MCL1 g and cBID r did not bind to this type of liposome according to the liposome float-up assay (Fig. 2B). Minimal FRET signal was also detected when NBD-MCL1 was mixed with Rho-cBID plus 100% CL LUV/ SUV (Fig. 9B), in agreement with FCCS-based data obtained with 100% CL GUV (Fig. 5C). Importantly, substantial FRET signal was obtained in mixtures containing 0% CL SUV or 30% CL SUV, indicating that NBD-MCL1 and Rho-cBID assemble into stable heterodimeric complexes in MOM-like liposomes and in mitochondrial CS-like liposomes possessing high membrane geometrical curvature (Fig. 9, A and B). Of note, the degree of FRET was clearly inferior in 30% CL high curvature SUV to that in 30% CL low curvature LUV (Fig. 9, A and B).
Last, we used this FRET-based assay to evaluate the interaction between the BCL2-like structural fold of BAK and cBID in the presence of different types of liposomes. In stark contrast with results obtained with the NBD-MCL1⅐Rho-cBID pair, virtually no FRET signal was detected for the NBD-BAK⅐Rho-cBID pair under any condition examined (Fig. 9, C and D). These results strongly suggest that NBD-BAK does not form stable complexes with Rho-cBID, irrespective of membrane CL content and geometrical curvature.

Discussion
Understanding the role of the MOM environment in the function of the BCL2 protein family is one of the most demanding and challenging tasks of current apoptosis research (1)(2)(3)(4). In this work, we used multiple fluorescencebased techniques and minimalist model systems to advance our understanding of MCL1 and BAK membrane activities. We report that BCL2-like structural folds of MCL1 and BAK lacking the MA domain display membrane interaction FIGURE 6. cBID r , CL, and curvature promote BAK g binding to liposomal membranes. A, BAK g alone (250 nM) or together with cBID r (250 nM) was incubated in the absence (SOL.) or presence of SUV-type liposomes (250 M) containing different amounts of CL. Other conditions were as explained in the legend to Fig.  2A. B, representative images of BAK g recruitment to GUV containing different amounts of CL, in the absence or presence of cBID r . BAK g and cBID r concentrations were 200 and 50 nM, respectively (30% CL GUV assays), and 200 and 20 nM (100% CL GUV assays). Other conditions were as explained in Fig. 2C. C, from the confocal images, the extents of BAK g and cBID r binding to GUV were measured as described in the legend to Fig. 2D. Error bars, S.D. modes sharing certain similarities but also displaying important differences.
Recent studies indicate that the localization of multiple BCL2 family proteins (including MCL1 and BAK) at the MOM is more dynamic than previously anticipated and depends on the physiological status of the cell, with these proteins continuously being retrotranslocated from the MOM into the cytosol under healthy conditions, whereas apoptotic stimulation leads to general accumulation of BCL2 family proteins at the MOM (1,3,4). It is becoming clear that retrotranslocation processes taking place under non-apoptotic conditions are governed by heterodimerizing interactions between BCL2-type and BAX-type proteins involving their C-terminal MA domains (14 -18). However, how BCL2 family proteins become generally accumulated at the MOM downstream of apoptosis triggering remains poorly understood, although different lines of evidence indicate that BCL2-like structural folds of multi-BH motif BCL2 family proteins can be implicated in this process (9,11,12,13). Here, we used model membrane systems to evaluate whether three distinct apoptosis-related factors (cBID, CL, and membrane geometrical curvature) affect membrane recruitment of BCL2-like structural folds belonging to MCL1 and to BAK as well as to gain more mechanistic insight into each one of these processes.
Regarding cBID, we found that this apoptogenic BH3-only protein promotes membrane recruitment of the BCL2-like structural fold of MCL1 through a canonical BH3-into-groove interaction mechanism, which has been extensively characterized in solution-based studies (1). A similar, although apparently not identical, protein-protein interaction mechanism also accounts for cBID-mediated membrane recruitment of the BCL2-like structural fold of BAK. These data are consistent with the "membrane-embedded" model for BCL2 protein family action stating that upon apoptosis triggering, mitochondriaassociated BH3-only proteins can act as receptors for recruiting BAX-type and BCL2-type proteins into the MOM (4,25,26). Of note, we also showed that interaction with cBID r triggers membrane insertion of BAK g but not MCL1 g . This is in contrast with FIGURE 7. Different membrane association modes of BAK g . A, structural representation of BID SAHB BH3 domain (red) bound to BAK⌬C21 (green), highlighting as spheres cBID Asp 95 and BAK Arg 127 residues implicated in BID BH3-BAK groove interaction (PDB code 2M5B). B and C, effect of cBID r D95A and BAK R127E mutations on protein binding to liposomes containing different amounts of CL. cBID r D95A recruitment to SUV and GUV was assessed as described in Fig. 3, B-D, whereas BAK R127E recruitment to liposomes was assessed by immunoblotting and densitometric quantification. BAK g and cBID r concentrations were 250 and 250 nM, respectively, in assays with SUV, 200 and 50 nM in assays with 30% CL GUV, and 200 and 20 nM in assays with 100% CL GUV. D, alkali extractability assay. Proteins incubated with liposomes were treated with (ϩAlkali) or without (ϪAlkali) (100 mM Na 2 CO 3 , pH 11.5), and their partition was assessed as described in Fig. 2, A and B. Error bars, S.D. JULY 3, 2015 • VOLUME 290 • NUMBER 27 another prediction of the "membrane-embedded" model stating that interaction with BH3-only proteins triggers membrane embedding of the ␣5␣6 region of BCL2-type proteins, which subsequently functions as a non-canonical surface for binding to and inhibiting BAX-type proteins (4). The latter proposal originated from observations made with BCL2 (11), but it remains to be proven that this phenomenon can be extended to BCL2-type proteins other than BCL2 itself.

Comparison of Membrane Interaction Modes of MCL1 and BAK
Concerning CL, we showed that this mitochondrion-specific lipid also promotes liposome association of BCL2-like structural folds corresponding to both MCL1 and BAK. Although the physiological relevance of the latter set of results can be put into question due to the high CL amounts required to observe such effects, recent studies indicate that CL and its derivatives become enriched at the MOM early during the apoptotic process (34,35). Interestingly, further evidence indicates that lipids other than CL accumulate at the MOM during apoptosis and interact selectively with specific BCL2 family members (22).
Last, our studies also revealed that high geometrical curvature promotes membrane association of BCL2-like structural folds belonging to both MCL1 and BAK. Certain proteins display the ability to sense membrane geometrical curvature through their intrinsic shape (38), but BCL2-like structural folds of MCL1 and BAK are not intrinsically curved. On the other hand, all multi-BH motif BCL2 family members contain amphipathic helices with potential for sensing membrane geometrical curvature by inserting within curvature-created lipid packing defects (38). Of note, specific lipids and curvature-created lipid packing defects are both increasingly recognized as contributing factors in membrane relocalization events involving not only cytosolic proteins translocating to intracellular membranes, but also membrane-integrated proteins relocalizing within specialized membrane regions (i.e. intramembrane protein sorting) (39 -41).
Based on these collective observations, it is tempting to speculate that MOM-localized BH3-only proteins, specific MOM lipids, and MOM geometrical curvature can all contribute to mitochondrial accumulation of multi-BH motif BCL2 family proteins observed in the course of the apoptotic process. Nevertheless, it is clear that further experiments with cellular systems are required to test the validity of this hypothesis.
Another important finding of our study is the identification of two different ways by which membrane association affects MCL1⅐cBID complex formation. On the one hand, our SFCCS and FRET results indicate that, relative to the situation found in solution, the likelihood of MCL1⅐cBID heterodimerization increases severely in the presence of mitochondrial CS-like GUV/LUV and less prominently in the presence of mitochondrial CS-like SUV. Based on previous observations, it is likely that liposomes with a mitochondrial CS-like membrane composition generally stimulate MCL1⅐cBID complex formation by triggering exposure of the cBID BH3 motif (42,43). However, why is the level of MCL1⅐cBID heterodimerization higher in mitochondrial CS-like GUV/LUV than in mitochondrial CSlike SUV? We note that the alkali-extractable fraction of MCL1 g is notably larger in the former low curvature liposomes than in the latter high curvature vesicles (Fig. 3E). On the one hand, MCL1 g may adopt a peripheral membrane-adsorbed conformation in low curvature CS-like LUV/GUV that preserves the BH3-binding groove of the molecule intact and thereby maximizes MCL1 g ⅐cBID r heterodimerization. On the other hand, partial membrane insertion of MCL1 g in high cur- vature CS-like SUV may alter the BH3-binding groove of MCL1 g in a manner that diminishes cBID r binding. Following the same rationale, minimal MCL1 g ⅐cBID r heterodimerization detected in CL SUV, LUV, and GUV could be explained by destruction of the BH3-binding groove of MCL1 g due to extensive integration of the protein into pure CL liposomal membranes. Alternatively, or in addition, high membrane geometrical curvature and/or CL content may cause membrane insertion and hindering of the cBID r BH3 motif. In fact, it has been reported that cBID inserts the hydrophobic face of its BH3 motif into the hydrophobic interior of highly curved anionic micelles (43).
Independently of these open questions, our findings raise the exciting possibility that MCL1 heterodimerization with other BCL2 family partners could be dynamically modulated by changes in lipid composition and/or geometrical curvature occurring at localized sites of the MOM. Furthermore, it can be hypothesized that changes in mitochondrial membrane lipid composition or geometrical curvature may affect MCL1 interaction with non-BCL2 family partners (44,45). In this context, we propose that mitochondrial membrane lipid composition and geometrical curvature should be considered as "active" parameters in the BCL2 family interactome, in the sense that they could contribute to the self-organization of reactions between BCL2 family members at the MOM level.
Interaction between BAK and cBID is commonly described as a transient "hit-and-run" process, mainly to account for the observation that despite the fact that cBID triggers functional BAK activation, cBID does not form complexes with BAK in detergent-solubilized mitochondrial membrane extracts (1,3,4). However, it remained unproven whether this is a property displayed by the BAK molecule in an unperturbed lipid bilayer membrane environment. Our SFCCS and FRET results now show that the BCL2-like structural fold of BAK does not form a stable complex with cBID in a mitochondrial-like lipid bilayer membrane environment. The molecular details of this phenomenon remain to be fully elucidated. Nevertheless, based on our own results as well as observations made by other groups, a likely scenario is as follows (1,10,46): (i) initially, membranebound cBID and BAK heterodimerize through a canonical BH3-into-groove mechanism; (ii) this event destabilizes the BAK solution fold, leading to disengagement of BID BH3 from the BAK groove and exposure of the BAK BH3 motif; (iii) the exposed BAK BH3 motif engages into the groove of another BAK molecule to form a stable BAK homodimer. One important question remaining is why cBID-BAK interaction is more destabilizing than cBID-MCL1 interaction at the membrane level. Another prominent question remaining is whether the manifold membrane interaction modes described here for the BCL2-like structural folds of MCL1, BAK, and cBID can be extended to other BCL2 family members. The reconstituted systems and techniques described in this study may provide powerful tools with which to continue elucidating important mechanistic aspects of BCL2 family proteins, particularly in the context of a lipid bilayer membrane milieu.