Neutron reflectometry and NMR spectroscopy of full-length Bcl-2 protein reveal its membrane localization and conformation

B-cell lymphoma 2 (Bcl-2) proteins are the main regulators of mitochondrial apoptosis. Anti-apoptotic Bcl-2 proteins possess a hydrophobic tail-anchor enabling them to translocate to their target membrane and to shift into an active conformation where they inhibit pro-apoptotic Bcl-2 proteins to ensure cell survival. To address the unknown molecular basis of their cell-protecting functionality, we used intact human Bcl-2 protein natively residing at the mitochondrial outer membrane and applied neutron reflectometry and NMR spectroscopy. Here we show that the active full-length protein is entirely buried into its target membrane except for the regulatory flexible loop domain (FLD), which stretches into the aqueous exterior. The membrane location of Bcl-2 and its conformational state seems to be important for its cell-protecting activity, often infamously upregulated in cancers. Most likely, this situation enables the Bcl-2 protein to sequester pro-apoptotic Bcl-2 proteins at the membrane level while sensing cytosolic regulative signals via its FLD region. Through neutron reflectometry and NMR spectroscopy studies, Mushtaq et al study the full-length Bcl-2 protein reconstituted in lipid bilayers. They find that, in contrast to previously studied truncated, soluble protein versions, intact Bcl-2 is mainly embedded in the membrane with its regulatory loop highly flexible.

P rogrammed cell death, also called apoptosis, is essential for embryonic development, tissue homeostasis and removal of harmful cells in mammals and many other multi-cellular organisms 1,2 . Following activation of the intrinsic apoptotic pathway by intracellular stress, mitochondria play major roles in the process through permeabilization of the mitochondrial outer membrane (MOM) system and consequent release of apoptotic proteins such as cytochrome c from their intermembrane spaces 3 . This finally paves the irreversible way towards cellular selfdestruction 4 . A key adaptive mechanism that protects healthy cells from inappropriate death, but eliminates harmful cells, is tight regulation (positive and negative) of this apoptotic pathway and MOM integrity by members of the Bcl-2 (B-cell lymphoma 2) protein family 3,5,6 . Pro-and anti-apoptotic proteins of this family meet at the MOM and their net interactions determine the host cell's fate: maintenance of intact MOM and cell survival or MOM perforation and cell death. The outcome is governed by their relative abundances and affinities to each other [7][8][9] , which are strongly affected by the local membrane environment 10 . Disturbances of this regulatory interplay can cause various pathological disorders, including cancer 11,12 .
The main Bcl-2 members are multidomain proteins and are generally thought to regulate commitment to apoptosis by controlling the MOM's integrity 13 . Therefore, these proteins must be membrane-active. They can be classified as 'tail-anchored membrane proteins' 14,15 , since they contain beside their main globular fold composed of amphitropic domains, also a single transmembrane domain (TMD) at their C-terminus, which is thought to facilitate their binding and localization to their target membranes. The post-translational mechanism by which these proteins translocate upon ribosomal generation to their specific intracellular membranes is not fully understood yet 15,16 . Nevertheless, various Bcl-2 proteins can become soluble by hiding their TMD inside their hydrophobic BH3 binding groove motif. Subsequent exposure of the TMD enables them to insert into their target membrane and adapt an active membrane-embedded conformational state 15,17 . This way apoptotic members such as Bax are kept soluble in the cytosol and only migrate to, and permeabilize the MOM following activation and exposure of their TMD, leading to partial membrane penetration 9,18 . Even the anti-apoptotic Bcl-x L is found in various soluble and membraneanchored subpopulations in cells 16,19 . In contrast, despite the structural similarity, the anti-apoptotic, name-giving Bcl-2 protein itself is only found firmly anchored to intracellular organellar membranes, most prominent to the MOM system 14,16 . Intact human Bcl-2 was found to be very insoluble (presumably due to a fully exposed TMD), and therefore its subcellular translocation to the MOM is not yet clear. However, it is suggested that this process actively involves other proteins like e.g. mitochondrial chaperones 16,20 . In the membrane Bcl-2 protects healthy cells simply by sequestering apoptotic proteins to prevent perforation of the MOM and release of apoptotic factors such as cytochrome c 10 . Bcl-2 has a specific groove region which is central in restraining these apoptotic proteins (Bax, Bak, and BH3 only proteins) by binding to their Bcl-2 homology 3 (BH3) motifs. This cell-protection mechanism also plays a notorious role in tumorigenesis and cancer treatment resistance, as the Bcl-2 protein is often upregulated by mechanisms that block death signals 12 .
Nevertheless, all these anti-apoptotic Bcl-2 proteins shouldbased on their "tail-anchored" sequence homologies-adopt a globular fold composed of six amphipathic helical domains wrapped around two more hydrophobic central helices and the membrane-anchoring TMD. This combination provides them with the conformational flexibility to transit from a cytosolic-like to a loosely membrane-tethered state and finally to a firmly membrane-integrated conformational state required to exert their function 14,15 . However, the molecular basis of this cell-protection functionality of anti-apoptotic Bcl-2 proteins is poorly understood, due to lack of knowledge of their membrane-associated state and topology 10,21 .
Recent structural work on soluble anti-apoptotic Bcl-x L indicated that membrane-anchorage by its C-terminus alone is not sufficient for its inhibitory functionality 19 . Thus, a membraneembedded conformation is also presumably required for those multidomain anti-apoptotic Bcl-2 proteins to function 15 . However, we lack molecular understanding of this mechanism due to the challenges and inherent complexities associated with these conformationally highly flexible proteins in their cellular membrane environment. Fig. 1 Schematic diagram of approaches used to probe the Bcl-2 protein's structure and dynamics in a membrane environment. NR (neutron reflectometry) provides a global picture of the membrane, including locations of its molecular compounds and their volumetric contributions, as shown in this schematic diagram of the Bcl-2 protein (in tail-anchored and membrane-embedded positions). NMR provides complementary structural and dynamic information at atomic to entire protein molecule levels. The schematic Bcl-2 structure shown is based on a model of Bcl-2 (derived from PDB_ID 1YSW), including its fully flexible loop domain (FLD).
To address the lack of molecular understanding of the membrane-associated functionality of anti-apoptotic Bcl-2 proteins we have studied both the location and the structural and dynamic characteristics of the full-length human Bcl-2 protein in a mammalian membrane-mimicking environment. For this, we used two complementary techniques: neutron reflectometry (NR) and NMR spectroscopy, as schematically illustrated in Fig. 1. NR has become a powerful technique for probing biological membranes of varying complexity in recent years since it can provide abundant information on the structure and composition of entire membrane systems and their molecular components 22,23 .
Inter alia, it can provide indications of proteins' distributions across a lipid membrane and thus their locations in it. This type of global information is not accessible by other methods such as x-ray crystallography and NMR spectroscopy. However, NMR can provide-at atomic resolution-structural and dynamic descriptions of the behavior of specific molecular (and even intramolecular) components, including lipids and proteins 24,25 . Solid-state NMR approaches are especially suitable for probing unmodified, full-length proteins in their membrane-complexes at near-physiological conditions [26][27][28] .
Since full-length (239 aa) human Bcl-2 protein is insoluble and difficult to produce, soluble truncated and chimeric (166 aa) forms have been used in most studies [29][30][31] . As these forms lack key functional features, like the membrane-anchoring C-terminus and regulatory FLD, we developed a protocol to produce mg amounts of fully functional, intact human Bcl-2 protein 32 . NR analysis of this protein reported here, revealed the location and compactness of the intact Bcl-2 protein in its membrane environment, with most of the protein embedded in the membrane like an integral membrane protein. Complementary solid-and liquid-state NMR experiments corroborated this depiction. NMRderived data also indicate that the protein's regulatory (ca. 58 aa) loop domain is highly flexible, consistent with a location at or near the membrane interface, while the main protein body is deeply buried and motionally restricted in the membrane.

Results
Neutron reflectometric evidence of the Bcl-2 protein's membrane location. For NR analyses, full-length human Bcl-2 protein was reconstituted in supported L-α-dimyristoylphosphatidylcholine (DMPC) lipid bilayers at a 70:1 lipid-to-protein molar ratio. The Quartz Crystal Microbalance with Dissipation (QCM-D) technique was used to develop a protocol for membrane deposition and protein incorporation with NR reflectometry data (Fig. 2a) confirming that well-aligned supported DMPC/Bcl-2 and pure DMPC bilayers could be generated.
The protein and lipids in the bilayers could be resolved by isotopic labeling of the individual membrane components (protonated/fully deuterated Bcl-2 and DMPC, and vice versa) due to the large difference in neutron scattering length density of hydrogen and deuterium. To obtain a full range of membrane depth profiles ( Fig. 2 Fig. 2a. The profiles indicate that in both systems the membranes have very high quality, and have clearly differing scattering patterns that directly indicate the presence, or absence, of embedded Bcl-2 protein. The analysis was carried out using Abelès matrix formalism 33 , with the proteo-lipid-membrane structure described using a series of homogeneous layers across the bulk interface. The obtained NR signal is modeled as the sum of contributions from each layer and the reflectivity arising from the bulk interface, which forms the SLD profiles across it, as shown in Fig. 2b. The presented SLD profile corresponds to the best model fit of the data and shows their high quality. It clearly indicates that the main part of Bcl-2 resides in the membrane's hydrophobic interior (where its volumetric contribution is ca. 9%; Table 1) and the rest (presumably its outer parts, including the flexible loop region) is located in the headgroup/interface region (volumetric contribution: ca. 7%). These results support the general model, with most of the protein deeply embedded in the host membrane and its more amphipathic/hydrophilic regions located at the interface region. The NR profiles in Fig. 2 do not show any protein density outside the membrane.
Solid-state NMR: behavior of Bcl-2 in its membrane setting. Solid-state NMR experiments with dynamic filtering 24,34 support the NR-derived model of a compact protein body buried in the host membrane. These involved cross-polarization (CP) NMR experiments to visualize motionally restricted (often membraneembedded) protein parts and insensitive nucleus enhancement polarization transfer (INEPT) NMR experiments to probe very dynamic and flexible protein segments, as usually found in aqueous or extra-membranous environments. This approach has been recently applied in analyses of GPCR (G-protein coupled receptors) membrane proteins and membrane-associated Bax protein 24,34 .
As shown in Fig. 3a, strong, overlapping 13 C CP MAS NMR signals acquired at 278 K indicate that the vast majority of Bcl-2 residues in the DMPC membranes are dynamically restricted. This confirms the main protein body's location in a tight membrane hydrophobic core, which severely restricts reorientation dynamics of the amino acid residues located in it. Dynamically restricted residues of the Bcl-2 body seem to form a helix bundle across our experimental temperature range region. Substantial changes in the NMR spectra with temperature increases (indicated by sharp lipid NMR signals) only appear at 313 K (Fig. 3a, the spectrum of protein-free DMPC lipid bilayers for comparison). This is due to the melting of the DMPC bilayer matrix above 297 K into its biologically relevant liquid-crystalline phase.
NMR insights into the location and dynamics of Bcl-2's regulatory FLD region. The Bcl-2 protein contains a nearly 60 aa long intrinsically disordered region: the FLD between its α1 and α2 helices (aa 32-90). This domain is one of the functional hotspots of Bcl-2 and is thought to act as an autoregulatory molecular switch similar to the loop region of its close relative Bclx L 35-37 . The FLD sequence of Bcl-2 is more hydrophilic than the main body of the protein and presumably located in the wateraccessible membrane surface region 35 . Since this domain represents only a minor, loosely organized part of the protein with a low volume fraction, as manifested by smearing of the layer profile (Fig. 2b), it would be difficult to detect by NR outside the membrane. This difficulty is exacerbated by roughness (hetereogeneity in interfacial position) of the planar membrane, being 4 Å for the d-Bcl-2 (Bcl-2 containing) membrane. However, if this intrinsically disordered loop region displays high conformational flexibility, its residues can be discerned by NMR with dynamic filtering, as described above. As expected, the mobile loop residues emerge as sharp NMR resonances in the 13 C INEPT MAS NMR spectra, even at low temperatures (Fig. 3b, 278 K), in contrast to the motionally restricted residues of the protein body visualized by CP NMR (Fig. 3a,c). Upon increasing the temperature to 313 K, these 13 C INEPT NMR signals become more pronounced and additional sharp protein signals arise due to a further increase in local conformational dynamics. To further visualize these sharp resonances unambiguously in the NMR spectra, and identify the corresponding individual flexible residues in the FLD region, we also carried out solid-state 2D CP 13 C-13 C DARR and INEPT-based 13 C-1 H HETCOR NMR experiments (Fig. 3c, d). These mobile signals are not present in the DARR spectrum, but they are highly visible and well-resolved in two dimensions in the HETCOR spectrum. To assign the residues additional solution NMR experiments were performed on U-[ 13 C, 15 N, 2 H] labeled Bcl-2 protein in membranemimicking dodecylphosphocholine (DPC) micelles. To verify that the protein was fully functional under those conditions we carried out titration studies using a Bax-derived BH3 peptide known to bind to the main groove region of Bcl-2 responsible for recognizing apoptotic proteins via their BH3 motifs 38 , as well as the Nur77 peptide known to bind to the flexible loop domain of Bcl-2 protein 35 . As seen in the NMR titration series (Supplementary Figs. 3, 4) the protein is fully functional by binding firmly to those peptides. As expected for the interaction with the Bax-BH3 peptide major chemical shift perturbations (CSP's) could be observed, most pronounced for residues plausibly directly involved in the binding event (s. numbered residues in Supplementary Fig. 3) and less pronounced for most other residues of Bcl-2. This clearly indicates that binding of the peptide to the central part of the Bcl-2 protein, its extended BH3 domainbinding groove, causes major implications across the entire protein fold. In contrast, binding of the Nur77 peptide to the flexible loop domain of Bcl-2 is causing only local changes, as visible in CSP of residues belonging mainly to the loop region while the remaining protein core is much less or not at all affected ( Supplementary Fig. 4). Using fully functional Bcl-2 protein in DPC micelles we were finally able to identify nearly 60 amino acids ( Supplementary Fig. 5) positioned in the predicted loop region of the 26 kDa protein in its membrane-mimicking micellar environment. Based on these data, various sharp NMR resonances in the HETCOR spectra from membrane-embedded Bcl-2 could be residue-specifically identified (as indicated in Fig. 3d) as parts of the intrinsically disordered loop domain (FLD).
To confirm that these residues are flexible and not dynamically restricted, unlike the main protein body, we conducted 15 N T 2 relaxation experiments on full-length 13 C/ 15 N uniformly labeled Bcl-2 protein in DPC micelles using a 410 ms CPMG (Carr-Purcell-Meiboom-Gill) pulse sequence 39 . The black spectrum (Fig. 4a) shows signals originating from the rigid protein body and loop region, while the red spectrum only displays signals from the loop region due to its slower relaxation. Those signals are not well dispersed, as expected for residues of IDP-like proteins 40 . Thus, we can unambiguously differentiate between NMR signals originating from residues of the regulatory FLD of the protein and the rigid protein core. The differentiation and conclusions are validated by comparison of these spectra with those of Bcl-2 protein in larger Brij-35 micelles (Fig. 4b).  In this environment, only the sharp signals from the loop regions with their typical IDP dispersion are visible in 1 H-15 N TROSY-HSQC (transverse relaxation-optimized spectroscopyheteronuclear single quantum coherence) spectra, due to slower relaxation rates, while most signals of the protein core are broadened beyond detection. Signals from the loop residues are also visible as sharp lines in the corresponding 13 C solid-state MAS NMR spectra of the Bcl-2 proteoliposomes (Fig. 3b, d).

Discussion
Implications for Bcl-2 protein functionality in its MOM. Our combined NR and NMR studies provide detailed molecular insights into the basic structural and dynamic features of Bcl-2 protein. Knowledge of these features is essential to elucidate its cell protection activity at the MOM, where Bcl-2 proteins engage in the regulation of apoptosis. In a physiological lipid membrane environment, full-length Bcl-2 behaves as a membrane-embedded protein (Fig. 5), which is presumably essential for its functions at the mitochondrial membrane, as suggested for anti-apoptotic Bcl-2 family members 14,15,41 . As seen here by NR, the main protein body (without its loop region) is buried in the membrane, where motions of its residues are restricted, as highlighted by solid-state 13 C CP NMR experiments (Fig. 3a, b). Solution NMR experiments with Bcl-2 in micelles of varying size confirm the restricted dynamics of these residues (broad lines). In contrast, residues of the very flexible FLD loop region have fast dynamics and high conformational flexibility, manifested by narrow NMR signals (Figs. 3, 4) in both solid-state and solution NMR experiments, due to much longer spin-spin relaxation times caused by fast intramolecular dynamics.
The picture of a fully membrane-embedded Bcl-2 protein body is not surprising since the Bcl-2 protein consists of two central hydrophobic helices (α5, α6) surrounded by a further six hydrophobic/amphiphilic helices, and is linked to a hydrophobic transmembrane anchoring helix region (α9). Since all of these helical structures have affinities for the insides of hydrophobic membrane environments, to various degrees 29,42 , a location inside a lipid bilayer with only the loop region sticking out into  the aqueous exterior is consistent with expectations. MD simulations comparing truncated versions with our physiologically intact Bcl-2 protein indicate that it is slightly less compact with the large flexible loop region, but still properly folded 36 .
Clearly, the function of Bcl-2 as an integral membrane protein observed here must be reconciled with the common view of the nature of anti-apoptotic Bcl-2 proteins' membrane association as deduced from the observation of a soluble member of the family, the Bcl-x L protein and its behavior upon anchoring to lipid nanodiscs 19 . This involves tail-anchored membrane attachment via their C-terminal hydrophobic transmembrane helix 14,15 , while the N-terminal head domain remains in the cytosol by retention of its water-accessible globular fold. Nevertheless, helices 1 and 2 of Bcl-x L are reportedly close to the membranes in these systems, indicating a propensity for membrane association or insertion. Not surprisingly, previous studies using truncated, soluble Bcl-2 and Bcl-x L variants were both found to be able to form channels in membranes, most likely due to similar features as pore-forming domains of colicins and diphtheria toxins 43,44 . However, this pore formation seems to be transient without a major effect on the MOM, but presumably prevents permeabilization of those membranes by apoptotic proteins like Bax. While those studies provide no direct information about the Bcl-2 location and exact arrangement in the membrane, they show that Bcl-2 proteins can insert into membranes and form water-accessible pores. Our findings show that in contrast to truncated constructs, the full-length native Bcl-2 protein is embedded firmly into a membrane with its-by helices composed -main fold residing in the membrane interior and only the loop region being water-accessible. Our model (Fig. 5) provides a fundamental understanding of the unique capability of Bcl-2 protein to recognize apoptotic proteins at the membrane level while accessing via the loop region regulative signals from the cytosol.
Generally, the N-terminal domains of these Bcl-2 members can be water-accessible or membrane-bound. Biophysical studies on various anti-apoptotic Bcl-2 proteins have indicated that they can undergo conformational changes from a cytosolic accessible compact globular fold to a membrane-inserted state 14,41,45 . As summarized by a graphic model (Fig. 1) and accompanying text in a previous publication 15 , this conformational plasticity is essential for the proteins' adoption of a membrane-embedded structure, which might be required for its cell-protective functions at the MOM.
Despite intact human Bcl-2 being very insoluble and its translocation mechanism from a cytosolic milieu to the mitochondrial target membrane is not yet clear, our observations of the Bcl-2 protein in an experimental lipid membrane settings are consistent with such conformational plasticity and final membrane-embedment (Fig. 5). By monitoring the behavior of full-length Bcl-2 (239 aa), and a soluble variant (Bcl-2 ΔTM; aa 1-207) with the C-terminal anchor removed, in the presence and absence of DPC detergent we could observe this type of restructuring of the Bcl-2's globular fold into a membrane-like state. As seen in Supplementary Fig. 6, Bcl-2 ΔTM displays a typical solution-like structure with a broad distribution of Trp signals ( Supplementary Fig. 6b), as observed for soluble truncated Bcl-2 protein 30 , and its Bcl-x L 19,46 and Bcl-w relatives 45 .
When soluble Bcl-2 ΔTM is treated with increasing concentrations of DPC to mimic a membrane environment we observe a structural transition from a soluble to membrane-like embedded state using 1 H-15 N-TROSY-HSQC NMR. Increasing amounts of detergent induce changes in the overall fold and general changes across the entire NMR spectrum (especially in the Trp region) in the major population of the Bcl-2 ΔTM protein. Refolding Bcl-2 ΔTM upon purification directly into DPC micelles results in identical spectral behavior for the entire population, as also observed for intact, fully Bcl-2 protein in DPC micelles (Supplementary Figs. 7 and 8). The detergent environment causes major conformational changes towards an active protein state, as demonstrated by successfully binding Bax-derived BH3 peptides to both the full-length Bcl-2 and the Bcl-2 ΔTM proteins in DPC micelles (Supplementary Figs. 3,9). This transition behavior reflects the propensity of other Bcl-2 proteins, such as Bcl-x L and Bcl-w to adopt structures suitable for functioning in a membrane. Following the addition of the membrane-mimicking detergent DPC, also their N-terminal head domain with its common globular fold adopted such structures [45][46][47] . This change in the globular fold was clearly visible in all the 1 H/ 15 N NMR spectra acquired in the cited studies, especially in contributions of the Trp residues, which reflect dramatic changes in their local environment. Presumably, full-length Bcl-2 is insoluble due to its hydrophobic TM domain extended away from the main protein body and not hidden in its BH3 binding groove region as in other Bcl-2 proteins which can exist as soluble cytosolic subpopulations 16 .
As can be seen in the solution NMR spectra (Fig. 4), the main N-terminal head domain (a common globular fold) of the entire Bcl-2 protein becomes dynamically restricted in detergent micelles while the FLD region remains highly flexible in the functional state where this loop region successfully binds the Nur-77 peptide (Supplementary Fig. 4). Our observations show that this arrangement of full-length Bcl-2 is preserved following the replacement of its micellar environment by native-like lipid membranes. NR showed that the protein body was buried in the membrane and remained dynamically restricted (as shown by  14 , comparison with the related Bcl-x L protein (which has conformational plasticity ranging from soluble to embedded states 15,19,41 ) and the NR/NMR results reported here. For membraneinserted α1 to α4 helices (gray cloud) no detailed information is available. b Locations of the Bcl-2 main body within the membrane boundaries and solvent-accessible regulatory loop region (FLD) at the membrane surface, observed by NR and NMR, respectively. The main protein body is dynamically restricted while the FLD region is highly flexible.
solid-state NMR), while the loop region remained flexible outside the membrane. Various residues and their positions could be identified from solution NMR-based residue assignment in the Bcl-2/micelle systems ( Supplementary Fig. 5), indicating that the protein adopted similar conformations in both membranemimicking environments. While NMR experiments (Supplementary Fig. 4) showed a clear binding of the Nur77 peptide to the FLD, this binding event did not induce any specific structural features into this intrinsically disordered extended loop and the remaining protein body was not affected. This indicates that the regulative loop region can act in a spatially decoupled manner from the main protein core with its BH3-BH1 groove region. Presumably, as proposed for Bcl-x L inhibition of Bax 48 , the loop provides the flexibility for the helices α1 from both proteins to form a second binding interface in addition to the main Bcl-x L groove interface which sequesters BH3 helix of Bax. We found that the full-length Bcl-2 protein resides in host membranes in a fully embedded conformation as a multidomain spanning integral membrane protein. This not only reflects the conformational plasticity required for proper functioning 15 , but also the location required for blocking cell-killing Bcl-2 family members like Bax. Cytosolic Bax proteins can translocate to the MOM where they partially penetrate and form oligomeric membrane pores, thereby releasing apoptotic factors and ensuring cellular death 34,49 . Bcl-2 is thought to block membrane perforation by Bax (and presumably its apoptotic relatives) by engaging with the BH3 domain, after partial membrane penetration, thus sequestering the apoptotic protein-membrane boundary before any dimerization and subsequent membrane damage occurs 50 .
To restrain apoptotic Bcl-2 proteins, the Bcl-2 protein itself has an extended groove interface formed by its BH3, BH1, and BH2 domains. By engaging four complementary pockets of the groove region, the Bcl-2 protein is binding four or more hydrophobic residues on consecutive helical turns of these BH3 motifs of apoptotic Bcl-2 proteins 31 . This design enables strong engagement with variable BH3 motifs from different apoptotic multidomain proteins but also from BH3 only proteins; something clearly confirmed by our NMR titration studies using Bax-BH3 and Bim-BH3 peptides and various Bcl-2 constructs where the BH1-BH3 groove region was conserved ( Supplementary Fig. 3, 9, 10 and 11). In all cases, the presence of BH3 peptides caused significant CSPs for residues presumably in direct contact with those peptides in the groove region. In addition, the entire protein experienced a structural adaption in response to the bound ligands, as visible in more moderate CSPs across most protein residues. This type or adaption ("structural breathing") is most likely the reason why the Bcl-2 protein can bind various BH3 motifs of diverse Bcl-2 family members with high affinity, even if mutations in this binding pocket occur which dramatically reduce the affinity of rigid cancer drugs like venetoclax targeting this groove region 31 . The hydrophobic groove region of Bcl-2 is central for blocking cell death and can most likely exert direct recognition with high affinity of varying BH3 death domains of apoptotic proteins via a binding interface facilitated by structural flexibility.
The membrane environment plays a crucial role by changing proteins' local abundances and their affinities through twodimensional crowding, while keeping the sequestration spatially separate from the FLD region, which regulates Bcl-2's conformation and activities by bundling numerous cytosolic signals 51 . In this manner, Bcl-2 protects healthy cells by sequestrating any transiently membrane-bound Bax at the MOM 52 . Many tumors escape therapy-induced death (often via activation of apoptotic Bcl-2 family members) by upregulation of this Bcl-2 protein and its cell-protecting relatives 12,53 , all of which presumably function in a similar manner to Bcl-2. Our model of Bcl-2 fully embedded in its host membrane provides the molecular basis for the protein's cell-protecting activity and may facilitate the discovery of drugs targeting currently non-treatable tumors.
A glycerol stock of transformed BL21(DE3) Rosetta™ cells with Bcl-2 encoded in a pET-15b vector (Novagen) was ordered from GenScript (Leiden, Netherlands). The cells were cultured in media with step-wise increases in the concentration of deuterium oxide (Armar Isotopes GmbH, Germany). Initially, cells from the glycerol stock were grown in 20 ml of LB media (10 g Bacto tryptone, 5 g yeast extract, 5 g NaCl per liter, pH adjusted to 7.5, and sterile filtered) overnight at 37°C. The following day 500 µl of the resulting culture was transferred into M9 media prepared as above, but included 50% D 2 O. Cells were grown at 37°C for an additional day, then 3 ml of the resulting culture was collected and centrifuged at 4400 × g for 10 mins. The obtained pellet was transferred into M9 medium containing 75% D 2 O, and the procedure was repeated, with step-wise increases to 90% and finally 99.8% D 2 O. The cells grown in 90% and 99.8% D 2 O were left to incubate for an extra day at 37°C before harvesting, due to their slower growth rate. The final batch for deuteration was prepared by transferring 100 ml of the adapted culture to 900 ml of M9 medium in 99.8% D 2 O. The cells were cultivated at 37°C, induced by adding 1 mM IPTG upon reaching OD 600~0 .6, and further grown overnight. The following day deuterated Bcl-2 protein was harvested by centrifuging the culture at 4400 × g for 30 min, then resuspending the cells in 20 ml 50 mM Tris, pH 8.0, and freezing them at −80°C until purification. Bcl-2 protein was labeled with either 13 C or 15 N using the same routines, but replacing the glucose or NH 4 Cl with the same amount of isotope-enriched variants.
Preparation of truncated Bcl-2 variants. Primers for constructing the Bcl-2 ΔTM truncation were ordered from Eurofins Genomics (Ebersberg, Germany), which introduces a stop codon to terminate the protein at position 207, thus disabling the membrane-binding region of the protein. Site-directed mutagenesis was performed using the Quikchange approach (Stratagene) with oligonucleotide forward primer 5′-GGACCTTCGATGCGTTAACTGTTTGATTTTTCG-3′ and reverse primer 5′-CGAAAAATCAAACAGTTAACGCATCGAAGGTCC-3′, respectively. Proper insertion of the stop codon TAA was verified by DNA sequencing (Eurofins Genomics).The Bcl-2 truncated variant Bcl-2 ΔN(1-82) was purchased from GenScript® (Leiden, Netherlands) and sub-cloned into a pET-15b expression vector (Novagen). The construct is aimed to delete the first 82 N-terminal amino acids, starting the N-terminus with residues MGP, while it maintains the Cterminal residues. Worth noting that all Bcl-2 variants carry an additional three Nterminal residues, GSH, due to the design of the pET-15b expression vector. Expression and purification of both mutants were carried out in an identical manner as for the full-length protein and the Bcl-2 ΔTM mutant. Soluble Bcl-2 ΔTM protein was purified directly from the minor soluble fraction after sonication and centrifugation, and then further purified using identical procedures. For concentration measurements, ε 280 = 44920 M −1 cm −1 was used for full-length Bcl-2, for the mutant Bcl-2 ΔTM, ε 280 = 37930 M −1 cm −1 , and for the Bcl-2 ΔN(1-82) mutant ε 280 = 33460 M −1 cm −1 , respectively.
In vitro binding assays for Bcl-2 proteins. The functionality of the Bcl-2 protein in both Brij-35 and DPC micelles was verified as recently described 32 . We checked the full-length protein's functionality in these environments by monitoring chemical shift perturbations using 15 N TROSY-HSQC NMR (Supplementary Figs. 3,  4) for two purposes. First, to determine the affinity of Bax-BH3 peptide (which is known to bind Bcl-2 32 and mimics the cell-death inducing Bcl-2 homology domain 3, BH3 of Bax) to the hydrophobic groove of Bcl-2 38 . Second, to determine the binding and affinity of the Nur77 peptide, which is known to bind to Bcl-2's loop region and trigger Bcl-2's activities 35 , to check the functionality of the regulatory loop region (residues 32-90 here), which are removed in truncated soluble Bcl-2 variants 29,30 .
Reconstitution of Bcl-2 protein in lipid bilayers. Bcl-2 protein was reconstituted using a protocol developed by us 21 , with some modifications as follows: A stock suspension of DMPC vesicles was prepared by adding 5 mM DMPC to 20 mM NaPi, 50 mM NaCl, 2 mM DTT, 1 mM EDTA at pH 7.4, then sonicated using a ultrasonic cleaner (VWR, USA). Mixed micelles were prepared by solubilizing DMPC or DMPC-d 67 vesicles in 0.05 % (w/v) Brij-35 (Sigma-Aldrich), together with a stock solution containing pure detergent-solubilized Bcl-2 protein until the solution became clear. The concentration of DMPC lipids and protein was tuned to generate a 1:70 (mol/mol) protein:lipid ratio. Brij-35 in the proteoliposome mixture was removed by adding approximately 1 ml of Biobeads™ (Biorad) to 10 ml of protein solution, and slowly set to mix on a rocking table at 4°C overnight. In the following days, the Biobeads™ were removed and this process was repeated until a cloudy solution emerged. Proteoliposomes were purified using a 25% and 40% (w/v) sucrose gradient, prepared by dissolving sucrose in 20 mM NaPi, 50 mM NaCl, 1 mM EDTA, pH 7.4. The sample was centrifuged at 75000 × g for 60 min at 4°C using a SW-41 Ti rotor together with an Optima L 90 K Ultracentrifuge (Beckman Coulter, USA). Isolated proteoliposomes were further purified by adding buffer and centrifugation using a Micro Star 17 tabletop centrifuge (VWR, USA) for 20 mins x2, and the sample was subsequently used in solid-state NMR or neutron scattering experiments. Analysis of the proteoliposome pellet by SDS-PAGE confirmed the presence of Bcl-2 protein with the correct molecular size. Fully deuterated DMPC and DMPC-d 67 were purchased from Avanti Polar Lipids (Alabaster, USA).
Neutron reflectometry measurements of silicon-supported membranes. Silicon single crystals (50 × 80 × 15 mm; Sil'tronix, France) with a 80 × 50 mm (111) face polished to 3 Å rms roughness were assembled into purpose-built solid/liquid flow cells with the polished face of the crystal in contact with the liquid reservoir of the liquid trough within the cell.
NR measurements were acquired with the SURF reflectometer (which uses a white beam of neutrons with wavelengths ranging from 0.5 to 7 Å) at the STFC ISIS Pulsed Neutron and Muon Source, Rutherford Appleton Laboratory, Oxfordshire, UK. The silicon crystal-containing solid/liquid flow cells were placed in the sample position in the instrument and connected to a L7100 high-performance liquid chromatography pump (Merck, Hitachi), for changing the H 2 O/D 2 O ratio of the solution in the cell. The following ratios were used: 100%, 80%, 38%, 0% D 2 O/H 2 O buffer (20 mM NaPi, 50 mM NaCl, 1 mM EDTA, pH 7.4).
In specular NR experiments, intensity is measured as a function of the angle and/or wavelength of the beam relative to the sample surface (46,47), expressed as the momentum transfer vector, Qz = (4π sin θ)/λ, where λ is wavelength and θ is the incident angle). The white beam instruments can probe a wide range of Q z space at a single angle of reflection due to the use of a broad neutron spectrum. To obtain reflectivity data across a Q z range of~0.01 to 0.3 Å −1 , glancing angles of 0.35°, 0.65°, and 1.5°were used.
During the experiments, the bare silicon/water interface was initially examined before the Bcl-2/DMPC vesicles were deposited by vesicle rupture onto the silicon interface. The composite interfacial structure was then examined under multiple solution isotopic contrast conditions so the relative distributions of both the lipid and protein components within the membranes could be resolved by simultaneous analysis of all the contrasts.
Neutron reflectometry data analysis. The magnitude of the coherent neutron scattering length of nuclei varies randomly across the periodic table 54 . Some isotopes, most usefully the hydrogen isotopes protium (99.98% natural abundance) and deuterium (0.015% natural abundance), have different neutron scattering lengths. Differential hydrogen isotope labeling of samples for neutron scattering experiments is commonly used to collect a series of datasets pertaining to samples under chemically similar but isotopically different conditions. Isotopic labeling of the bulk solution and sample to produce multiple scattering 'contrasts' is advantageous when analyzing complex structures. This is because multiple datasets strongly limit the number of potential structural solutions when they are simultaneously analyzed and allow resolution of distributions of individual components within analyzed complexes if a suitable labeling strategy is employed 55 .
NR data were analyzed using RasCal software (A. Hughes, ISIS Spallation Neutron Source, Rutherford Appleton Laboratory), which fits layer models describing the interfacial structure calculated using Abelès matrix calculation 33 to experimental data. In this approach, the interface is described as a series of slabs, each characterized by its nSLD, thickness and roughness. The reflectivity for the model's starting point is then calculated and compared with the experimental data.
The final datasets consisted of four individual reflectivity profiles obtained from the membrane-coated silicon/water interface collected in 100% D 2 O, 80% D 2 O and 20% H 2 O, silicon-matched water (Si-MW, 38% D 2 O and 62% H 2 O) and 100% H 2 O. The membrane was composed of hydrogenous (i.e. natural abundance hydrogen) DMPC and deuterated (d-)Bcl-2 protein. The difference in isotope labeling of the membrane constituents allows the protein and lipid components to be highlighted in different solution contrasts. Specifically, the D 2 O solution contrast was sensitive to the structure of the lipid component of the membrane due to the large difference in SLD (ρ) between the lipid and D 2 O (Table 1). Similarly, the H 2 O solution isotopic contrast was sensitive to the protein component of the membrane due to the strong difference between these two components.
For modeling the interfacial structure, the known scattering length densities of the protein and lipid components were used as constraints and the membrane was modeled as a five-layer structure. The layers of the interfacial structure (from silicon to solution) were: SiO 2 , the inner headgroup region, bilayer hydrophobic tail region and outer headgroup region. These layers were fitted with the same SLD, thickness and roughness across all four solution isotopic contrasts. The membrane structures at the silicon surface (and immersed in the bulk solution) were fitted with varying SLDs between contrasts to account for the presence of water within the layers and labile hydrogen exchange in different solution isotope mixtures. Interpretation of the interfacial structure from the resulting SLD profile informed further fitting. The positions of each component in the protein/lipid membrane were linked with the known SLDs of the components and the solution isotopic contrast to determine the volume fractions of the protein and lipid components within the model membrane 55 .
Solid-state NMR spectroscopy analysis of full-length Bcl-2 in proteoliposomes. All 1 H and 13 C MAS NMR spectra were acquired using a 3.2 mm HCN MAS Bruker probe at an 850 MHz Avance III HD spectrometer (Bruker, Switzerland) and proteoliposomes in NMR buffer (20 mM NaPi, 50 mM NaCl, 1 mM EDTA, 1 mM DTT, pH 7.4). 1 H MAS NMR spectra were acquired using a single π/ 2-pulse with a 3.1 μs ( 1 H) duration, and 1.5 s repetition delay and a sample spinning rate of 15 kHz. One-dimensional 13 C solid-state CP and INEPT NMR spectra were acquired to obtain information on both the rigid and more flexible parts of the protein 24 .
The CP NMR experiments were performed at varying temperatures with a contact time of 0.5 ms, an RF field of 50 kHz at the 13 C frequency and a 1 H ramp power from 34 to 69 kHz, and 63 kHz of decoupling power. For refocused INEPT and direct polarization experiments, the 13 C pulse duration was typically 3.35 µs with the decoupling lowered for INEPT (57 kHz swft ppm) while the acquisition time increased from 19 ms to 37 ms. The 13 C chemical shift of 38.5 ppm for adamantane was used as an external reference. In two-dimensional NMR experiments, 13 C CP MAS NMR spectra were acquired using a π/2-pulse with 3.1 μs ( 1 H) duration followed by 1 H ramped CP at 13 C (79 kHz; 40-81 kHz ramp) for 1 ms, a 1 H SPINAL-64 decoupling at 81 kHz, and a 1.5 s repetition delay at a sample spinning rate of 15 kHz. 2D 13 C-13 C DARR NMR experiments were carried out with 25, 50, and 100 ms mixing times using a 15 kHz MAS rate.
Solution state NMR spectroscopy of full-length Bcl-2 protein in micelles. 1 H-15 N TROSY experiments were performed using 0.3 mM 13 C, 15 N-labeled full-length Bcl-2 in 5 mM DPC micelles (20 mM NaPi, 20 mM NaCl, 2 mM TCEP at pH 6.0) and 328 K for backbone assignments, and 15 N-labeled 0.3 mM Bcl-2 and 0.25 mM Bcl-2 ΔTM in 5 mM DPC micelles (20 mM NaPi, 20 mM NaCl, 2 mM TCEP at pH 6.0) and 310 K for Bax-peptide titration experiments. 1 H-15 N TROSY spectra using a 15  The pulse programs were obtained from the Bruker TopSpin 3.6.1 library. These data were processed with free NMRPipe and NMRDraw softwares 56 , and visualized and analyzed with the programs CCPN2.1.5 57 and Sparky 58 . All NMR measurements for backbone chemical shift assignments were acquired at 328 K using an Avance 850-MHz NMR machine with a tripleresonance cryogenic probe (Bruker, Germany).
Statistics and reproducibility. The structure of the Bcl-2:DMPC complex was examined in triplicate by NR. Twice using a labeling strategy of d-Bcl-2:h-DMPC (Fig. 2) and once with h-Bcl-2:d-DMPC (Supplementary Fig. 2). Model parameter error estimation from the fits of these NR datasets was conducted using Rascals Bayesian Error estimation routines, with the log-likelihood function described in terms of chi-squared 22 . Marginalized posteriors were obtained using a Delayed Rejection Adaptive Metropolis algorithm 22 , and the best-fit parameters taken as the distribution maxima; the uncertainties presented here are from the shortest 95% confidence intervals of each distribution.