Structural insights into apoptotic regulation of human Bfk as a novel Bcl-2 family member

Graphical abstract


Introduction
Proteins of the B cell lymphoma-2 (Bcl-2) family are involved in the intrinsic apoptotic pathway that releases cytochrome C into the cytoplasm by regulating mitochondrial outer membrane permeabilization (MOMP) [1,2]. Bcl-2 family proteins harbor up to four highly conserved regions called Bcl-2 homology (BH) domains, although overall sequence identities of Bcl-2 family proteins are low [3,4]. Among the four BH domains, a BH3 domain is present in all Bcl-2 family proteins and governs interactions with the hydrophobic cleft of other Bcl-2 family proteins, which is also known as a BH3 and C-terminus binding groove (BC groove) [5].
Anti-apoptotic proteins inhibit the activation of pro-apoptotic pore-formers by sequestering the BH3 domains of pro-apoptotic proteins in their BC grooves with high affinity [8,9]. Otherwise, pro-apoptotic pore-formers undergo structural rearrangements upon the release from anti-apoptotic proteins or the transient binding of BH3-only proteins, which result in oligomerization and pore formation within the mitochondrial outer membrane [10]. Recent mutation studies on Bak, a pro-apoptotic pore-former, showed that the BH1 and BH3 domains play important roles in the activation and oligomerization of Bak and further MOMP [11].
In contrast to multi-domain Bcl-2 family proteins, pro-apoptotic BH3-only proteins are intrinsically disordered and, except for Bid, exhibit pro-apoptotic activity without forming the Bcl-2 fold [12]. Bid is a BH3-only protein known to induce the activation of proapoptotic pore-formers [11,13], although it remains debated whether the direct binding of BH3-only proteins to Bax/Bak is essential to Bax/Bak activation or whether they play an indirect role in the activation by neutralizing only anti-apoptotic Bcl-2 proteins, such as Bcl-X L or Mcl-1 [14]. Full-length Bid maintains the Bcl-2 fold in the cytosol until it is cleaved by caspases [15,16]. The cleavage of Bid (p22) releases an N-terminal fragment (p7) upon contact with a hydrophobic membrane, whereas the truncated Bid (tBid; p15) undergoes a conformational change to embed it in the mitochondrial outer membrane for further recruitment and activation of Bcl-2 family pore-formers [17,18]. Recently, tBid is also suggested to mediate MOMP even in absence of Bax and Bak [19].
Some proteins in the Bcl-2 family do not meet the general classification criteria described previously herein owing to their ambiguous functions and compositions of BH domains. Bcl-2 family kin (Bfk) is a novel Bcl-2 family protein that harbors a unique combination of BH2 and BH3 domains [20]. Human BFK is predominantly expressed in the gastrointestinal tract, and its enforced expression weakly promotes apoptosis [20]. Additionally, BFK expression is significantly downregulated in tumors isolated from various gastrointestinal tissues [21,22]. A truncated form of Bfk (tBfk), following cleavage of an N-terminal fragment by caspases, elicits stronger pro-apoptotic activity than full-length Bfk [21]. Moreover, tBfk induces apoptosis in a Bax/Bak-dependent manner, whereas co-expression with Bcl-X L and Bcl-W reverses the tBfkmediated apoptotic activity [23]. Despite the biological significance of Bfk in the onset of apoptosis, the structural and molecular mechanisms underlying apoptotic regulation by Bfk remain elusive.
Here, we report the first crystal structure of human Bfk at 2.70 Å resolution, as a representative of Bcl-2 family proteins comprising a novel combination of BH2 and BH3 domains. The overall structure of Bfk adopts the Bcl-2 fold, similar to those of multidomain Bcl-2 family proteins. However, structural, biophysical, and computational analyses of Bfk revealed significant similarities with Bid, a BH3-only protein, thus implying that Bfk induces apoptosis activation by adopting the BH3-releasing conformation. These findings provide insights into the apoptotic mechanism of action of Bfk and a valuable foundation for the development of a new therapeutic target to regulate apoptosis.

Cloning, expression, and purification of Bfk
A full-length human BFK (residues 1-163) encoded from the BCL2L15 gene was amplified using polymerase chain reaction and cloned into the expression vector pET-28a(+) (Novagen, Madison, WI, USA) between Nde Ⅰ and Xho Ⅰ restriction sites to contain an N-terminal hexahistidine (His 6 ) tag (MGSSHHHHHHSSGLVPRGSH). The BFK cloned plasmid was transformed into OverExpress TM C43 (DE3) (Lucigen, Middleton, WI, USA), an Escherichia coli strain. The transformed cells were cultured in Lysogenic Broth media containing 30 lg/mL kanamycin at 37℃ until OD 600 reached 0.6, and were added of 0.5 mM isopropyl b-D-1-thiogalactopyranoside (IPTG) to induce overexpression. After incubation for an additional 16 h at 20℃, the cells were harvested by centrifugation at 6,000 Â g for 10 min. The pelleted cells were resuspended in buffer A (500 mM NaCl, 35 mM imidazole, and 20 mM Tris-HCl at pH 7.5) containing 1 mM phenylmethanesulfonylfluoride and lysed by sonication. The lysate was centrifuged at 35,000 Â g for 60 min at 4℃. The supernatant was filtered through a 0.45 lm syringe filter device (Sartorius, Göttingen, Germany) to remove cell debris and any precipitated proteins. For affinity chromatography, the filtrate was applied onto 5 mL HiTrap chelating HP column (GE Healthcare, Chicago, IL, USA) which had been pre-charged with Ni 2+ and equilibrated with buffer A. The retained proteins were eluted with a gradient of buffer B (500 mM NaCl, 1 M imidazole, and 20 mM Tris-HCl at pH 7.5). For size exclusion chromatography, the eluent was applied onto a HiLoad 16/600 Superdex 75 pg column (GE Healthcare, Chicago, IL, USA) which had been equilibrated with buffer C (200 mM NaCl and 25 mM MES-NaOH at pH 6.0). The purified Bfk was concentrated to 15 mg mL À1 using an Amicon Ultra-15 Centrifugal Filter Unit (Merck Millipore, Darmstadt, Germany).
For selenomethionine (SeMet)-derived proteins, the recombinant Bfk proteins were overexpressed in E. coli strain B834(DE3) using the media containing M9 minimal salts (Sigma-Aldrich, Darmstadt, Germany) and amino acid mix containing Lselenomethionine (Sigma-Aldrich, Darmstadt, Germany). The protein was expressed and purified as for the native Bfk.

Crystallization, data collection, and structure determination
Purified Bfk proteins were crystallized at 22℃ using the sittingdrop vapor diffusion method by mixing 0.5 lL proteins and 0.5 lL crystallization solution. Initial crystals were obtained under commercial crystallization screening conditions containing 0.1 M Bis-Tris at pH 7.5 and 25% polyethylene glycol (PEG) 3,350 of the Shot-Gun 1 kit (Molecular dimension, Cambridge, England). The crystals suitable for data collection were optimized at 16℃ using the microseeding method. The micro-seeds of crystals were prepared from the initial crystals of Bfk using Seed Bead TM Kits (Hampton Research, Aliso Viejo, CA, USA) according to the manufacturer's instructions. The SeMet-derived crystals were obtained in the same manner as native Bfk. The crystals were cryoprotected in the crystallization solution supplemented with 17% glycerol and flash-frozen in a 100 K nitrogen gas stream for exposure to X-ray beams. The X-ray diffraction data for native Bfk was collected to 2.70 Å using an Eiger 9 M detector system (Dectris Ltd., Baden, Switzerland) at beamline 5C experimental station at Pohang Light Source, Korea. To solve the phase problem, single-wavelength anomalous diffraction (SAD) data of SeMet-derived Bfk was collected to 2.45 Å at the same beamline. The raw data were indexed and scaled using the HKL2000 program suite [24] and XDS [25]. SAD phases were calculated with Autosol in the PHENIX software suite [26], and further improved by density modification using the automatic model building program Resolve [27]. Then, the model was used as a template for molecular replacement method using MolRep [28] to obtain phases for the diffraction data collected from the native Bfk crystals. The resulting models were further refined using iterative cycles of model building with Coot [29] and Refmac5 [30] in CCP4i program suite [31]. Since the diffraction data of the Bfk native crystal was analyzed as having twin fractions by Xtriage of the PHENIX software suite, which caused a subsequent problem with abnormally high R-factors, the intensity-based twin refinement of Refmac5 was applied. After the refinement with twin fractions of 0.543 and 0.457 for the twin operators of (h, k, l) and (-h, -k, l), respectively, the R-factors and quality of the electron density were suitably improved. All refinement steps were checked using an R free value [32] calculated for a randomly chosen 5% of reflections and the reliability of refined models was validated using MolProbity [33] and the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB) validation server. The statistics of data collection and refinement are summarized in Table 1.

Protein preparations for Bcl-2, Bcl-X L , and Bak
The genes, encoding Bcl-2 (residues 1-207), Bcl-X L (residues 1-209), and Bak (residues 23-186), were each cloned into the expression vector pET-28a(+) (Novagen) between Nde Ⅰ and Xho Ⅰ restriction sites to contain the N-terminal His 6 tag. The cloned plasmids of Bcl-2, Bcl-X L , and Bak were each transformed into E. coli Rosetta TM 2(DE3), Rosetta TM 2(DE3), and Rosetta TM 2(DE3) pLysS strains, respectively. Bcl-2-or Bcl-X L transformed cells were cultured in Lysogenic Broth media containing 30 lg/mL kanamycin at 37℃ until OD 600 reached 0.6. Overexpression of the cells were induced by adding 0.5 mM IPTG to induce overexpression, followed by further incubation for additional 16 h at 20℃. Bak-transformed cells were identically cultured in Terrific Broth media. The purification steps were identically performed as described above. The size exclusion chromatography was performed using buffer D (150 mM NaCl, 20 mM Tris at pH 7.5, and 0.5 mM TCEP). The proteins of Bcl-2, Bcl-X L , and Bak were buffer-exchanged to the HBS-EP buffer containing 10 mM HEPES at pH 7.5, 150 mM NaCl, 0.05 mM ethylenediaminetetraacetic acid (EDTA), and 0.005% Tween 20 for further analyses.

Surface plasmon resonance analyses
The binding affinities of the synthetic BH3 peptide derived from Bfk (residues 47-67; SFDVAIIAGRLRMLGDQFNGE) or the fulllength Bfk with Bcl-X L , Bcl-2, and Bak were investigated by surface plasmon resonance (SPR) kinetics experiments using the Biacore T200 apparatus (GE Healthcare, Chicago, IL, USA). The Bfk BH3 peptide and the full-length Bfk prepared in 10 mM sodium acetate buffer (at pH 5.0 and pH 3.6, respectively) were immobilized on a CM5 sensor chip with the HBS-EP buffer using the amine coupling kit The sensor chip was regenerated using 5 mM NaOH for 10 s between cycles. Biacore T200 evaluation software (GE Healthcare, Chicago, IL, USA) was used to calculate the kinetics data using the 1:1 binding model.
The binding affinities of the synthetic BH3 peptide derived from Bid (residues 79-100; EDIIRNIARHLAQVGDSMDRS) as a representative BH3 peptide with Bfk, Bcl-X L , Bcl-2, and Bak were investigated by SPR affinity experiments using the Biacore T200 apparatus (GE Healthcare, Chicago, IL, USA). The Bid BH3 peptide in 10 mM sodium acetate at pH 4.0 was immobilized on a CM5 sensor chip as described above. Reference flow cells were treated identically without the ligands. The responses for the reference flow is the intensity of reflection h, R h is the sum over all reflections, and R i is the sum over i measurements of reflection h. c Friedel pairs were treated as separate observations. d R free = R||F obs |-|F calc ||/R|F obs |, where R free is calculated for a randomly chosen 5% of reflections that were not used for structure refinement. R work is calculated for the remaining reflections. e Twin fractions for the twin operators of (h, k, l) and (-h, -k, l), in order, are calculated during the intensitybased twin refinement of Refmac5 in the CCP4i suite. f Values obtained using MolProbity. cells were subtracted from each sample. Bcl-2 and Bcl-X L at concentrations of 3.13, 6.25, 12.5, 25.0, 50.0, 100, and 200 nM, and Bak at concentrations of 1.25, 2.50, 5.00, 10.0, 20.0, and 40.0 lM were injected over the chip at a rate of 30 lL min À1 for 90 s, followed by dissociation for 300 s in multi-cycle reactions. The sensor chip was regenerated using 5 mM NaOH for 10 s between cycles. Biacore T200 evaluation software (GE Healthcare, Chicago, IL, USA) was used to calculate the kinetics data using the 1:1 binding model.

Molecular dynamics simulation
Protein models for full-length or truncated forms of Bfk and Bid are prepared using Protein Preparation Wizard [34] in Schrödinger software with OPLS4 force field. In particular, an unmodelled loop in the crystal structure of Bfk were generated by Prime [35] in Schrödinger software. All molecular dynamics (MD) simulations were performed using the Desmond [36] in Schrödinger software. Periodic boundary conditions using orthorhombic boxes buffered at 10 Â 10 Â 10 Å or 15 Â 15 Â 15 Å distances were applied for explicit solvent simulations. The system was solvated with water adopting TIP3P water model and 150 mM NaCl after it was neutralized with sodium (or chloride) ions to electrically balance the system. The solvated system containing protein was energyminimized and relaxed for 100 ps by the minimization step of Desmond using OPLS2005 force field. MD simulations were conducted in the NPT (isothermal and isobaric simulation) ensemble, where Martyna-Tobias-Klein method [37] and Nose-Hoover thermostat algorithm [38] were used for isotropic pressure (1 atm) and constant temperature (300 K), respectively. Total 100-300 ns simulations were run and saved as trajectories at 100-300 ps intervals with two independent replicates. The trajectories were analyzed using Simulation Interaction Diagram in Desmond and VMD (Visual Molecular Dynamics) 1.9.3 [39].

Data deposition
The coordinates and structure factors of SeMet-derived and native Bfk are available in the Protein Data Bank under accession codes 7CCM and 7CCL, respectively.

The overall structure of human Bfk adopts the Bcl-2 fold of multidomain Bcl-2 family proteins
Bfk has a unique domain composition involving only BH2 and BH3 domains compared with multi-domain Bcl-2 family proteins that harbor three (BH1-3) or four (BH1-4) domains with or without a C-terminal transmembrane helix (Fig. 1a). To gain insight into the molecular basis for the function of human Bfk, we determined the crystal structure of human full-length Bfk at 2.70 Å resolution using a single-wavelength anomalous dispersion method with selenomethionine-substituted crystal of Bfk (Table 1). Since overall structures and positions of key residues of native and selenomethionine-derived Bfk are nearly identical (a Ca rootmean-square deviation value of 0.184), hereinafter we will discuss the features of the native Bfk structure for clarity (Table 1). Interpretation for the native structure with the relatively high B-factor (an average value of 70.7 Å 2 ), probably due to the crystal quality and intrinsically flexible nature of its fold, was complemented and confirmed by comparing with the selenomethioninederived structure (an average protein B-factor of 42.8 Å 2 ). The Bfk structure consists of two central hairpin-shaped hydrophobic a-helices (a5-a6) encompassed by six amphipathic a-helices (a1-a2, a3-a4, and a7-a8) (Fig. 1b). The overall structure of Bfk adopted the Bcl-2 fold typical of multi-domain Bcl-2 family proteins, which is consistent with the results of structural similarity analysis using the Dali server [40] where the overall structure of Bfk resembles that of Bcl-X L (Z-score of 11.6; PDB code 6RNU), Mcl-1 (Z-score of 11.4; PDB code 6QGD), Bcl-2 (Z-score of 10.3; PDB code 6QGH), Bak (Z-score of 8.4; PDB code 2M5B), and Bax (Z-score of 7.8; PDB code 5 W60) ( Supplementary Fig. S1). However, there were subtle differences from the other structures; for example, Bfk contains an additional alpha helix (a1 0 ) following helix a1 and a helix a8 shorter than those of the multi-domain proteins ( Fig. 1b and Supplementary Fig. S1). Furthermore, noticeable structural differences were found among them in the conformations of helices a2-a4 (Fig. 1b,c and Supplementary Fig. S1), as described later in detail.

The Bfk structure lacks the typical BC groove, albeit with the Bcl-2 fold
Multi-domain Bcl-2 family proteins typically form a large hydrophobic cleft, also known as the BC groove, surrounded by helices a2-a5 (Fig. 2b), which enables their direct interaction with the BH3 domain of other Bcl-2 family proteins to regulate intrinsic apoptotic signals [5]. The Bfk structure with the canonical Bcl-2 fold was found to possess helices a2-a5 at similar positions as shown in multi-domain proteins, but lacked the BC groove owing to a distinct conformation of helices a3-a4, with a small angle between them and a long helix a4 of up to 20 residues (Fig. 2a). In the structure of Bfk, helices a3-a4 with an angle much smaller than those in the multi-domain proteins cannot make a large cleft between them, and the helix a4 that is long enough to reach helix a2 blocks a groove extension via hydrophobic contact between Trp99 on helix a4 and Phe64 on the a2-a3 loop with a distance of 3.5 Å (Fig. 2a). Interestingly, the helix a2 is positioned to lock off the top of the BC groove, which completely blocks the P3 and P4 among four pocket regions (P1-P4) present in the typical BC groove of other Bcl-2 family proteins ( Supplementary Fig. S2). In contrast, the structure of Bcl-X L , a representative multi-domain Bcl-2 family protein, has a large angle between helices a3-a4 and a short helix a4, thereby forming a large BC groove that harbors P1-P4 pockets responsible for hydrophobic contacts with other BH3 domain [41,42] (Fig. 2b and Supplementary Fig. S2).
When we measured the angle (x°) between helices a3-a4 and the length (Å) of helix a4 using structures of Bcl-2 family proteins in the BH3-unbound state, the differences between Bfk and other Bcl-2 family proteins were clearly compared ( Fig. 2c and Supplementary Fig. S3). The structure of Bfk exhibited a remarkably small angle between helices a3-a4 and a longer helix a4 (19.6°and 30.0 Å, respectively), whereas the structures of multi-domain Bcl-2 family proteins, including Bcl-2, Bcl-X L , Bcl-W, Mcl-1, Bfl-1, Bax, Bak, and Bok, show large angles between helices a3-a4 (>32.4°) and a shorter helix a4 (15.8-20.9 Å) (Fig. 2c). These findings indicate that the Bfk structure, albeit with the Bcl-2 fold, does not bear the BC groove to accommodate a BH3 domain from other Bcl-2 family proteins.
In addition, owing to the absence of a BH1 domain, Bfk also lacks a NWGR motif that in general is conserved in multi-domain Bcl-2 family proteins (Fig. 2d), which is known to play a key role in the recognition of a BH3 domain from other Bcl-2 family proteins [43]. This led us to speculate that owing to the absence of both the BC groove and the NWGR motif, Bfk cannot recognize a BH3 domain of other Bcl-2 family proteins. In support of this speculation, using a surface plasmon resonance (SPR) method, we measured the binding ability of the BH3 domain from Bid as a representative BH3 peptide to Bfk in comparison with that of other Bcl-2 family proteins such as Bcl-X L , Bcl-2, and Bak. Bfk showed no appreciable affinity for the BH3 peptide at concentrations up to 40 lM, whereas Bcl-X L , Bcl-2, and Bak expectedly had binding affinity with dissociation constant (K D ) values of 87 (±1) nM, 44 (±4) nM, and 37 (±14) lM, respectively ( Fig. 2e and Supplementary Fig. S4). Therefore, unlike other multi-domain Bcl-2 family proteins, full-length Bfk does not possess the NWGR motif and the BC groove suitable for the recognition of a BH3 domain from other Bcl-2 family proteins, implying that the fulllength form of Bfk remains inert with respect to apoptotic regulation.
3.3. The a1 0 -a2 loop, cleavable by caspases, is anchored to a novel pocket on the other Bfk molecule A part (residues 32-42) of the loop between helices a1 0 and a2 (a1 0 -a2 loop; residues Ser31-Asp49) could not be modeled due to the lack of electron densities, presumably reflecting the flexible nature of the region (Fig. 1b). The sequence analysis of Bfk revealed the presence of a cleavage site (Asp41-Ser42 following a DEVD motif) targeted by caspase-3/7 in the a1 0 -a2 loop region, as evidenced by peptide and proteomic studies for substrate preference motifs of caspases [23,44]. Notably, in the Bfk crystal structure, the Gly43-Ser47 region of the a1 0 -a2 loop that is located immediately adjacent to the cleavage site was clearly visible through stabilization via intermolecular interactions among three monomeric molecules in the crystallographic asymmetric unit. That is, the loop region of one molecule was shown to be docked to an uncharacterized pocket on the other molecule, which was mainly formed by helix a2, the a4-a5 loop, helix a5, and helix a8 (Fig. 3a-c). Indeed, the loop residues between helices a1 0 and a2 formed extensive interaction networks via several hydrogen bonds and van der Waals contacts in the novel pocket, where the sidechain of Glu44 was determined to form two hydrogen bonds with the sidechains of Glu162 and Arg110 from the other molecule (Fig. 3b). In addition, the mainchains of Glu44, Pro45, and Cys46 and the sidechain of Ser47 made polar interactions with the sidechains of Ser105, Asp103, and Gln63 (Fig. 3b). In the middle of the unknown pocket, Pro45 was found to occupy a hydrophobic region surrounded by helices a2, a4, and a5 from the other molecule (Fig. 3b,c). In molecular dynamics (MD) simulations using two interacting molecules of Bfk, these extensive intermolecular interactions between the loop region and the novel pocket were constantly sustained over a 300 ns simulation (Fig. 3d). Moreover, the comparison of root-mean-square fluctuation (RMSF) values from the MD appeared to be stabilized when bound to another Bfk molecule (the red line in Fig. 3e), possibly allowing the next cleavage site to be positioned and poised for the attack of caspases. Considering that the formation of truncated Bfk through caspase-mediated cleavage is necessary to become a pro-apoptotic protein [23], the anchoring of the loop region adjacent to the cleavage site to the novel pocket from the other molecule might contribute to Bfk activation for pro-apoptotic activity.

The helix a5 specific for Bfk and Bid contributes to their characteristic Bcl-2 fold formation
Bid is the only protein that has the helix a4 with a length comparable to that of Bfk among Bcl-2 family proteins (Fig. 2c). In addition, structures of both Bfk and Bid share the unique Bcl-2 fold lacking a canonical BC groove [45] (Supplementary Fig. S1). Bfk has a high sequence similarity (of 52.9% and 17.8% sequence identity) with Bid, showing that the two commonly contain the caspase cleavage sites on the flexible loop before the BH3 domain of helix a2 (Fig. 4a). Moreover, Bfk and Bid have helix a5 with a highly con- served sequence, whereas multi-domain Bcl-2 family proteins instead share a BH1 domain in this region (Fig. 4a,2d). Since the helix a5 comprising the BH1 domain in multi-domain Bcl-2 family proteins plays a central role in forming the solvent accessible hydrophobic BC groove in the Bcl-2 fold [46] ( Supplementary  Fig. S1), Bfk and Bid, which lack the BH1 domain component, instead possess a characteristic helix a5, forming a Bcl-2 fold distinct from the multi-domain Bcl-2 family proteins. Indeed, when analyzing structures of Bfk and Bid, it was shown that the intrinsic properties of helices a5 of Bfk and Bid influence the conformation of their peripheral helices a3-a4 in a similar way, obstructing the BC groove in both structures (Fig. 4b,c). In detail, the structures of Bfk and Bid revealed that they share aliphatic amino acids on helix a5 with similar conformations; Leu113, Ala114, Val115, Val117, Leu119, and Leu120 in the Bfk structure corresponded to Leu149, Val150, Leu151, Leu153, Leu154, and Ala155 in the Bid structure, respectively (Fig. 4d,e).
In both Bfk and Bid, the helix a5 contains positively charged residues of Arg110 and Lys146 at the N-terminus of the helix, respectively, instead of a tryptophan residue (such as Trp144 of Bcl-2), which is highly conserved in multi-domain Bcl-2 family proteins as one residue of the NWGR motif in the BH1 domain (Fig. 4f,g and Supplementary Fig. S6). As a result, Arg110 of Bfk established hydrophilic interactions with Gln155 and Glu162 of Bfk on the Cterminal a7-a8 helices, and similarly, Lys146 of Bid interacted with Asp195 and Asn192 of Bid (Fig. 4f,g). In contrast, the corresponding tryptophan residue in multi-domain Bcl-2 family proteins makes structurally conserved hydrophobic contacts with a WIxxxGGW motif of BH2 domain at helices a7-a8 (Supplementary Fig. S6a), which is essential for the Bcl-2 family heterodimerization-mediated regulation of apoptosis [46]. In order to investigate a structural impact of the distinct residue in the helix a5 of Bfk and Bid, we conducted a MD simulation using mutant structures (Bfk R110A, Bfk R110W, Bid K146A, and Bid K146W) and observed that the interaction networks between helix a5 and helices a7-a8 mediated by Arg110 of Bfk or Lys146 of Bid significantly disappeared in the mutant forms ( Supplementary   Fig. S7). Owing to such characteristics of helix a5, it was suggested that Bfk and Bid form their own core structures that differ from the BC groove-preserving multi-domain Bcl-2 family proteins, probably for further Bid-like action releasing the BH3 domain. Therefore, the full-length structures of Bfk and Bid, at first glance, seem to adopt a fold similar to those of other multi-domain proteins, but they completely differ in the way they form their inner core, thereby remaining inactive and unable to interact with other Bcl-2 family proteins until they are further activated by caspases.

The truncated form of Bfk is suggestive of its active characteristics
Bfk and Bid have in common the pseudo-BH4 domain containing an ECIxNxLxxxFL sequence (where Â represents any amino acid) near the N-terminus (Fig. 4a), which has been known to suppress pro-apoptotic activity until the full-length Bid is cleaved by caspases and becomes a truncated form (tBid) by releasing the N-terminal fragment [15,16,45]. In the structures of Bfk and Bid, the N-terminal regions containing the pseudo-BH4 domain were found to have an additional helix a1 0 following helix a1, although their orientations were different (Fig. 5a,b). The helices a1 and a1 0 were shown to cover the core region of the Bcl-2 fold; Ile14, Ile18, and Phe22 on helix a1 of Bfk (equivalent to Ile16, Leu20, and Phe24 of Bid) established hydrophobic contacts with the groove formed by helices a2, a5, and a6 (Fig. 5a,b). Moreover, the conserved residues Glu12, Cys13, and Asn16 of Bfk (equivalent to Glu14, Cys15, and Asn18 of Bid) on the opposite side of helix a1 were equally exposed to solvent, which was determined to contribute to similar features of the surface formed by helix a1 (Fig. 5a,b). Considering that the helices a1 and a1 0 are bound to the core structure in full-length Bfk and Bid, as proposed in tBid formation [17,18], the detachment of the helices should be preceded after caspase cleavage, thus exposing and releasing the helix a2 containing the BH3 domain that is actually involved in the pro-apoptotic activity of Bfk.
To determine whether the flexibility of helix a2 in the truncated form could be increased compared with that of the full-length form, we performed MD simulations using full-length and truncated structures of Bfk and Bid. A comparison of RMSF values between them showed that the helix a2 became more flexible in the truncated forms of both Bfk and Bid (Fig. 5c,d), implying that the helix a2 constituting the BH3 domain can be easily released for the interaction with other Bcl-2 family proteins. Since we could not obtain a stable truncated Bfk (tBfk) in aqueous solution, we measured the binding affinity of the Bfk BH3 peptide (residues 47-67; SFDVAIIAGRLRMLGDQFNGE) instead of that of tBfk or full-length Bfk versus those of other Bcl-2 family proteins to confirm that the released BH3 domain can bind to anti-apoptotic Bcl-2 proteins or pro-apoptotic pore-formers. Indeed, whereas the full-length Bfk exhibited no binding affinity for Bcl-2, Bcl-X L , or Bak (the orange lines in Fig. 5e), the Bfk BH3 peptide showed strong binding affinity for Bcl-X L and Bcl-2 with K D values of 36.9 nM and 17.4 nM, respectively, and weak binding affinity for Bak with a K D value of 70.0 lM (the green lines in Fig. 5e and Supplementary Fig. S8). Collectively, full-length Bfk that lacks the BC groove remains inert with respect to apoptotic regulation, but upon cleavage by caspases and detachment of the N-terminal fragment, the truncated form of Bfk exhibits its pro-apoptotic characteristics by enabling the BH3 domain to interact with other Bcl-2 family proteins.

Discussion
Human Bfk, belonging to the Bcl-2 family, is predominantly expressed in tissues of the gastrointestinal tract and its expression is substantially reduced in tumors of the gastrointestinal tract such as the colon, small intestine, and stomach [20,21]. Additionally, since upregulated Bfk expression promotes apoptosis, human Bfk has been generally regarded as a pro-apoptotic protein [20]. However, the mechanism underlying Bfk-mediated apoptotic regulation has been elusive, given that the unique domain composition of pro-apoptotic Bfk (harboring BH2 and BH3 domains) does not correspond to any of the pro-apoptotic protein subgroups among Bcl-2 family proteins. Indeed, Bfk is neither a pore-former nor a BH3-only protein. Moreover, a three-dimensional structure of Bcl-2 family protein comprising only BH2 and BH3 domains has not been reported to date. In this study, we report the crystal structure of human Bfk, as the first structure among Bcl-2 family proteins containing BH2 and BH3 domains. The overall structure of Bfk exhibited a Bcl-2 fold with high structural similarity to those of multi-domain Bcl-2 family proteins but lacked the BC groove due to the absence of the BH1 domain. Rather, Bfk-specific key residues of helix a5 and the way of forming the core structure in the Bfk structure resembled those of Bid, a BH3-only protein in the Bcl-2 family, suggesting that Bfk exerts pro-apoptotic activity in a manner similar to Bid.
Bid is a Bcl-2 family protein that bridges the intrinsic and extrinsic apoptotic pathways, as tBid translocates to the mitochondrial outer membrane to initiate MOMP, following cleavage of fulllength Bid by caspase-3 and caspase-8 in the cytosol [45,47]. It was revealed that tBid forms an extended structure with a C-shaped configuration and embeds into the mitochondrial outer membrane via helices a5-a6, thereby enabling the efficient recruitment of cytosolic Bax or Bak to the mitochondrial outer membrane for fur-ther pore formation [17,18,48]. In the Bfk structure, Bfk possesses the helix a5 that shares distinct properties with Bid, which is distinguished from helices a5 constituting the BH1 domain conserved in other multi-domain Bcl-2 family proteins (Fig. 4). In particular, helix a5 of Bfk and Bid contains predominantly aliphatic residues at the center and positively charged residues (Arg110 in Bfk and Lys146 in Bid) in the N-terminal direction (Fig. 4). In addition, the helix a5 in both Bfk and Bid structures is covered by helix a1 containing the pseudo-BH4 domain, of which the invariant residues make hydrophobic contacts between helices a1 and a5 ( Fig. 5f,g), implying that they share structural and sequential features in the formation of their distinct folds. It is interesting that tBid utilizes helices a5-a6 to associate with the mitochondrial outer membrane instead of a transmembrane helix that other Bcl-2 family proteins have at C-terminus in many cases [6]. In this regard, tBfk that lacks a transmembrane region but shares a similar motif in the helix a5 with tBid might also adopt its helix a5-a6 region for a subcellular localization and association with a membrane. Therefore, we suggest that Bfk and Bid have the same mechanism underlying the formation of the truncated conformation and association with MOM, in which tBid dissociates from helices a1-a1 0 upon caspase cleavage and contact with mitochondrial membranes [49][50][51] and its exposed helix a5 embeds into mitochondrial outer membranes [13,17,18]. However, the cellular localization of tBfk has not yet been identified and the cell-based evidence for the unfolding of tBfk in the presence of a membrane to elicit a pro-apoptotic signal remains lacking. Therefore, our speculation of the membrane-associated subcellular localization of tBfk to promote apoptosis needs to be addressed through further studies.
Since the truncated Bfk, through cleavage by caspases, becomes a pro-apoptotic protein [23], recognition of the cleavage site on the flexible loop of Bfk by caspases can be one of the rate-limiting steps for further activation. In the Bfk crystal structure, we observed the intermolecular interactions between two adjacent molecules in the crystal structure of monomeric Bfk, especially through the flexible a1 0 -a2 loop containing a caspase cleavage site. Accordingly, it is assumed that the loop part next to the cleavage site is fixed, so that the caspases can easily recognize the exposed site. Together with MD simulation, the a1 0 -a2 loop bound to another Bfk molecule was shown to be considerably stable. Such interactions between two Bfk molecules could be an artifact due to crystallization under high concentrations, but since a dimeric form was also observed during the gel filtration step ( Supplementary Fig. S9), however minor, it is possible that the loop stabilization through the interplay between two Bfk molecules could play a role in the truncation and activation processes.
tBfk would be structurally rearranged to an active conformation that exposes its BH3 domain to interact with other Bcl-2 family proteins. The comparative MD simulation-based studies on Bfk and Bid showed that the truncated forms from both Bfk and Bid exhibit elevated fluctuations of the BH3-containing helix a2 (Fig. 5c,d), which tend to expose the BH3 domain. Furthermore, as tBfk was reported to exhibit pro-apoptotic activity in a Bax/ Bak-dependent manner [23], we demonstrated that the Bfk BH3 domain is capable of binding to Bak (K D = 70 lM) (Fig. 5).
Therefore, tBfk would be induced to release the BH3-containing helix a2 for further activation of Bax/Bak-mediated apoptosis through conformational changes as shown with tBid. Interestingly, kinetic constants such as on and off rates of Bfk BH3 against Bcl-X L , Bcl-2, and Bak in our SPR experiments are shown to be fast compared to other BH3 domains [52][53][54]. In our comparative SPR experiments between Bfk BH3 and Bid BH3 against Bcl-2 family proteins, the Bfk BH3 tends to exhibit slightly faster off rates (Supplementary Figs. S8a and S4a). The fast on and off rates of Bfk BH3 might be advantageous for a 'hit and run' mechanism that is a possible model for an effector activation [55]. Although whether the direct binding of BH3-only proteins such as Bid activates the pore formation of Bax/Bak into MOM remains debated, many studies have shown that the Bid BH3 domain induces the oligomerization of Bax/Bak essential for further pore formation via transient binding with sub-micromolar affinity [6,56,57]. Although our study has shown the direct binding of Bfk BH3 domain to Bak with a fast off rate, it is not enough to fully demonstrate the role of Bfk BH3 as a direct activator because it might act as a sensitizer, like Bad, which elicits pro-apoptotic activity but does not directly activate the pore forming of Bax and/or Bak. Thus, the Bfk BH3 domain needs to be further investigated in the cellular experimental system with respect to whether it could promote MOMP by activating Bcl-2 family pore-formers. Although Bfk resembles Bid in various aspects, the BH2 domain at the C-terminus of Bfk does not exist in Bid. In a previous study on Bcl-G, another Bcl-2 family protein harboring BH2 and BH3 domains, the autorepression of pro-apoptotic signaling was suggested as a function of the BH2 domain because the proapoptotic activity of Bcl-G following deletion of the BH2 domain was increased [58]. Structural analysis of Bfk in comparison with the BH3-only Bid or multi-domain Bcl-2 family proteins revealed that the BH2 domain on the C-terminal helices a7-a8 of Bfk establishes interaction networks with helix a5 in the Bfk core region, contributing to the structural integrity of the core Bcl-2 fold (Fig. 4f). Indeed, MD simulations with tBfk and tBid with and without the BH2 domain, respectively, showed that the BH2-containing helices a6-a8 of tBfk were less flexible than those of tBid (Fig. 5c,   d). Thus, the BH2 domain of Bfk is thought to attenuate the structural rearrangement when the Bcl-2 core fold is wound, leading to a delay in the rapid exposure of the BH3 domain and thus reducing pro-apoptotic signals. This is supported by the increased apoptosis observed upon deletion of the BH2 domain from Bcl-G [58], although its concomitant molecular basis could not be verified due to a lack of structural knowledge of Bcl-G. The presence of the BH2 domain in Bfk appears to strictly regulate its proapoptotic activity compared to that with the BH3-only protein Bid. Human Bfk is mainly expressed in the gastrointestinal tract, where stem cells undergo rapid proliferation, differentiation to terminal mature cells, and cell death by apoptosis [59]. This suggests a reasonable speculation that Bfk, mainly expressed in gastrointestinal tract, induces apoptosis under stringent regulation through the BH2 domain.
Recently, AlphaFold is emerging as the most powerful tool to predict a protein structure based on the sequence. When we compared the crystal structure of Bfk with the AlphaFold-predicted structure available in the EBI database, despite the high similarity (a Ca root-mean-square deviation value of 0.741) in their overall structures, there are structural differences between them, which might be related to functional aspects. Most importantly, the sidechain of Arg110 on helix a5 in the crystal structure makes a hydrogen bond with Gln155 of helix a7, whereas the one of Arg110 in the AlphaFold-predicted structure is directed toward helix a2 by forming a hydrogen bond with Gln63 ( Supplementary Fig. S10).
In the crystal structure, the hydrogen bonds of Arg110 with Gln155 and Glu162, are thought to enhance an interaction network of a5 with helices a7-a8, which belong to a region for BH2 domain ( Supplementary Fig. S10). Therefore, the conformation of Arg110 seen in the crystal structure that might affect the structural integrity is noteworthy. Another noticeable difference between the crystal structure and the AlphaFold structure is the helix a1 0 next to helix a1 (Supplementary Fig. S10). All Bcl-2 family proteins including Bfk have the long loop of the flexible nature between helices a1 and a2, so that most Bcl-2 family proteins and even the structure predicted by AlphaFold present this region as a loop or as in not modeled. Interestingly, Bfk and Bid contain additional helix a1 0 before a flexible loop starts, and the first crystal structure of Bfk confirms the existence of such helix a1 0 and reveals its difficult-to-predict characteristic structure in detail. Considering that the region for helix a1 0 is encompassed by several helices of symmetrical molecules and exhibits a large RMSF during MD simulation, the possibility that the helix a1 0 could be caused by crystal contact cannot be excluded, but it might be valuable to compare them and further describe the additional information that can only be obtained through experimental structural determination.
In conclusion, whereas functional mechanisms of Bcl-2 family proteins in the intrinsic pathway of apoptosis have been actively studied, a molecular basis for Bcl-2 family proteins that belong to an orphan member, owing to its unique composition of BH2 and BH3 domains, has been remarkably in veil. In this study, we first reported the crystal structure of human Bfk as an orphan member among the Bcl-2 family proteins, which revealed that the BC groove is absent and its P1-P4 pockets remains blocked by the unique conformation of helices a2-a4. In addition, we opened the possibility of a Bid-like apoptotic mechanism for Bfk via structural, biophysical, and computational analyses. Our findings will provide insight into the molecular basis underlying apoptotic regulation by Bfk as a novel Bcl-2 family protein.