Structural basis of BFL-1 for its interaction with BAX and its anti-apoptotic action in mammalian and yeast cells.

BFL-1 is the smallest member of the BCL-2 family and has been shown to retard apoptosis in various cell lines. However, the structural basis for its function remains unclear. Molecular modeling showed that BFL-1 could have a similar core structure as BCL-xL, consisting of seven alpha helices, although both proteins share only the conserved BCL-2 homology domains (BH1 and BH2 domains), but otherwise have very limited sequence homology, particularly in the N-terminal region. We demonstrated in the yeast two-hybrid system that BFL-1 interacts strongly with human BAX but is not able to form homodimers nor to interact with human BCL-2 or BCL-xL. Overexpression experiments in REF52 rat fibroblasts showed that BFL-1 conferred increased resistance to apoptosis induced by serum deprivation. BFL-1 had also the ability to neutralize BAX lethality in yeast. BAX requires the BH3 domain for interaction with BFL-1. However, the minimal region of BFL-1 for the interaction with BAX in coimmunoprecipitation experiments was not sufficient to protect cells from apoptosis. Further examination of BFL-1 and several other anti-apoptotic proteins suggests a more general type of structure based on structural motifs, i.e. a hydrophobic pocket for the binding of proapoptotic proteins, rather than extended sequence homologies.

BFL-1 is the smallest member of the BCL-2 family and has been shown to retard apoptosis in various cell lines. However, the structural basis for its function remains unclear. Molecular modeling showed that BFL-1 could have a similar core structure as BCL-xL, consisting of seven ␣ helices, although both proteins share only the conserved BCL-2 homology domains (BH1 and BH2 domains), but otherwise have very limited sequence homology, particularly in the N-terminal region. We demonstrated in the yeast two-hybrid system that BFL-1 interacts strongly with human BAX but is not able to form homodimers nor to interact with human BCL-2 or BCL-xL. Overexpression experiments in REF52 rat fibroblasts showed that BFL-1 conferred increased resistance to apoptosis induced by serum deprivation. BFL-1 had also the ability to neutralize BAX lethality in yeast. BAX requires the BH3 domain for interaction with BFL-1. However, the minimal region of BFL-1 for the interaction with BAX in coimmunoprecipitation experiments was not sufficient to protect cells from apoptosis. Further examination of BFL-1 and several other antiapoptotic proteins suggests a more general type of structure based on structural motifs, i.e. a hydrophobic pocket for the binding of proapoptotic proteins, rather than extended sequence homologies.
Apoptosis plays an important role in the development of multicellular organisms and in various pathological processes (1). Proteins of the BCL-2 family are important regulators of apoptotic cell death (2). In mammalian cells, apoptosis can be inhibited by overexpression of anti-apoptotic members of the family, including BCL-2, BCL-xL, MCL-1, BCL-w and A1. In contrast, proapoptotic members of the family, such as BAX, BAK, and BCL-xS, promote apoptosis and can antagonize the protective effects of BCL-2 and BCL-xL. Interestingly, some functional aspects of the BCL-2 family proteins are conserved in yeast. Expression of BAX and BAK in the budding yeast Saccharomyces cerevisiae and in the fission yeast Schizosaccharomyces pombe can prevent cell growth and depending on growth conditions be lethal (3,4). BCL-2, BCL-xL, and MCL-1 are able to suppress the lethal activity of BAX in yeast (3,5,6), analogous to findings in mammalian cells. Despite intensive research the molecular mechanisms by which the BCL-2 proteins regulate apoptosis are not yet fully understood.
Several domains of the BCL-2 family proteins are evolutionarily conserved. These regions are described as BCL-2 homology (BH) 1 domains. BH1 and BH2 domains are common throughout the family for anti-apoptotic members (7,8). The most thoroughly studied anti-apoptotic members, BCL-2 and BCL-xL, share the BH3 and BH4 domains (9,10). The proapoptotic members of the family, such as BAX and BAK, also share the BH3 domain, which, however, has distinguished features as compared with that of the anti-apoptotic members (9,11,12). Numerous studies indicate that the members of the family are able to interact with each other via regions including the homology domains, and it is generally accepted that proteinprotein interactions of the BCL-2 family members may be one of the most important mechanisms in the regulation of apoptosis (13).
A1 is a BCL-2-related protein. It was originally identified as a murine hematopoietic-specific, granulocyte-macrophage colony-stimulating factor-inducible gene product (14). Human BFL-1 can be considered as a homologue of mouse A1, because these two proteins share about 72% amino acid identity (14,15). BFL-1 appears to be induced by the inflammatory cytokines, tumor necrosis factor and interleukin-1 (15), and it has been shown to be a direct transcriptional target of nuclear factor-B (16 -18). Like BCL-2, BFL-1 can prolong cell survival; it retards tumor necrosis factor-induced apoptosis in the human dermal microvascular cell line, HMEC-1 (15), and p53induced apoptosis in the primary rat kidney cells (19). However, the function of BFL-1 seems to be distinct from that of BCL-2, because BFL-1 permits cell proliferation (19,20). Mutational analysis of BFL-1 revealed that mutations within the BH1 and BH2 domains abolished the anti-apoptotic activity, whereas the N-terminal domain contributed to the proliferative activity of BFL-1 (20). The proposed BH3 domain, however, did not promote anti-apoptotic activity (20).
Because the structure-function relationship of the BFL-1 protein is still unclear, we have used molecular modeling to investigate the possible structure of BFL-1 and further dissected the structure and function of BFL-1 on the basis of structural similarities with other members of the BCL-2 family.

EXPERIMENTAL PROCEDURES
Molecular Modeling-After manual optimization to obtain different alignments of the sequences within Insight II (Molecular Simulations, * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  Inc.), models of human BFL-1 were built using the program Modeller (21,22) and examined. Further optimization was done manually using the graphics program O (23). Verification of the quality of the models was done using stereochemistry values calculated in Modeller and profile analysis of the protein sequence threaded onto its own threedimensional template (24). Further analysis with Procheck (25) and tools within the program O helped to confirm these results.
Plasmid Constructions-Several cDNA sequences encoding BFL-1 (14) were identified in the expressed sequence tag (EST) data base of GenBank TM using a Blast (NCBI) search. EST cDNA clone N28416 was purchased from Research Genetics, Inc. and sequenced. The cDNA was subcloned in-frame into the mammalian expression vector pcDNA3.1/His (Invitrogen), yeast expression vector pYES2 (Invitrogen), and the yeast two-hybrid plasmids pLex-a and pVP16 (26). To construct the deletion mutants M1 to M7 of BFL-1, EST clone N28416 was used as a template for polymerase chain reaction. Forward and reverse primers are listed in Table I. All fragments generated by polymerase chain reaction were cloned in-frame into the EcoRI/NotI sites of the mammalian expression vector pcDNA3.1/His (Invitrogen) and confirmed by sequencing. For yeast two-hybrid studies, BAX constructs were as described earlier (3,27). Expression of the Bcl-2 family members in yeast was confirmed by Western blotting using antibodies against LexA (CLONTECH) (27). Clone T81750 of GenBank TM was identified as human BAG by a Blast search. The EST cDNA clone was sequenced and used for cloning.
Cell Lines, Cell Culture, Transfection, and Western Blot Analysis-REF52 rat fibroblasts were obtained from the American Type Culture Collection and cultured as described (27). For transfection, REF52 cells were seeded at 10 -30% confluency (ϳ5 ϫ 10 4 cells) in 12-well cell culture plates and grown for 24 h. Cells were then transfected with a total of 2 g of plasmids. Transfections were performed successfully with either LipofectAMINE Reagent (Life Technologies, Inc.) or Superfect (Qiagen). Mouse pro-B cell line FL5.12 was obtained from IDUN Pharmaceuticals, Inc., San Diego, CA. The culture and transfection of FL5.12 were as described (7). For analysis of the expression and stability of the BFL-1 deletion protein M5, REF52 cells were transfected with the pcDNA3.1/His construct and selected with G418. 10 7 cells were collected before and after serum deprivation, and cell lysates were subjected to electrophoresis. The expression level of the M5 mutant protein was examined by Western blotting using a monoclonal anti-His antibody (Milan Analytica AG).
Cell Viability Assay-REF52 cells were transfected with 1 g of plasmids encoding the green fluorescent protein (pEGFP-N1, CLON-TECH) together with 1 g of pcDNA3 carrying genes to be tested. The pcDNA3 plasmid served as control. Apoptotic cells were round up after 3 h of serum deprivation (29). Green fluorescent cells were monitored, and normal flat and round apoptotic cells were counted by microscopic examination. For interleukin-3 deprivation experiments, transfected FL5.12 cells were washed three times in serum-free medium to remove interleukin-3 and cultured at 5 ϫ 10 5 cells/ml in triplicate. At various time points at least 100 cells from each individual culture were analyzed by trypan blue exclusion staining. Cell survival is expressed as percentage of surviving cells/total number of cells, given with the standard deviation of the assay (49).
Coimmunoprecipitation Analysis-Purified 6His-BAX protein, produced in Escherichia coli, was kindly provided by Dr. G. Fendrich (Novartis). BFL-1 and its deletion mutants were expressed in vitro by the TNT system (Promega); appropriate cDNAs were sub-cloned into pcDNA3.1/His as described above. For in vitro transcription/translation, 2 g of plasmid DNA were linearized (creating 5Ј-overhangs) and used with TNT wheat germ lysates (Promega) in the presence of [ 35 S]methionine. 6His-BAX fusion protein (10 g) was mixed with 20 l of in vitro translated 35 S-labeled proteins for 1 h on a rotator in NETN buffer (20 mM Tris, pH 8.0, 100 mM NaCl, 1 mM EDTA and 0.5% Nonidet P-40). The mixtures were then diluted to 200 l with Tris-buffered saline/0.5% Nonidet P-40. A complete protease inhibitor mixture (Roche Diagnostics) was added. The solution was precleared with 2.5% (v/v) Sepharose 4B (Amersham Pharmacia Biotech) for 45 min. After removal of the gel by centrifugation at 1000 rpm, the supernatant was incubated with 5 g of a polyclonal anti-human BAX antibody (Pharmingen) for 1 h. Subsequently, 25 l of a 50/50 slurry of protein G-fast flow-Sepharose (Amersham Pharmacia Biotech) were added, and the incubation was continued for another hour. The protein G-fast flow-Sepharose was collected by low speed centrifugation (3000 rpm), washed three times with Tris-buffered saline/1% Nonidet P-40 and once with Tris-buffered saline, resuspended in Laemmli buffer, and heated to 100°C for 5 min. After low speed centrifugation, the resulting supernatants were applied to 15% SDS-polyacrylamide gel electrophoresis. Gels were dried and exposed to a phosphorimage screen (Molecular Dynamics). All experiments were performed at 4°C.
Yeast Growth Assay-The yeast S. cerevisiae L40 strain (26) was grown and transformed as described previously (26). Yeast cells were grown on SC plates (28) lacking tryptophan, leucine, uracil, and histidine at 30°C. Cells expressing interacting genes or gene fragments formed colonies within 2-3 days.
Spot Tests in BAX Lethality Experiments-The yeast strain Bi (3) was used in BAX lethality experiments. Cells were suspended in water, and then 5 l of aliquots were dropped onto the appropriate agar plates. The growth media used were mainly as described by Sherman (28). YEP being a rich medium and SD a defined medium with NH 4 SO 4 as a nitrogen source (3). For simplicity all defined media have been described simply as SD and do not have the supplements added specified; the supplements being all those required for growth of the specific strain. Where strains carry a plasmid, an appropriate selection was used (i.e. minus uracil with all episomal plasmids and minus leucine with the integrant). The carbon sources were 2% (w/v) glucose, which prevents BAX expression, or 2% galactose, which induces BAX expression. Supplements were added from 5 mg/ml stock solutions to give the final concentrations of: 20 mg/liter of lysine and histidine-HCl, and 30 mg/liter of leucine and uracil (where needed). Agar was added to a final concentration of 2%.

Molecular
Modeling of BFL-1-BFL-1 is one of the smallest anti-apoptotic members identified so far in the BCL-2 family (14), comprising 175 amino acids with a putative transmem- Interaction between BFL-1 and BAX FIG. 1. Molecular modeling of BFL-1. A, amino acid sequence alignment of BFL-1 with BCL-2 and BCL-xL. The ClustalW Interactive Multiple Sequence Alignment program from the European Bioinformatics Institute (UK) was used to align the amino acid sequences of BFL-1 (Swiss-Prot Accession number Q16548), BCL-2 (Swiss-Prot Accession number P10415), and BCL-xL (Swiss-Prot Accession number Q07817). The conserved domains BH1 and BH2 are indicated. The helices ␣1-␣7 were represented according to Muchmore et al. (30). B, superposition of a model for BFL-1 (highly homologous core region in yellow, N-terminal region in light gray), with the crystal structure of BCL-xL (30) (light blue). The BH1 (blue), BH2 (blue), BH3 (red), and BH4 (green) regions of BCL-xL are labeled. The numbering is from the BFL-1 sequence. The sequence alignment, and hence the model, is well defined between residues 68 and 148, whereas the parts for residues 6 -22 and 36 -68 are more ambiguous. For this model brane domain. Preliminary sequence alignment of BFL-1 with the well characterized apoptosis-suppressing family members, BCL-2 and BCL-xL, shows that BFL-1 contains the conserved BH1 and BH2 domains but has only very limited homology with the other domains. Therefore, we wanted to explore the structural basis for the cell death suppressor activity of BFL-1. Models of BFL-1 were built based on the three-dimensional structure of BCL-xL (30,31). The homology between both proteins in the region of the BH1 and BH2 domains (helices 4 -7) is high and therefore allowed to build this part of the model with confidence (Fig. 1A). The core structure shows hydrophobic patches on the surface, which would be covered by the structure of the remaining nonhomologous parts of the sequence. Several models for the N-terminal part of the sequence were built to test different alignments of the sequences obtained by manual optimization because of the low homology. The most reasonable models required a sequence alignment similar to that depicted in Fig. 1A, where three helices were built to resemble the N-terminal structure of BCL-xL (Fig. 1B). If helices were built for BFL-1 based on the sequence alignment, which suggests that helix 3 does not exist (20), there was no stereochemically reasonable way to connect helices 2 and 4 as there are only 7 residues to cover more than 20 Å. Therefore, it was reasonable to model the BFL-1 sequence on the BCL-xL fold, which includes all 7 helices. Accordingly, residues aligned at positions 1, 5, and 6 of the core BH3 domain (8), which are highly conserved in BH3 domains of the BCL-2 family members, are not conserved in BFL-1. We therefore divide the N-terminal sequence into structural homology regions: domain A for helix 1 (amino acids 1-32), domain B for helix 2 (amino acids 36 -49), and domain C for helix 3 (amino acids 53-62).
Interaction of BFL-1 with BAX-The yeast two-hybrid system was used to explore the interactions of BFL-1 with other BCL-2 family members, like BAX (27,32). cDNAs encoding human BFL-1, BAX, and truncations of BAX were cloned into the yeast two-hybrid vectors pLex-a and pVP16. The cloning The BAX truncations were fused to VP16. B, full-length BFL-1, fused to LexA and BAX (amino acids 1-172, BAX), as well as the BAX truncations A-H, fused to VP16, were tested as pairs in the yeast two-hybrid system. ␤-Galactosidase activity was measured by the liquid assay as described under "Experimental Procedures." more than 92% of the residues lie in the most favored regions of the Ramachandran plot (25) and there are no residues in the disallowed regions. Verification of the model with the three-dimensional profile method (24) showed that only one small region may not be well modeled. This region lies at the end of helix three and also includes the loop connecting helix three and helix four. It is clear that the local conformation of the structure here is quite different from BCL-xL because there is a deletion in the loop region. led to in-frame fusions of the proteins with the DNA-binding protein, LexA, and the VP16 transcription activation domain (26), respectively. The putative transmembrane domain of BAX was omitted to avoid complications in the two-hybrid system. These constructs were used to transform cells of yeast strain L40 in combinations of two to test the interaction of BFL-1 with BAX. The plasmids were also introduced singly into L40 and tested without an interacting partner to see whether the fusion protein on its own had intrinsic transcription activation activity (autoactivation). The protein-protein interactions were determined as growth in the absence of histidine and by ␤-galactosidase activity. Cotransformation of cells with BFL-1 and BAX resulted in activation of the his3 and LacZ reporter genes. The interaction could be detected with both the qualitative ␤-galactosidase plate assay and the quantitative ␤-galactosidase liquid assay, which detects only strong interactions. BFL-1 showed no interaction with BCL-2, BCL-xL, or BAG, nor was homodimerization of BFL-1 observed (not shown) indicating that BFL-1 in solution is preferentially in a monomeric form, as is BCL-xL.
The growth test and ␤-galactosidase plate assay indicated that BFL-1 interacted, in addition to full-length BAX, also with several BAX truncations. Interactions of eight different BAX truncations ( Fig. 2A) with full-length BFL-1 were tested. The experiment was performed in two different setups, BFL-1 fused to LexA and the BAX truncations to VP16 and vice versa. The growth test and the ␤-galactosidase plate assay showed that all BAX truncations containing the BH3 domain interacted with BFL-1, and those lacking BH3 did not (not shown). The results were the same in both setups, although the LexA fusions of some BAX truncations showed autoactivation (27).
The quantitative ␤-galactosidase liquid assay (Fig. 2B) confirmed the results from the growth and ␤-galactosidase plate assays. BAX truncations A, C, E, F, and G showed strong interactions with BFL-1. The interaction of BFL-1 with the BAX truncation C, which contains essentially only the BH3 domain, was very strong. The only common domain in the truncations, which was necessary and sufficient for a strong interaction with BFL-1, was BH3 (Fig. 2). The interactions of BAX and its truncations with BFL-1 were roughly as strong as the earlier reported interactions of BAX with BCL-2 (27). The fact that BAX appears to interact similarly with BFL-1 as it does with BCL-2, BCL-xL, and BAX (27) supports the idea that BFL-1 could have a three-dimensional structure similar to that FIG. 3. Binding of BFL-1 deletion mutants to 6His-BAX analyzed by immunoprecipitation. A, deletion mutants of BFL-1. BFL-1 represents the schematic organization of the full-length BFL-1 protein (amino acids 1-175). M1-M7 indicates seven deletion mutants. Domains A, B, and C, the homology regions BH1 and BH2 and the ␣ helical regions ␣1-␣7 are indicated as defined in Fig. 1. B, BFL-1 and the BFL-1 deletion mutants were cloned in pcDNA3.1/His, transcribed, and translated in the TNT system in the presence of [ 35 S]methionine. An aliquot was analyzed by SDSpolyacrylamide gel electrophoresis. CRTL is the vector control sample. C, 6His-BAX fusion proteins (10 g) were mixed with 20 l of in vitro translated 35 S-labeled proteins in NETN buffer and then incubated with 5 g of a polyclonal anti-human BAX antibody. Subsequently, protein G-fast flow-Sepharose was added to obtain the precipitated proteins. The resulting samples were applied to 15% SDSpolyacrylamide gel electrophoresis. Gels were dried and exposed to a phosphorimage screen.
of BCL-xL, which consists of 7 helices that create a hydrophobic pocket for BH3 binding.
The interaction of BFL-1 with BAX was seen only when BFL-1 was fused to LexA and BAX to VP16. In the other orientation no interaction was seen in the ␤-galactosidase liquid assay. This is most probably because of the fact that the interaction requires a conformational change, which in some cases is inhibited by the fusion of the proteins to LexA or VP16. We have seen earlier the same phenomenon with BAX and BCL-xL (27).
Identification of BFL-1 Sequences That Mediate the Interaction with BAX-The results from the yeast two-hybrid assays prompted us to further identify the domains of BFL-1 necessary for the interaction with BAX. Full-length BFL-1 and seven different BFL-1 truncations (M1-M7, Fig. 3A) were synthesized separately in an in vitro transcription-translation TNT system (Promega) in the presence of [ 35 S]methionine (Fig. 3B). The empty vector (pcDNA3.1/His) served as a control. Fulllength 6His-tagged human BAX protein was purified from a bacterial expression system. The physical interaction with BAX was assayed by coimmunoprecipitation using an anti-human BAX polyclonal antibody. Fig. 3C demonstrates that BFL-1 and the truncations M1, M3, and M5 could be coimmunoprecipitated with BAX, indicating that these truncations interacted with BAX. The common domains of the truncations M1, M3, and M5 were BH1 and domains B and C (helix 2 and 3; see Fig.  1). However, BH1 alone (M7) or domains B and C alone (M6) did not show any interaction (Fig. 3C). Also the control protein did not bind to BAX. Therefore, the minimal domains of BFL-1 that were required for its interaction with BAX were BH1 plus domains B and C. These domains form the majority of the hydrophobic pocket in BCL-xL that binds the BH3 peptide (33).

Suppression of Cell Death by Full-length BFL-1 but Not by Mutant M5 in REF52 Cells-Suppression of apoptosis by
BFL-1 was studied in REF52 cells, which undergo apoptosis upon serum deprivation (29), round up at an early apoptotic stage and detached from the surface of the dish (27). DNAs coding for full-length BFL-1 and for truncation M5 were cloned into the pcDNA3 expression vector. The BFL-1 constructs were transfected into REF52 rat fibroblasts together with a vector (pEGFP-N1) encoding the green fluorescent protein (GFP). GFP served as marker for successfully transfected cells. 15 h after transfection, the cells were deprived of serum, and 3 h later apoptotic REF52 cells were counted. Fig. 4 shows that after 3 h of serum deprivation, about 50% of cells transfected with the control plasmid (pcDNA3) were apoptotic. Approximately 20 and 10% of the transfected cells could be rescued by BCL-xL and BFL-1 expression, respectively, whereas the expression of BAX increased apoptosis by 25%. The truncated BFL-1 construct, M5, which contains BH1 and domains B and C, did not have statistically significant activity in REF52 cells. M5 showed neither protecting nor killing activity. The lack of a cellular effect was not because of low level or short lived M5 protein; the expression level seen on Western blots was not affected by serum deprivation (data not shown). Similar results were obtained in mouse FL5.12 cells, a murine lymphoid progenitor cell line, in which apoptosis is induced by interleukin-3 deprivation (not shown).
Expression of BFL-1 Rescues BAX Lethality in Yeast-It has been shown that the expression of BAX prevents cell growth and is lethal in the budding yeast, S. cerevisiae (3,34), and the effects of BAX on yeast mitochondria appear to be similar to those of mammalian cells (35). Furthermore BAX effects can be attenuated by expression of BCL-2 and BCL-xL, in a similar manner to that seen in mammalian cells, although the downstream targets of the proteins may not be identical (3). We investigated whether BFL-1 can also protect yeast from BAX expression. BFL-1 was cloned into the pYES episomal vector under the control of the galactose-inducible GAL1 promoter (see "Experimental Procedures"). This plasmid was used to transform Bi, a yeast strain that expresses human BAX-␣ protein from an integrated copy of an artificial gene (3), and the cells were spot tested. On glucose containing medium, BAX expression is repressed and all transformants were able to grow (Fig. 5A). On galactose plates BAX expression is induced. After 2 days no growth could be seen on galactose-containing medium with any of the clones transformed with the BFL-1 or BAG constructs, whereas a transformant expressing BCL-xL was clearly grown (Fig. 5B). However, clear (although weak) growth was seen after 72 h with the BFL-1 transformant. After 7 days (Fig. 5C) growth was obvious with co-expression of BFL-1 (the lower 3 spots in column 1) and BAG (the lower 3 spots in column 2). The specificity of the results was confirmed by using irrelevant clones from a library screening (LT1, LT2, and LT3), none of which supported growth even after 168 h (Fig. 5C). Thus, BFL-1 was able to suppress BAX lethality, but the rescue effect was weaker than that of BCL-xL, consistent with the observation from another yeast system that mouse A1 was a weaker suppressor of apoptosis than BCL-xL (36).

Structural Basis of BFL-1 for Its Interactions and Its Function-
The sequence alignment of BFL-1 with class I and II apoptosis-suppressing family members (8) shows that BFL-1 contains the conserved BH1 and BH2 domains but has only very limited homology with the other domains. Our results from molecular modeling indicate that despite the low sequence homology in the N-terminal region, BFL-1 could still be structurally homologous to BCL-xL, such that seven helices exist in BFL-1 as in the three-dimensional structure of BCL-xL (30). This means that a hydrophobic pocket, which in BCL-xL is known to bind the BAK-BH3 peptide (33), is formed mainly by helices 2, 3, and 5. Based on this model and making the assumption that homodimerization occurs between BCL-2 family proteins in the manner observed for the binding of the BAK-BH3 peptide to BCL-xL, homodimerization of BFL-1 would be unfavorable because there is a lysine at position 46 of BFL-1 instead of a negatively charged residue (Asp-95 in BCL-xL), which would be necessary to interact with Arg-88 of BFL-1 (Arg-139 in BCL-xL) (33). Furthermore, if the binding of BH3 peptides to BFL-1 is structurally similar to the binding of BAK-BH3 to BCL-xL, predictions can be made about the binding of different BH3 ligands to BFL-1 based on the most important interactions observed in the BCL-xL complex structure (33). Hydrophobic interactions are rather well conserved for all potential ligands, but one of the hydrophilic interactions (Glu-129 and Arg-76 in BCL-xL and BAK, respectively), is different because of the presence of a lysine at the corresponding position in BFL-1 (Lys-77). Based on this it is predicted that BFL-1 would bind BAX-BH3 better than the other BH3 peptides, because they do not have such a favorable residue in the appropriate position (Glu-61 in BAX) for the interaction with this lysine (e.g. Leu in BCL-2, Gln in BCL-xL, Asn in BFL-1, Arg in BAK). However, BAX-BH3 also binds to BCL-2, which has a glutamate residue in the position of Lys-77 of BFL-1, so this hydrophilic interaction can only be a small contribution to the binding specificity. Site-directed mutagenesis of, for example, Lys-77 to an amino acid with a neutral or negatively charged side chain to see if it changed the specificity of BFL-1 for BH3 peptides or of Lys-46 to see if homodimerization could occur would be a way of confirming the importance of these residues.
The model presented here is strongly supported by experimental data from two-hybrid studies and immunoprecipitation analysis. By studying the interactions of BFL-1 with itself, with full-length BAX and with eight different BAX truncations ( Fig. 2A) we found that BFL-1 is not able to form homodimers but strongly binds BAX. The BAX-BFL-1 interaction was roughly as strong as the interaction of BAX with BCL-2. Only one domain, BH3, was required and sufficient for interaction with full-length BFL-1 (Fig. 2B). An analysis of the BFL-1 domains required for interaction with BAX demonstrated that the region containing helices 2-5 (M5) was the minimum sequence that allowed coimmunoprecipitation with BAX (Fig.   3C), suggesting that this region indeed forms a hydrophobic pocket, like that observed for BCL-xL, for the binding of BH3 from BAX and may function analogous to a receptor for the binding of proapoptotic proteins.
Although the BFL-1 deletion mutant M5 can bind to BAX, the analysis in REF52 cells revealed that it is not sufficient for the anti-apoptotic activity of BFL-1 nor proapoptotic activity (Fig. 4). This result is consistent with observations obtained in cell-free systems, which showed that only domains from apoptotic members of the family had apoptosis-inducing activity (37). Furthermore, the analysis in a yeast system showed that BFL-1 was able to suppress BAX lethality (Fig. 5), but the rescue effect was markedly weaker than that of BCL-xL despite their similar level of interaction with BAX. The anti-apoptotic activity of BFL-1 does not directly correlate with the strength of its binding to BAX, which means that the binding of BFL-1 to BAX alone is not sufficient for its anti-apoptotic function.
It has been suggested that BFL-1 lacks the cell proliferationrestraining activity (20) that other anti-apoptotic BCL-2 family proteins possess. We did not observe proliferation in either wild type BFL-1 or the M5 mutant in a standard apoptosis assay. It would be interesting to examine this in an appropriate cell proliferation assay as has been done by D'Sa-Eipper and Chinnadurai (20).
General Structural Organization for Anti-apoptotic Proteins of the BCL-2 Family-According to our model, the amino acid residues aligned to positions 1, 5, and 6 of the BH3 domain (8), highly conserved in BCL-2 family members, are not conserved in BFL-1. It is not only BFL-1 that has rather weak conservation in helices 1-3, which are the domains A, B, and C in our model. Several other anti-apoptotic proteins, such as the mouse BCL-2-related protein A1 (14) and various viral BCL-2 homologs, such as open reading frame-16 of human herpesvirus-8 FIG. 6. Domain organization of the anti-apoptotic members of the Bcl-2 family. Domains (not to scale) with structural similarities are labeled accordingly. BH indicates Bcl-2 homology domains. TM, transmembrane domains. The domains A, B, and C of the class III proteins, which were predicted to be ␣-helical segments, exhibit weak or no sequence homology to the BH4 and BH3 domains of the class I and II proteins but exhibit structural homology.
FIG. 5. Inhibition of BAX by BFL-1 in yeast. The BAX-expressing yeast strain Bi (3) was transformed with episomal vector pYES carrying the genes for BFL-1, BAG, and BCL-xL, respectively, under the control of the galactose-inducible GAL1 promoter. Separate transformants were spotted onto SD agar plates (see "Experimental Procedures") either containing glucose (A) or galactose (B and C). The spotting array was as given in the left panel. As controls three randomly chosen cDNA clones (LT1, LT2, and LT3) were included. Growth of yeast cells was recorded after 2 days (A and B) and 7 days (C) at 30°C. (40) and Herpesvirus Saimiriine (41,42), KS-BCL-2 of Kaposi sarcoma-associated virus (43), LMW5-HL of African swine fever virus (44), EBV-BHRF1 of Epstein-Barr virus (45,46), and nr-13 of Rous sarcoma virus (47) also contain conserved BH1 and BH2 domains but do not have recognizable BH3 domains. Analysis of secondary structure predictions for all these sequences revealed that the N-terminal region has a high helical propensity (data not shown). It is likely that three helices could exist in this region of the three-dimensional protein structure. No matter how low the conservation in the region of the BH4 or BH3 domains is, these proteins protect cells very efficiently from apoptosis (38 -48), suggesting that the overall fold with seven helices might be an important structural element for the BCL-2 family anti-apoptotic function. Based on the domain organization we, therefore, propose a new class, class III, of anti-apoptotic members of the BCL-2 family, which contain the conserved BH1 and BH2 domains only and structural, rather than sequence homology at their N-terminal part. This is in addition to class I and II anti-apoptotic members (8). Class III includes human BFL-1, mouse A1, and viral proteins nr-13, open reading frame-16, LMW5-HL, and EBV-BHRF1 (Fig. 6).
In summary, we have demonstrated by two-hybrid analysis and coimmunoprecipitation that BFL-1 cannot interact with itself nor with other apoptosis-suppressing proteins of the BCL-2 family but interacts strongly with the proapoptotic family member BAX. The interaction of BFL-1 with BAX is at least as strong as that of BAX with BCL-2. The BH3 domain of BAX is sufficient for interaction with BFL-1. BFL-1 requires its BH1 region and domains B and C to interact with BAX. BFL-1 is less potent than BCL-xL in its apoptosis-suppressing activity both in mammalian and yeast cells. The experimental results are in agreement with the predictions from our structural model for BFL-1, which consists of seven helices as in BCL-xL. By examination of all known anti-apoptotic proteins in the BCL-2 family, we propose that proteins sharing only the BH1 and BH2 domains, but none of the other homology regions, consist of a new class, class III, in which they share the importance of structural homology rather than sequence homology in their N-terminal region.