Proapoptotic Protein Bax Heterodimerizes with Bcl-2 and Homodimerizes with Bax via a Novel Domain (BH3) Distinct from BH1 and BH2*

Most members of the Bcl-2 protein family of apoptosis regulating proteins contain two evolutionarily conserved domains, termed BH1 and BH2. Both BH1 and BH2 in the Bcl-2 protein are required for its function as an inhibitor of cell death and for heterodimerization with the proapoptotic protein Bax. In this report, we mapped the region in Bax required for heterodimeriza- tion with Bcl-2 and homodimerization with Bax, using yeast two-hybrid and in vitro protein-protein interac- tion assays. Neither the BH1 nor the BH2 domain of Bax was required for binding to the wild-type Bcl-2 and Bax proteins. Moreover, Bax ( (cid:68) BH1) and Bax ( (cid:68) BH2) mutant proteins bound efficiently to themselves and each other, further confirming the lack of requirement for BH1 and BH2 for Bax/Bax homodimerization. Bax/Bax ho- modimerization was not dependent on the inclusion of the NH 2 -terminal 58 amino acids of the Bax protein in each dimerization partner, unlike Bcl-2/Bcl-2 homodimers which involve head-to-tail interactions be- tween the region of Bcl-2 where BH1 and BH2 resides, and an NH 2 -terminal domain in Bcl-2 that contains an- other domain BH4 which is conserved among antiapoptotic members of the Bcl-2 family. Similarly, het- erodimerization with Bcl-2 occurred without the NH 2 terminal domain of either Bax or Bcl-2, suggesting a tail-to-tail this 15-amino acid region abolished the abil- of Bax to dimerize and to heterodimerize with Bcl-2. The suggest that the structural fea- tures of Bax and Bcl-2 that allow them to participate in homo- and heterodimerization phenomena are mark- edly different, despite their amino-acid sequence similarity. autoradiography were accomplished by previously published methods (10, 14–17). Preparation and characterization of all human Bcl-2 yeast two-hy- brid and GST-bacterial expression plasmids and their encoded proteins, as well as a variety of negative control plasmids used for these studies, have been described (10, 14–17).

physiological conditions and which, when dysregulated, can contribute to several diseases including cancer, autoimmunity, AIDS, and ischemia-associated tissue loss (reviewed in Refs. 1 and 2). The Bcl-2 family proteins regulate a distal step in an evolutionarily conserved pathway for programmed cell death (1,3). Several members of the Bcl-2 protein family can form physical interactions with each other in a complicated network of homo-and heterodimers (4 -6). Although many details remain unclear at present, in general, the ratio between antiapoptotic proteins such as Bcl-2 relative to pro-cell death proteins such as Bax determines the ultimate sensitivity of cells to various apoptotic stimuli (7).
With the exception of some relatively nonabundant isoforms that arise through alternative mRNA splicing mechanisms, essentially all known members of the Bcl-2 protein contain two conserved regions of amino acid similarity, which we have previously termed Bcl-2 domains (BD) B and C but which are better known as BH1 and BH2 (4,5,8). In addition, the antiapoptotic proteins Bcl-2, Bcl-X L , Mcl-1, A1, Nr-13, and Ced-9 all contain an additional region of homology near their NH 2 termini, comprising a domain which we have termed BD-A but hereinafter refer to as BH4 for Bcl-2 homology-4 domain. Mutagenesis studies have shown that deletion of the BH1 or BH2 domain of Bcl-2 as well as certain amino acid substitutions in these conserved domains abolish Bcl-2 function as a suppressor of cell death and also abrogate the ability of Bcl-2 to form heterodimers with Bax (9,10). Similar mutations in the BH1 and BH2 domains of the antiapoptotic protein Bcl-X L have the same effects on function and Bax binding (6). These observations suggest that for Bcl-2 and Bcl-X L to suppress apoptosis, they must be able to heterodimerize with Bax. However, the situation is likely to be more complicated because deletion of the BH4 domain of Bcl-2 also destroys function but does not interfere with Bax binding (10,11).
In contrast to Bcl-2, essentially nothing is known at present about structure-function relations in the Bax protein as pertains to homodimerization with itself and heterodimerization with Bcl-2. In this report, we map for the first time a dimerization domain in the Bax protein and show that this novel domain, which we have termed BH3, is distinct from BH1 and BH2.

MATERIALS AND METHODS
A murine bax cDNA (12) was employed for mutagenesis experiments. Mutations were created using polymerase chain reaction-assisted methods and specific primers, essentially as described (5, 10, 13) (details available upon request), and their DNA sequences were confirmed by routine dideoxy sequencing methods. Bax mutants were expressed as fusion proteins either with an NH 2 -terminal LexA DNA binding domain in pEG202 for yeast two-hybrid experiments or with an NH 2terminal GST 1 domain in pGEX-4T-1 for production of recombinant * This work was generously supported by CaP-CURE, Inc. 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 (5,10). All versions of Bax and Bcl-2 employed here lacked the COOH-terminal transmembrane domains (TM) (residues 172-191 in Bax; 219 -239 in Bcl-2), thus avoiding problems with targeting of proteins to the nucleus for two-hybrid experiments and with solubility of bacteria-produced proteins.
The procedures for transformation of yeast (EGY-191 strain), induction of expression of Bax and Bcl-2 fusion proteins containing NH 2terminal transactivation (TA) domains using galactose-containing media (repression on glucose), and performance of two-hybrid reporter gene assays by auxotrophic growth on leucine-deficient media as well as by ␤-galactosidase-based colorimetric filter assays have been described in detail (5,10). Expression of all LexA-Bax mutants was confirmed by immunoblotting using a rabbit anti-LexA antiserum (generous gift of Erica Golemis, Philadelphia, PA).
Preparation and characterization of all human Bcl-2 yeast two-hybrid and GST-bacterial expression plasmids and their encoded proteins, as well as a variety of negative control plasmids used for these studies, have been described (10, 14 -17).

RESULTS AND DISCUSSION
For initial explorations of the regions within the Bax protein required for homodimerization with Bax and heterodimerization with Bcl-2, a series of NH 2 -terminal and COOH-terminal truncation mutants were constructed in the two-hybrid plasmid pEG202 and expressed in yeast as fusion proteins with NH 2 -LexA DNA binding domains (Fig. 1). These LexA-Bax proteins were then tested by yeast two-hybrid assays for interactions with "full-length" Bax and Bcl-2 proteins (missing TM domains only), which were expressed as fusions with a NH 2transactivation domain (TA) under the control of a Gal-1 promoter using the plasmid pJG4 -5 (18).
The essentially full-length Bax protein (amino-acids 1-171; i.e. missing only residues 172 3 COOH terminus to exclude TM domain) produced strong two-hybrid interactions with both Bax and Bcl-2 when plated on galactose-containing medium (induces Gal-1 promoter in pJG4 -5), whereas little or no growth on leucine or positivity in ␤-galactosidase colorimetric assays occurred when cells where plated on glucose-containing medium (represses Gal-1 promoter). COOH-terminal truncation mutants of Bax containing only residues 1 3 159 or 1 3 117 retained the ability to interact strongly with both Bax and Bcl-2, whereas a mutant that consisted only of amino acids 1 3 68 did not produce two-hybrid interactions with either fulllength Bax or Bcl-2 ( Fig. 1). Next, a series of NH 2 -terminal truncation mutants were tested. Deletion of the first 58 amino acids of Bax had no effect on its ability to form two-hybrid interactions with either Bax or Bcl-2. In contrast, removal of the first 99 or of the first 150 amino acids of Bax abolished all reactivity with both Bax and Bcl-2 ( Fig. 1).
Because this result suggested that the dimerization domain of Bax lies between residues 59 and 117, a series of Bax mutants which contained only the 59 -117 region or various subfragments thereof were tested for interactions with Bax and Bcl-2. The Bax-(59 -117) fragment produced strong two-hybrid interactions with both Bax and Bcl-2, which were comparable in strength to those seen with full-length Bax. In contrast, a smaller fragment of Bax containing only amino acids 69 -117 completely lacked reactivity with both Bax and Bcl-2 ( Fig. 1). Since the 59 -117 region of Bax retains the well-conserved BH1 domain, a smaller fragment of Bax was produced which lacked this domain, Bax-(59 -101). This protein, however, resulted in high amounts of background lacZ reporter gene transactivation, and thus could not be tested reliably by two-hybrid assay. A smaller fragment consisting of residues 69 -101 was devoid of all reactivity with Bax or Bcl-2 in two-hybrid assays.
As an alternative to the Bax-(59 -101) construct, an internal deletion mutant of Bax was produced which specifically lacked the sequences encoding the BH1 domain, Bax (⌬BH1). In addition, to confirm the apparent lack of requirement for BH2, an internal deletion mutant was also prepared which specifically lacked this domain, Bax (⌬BH2). Both the Bax (⌬BH1) and the Bax (⌬BH2) mutants produced strong two-hybrid interactions with full-length Bax and Bcl-2, which were comparable in strength to the wild-type Bax protein. These data argue that neither the BH1 nor the BH2 domain of Bax is required for homodimerization with wild-type Bax or heterodimerization with Bcl-2. With the exception of the Bax-(59 -101) protein and a Bax-(151-171) mutant, both of which suffered from background problems, none of the Bax mutants described here resulted in significant reporter gene activation with tested with pJG4 -5-Ha-Ras or the pJG4 -5 parent plasmid lacking an insert.
The mapping experiments described above suggested that the Bax dimerization domain resided between residues 59 and 101, but we were unable to test this through two-hybrid assays. For this reason, we explored the binding properties of Bax-(59 - The structure of the wild-type mouse Bax protein is depicted (top), showing the BH1, BH2, and TM (transmembrane) domains. Also shown are the structures of the various Bax deletion mutants which were expressed with NH 2 -terminal LexA DNA binding domains in pEG202 and tested by two-hybrid assays for specific interactions with Bax and Bcl-2 TA-fusion proteins which were produced as full-length proteins (with exception of removal of their TM domains), with TA domains fused to their NH 2 termini in pJG4 -5 (18). Reactions were scored as positive (ϩ) if galactose-inducible growth on leucine-deficient plates occurred within 5 days as well as conversion of X-gal substrate yielding an unambiguous blue color within 1 h, and if no substantial growth on leucine-deficient medium and no blue color development were observed when tested against pJG4 -5-Ras and pJG4 -5 without an insert. Some Bax mutants produced high background (HB) and therefore could not be evaluated, i.e. growth on leucine-deficient medium and conversion of X-gal substrate (blue color) occurred when cells were plated on both galactose (activates Gal-1 promoter that drives expression of TA-Bax and TA-Bcl-2) and glucose (represses Gal-1 promoter), and/or positive reactions were produced with negative control plasmids, pJG4 -5-Ras and pJG4 -5. 101), as well as several other Bax mutants through in vitro binding assays where Bax mutants were expressed as GST fusions in bacteria and the purified proteins were immobilized on glutathione-Sepharose and tested for specific binding to in vitro translated, [ 35 S]methionine-labeled Bax and Bcl-2. Alternatively, in some experiments, Bax mutants were in vitro translated and tested for binding to wild-type Bax or Bcl-2 GST-fusion proteins. As shown in Fig. 2A, a GST-Bax-(59 -101) fusion protein bound to 35 S-Bax with comparable efficiency to an essentially full-length GST-Bax fusion protein (lacking only the TM domain). This 59 -101 fragment of Bax also bound effectively to in vitro translated Bcl-2 protein. The specificity of the binding was confirmed by use of control GST and GST-CD40 cytosolic domain fusion proteins, as well as by failure of the GST-Bax-(59 -101) protein to interact with other irrelevant in vitro translated proteins, including R-Ras, Raf-1, and baculovirus p35 ( Fig. 2A and data not shown).
Because the Bax-(59 -101) fragment lacks the BH1 and BH2 domains, in vitro binding experiments were performed to confirm the lack of dependence on the these conserved domains for dimerization with Bax and Bcl-2. In vitro translated full-length Bax was compared with Bax (⌬BH1) and Bax (⌬BH2) for binding in vitro to GST-Bax and GST-Bcl-2. Both the Bax (⌬BH1) and Bax (⌬BH2) mutants bound to GST-Bax and GST-Bcl-2 with efficiencies comparable to wild-type Bax, but did not bind to GST nonfusion, GST-CD40, or other control GST-fusion proteins ( Fig. 2B and data not shown). In addition, the Bax (⌬BH2) and a Bax (⌬1-58) mutant were also capable of binding in vitro to a GST-Bax-(59 -101) fusion protein, further indicating that the 59 -101 region of Bax can bind to Bax independently of the BH2 and NH 2 -terminal domains of Bax (Fig. 2C). Interestingly, however, this 42-amino acid fragment of Bax (residues 59 -101) was not capable of binding to Bax (⌬BH1) in vitro. This result suggests that the BH1 domain, although perhaps not directly required for Bax homodimerization, may be necessary to facilitate the formation of an optimal binding site for the 59 -101 Bax fragment (see below). In this regard, because the BH1 domain is located immediately adjacent to the 59 -101 region, it is possible that BH1 is necessary for proper folding or contextual presentation of the 59 -101 domain for homodimerization.
Finally, because the two-hybrid experiments above suggested an important role for residues 59 -69 of Bax for dimerization with Bax and Bcl-2, a Bax-(59 -117) fragment (which formed two-hybrid interactions with Bax and Bcl-2 in twohybrid experiments) was compared with a Bax-(69 -117) fragment (which did not react with Bax or Bcl-2 in yeast). When expressed as GST-fusion proteins and tested for binding in vitro to in vitro translated 35 S-Bax and 35 S-Bcl-2, the GST-Bax-(59 -117) protein bound in vitro to Bax and Bcl-2 with efficiencies comparable to the essentially full-length GST-Bax protein which was missing only the TM domain ( Fig. 2D and data not shown). In contrast, the GST-Bax-(69 -117) protein failed to bind or, at best, bound very little of the 35 S-Bax or 35 S-Bcl-2 proteins in vitro, confirming the two-hybrid results which suggested that residues 59 -69 of Bax are required for dimeriza- FIG. 2. In vitro binding assays demonstrate dependence on 59 -101 region of Bax and independence of BH1 and BH2 for dimerization of Bax mutants with wild-type Bax and Bcl-2. GSTfusion proteins (5 g (A, B, and C); 10 M (D)) immobilized on glutathione-Sepharose (10 l) were tested for binding to [ 35 S]methionine-labeled in vitro translated proteins (10 l of reticulocyte lysates primed with 1 g of plasmid DNA) as described (10,17). Proteins that associated with GST fusions were analyzed by SDS-polyacrylamide gel electrophoresis (12% gel) and radiofluorography. Staining of gels with Coomassie Blue dye confirmed loading of similar amounts of mostly intact GST-fusion proteins in all experiments (not shown). As a control, 1 l of reticulocyte lysates containing in vitro translated proteins was run directly in gels (IVT Control). In A and D, full-length mouse Bax or human Bcl-2 proteins were produced by in vitro translation, whereas in B and C full-length Bax or Bax (⌬BH1), Bax (⌬BH2), and Bax (⌬1-58) mutants were translated in vitro.
An amino acid sequence alignment was performed for the 59 -69 region of Bax and several other known members of the Bcl-2 family, including Bcl-2, Bcl-X, Mcl-1, Ced-9, Bak, Bad, Bik, and Nr13. Significant homology was found in this region of Bax with the Bak, Bcl-2, Bcl-X, and Mcl-1 proteins (Fig. 3A), suggesting the existence of a functionally important, previously unrecognized conserved domain. We have termed this homologous region the Bcl-2 homology domain-3 (BH3). Thus, four evolutionarily conserved domains are envisioned in Bcl-2 family proteins: the BH4 domain (previously termed the A-box (5,8,10) which is found in the antiapoptotic members of the Bcl-2 family (Bcl-2, Bcl-X L , Mcl-1, Ced-9, A1, and Nr-13); the BH3 domain described here, and the BH1 and BH2 domains (Fig. 3B).
To confirm the importance of the BH3 domain for dimerization of Bax with itself and with Bcl-2, a Bax (⌬BH3) deletion mutant was tested for binding to Bax, Bax (⌬BH3), and Bcl-2 by two-hybrid assays. Although immunoblot analysis confirmed that the Bax (⌬BH3) protein was produced at levels comparable to the wild-type Bax protein (both when fused with a LexA-DNA binding domain or a B42 transactivation domain) (not shown), the Bax (⌬BH3) mutant failed to react with wildtype Bax, itself, or with Bcl-2 in two-hybrid assays (Fig. 4A). Thus, the BH3 domain is required for Bax homodimerization and for heterodimerization with Bcl-2. Interestingly, an isoform of Bax (Bax-␦) has recently been described which arises because of alternative splicing (19). The predicted Bax-␦ protein lacks residues 30 -77 where the BH3 domain resides and thus should be incapable of dimerizing with either Bcl-2 or Bax, based on the findings shown here.
Our previous analysis of the domains required for Bcl-2/Bcl-2 homodimerization suggested an anti-parallel, head-to-tail model wherein the NH 2 -terminal domain of Bcl-2 where the BH4 domain resides (residues 1-81) binds to structures contained in a downstream region of Bcl-2 where the BH1, BH2, and BH3 domains are located (amino acids 83-218) (5). Furthermore, those studies showed that physical interactions of the NH 2 -terminal Bcl-2 domain (residues 1-81) and downstream region in Bcl-2 (amino acids 83-218) require the simultaneous presence of both the BH1 and BH2 domain in the downstream region of Bcl-2 and the BH4 domain in the NH 2terminal segment of Bcl-2 (10). However, if one dimerization partner was missing BH1 and BH2 (but retained BH4) and the other was lacking BH4 (but still had BH1 and BH2), then interactions could occur (10). This antiparallel fashion in which Bcl-2 homodimerizes promoted us to further explore the nature of Bax/Bax homodimers by testing binding of additional Bax mutants to each other by two-hybrid assays. As shown in Although not shown here, all LexA-Bax mutants were also tested for reactivity with TA-Bcl-2, TA-Ras, and TA revealing that all mutants which reacted with TA-Bax also were positive with TA-Bcl-2 but did not react with TA-Ras or TA. In addition, all TA-Bax mutants were also tested for reactivity with LexA-Fas and LexA, revealing no two-hybrid interactions with these control proteins. In B, in vitro binding studies were performed using GST-fusion proteins and a [ 35 S]methionine-labeled, in vitro translated mutant of Bax containing residues 59 -191. nal 58 amino acids of Bax are not required for homodimerization of Bax (⌬N) to Bax (⌬N) mutants, implying that Bax/Bax homodimerization occurs via a tail-to-tail interaction. Fig. 4A shows in addition that the Bax (⌬BH1) and Bax (⌬BH2) internal deletion mutants were able to form strong two-hybrid interactions with themselves and each other, further confirming the data presented in Fig. 2C which suggested an absence of a requirement for BH1 or BH2 for homodimerization of Bax (Fig.  4A). The ability of the Bax (⌬BH1) mutant in particular to homodimerize with itself supports the idea that if BH1 is involved in Bax homodimerization, it plays only an indirect or facilatory role compared to BH3. These same mutants of Bax (Bax (⌬N), Bax (⌬BH1), Bax (⌬BH2)) also formed two-hybrid interactions with Bcl-2 with strengths comparable to the wildtype Bax protein (not shown). The specificity of these interactions was confirmed by use of various irrelevant proteins ( Fig.  4A; see legend).
Finally, the structural features of Bax/Bcl-2 heterodimers were explored by testing the binding of Bcl-2/mutants to Bax mutants. As shown in Fig. 4B, an in vitro translated N-truncation mutant of Bax missing the first 58 amino acids bound with comparable efficiencies in vitro to an essentially fulllength GST-Bcl-2 protein (missing only TM domain; i.e. has residues 1-218) and a GST-Bcl-2 N-truncation mutant missing the first 82 amino acids of the Bcl-2 protein (83-218) but failed to bind to a C-truncation mutant of Bcl-2 comprised only of Bcl-2 residues 1-81 fused to GST. In contrast, in vitro translated 35 S-Bcl-2 bound to both the GST-Bcl-2-(1-81) and GST-Bcl-2-(83-218) proteins employed for this experiment (not shown), confirming the integrity of the GST-Bcl-2-(1-81) protein despite its failure to bind to Bax. The specificity of the in vitro interactions of the Bax (⌬N) protein (residues 59 3 COOH terminus (191)) with GST-Bcl-2 and GST-Bax was confirmed by use of GST, GST-CD40, and GST-R-Ras control proteins. Unlike Bcl-2/Bcl-2 homodimerization, therefore, the heterodimerization of Bcl-2 with Bax appears to occur in a parallel, tail-to-tail fashion in which the NH 2 -terminal domains of neither Bcl-2 nor Bax are required.
Taken together, these observations suggest that dimerization of Bax with itself and with Bcl-2 occurs independently of the well-conserved BH1 and BH2 domains. Thus, the structural features of Bax that allow it to physically interact with other members of the Bcl-2 family are strikingly different from Bcl-2, which does require BH1 and BH2 for heterodimerization with Bax, as well as for homodimerization with itself, at least when testing for binding of mutant Bcl-2 to mutant Bcl-2 (9, 10). Moreover, the data presented here indicate that another region in the Bax protein, where the BH3 domain resides, is both necessary and sufficient for binding to the wild-type Bax and Bcl-2 proteins and thus defines a novel dimerization domain for this family of apoptosis-regulating proteins. The finding that this region of Bax shares strong amino acid sequence homology with some other members of the Bcl-2 protein family, including proapoptotic (Bak, Bik) and antiapoptotic (Bcl-2, Bcl-X L , Mcl-1) proteins, also raises the possibility that the BH3 domain may be required for interactions among other Bcl-2 family proteins. In this regard, while this paper was under review, a report appeared showing that the region of Bak that contains BH3 is sufficient both for heterodimerizing with Bcl-X L and for promoting apoptosis in mammalian cells (20). It remains to be determined however whether the BH3 domain of Bak promotes apoptosis by directly engaging the cell death pathway (i.e. effector domain) versus by acting as a decoy (i.e. regulatory domain) that ties up antiapoptotic protein such as Bcl-X L , thereby preventing them from forming effective interactions with the full-length endogenous Bak or Bax proteins. In addition, a new member of the Bcl-2 protein family was described Bik that: (a) promotes apoptosis, (b) contains a region with strong homology to the BH3 domain but lacks the BH1 and BH2 domains (Fig. 3, A and B), and (c) which binds to Bcl-2 in a BH3-dependent manner. When taken together with the data presented here, therefore, these findings confirm the functional importance of the BH3 domain. It will be interesting in future investigations to determine whether the BH3-mediated homodimerization of Bax with itself is necessary for promotion of apoptosis by the Bax protein.