Sequence-specific Dimerization of the Transmembrane Domain of the “BH3-only” Protein BNIP3 in Membranes and Detergent*

Mitochondria-mediated apoptosis is regulated by proteins of the Bcl-2 superfamily, most of which contain a C-terminal hydrophobic domain that plays a role in membrane targeting. Experiments with BNIP3 have implicated the transmembrane (TM) domain in its proapoptotic function, homodimerization, and interactions with Bcl-2 and Bcl-xL. We show that the BNIP3 TM domain self-associates strongly in Escherichia coli cell membranes and causes reversible dimerization of a soluble protein in the detergent SDS when expressed as an in-frame fusion. Limited mutational analysis identifies specific residues that are critical for BNIP3 TM self-association in membranes, and these residues are also important for dimerization in SDS micelles, suggesting that the self-association observed in membranes is preserved in detergent. The effects of sequence changes at positions Ala176 and Gly180 suggest that the BNIP3 TM domain associates using a variant of the GXXXG motif previously shown to be important in the dimerization of glycophorin A. The importance of residue His173 in BNIP3 TM domain dimerization indicates that polar residues, which have been implicated in self-association of model TM peptides, can act in concert with the AXXXG motif to stabilize TM domain interactions. Our results demonstrate that the hydrophobic C-terminal TM domain of the pro-apoptotic BNIP3 protein dimerizes tightly in lipidic environments, and that this association has a strong sequence dependence but is independent of the identity of flanking regions. Thus, the transmembrane domain represents another region of the Bcl-2 superfamily of proteins that is capable of mediating strong and specific protein-protein interactions.

The Bcl-2 superfamily of proteins plays a central role in regulating mitochondria-mediated apoptosis; the Bax subfamily promotes apoptosis, whereas the Bcl-2 subfamily protects against apoptosis (reviewed in Ref. 1). Protein-protein interactions between Bax and Bcl-2 subfamily members help determine cell fate (2,3) and have accordingly been the subject of intensive study. Four regions of sequence homology, the BH1 through BH4 domains, contribute to the structure and function of these proteins, and the BH3 domain is particularly implicated in heterodimerization events (4 -9). Peptide and small molecule inhibitors of these protein-protein interactions can modulate apoptosis, further demonstrating the functional and pharmaceutical importance of these contacts (10 -13).
Differentiation and development in metazoans requires celland signal-specific inputs to a functional Bcl-2/Bax checkpoint. The pro-apoptotic "BH3-only" proteins, which show homology to the Bcl-2 superfamily through the BH3 domain alone (14), are expressed or activated in specific tissues and in response to certain stimuli, making them candidates for connecting diverse signaling pathways to the ubiquitous apoptosis effector machinery (15,16). BNIP3 (Bcl-2/19-kDa interacting protein 3) is a BH3-only protein whose expression and pro-apoptotic activity is induced following hypoxia (17)(18)(19)(20). Conflicting reports indicate that the apoptotic effect of BNIP3 and its interaction with Bcl-2 and homologs either depend upon (21) or are independent of (22,23) the BNIP3 BH3 domain. However, both the proapoptotic activity of BNIP3 and its interaction with Bcl-2 have been shown to depend upon the C-terminal hydrophobic TM 1 domain (21, 22, 24 -26), and the TM domain is required for BNIP3 homodimerization in both SDS-PAGE and in yeast twohybrid analysis (24). Because these observations suggest that the BNIP3 TM domain makes functionally important contacts in membranes, we tested the self-association properties of the BNIP3 TM domain. We find that the BNIP3 TM domain forms strong and specific dimers in membranes and in SDS-PAGE in the absence of any additional regions of BNIP3.
To probe the physical basis for BNIP3 TM domain dimerization in membranes, we targeted the polar residue His 173 and the GXXXG motif using site-directed mutagenesis because these sequence elements might be expected to give rise to strong helix-helix interactions. Strongly polar residues within hydrophobic TM domains have been shown to drive oligomerization in membranes and detergents (27)(28)(29)(30), and the GXXXG motif has been shown to mediate interactions between TMs in several biological systems (31)(32)(33)(34)(35)(36)(37). Mutational analysis shows that the TM domain interaction is mediated by both polar interactions and the motif AXXXG, and that the detergentsolubilized dimer is disrupted by the same mutations that disrupt the membrane-embedded complex. Our findings demonstrate that the BNIP3 TM domain can form stable dimers in membranes independent of other regions of BNIP3 and suggest that the C-terminal hydrophobic tails of this Bcl-2 family member may mediate functionally important protein-protein interactions in membranes.
XbaI and BamHI sites in the PCR products. Sequences were confirmed by automated dideoxynucleotide sequencing of the TM regions of the constructs; amino acid sequences are given in Table I. The staphylococcal  nuclease/BNIP3 TM domain fusion protein (SN/BNIP3TM) was generated from the previously described staphylococcal nuclease/glycophorin A TM domain fusion protein vector (pT7SN/GpA) (39) by ApaI/BamHI digestion and in-frame insertion at the ApaI site (which encodes a Gly-Pro linker) of DNA coding for amino acids 146 -194 of mammalian BNIP3. The entire coding sequence of pT7SN/BNIP3 was verified by automated dideoxynucleotide sequencing. The amino acid sequence of the BNIP3 region, with the predicted TM domain underlined, is GPRNTSV-MKKGGIFSAEFLKVFLPSLLLSHLLAIGLGIYIGRRLTTSTSTF.
Site-directed Mutagenesis-Site-directed mutagenesis was performed in pccBNIP3 and in pT7SN/BNIP3 using the QuikChange kit (Stratagene, La Jolla, CA). All mutations were confirmed by automated dideoxynucleotide sequencing of the TM regions of the constructs.
Expression of ToxRЈ(TM)MBP Constructs-Plasmids encoding ToxRЈ(TM)MBP chimerae were transformed into Escherichia coli NT326 cells (kindly provided by D. M. Engelman) and plated onto Luria Bertani (LB) plates (with 50 g/ml ampicillin); colonies were inoculated into LB medium (with 50 g/ml carbenicillin), and glycerol stocks were made at A 600 ϳ0.2 and stored at Ϫ80°C. LB cultures (with 50 g/ml carbenicillin) were inoculated from frozen glycerol stocks and grown at 37°C until approximately A 420 ϳ1.0, when culture densities were equalized by (at least 16-fold) dilution into fresh culture tubes. These cultures were again grown to A 420 ϳ1.0, and 6.0 A 420 units of cells were harvested by centrifugation and washed with 0.5 ml of sonication buffer (25 mM Tris-HCl, 2 mM EDTA, pH 8.0). Cells were then resuspended in 2 ml of sonication buffer and lysed by probe sonication. After removing an aliquot (100 l) for Western blot analysis, the remaining lysate was clarified by centrifugation at 13,000 ϫ g, and the supernatant was stored on ice until the spectrophotometric assay was performed. Normalization of lysate concentrations was confirmed by assaying total protein (Bio-Rad).
Spectrophotometric CAT Assay-The assay of Shaw (40) was used to detect chloramphenicol acetyltransferase activity in cell lysates. This method follows the release of coenzyme A upon transfer of an acetyl group from acetyl-CoA to chloramphenicol by quantifying the color change at 412 nm that occurs upon reaction of the free thiol of CoASH with 5,5Ј-dithiobis-(2-nitrobenzoic acid). 10 l of lysate was mixed with 250 l of reaction buffer (0.1 mM acetyl-CoA, 0.4 mg/ml 5,5Ј-dithiobis-(2-nitrobenzoic acid), 0.1 M Tris-HCl, pH 7.8), and the absorbance at 412 nm was acquired at 3-s intervals for a minimum of 2 min to establish a background rate of acetyl-CoA hydrolysis. 10 l of 2.5 mM chloramphenicol was then added with mixing, and the absorbance was followed for 1 min. Each lysate was assayed in triplicate, and the slopes of the recorded data were converted to enzyme activity units (⑀ ϭ 13,600 at 412 nm). Linear least squares fits to the data with high correlation coefficients and low variance indicated that the data being fit corresponded to regions for initial rates. Background acetyl-CoA hydrolysis was very low, rarely exceeding 1 unit of activity. Reported data are the averages from at least three separate experiments, and this method proved to give excellent reproducibility. All kinetics traces were acquired on a Cary 50 spectrophotometer.
Western Blots-Cell lysates were mixed 1:1 with 2ϫ SDS-PAGE sample buffer, heated to 90°C or 95°C for 5 min, separated on pre-cast 15% polyacrylamide gels (Bio-Rad), blotted onto nitrocellulose, detected with an anti-MBP primary antibody (New England Biolabs) or with an anti-staphylococcal nuclease antibody (prepared by Bethyl Industries), and visualized with an anti-rabbit horseradish peroxidase conjugate and ECL reagent (Amersham Biosciences).
Maltose Complementation Assays-For complementation assays, E. coli NT326 cells expressing ToxRЈ(TM)MBP constructs were grown overnight in liquid M9 minimal medium containing 0.4% glucose, washed, streaked on M9 minimal media plates containing 0.4% maltose as the only carbon source, and incubated for 2 days at 37°C. Plates were imaged using a NucleoTech GelExpert digital camera system.
Expression and Purification of Staphylococcal Nuclease Chimerae-Plasmid pT7SN/BNIP3 (or a mutant) was transformed into BL21(DE3) cells and plated onto LB plates (with 50 g/ml ampicillin). Colonies were inoculated into LB medium (with 50 g/ml carbenicillin), and glycerol stocks were made at A 600 ϳ0.2 and stored at Ϫ80°C. LB overnight cultures (with 50 g/ml carbenicillin) started from frozen glycerol stocks were used to inoculate 1-liter LB cultures, which were grown at 37°C to A 600 ϳ1.0, induced with 0.6 mM isopropyl-␤-D-thiogalactoside, and harvested by centrifugation after 3 h. Cells were resuspended in 1:20 culture volume of lysis buffer (20 mM Tris-HCl, 2 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, pH 8.0), subjected to three rounds of freeze-thaw, incubated with 0.1 mg/ml hen egg white lysozyme for 30 min on ice, and disrupted by sonication. CaCl 2 was added to 5 mM, and the lysate was incubated on ice for a further 30 min to allow the nuclease activity of SN/BNIP3TM to fragment nucleic acid. The soluble fraction of this lysate was removed by centrifugation (12,000 ϫ g for 10 min) and decantation. The pellet was extracted with lysis buffer plus 1 M NH 4 OAc, subjected to centrifugation, and the supernatant discarded. The pellet was extracted with lysis buffer plus the detergent Thesit (Fluka) at a concentration of 2% (w/v) and subjected to centrifugation, and the supernatant was again discarded. This pellet was extracted with lysis buffer plus 1 M NH 4 OAc and 2% Thesit, and the SN/BNIP3TM fusion protein was found in the supernatant following centrifugation for 30 min at 15,000 ϫ g. This supernatant was dialyzed against lysis buffer plus 0.1 M NH 4 OAc and 0.2% Thesit and clarified by centrifugation for 10 min at 12,000 ϫ g. SN/BNIP3TM fusion protein was purified from the supernatant by ion-exchange chromatography over DE52 and CM52 resins at pH 8.0 in the presence of 0.2% Thesit. Additional purification was achieved in some cases by reversed phase-high pressure liquid chromatography on a phenyl stationary phase (Zorbax SB-Phenyl) in a gradient of isopropyl alcohol/ acetonitrile (2:3) and water in the presence of 0.1% trifluoroacetic acid.
Purification of BNIP3 TM Peptide-SN/BNIP3TM protein purified by ion exchange in the presence of Thesit (above) was digested by addition of trypsin (1:40 weight ratio) and incubation overnight at 37°C. Time courses revealed that intact SN/BNIP3TM disappeared within 30 min. BNIP3 TM domain peptide was purified by reversed phase-high pressure liquid chromatography on a phenyl stationary phase (Zorbax SB-Phenyl) in a gradient of isopropyl alcohol/acetonitrile (2:3) versus water in the presence of 0.1% trifluoroacetic acid. Peptide identity was confirmed by matrix-assisted laser desorption ionization time-of-flight-mass spectrometry (Dr. T. D. Marriott, Rice Department of Chemistry), with experimental masses matching calculated masses to within 2 atomic mass units.

RESULTS
Homodimerization of BNIP3 as detected both on SDS-PAGE and in yeast two-hybrid experiments depends on the presence of the BNIP3 TM domain (24). We have used the TOXCAT system (38) to determine whether the isolated BNIP3 TM domain is sufficient to cause self-association. In this assay, the TM domain is inserted as an in-frame fusion between ToxRЈ (an N-terminal DNA-binding protein) and MBP (a C-terminal maltose-binding protein that is targeted to the periplasm). Expression of the ToxRЈ(TM)MBP construct in the E. coli membrane allows the read-out of TM-dependent dimerization of the ToxRЈ domain through the reporter gene chloramphenicol acetyltransferase (CAT), which is under the control of the ToxRresponsive promoter ctx. Proper insertion topology can be verified by the ability of periplasmic MBP to restore to mutant E. coli the ability to grow on maltose as a sole carbon source. This approach has been used to characterize TM-TM interactions in several systems (34 -36, 41, 42).
BNIP3 TM Domain Dimerizes Strongly in E. coli Membranes-To assess the ability of the BNIP3 TM domain to self-associate in E. coli membranes, we transformed a series of ToxRЈ(TM)MBP constructs (Table I) into NT326 cells and measured the resulting levels of CAT expression in cell lysates using a spectrophotometric assay (40) (Fig. 1). The positive control, ToxRЈ(GpA)MBP, bears the self-associating TM domain of glycophorin A and causes strong expression of the CAT gene (38,39,43). The negative control, which is identical except for the disruptive mutation G83I in the TM domain, causes minimal CAT expression (31,38). The ToxRЈ(BNIP3)MBP construct, which bears the BNIP3 TM domain (including residues 164 -184 of mammalian BNIP3), shows CAT levels more than double those of the glycophorin A construct, indicating that the BNIP3 TM domain drives dimerization more strongly than does the glycophorin A TM domain. By contrast, the TM domains of two other BH3-only proteins, Hrk (44) and Blk (45), and the TM domain of the human insulin receptor (46,47) cause only slight expression of CAT as ToxRЈ-MBP fusion proteins, demonstrating that TM sequences of similar length, composition, and membrane targeting properties do not constitutively activate the TOXCAT system. Western blots of whole cell lysates detected with anti-MBP antibodies (inset in Fig. 1) demonstrate that the levels of ToxRЈ(TM)MBP fusion proteins in these cells are comparable, so differences in CAT expression are not due to variations in the amount of ToxRЈ DNA-binding domain. MBP complementation assays confirm that the topologies of the ToxRЈ-MBP fusions are as expected (Fig. 2). We conclude that the BNIP3 TM domain is capable of very tight self-association in E. coli membranes.

BNIP3 TM Domain Causes Dimerization of a Fusion Protein in SDS-
To produce milligram quantities of the BNIP3 TM domain for biophysical studies, we generated an in-frame fusion of BNIP3 TM to the C-terminal end of staphylococcal nuclease in the manner of Lemmon et al. (39) and expressed the SN/BNIP3TM fusion protein in BL21(DE3) cells. Fig. 3 shows that the purified protein migrates on SDS-PAGE predominantly as a dimer of 42 kDa; a monomer species can be detected by western blot analysis (Fig. 4B). This result suggests that the BNIP3 TM domain dimers that form in membranes also resist disruption by SDS, as has been noted for the single-span proteins glycophorin A (39) and phospholamban (48 -51), the por-ins and other ␤-barrel outer membrane proteins (52,53), and the E. coli ammonium transporter AmtB (54).
Sequence Specificity of Dimerization in Membranes and in SDS-The ability of the isolated BNIP3 TM domain to dimerize suggests that this self-association property may underlie the previously observed functional importance of the BNIP3 TM domain (21, 22, 24 -26). Inspection of the amino acid sequence of BNIP3 TM (Fig. 4A) and comparison with previously reported interaction motifs (27-33, 35-37, 55) reveal two potential sources for strong helix-helix interactions, the polar residue His 173 and the GXXXG motif. We have begun to test the roles of small residues and of His 173 in mediating dimer stability using site-directed mutagenesis and two different assays: TOXCAT for dimerization in membranes and SDS-PAGE for dimerization in detergents. Fig. 4 shows that, for wild type BNIP3 TM and seven mutants, the TOXCAT and SDS-PAGE assay results parallel one another, supporting the idea that the same interactions that stabilize the dimer in membranes also stabilize the dimer in detergent. Western blots (inset) indicate that the mutant constructs are expressed at similar levels, and complementation assays (data not shown) show that the mutations do not affect  Table I. the ToxRЈ(BNIP3TM)MBP topology, so the effects of individual mutations on CAT levels provide insight into the physical basis for dimerization. Mutation G180I causes disruption of dimerization in both TOXCAT and on SDS, whereas mutation G178I causes no reduction of dimerization in either assay. Even the minimal mutation G180A causes complete disruption of the SN/BNIP3TM dimer on SDS-PAGE and an 8-fold drop in CAT activity, indicating that the association is strongly dependent on Gly 180 . It is noteworthy that mutations to Gly 180 abolish dimerization despite the presence of His 173 , demonstrating that the polar residue at position 173 is not sufficient to cause strong dimerization of BNIP3 TM domain in membranes or in detergent. Mutations H173W and H173A show that this site is, however, also an important determinant of dimerization, because mutations at this position partially or completely abrogate dimerization on SDS-PAGE and reduce measured CAT activity nearly 4-fold. Mutations A176C and A176L demonstrate that this residue is probably also involved in TM-TM interactions, because cysteine supports dimerization in both environments, and the larger leucine is disruptive. Together, these data show that BNIP3 TM domain self-association is sensitive to conservative amino acid substitutions. A full sequence dependence of BNIP3 TM domain dimerization will be obtained by saturation mutagenesis of the TM region.
SN/BNIP3TM Associates Reversibly through TM-TM Interactions-To characterize further the self-association of SN/ BNIP3TM in SDS, we mixed the fusion protein with purified BNIP3 TM peptide before heating the samples to 90°C and resolving them on SDS-PAGE (Fig. 5). Mixing results in loss of the fusion protein dimer band at 42 kDa with the concomitant appearance of a new band at 26 kDa, or roughly the mass of the SN/BNIP3TM monomer plus one BNIP3 TM peptide monomer, which we assign as the protein-peptide heterodimer. The generation of this species demonstrates that the BNIP3 TM peptide can compete with the TM domain of SN/BNIP3TM in forming TM domain complexes.
Because this finding suggests that SN/BNIP3TM dimers are in equilibrium with monomers in SDS, rather than being kinetically trapped, we diluted the fusion protein in 4% SDS to determine whether the ratio of dimer to monomer exhibited a concentration dependence. Serial 2-fold dilutions of the wild type fusion protein and of a slightly disruptive mutant, H173W, were resolved on SDS-PAGE and detected by Coomassie Blue staining or by western blotting with an anti-staphylococcal nuclease antibody (Fig. 6). Wild type SN/BNIP3TM loaded at 80 g ml Ϫ1 (3.6 M) appears to be entirely dimeric on staining with Coomassie Blue, but a monomer band can be detected by western blotting. At 1.25 g ml Ϫ1 (55 nM), the monomer and dimer bands are approximately the same intensity. The H173W mutant fusion protein shows detectable monomer by Coomassie Blue staining at 80 g ml Ϫ1 (3.6 M), and at 10 g ml Ϫ1 (450 nM) the monomer and dimer bands are approximately the same intensity; at further dilutions the monomer band is the major species. We estimate the apparent dissociation constants of wild type and H173W SN/BNIP3TM in 4% SDS to be 55 and 450 nM, respectively, with the caveats that these data are in an artificial detergent and gel system.

DISCUSSION
Mitochondria-mediated apoptosis is regulated by interactions between members of the Bcl-2, Bax, and BH3-only families of proteins (2, 15, 56 -58), and several lines of evidence suggest that protein-protein interactions at or within membranes have functional consequences for apoptosis (3, 58 -61). BH3-only proteins appear to connect specific signaling pathways to the Bcl-2/Bax apoptosis checkpoint (15,16), but the nature of this regulation may be quite diverse, because some BH3-only proteins bind the anti-apoptotic Bcl-2 subfamily members (56,62), whereas others are capable of binding to either Bcl-2 or Bax (63,64). The BH3-only protein BNIP3 interacts with anti-apoptotic Bcl-2 subfamily members, including Bcl-x L , but neither the mammalian nor Caenorhabditis elegans orthologs of BNIP3 require a BH3 domain to bind to Bcl-x L (22,65). By contrast, the BNIP3 transmembrane domain has been shown to be necessary for homodimerization, proapoptotic activity, and heterodimerization with Bcl-2 and Bcl-x L (21, 22, 24 -26). The TM domain is also necessary for BNIP3 interaction with CD47 at the plasma membrane (66). To test its self-association properties, we studied the BNIP3 TM domain by using an in vivo helix-helix association assay called TOXCAT (38). We show that the 21-residue BNIP3 TM domain self-associates strongly in E. coli membranes, whereas the TM domains of two other BH3-only proteins, Hrk and Blk, do not interact. Combining these results with the previously reported (24) self-association of BNIP3 in a yeast two-hybrid assay, we conclude that the TM domain is both necessary and sufficient to cause homodimerization of BNIP3.
Determining the functional role for BNIP3 TM domain dimerization is challenging because deletion of the TM domain will affect not only self-association but also membrane targeting. Understanding the physical basis for BNIP3 TM domain dimerization in membranes might allow alteration of the selfassociation of the TM domain of BNIP3 without affecting subcellular targeting, thus permitting the relative functional importance of these two properties to be determined. Accordingly, we have begun to explore the sequence dependence of BNIP3 TM domain dimerization by targeting several sites in the BNIP3 TM domain for mutagenesis. Because our results with a staphylococcal nuclease/BNIP3 TM domain fusion protein show that the BNIP3 TM domain is sufficient to induce dimerization in the detergent SDS, we have used both in vivo and in vitro assays of self-association to analyze the mutants. These experiments have demonstrated an excellent correspondence between the effects of mutations on dimerization in membranes and in SDS-PAGE, suggesting that both assays are reporting on the same physical association.
We targeted position His 173 within the TM domain because of previous reports (27)(28)(29)(30) that polar side chains can drive helix-helix association. Because mutations at His 173 have graded effects on dimerization in both TOXCAT and SDS-PAGE assays, our results demonstrate that this polar residue does help to stabilize TM domain interactions. However, the histidine is not sufficient to drive strong association of BNIP3 TM domains because several mutations at positions 176 or 180 almost completely abolish dimerization in SDS-PAGE and in E. coli membranes. Thus, the interactions supported by His 173 within the BNIP3 TM domain depend strongly upon the flanking sequences, unlike the interactions observed between polar residues in model TM domains (27)(28)(29)(30).
The BNIP3 TM residues Ala 176 , Gly 180 , and Gly 184 form the sequence motif GXXXG and the variant AXXXG (55). Because the GXXXG motif has been shown to mediate TM domain interactions in glycophorin A (31)(32)(33), the Helicobacter pylori vacuolating toxin (34,35), and the ErbB growth factor receptors (36), we expected that mutations at these positions might affect BNIP3 dimerization. Consistent with this expectation, we found that changing Gly 180 to isoleucine or even alanine is strongly disruptive in TOXCAT and completely disruptive on SDS-PAGE. By contrast, the mutation G178I has no effect on dimerization, indicating that Gly 178 is not important to dimerization, whereas Gly 180 plays a key role in the association. Because the mutation A176L also disrupts dimerization significantly, residues Ala 176 and Gly 180 may be a variant of the GXXXG motif that permits close approach of the helices in the known structure of glycophorin A (37).
In an analysis of glycophorin A TM domain dimerization (31), the effects of single point mutations were used to identify a dimerization interface that was later confirmed by multiple substitutions (67) and by the NMR structure of the TM domain dimer (37). From our limited mutagenesis data, we identify residues His 173 , Ala 176 , and Gly 180 as participating in the BNIP3 TM domain dimer interface, with residue Gly 178 being located away from the interface. It has been shown previously that the GXXXG motif can occur in different contexts, with multiple copies of the motif aligned in tandem (34,35) or displayed on different faces of the TM helix, providing alternate interaction sites (36). Our identification of His 173 , Ala 176 , and Gly 180 as interfacial residues demonstrates that the previously described (27)(28)(29)(30) ability of polar residues to drive interactions between TM domains can be exploited by biology in the context of an AXXXG motif to give additional sequence specificity and stability to helix-helix interactions in membranes.
The reversible association of the BNIP3 TM domain demonstrated by peptide competition experiments (Fig. 5) and by dilution (Fig. 6)  nM) and slightly disruptive H173W (450 nM) indicate that BNIP3 TM and the mutant H173W both associate more tightly in SDS than does glycophorin A. This rank order of stability is also borne out by our TOXCAT measurements, demonstrating that the relative stability of different TM-TM domain interactions in membranes can be predicted from biochemical data in detergent.
Because glycophorin A associates more strongly in mild detergents than in SDS, as shown by ultracentrifugation (70) and resonance energy transfer (69), the absolute stabilities of TM-TM domain interactions from these experiments are not readily interpretable. The PAGE method used here may also suffer from systematic biases, such as concentrating of the sample during stacking. However, comparison of sequence variants under identical conditions should cancel detergentspecific influences and systematic errors to give the relative effect of the mutation on self-association. We therefore interpret our dilution series data to mean that the mutation H173W increases the BNIP3 TM domain dissociation constant (in 4% SDS) by a factor of about 8, which corresponds to ϳ1.4 kcal mol Ϫ1 of dissociation free energy. This large effect shows that the association is highly sequence-specific; note that several mutations affect BNIP3 TM dimerization more severely than H173W in both detergent and in membranes (Fig. 4).
Our work has identified the TM domain of BNIP3 as providing a strong and specific basis for dimerization of the fulllength protein in membranes, but the precise biological role for this dimerization is unclear. A functional role for BH3-only protein TM domain interactions is suggested by work with the protein Nix, a homolog that is 56% identical to BNIP3. Expression of a version of Nix that had no effect on apoptosis in KB cells blocked the pro-apoptotic effect of BNIP3 (but not E1A) expression in trans (26). Importantly, this suppression of BNIP3 function depended on the presence of the Nix TM domain (26). Nix is 80% identical to BNIP3 in the TM region, and thus the reported homodimerization of Nix is probably also mediated by the TM domain (25,26,71); note that residues His 173 , Ala 176 , and Gly 180 that we have shown to be critical to BNIP3 dimerization are conserved in Nix. The ability of nonfunctional Nix to inhibit BNIP3 in trans can be explained by a model in which the normal function of BNIP3 depends on homodimer formation through the TM domain, whereas heterodimerization with an inactive form of Nix through TM-TM interactions blocks BNIP3 function.
The observed dominant negative phenotype demonstrates that interactions between the membrane-embedded regions of the BH3-only proteins BNIP3 and Nix can modulate their pro-apoptotic function. Possible functional roles for membrane spanning domains of other Bcl-2 family members have also been reported; several lines of evidence indicate that Bax, which oligomerizes and forms pores in membranes (3,58,60,61), may make new protein-protein interactions in the context of a membrane (58, 59). The NMR structure of Bax suggests that a binding pocket occupied by the C-terminal hydrophobic tail of Bax becomes exposed only upon insertion of the hydrophobic tail into a target membrane, thus coupling subcellular localization to the regulation of protein-protein interactions (9). The ability of the BNIP3 TM domain to dimerize in bilayers may serve to ensure that membrane targeting precedes protein homodimerization. Accordingly, elucidating the chemical and structural basis for sequence-specific TM domain interactions in the BNIP3 system will be important to understanding BNIP3 function. Experiments to test the functional importance of these interactions, using mutations whose effects upon TM domain interactions are well characterized, will allow the rel-evance of these interactions to the regulation of programmed cell death to be determined.