The Single Transmembrane Domains of ErbB Receptors Self-associate in Cell Membranes*

Members of the epidermal growth factor receptor, or ErbB, family of receptor tyrosine kinases have a single transmembrane (TM) α-helix that is usually assumed to play a passive role in ligand-induced dimerization and activation of the receptor. However, recent studies with the epidermal growth factor receptor (ErbB1) and the erythropoietin receptor have indicated that interactions between TM α-helices do contribute to stabilization of ligand-independent and/or ligand-induced receptor dimers. In addition, not all of the expected ErbB receptor ligand-induced dimerization events can be recapitulated using isolated extracellular domains, suggesting that other regions of the receptor, such as the TM domain, may contribute to dimerization in vivo. Using an approach for analyzing TM domain interactions in Escherichia colicell membranes, named TOXCAT, we find that the TM domains of ErbB receptors self-associate strongly in the absence of their extracellular domains, with the rank order ErbB4-TM > ErbB1-TM ≈ ErbB2-TM > ErbB3-TM. A limited mutational analysis suggests that dimerization of these TM domains involves one or more GXXXG motifs, which occur frequently in the TM domains of receptor tyrosine kinases and are critical for stabilizing the glycophorin A TM domain dimer. We also analyzed the effect of the valine to glutamic acid mutation in ErbB2 that constitutively activates this receptor. Contrary to our expectations, this mutation reduced rather than increased ErbB2-TM dimerization. Our findings suggest a role for TM domain interactions in ErbB receptor function, possibly in stabilizing inactive ligand-independent receptor dimers that have been observed by several groups.

Members of the epidermal growth factor receptor, or ErbB, family of receptor tyrosine kinases have a single transmembrane (TM) ␣-helix that is usually assumed to play a passive role in ligand-induced dimerization and activation of the receptor. However, recent studies with the epidermal growth factor receptor (ErbB1) and the erythropoietin receptor have indicated that interactions between TM ␣-helices do contribute to stabilization of ligand-independent and/or ligand-induced receptor dimers. In addition, not all of the expected ErbB receptor ligand-induced dimerization events can be recapitulated using isolated extracellular domains, suggesting that other regions of the receptor, such as the TM domain, may contribute to dimerization in vivo. Using an approach for analyzing TM domain interactions in Escherichia coli cell membranes, named TOXCAT, we find that the TM domains of ErbB receptors self-associate strongly in the absence of their extracellular domains, with the rank order ErbB4-TM > ErbB1-TM Ϸ ErbB2-TM > ErbB3-TM. A limited mutational analysis suggests that dimerization of these TM domains involves one or more GXXXG motifs, which occur frequently in the TM domains of receptor tyrosine kinases and are critical for stabilizing the glycophorin A TM domain dimer. We also analyzed the effect of the valine to glutamic acid mutation in ErbB2 that constitutively activates this receptor. Contrary to our expectations, this mutation reduced rather than increased ErbB2-TM dimerization. Our findings suggest a role for TM domain interactions in ErbB receptor function, possibly in stabilizing inactive ligand-independent receptor dimers that have been observed by several groups.
The ErbB (or HER) family of growth factor receptor tyrosine kinases has four members: the epidermal growth factor (EGF) 1 receptor (ErbB1), ErbB2 (also known as HER2 or the Neu oncogene product), ErbB3 (HER3), and ErbB4 (HER4) (1). Each ErbB receptor has a large extracellular (EC) domain of 600 -630 amino acids, a single membrane-spanning ␣-helix, and an intracellular domain of ϳ500 amino acids that contains the tyrosine kinase domain plus regulatory sequences (2). It is now quite well established that activation of ErbB receptors involves their ligand-induced (or ligand-stabilized) oligomerization, which in turn leads to receptor trans-phosphorylation and activation within dimers or higher order oligomers (reviewed in Ref. 3). In the case of ErbB1, we and others have demonstrated that the isolated EC domain of the receptor dimerizes completely upon EGF binding (4 -10). Similarly, the isolated EC domain of ErbB4 oligomerizes strongly when it binds to its growth factor ligand neuregulin 1-␤ 1 (4). These findings have led to the argument that ErbB receptor activation results directly and solely from ligand-induced oligomerization of EC domains (3,11,12). In this view, the transmembrane (TM) and intracellular domains of receptor molecules need not contribute directly to receptor oligomerization, but are instead driven together by EC domain interactions in the process of receptor activation.
By contrast with this model, Tanner and Kyte (13) have argued that interactions between TM domains do contribute significantly to ErbB1 dimerization. In fact, they go so far as to argue from their data that ErbB1 TM domain interactions provide the primary driving force for receptor dimerization. According to their proposal the unliganded ErbB1 EC domain sterically inhibits TM domain-mediated dimerization of the receptor. Activation of the receptor by EGF is then proposed to occur when ligand binding induces conformational changes in the EC domain that relieve this inhibition and permit TM domain-mediated ErbB1 homodimerization (13). This model was prompted by the finding of Tanner and Kyte (13) that EGF-induced dimerization is far more efficient for a fragment of ErbB1 that contains both the EC domain and the TM domain (when studied in detergent solution) than it is for the EC domain alone. Several other pieces of data also support a role for TM domains in dimerization of bitopic cell-surface receptors. For example, a specific Val 3 Glu mutation within the TM domain of the Neu oncogene product (ErbB2) is well known to induce ErbB2 dimerization and activation (14,15). Although precisely how the Val 3 Glu mutation activates the receptor is not clear, it has been proposed to stabilize interactions between TM ␣-helices directly (16). Tzahar et al. (17) have also suggested a role for TM domain interactions in ligand-induced ErbB receptor heteromerization. Finally, it was recently shown that the single TM domain from the erythropoietin (EPO) receptor forms oligomers (18,19) and that this self-association may play a role in EPO receptor signaling across the membrane (19,20).
Although recent findings by our laboratory and others argue that EC domains are often sufficient to drive receptor dimerization, this may not always be the case. For example, neuregulin-induced homodimerization of the isolated ErbB3 EC domain cannot be detected in vitro (4,21). Furthermore, the ligand-induced heteromeric association of ErbB receptors detected in vivo cannot be recapitulated using isolated EC domains in vitro (4,21). Motivated by these observations, together with the reports of receptor TM domain self-association outlined above, we have initiated studies of TM domain interactions in the ErbB receptor family. Using TOXCAT, a system recently developed by the Engelman laboratory for analyzing TM domain interactions in the inner membrane of Escherichia coli (22), we find that the TM domains from ErbB receptors homo-oligomerize efficiently. Identification of mutations that disrupt association of ErbB receptor TM domains suggests that the interactions responsible are similar to those that mediate dimerization of the glycophorin A TM ␣-helix (22)(23)(24). Surprisingly, we also find that the Neu mutation destabilizes rather than stabilizes dimerization of the ErbB2 TM domain. Our results support a role for TM domains in stabilizing (possibly ligand-independent) dimerization of ErbB receptors and suggest that the Neu mutation may achieve its effects by altering the nature, rather than extent, of ErbB2 TM domain dimerization.
Site-directed Mutagenesis-Site-directed mutagenesis was performed using the QuikChange kit (Stratagene, La Jolla, CA) following the manufacturer's instructions or by using a standard four-primer PCR method (25). Mutated sequences were confirmed by automated dideoxynucleotide sequencing of TM domain inserts.
Expression of ToxR Chimera-Plasmids encoding ToxRЈ(ErbB-TM)MBP chimerae were transformed into E. coli MM39 cells (gift of W. Russ and D. M. Engelman), a malE-deficient bacterial strain. Cells were grown overnight in Luria Bertani broth (with 200 g/ml ampicillin), diluted 1/100 into fresh medium, and re-grown to mid-logarithmic phase (A 600 ϳ0.6). 0.3 A 600 units of cells were then pelleted and resuspended into 60 l of phosphate-buffered saline. Cells were lysed by addition of 30 l of 3x SDS-PAGE sample buffer with heating to 95°C for 2 min. Lysates were loaded onto 7.5% SDS-PAGE gels and immunoblotted using an anti-maltose-binding protein (MBP) antibody (New England Biolabs, Beverly, MA, 1:5000 dilution) followed by an antirabbit horseradish peroxidase conjugate and visualization with ECL (Amersham Biosciences, Inc.).
Maltose Complementation Assays-For maltose complementation, M9 minimal medium plates were prepared containing 5% maltose as the only carbon source. MM39 cells expressing the proteins of interest were streaked onto these plates and incubated for several days at 37°C. pMALc2 and pMALp2 (New England Biolabs) transformed cells, expressing cytoplasmic and periplasmic targeted MBP, respectively, were used as controls. Plates were imaged using a Kodak ImageStation 440.
Protease Accessibility Assays-Cultures expressing the relevant chimerae were grown to mid-logarithmic phase, and two identical samples containing the equivalent of 3 A 600 units of cells were pelleted for each culture. One sample was used to prepare spheroplasts, while the second was used as a whole-cell control. For spheroplast preparation, cells were first resuspended in 90 l of resuspension buffer (100 mM Tris, pH 8.2, 500 mM sucrose, 0.5 mM EDTA, 2.5 mg/ml lysozyme) and incubated at room temperature for 5 min. One volume of ice-cold water was then added, and the sample was mixed and placed on ice. MgCl 2 was next added to a final concentration of 10 mM. After incubation on ice for 5 min, spheroplasts were pelleted at 500 ϫ g for 5 min, and the supernatant (corresponding to the periplasmic fraction) was removed. The pellet (containing spheroplasts) was resuspended in 90 l of resuspension buffer and divided into three 30-l aliquots. To one aliquot was added 3 l of 10% Nonidet P-40 (to lyse the spheroplasts) plus proteinase K (to a final concentration of 50 g/ml). To the second aliquot was added 3 l of resuspension buffer plus proteinase K. The third aliquot was left untreated (except for addition of resuspension buffer to maintain volume). Proteinase K digestion was allowed to proceed for 30 min on ice. Protein was then precipitated from each sample by addition of 3.5 l of 100% trichloroacetic acid, and pellets were resuspended in SDS-PAGE loading buffer for electrophoresis and Western blotting with an anti-MBP antibody. Whole-cell samples were processed alongside spheroplast samples (with phosphate-buffered saline replacing resuspension buffer and water during the osmotic shock treatment).
In Vivo Dimerization Assays-For disc-diffusion assays, cultures were grown to mid-logarithmic phase, diluted 20-fold into Luria Bertani broth, and 500 l was plated on Luria Bertani agar/ampicillin plates. A 2-cm-diameter Whatman filter disc, prespotted with 60 l of a 90 mg/ml chloramphenicol (CAM) solution, was placed in the center of the plate, and the plate was incubated overnight at 37°C. On the following day the diameter of the circle (with the CAM disc at its center) that lacked bacterial growth was measured as the zone of inhibition (ZOI). Results from these assays are quoted as the net ZOI, from which has been subtracted the diameter of the filter disc.
For CAT assays, cell lysates were prepared essentially as described by Russ and Engelman (22), with the additional normalization of protein levels in cell lysates (using a Bio-Rad protein assay). CAT assays employed the Quan-T-CAT system (Amersham Biosciences, Inc.) according to the manufacturer's instructions. All experiments were performed in triplicate, alongside samples using pccKAN and pccGpAG83I (as negative controls), pccGpA-WT as a positive control, and standards of purified CAT enzyme (provided in the Quan-T-CAT kit).

RESULTS
To investigate the ability of ErbB receptor TM domains to self-associate in the plasma membrane of E. coli, we employed the TOXCAT system developed by Russ and Engelman (22). This system exploits two properties of the ToxR protein from Vibrio cholerae, namely that it is a specific transcriptional activator that requires dimerization to induce expression from its target promoters and that it is an integral membrane protein with modular structure (26). Fusion of the cytoplasmic domain of ToxR (ToxRЈ) to the strongly dimerizing TM domain from glycophorin A (GpA) (27,28) allows ToxRЈ to activate transcription from the ctx promoter (22,29). This finding has been exploited in several studies of TM domain interactions (18, 19, 30 -32). The TM domain of interest is fused between a NH 2 -terminal cytoplasmic ToxRЈ transcriptional activation domain and a periplasmic (assuming correct membrane insertion) MBP domain. Correct membrane insertion of the ToxRЈ (TM)MBP chimera can be determined by assessing both its ability to complement an MBP deficiency in mutant E. coli and its accessibility to protease digestion in spheroplasts. TM do- Note that two such potential motifs overlap toward the NH 2 terminus of the ErbB1, ErbB4, and GpA TM domains (only one is shaded: see Table  I for complete list of motifs). Glycines at which substitutions are found to significantly inhibit dimerization (in ErbB1-TM, ErbB2-TM, ErbB4-TM, and GpA-TM) are in white type (shaded black). main dimerization is assessed in the TOXCAT system by monitoring expression of CAT, which is under control of the ToxRresponsive ctx promoter on a reporter plasmid.
ErbB Receptor TM Domains Dimerize-By assessing the ability of ToxRЈ chimerae to activate CAT reporter expression in E. coli we found that each ErbB receptor TM domain homodimerizes significantly, even when compared with the strongly dimerizing GpA TM domain (GpA-TM). The ErbB receptor TM domains homodimerize in the rank order ErbB4-TM Ͼ ErbB1-TM Ϸ ErbB2-TM Ͼ ErbB3-TM. CAT expression in E. coli can be monitored using a disc diffusion assay (see "Experimental Procedures"), in which resistance to CAM around a disc impregnated with this antibiotic is assessed. As shown graphically in Fig. 2 (see also Ref. 22), only a small ZOI surrounds a CAM-impregnated disc placed on a plate of E. coli expressing a dimerizing ToxRЈ chimera (ToxRЈ(GpA-TM)MBP, for example), since these cells are CAM-resistant. The ZOI is much larger when the bacteria express a nondimerizing ToxRЈ chimera; either with no TM domain (No TM) or with a nondimerizing TM domain (the G83I mutant of GpA-TM (23)), since activation of CAT expression is reduced. As shown in Fig.  2, expression of ToxRЈ chimerae containing the ErbB1, ErbB2, or ErbB4 TM domains results in ZOI's similar in size to that seen for GpA-TM, indicating strong TM domain dimerization. A chimera containing the ErbB3 TM domain also appears to confer some CAM resistance, but significantly less than for the other ErbB TM domains or GpA-TM. The ZOI measured for the ErbB3 chimera is only slightly smaller than with the G83I mutant of GpA-TM in which dimerization is severely disrupted (23).
For a more quantitative comparison of TM domain dimerization we next assayed CAT activity directly in lysates of E. coli expressing the different chimerae, by measuring the transfer of 3 H acetyl groups to biotinylated CAM (see "Experimental Procedures"). As shown in Fig. 3, expression of the ToxRЈ(GpA-TM)MBP chimera resulted in a high level of CAT activity in E. coli lysates, which was reduced over 12-fold by the G83I mutation that disrupts GpA-TM dimerization. ToxRЈ chimerae containing the ErbB1 or ErbB2 TM domains induced approximately half the CAT activity seen for GpA-TM: the ErbB3 TM domain induced slightly less (approximately 40%) and the ErbB4 TM domain more (approximately 70%) activity.
In control experiments, we demonstrated by Western blotting that expression levels of the ToxRЈ chimerae containing ErbB receptor TM domains were essentially identical (Fig. 4A).
To demonstrate that the chimerae are correctly integrated into the membrane, each was shown to complement a malE deletion in E. coli, allowing a strain that does not express MBP (MM39) to grow with maltose as its sole carbon source (Fig. 4B). This can only occur if the MBP moiety of the chimera is present in the periplasm and so provides evidence that the chimera inserts correctly in the plasma membrane: with MBP in the FIG. 2. ErbB receptor TM domains dimerize. The ability of the ToxRЈ(ErbB-TM)MBP chimerae to dimerize and activate CAT transcription was tested using disc assays to assess chloramphenicol resistance (see "Experimental Procedures"). Results are expressed as the net diameter of the ZOI, for which the diameter of the filter disc is subtracted from the ZOI diameter. Data plotted are the means from four plates in a single assay (ϮS.D.). The most strongly dimerizing chimera (GpA-TM) gives the smallest ZOI, inducing the most significant CAM resistance. periplasm, ToxRЈ sequences in the cytoplasm and the TM domain spanning the inner membrane. Correct membrane integration was further supported by that fact that each chimera was accessible to partial proteinase K digestion in spheroplasts (but not in whole cells), while cytoplasmic proteins were not (representative data are shown in Fig. 5). In these experiments, nearly all of the expressed chimera was accessible to proteinase K in the spheroplast preparation, suggesting that almost all has acquired the correct transmembrane orientation.
Presence of Potential Dimerization Motifs in ErbB Receptor TM Domains-Three primary modes have been described for the strong noncovalent self-association of individual TM domains. In one, a heptad motif of leucines is thought to mediate oligomerization, presumably through side chain packing interactions similar to those seen in leucine zippers (18). The ErbB receptor TM domains contain no such heptad motifs, arguing against the relevance of this type of interaction for our studies. A second mode involves stabilization of TM domain oligomers by intramembraneous hydrogen bonds between polar side chains (32,33). The ErbB receptor TM domains contain no strongly polar side chains, allowing us to dismiss this as a possibility. The third mode, likely to be of most relevance for ErbB receptor TM domains, involves the so called GXXXG motif (30). The GXXXG motif is a central component of the dimerization interface for the GpA TM domain (28) and also appears to be critical for self-association of the M13 coat protein transmembrane segment (34). Furthermore, nearly all dimerizing sequences isolated from a TM domain library in a TOXCAT-based screen contained a sequence related to the GXXXG motif (30). Structural studies of the GpA-TM dimer have shown that the GXXXG motif increases the total area of the helix-helix interface by allowing close proximity of adjacent straight helices when they cross in a right-handed sense (28). In addition, the motif provides a flat surface against which side chains of other interfacial residues may pack.
Inspection of the sequences for the ErbB receptor TM domains (Fig. 1) shows several elements that closely resemble the GXXXG motifs found by Russ and Engelman (30) to drive TM domain dimerization. These motifs are listed in Table I. Sternberg and Gullick (35) previously pointed out the frequent occurrence of this general pentapeptide motif (P 0 -P 1 -P 2 -P 3 -P 4 ) in the TM domains of receptor tyrosine kinases, in which the first FIG. 4. Mutation of ErbB TM domains does not affect expression or membrane insertion of ToxR' chimerae. A, lysates from E. coli expressing the marked chimerae were normalized for protein content, run on 7.5% SDS-PAGE gels, and immunoblotted with an antibody to the COOH-terminal MBP domain. All chimerae tested expressed at essentially identical levels, except for the ErbB4 N 2 /C mutant (which actually showed enhanced apparent dimerization). Results for all mutants that were defective in oligomerization are shown, to confirm that loss of CAT transcription did not simply result from a reduction in the quantity of the transcriptional activator. B, MM39 (malE-deficient) E. coli expressing the indicated chimera were streaked on an M9 minimal media plate with maltose as the sole carbon source. Only cells in which MBP is periplasmic (and therefore in which the chimera is correctly inserted into the membrane) can survive with maltose as sole carbon source. This was true for all TM domain chimerae used in this study, indicating proper membrane integration. As controls, growth of cells expressing periplasmic MBP from pMAL-p2 is shown, as is the lack of growth of cells expressing only cytoplasmic MBP from pMAL-c2 (see "Experimental Procedures"). position (P 0 ) is occupied by a residue with a small side chain (G/A/S/T/P), position P 4 is occupied by alanine or glycine, and position P 3 is most often occupied by a residue with a large aliphatic side chain. The TM domains from ErbB1, ErbB2, and ErbB4 all possess at least two such potential motifs, one toward the amino terminus and one toward the carboxyl terminus of the domain (separated by ϳ3 turns of ␣-helix). Each motif is designated with the receptor name and respective terminus in Table I (for example, motif 1C is the ErbB1 COOH-terminal motif with sequence AXXXG). In the ErbB1 and ErbB4 TM domains there are two possible NH 2 -terminal motifs that overlap. These are denoted by the addition of a subscripted "1" or "2" to the motif name. For example, the ErbB1 NH 2 -terminal TXXXG pattern is motif 1N 1 , and the overlapping GXXXA pattern is motif 1N 2 (see Table I). The ErbB3 TM domain is exceptional in having only one possible motif, a TXXXG sequence (3N), that is positioned similarly to the GXXXG motif in GpA-TM (Fig. 1).

Mutations in Potential Dimerization Motifs Disrupt
Dimerization of ErbB1, ErbB2, and ErbB4, but not ErbB3 TM Domains-To assess the importance of each GXXXG motif in dimerization of ErbB receptor TM domains, the residue corresponding to the second glycine (equivalent to Gly 83 in GpA-TM) was mutated to valine (see Table I), and dimerization of the mutated TM domain was assessed using TOXCAT. In previous studies of GpA-TM dimerization, all substitutions at Gly 83 were found to be strongly disruptive (23). In the ErbB1, ErbB2, and ErbB4 TM domains, an equivalent mutation in at least one of the GXXXG motifs also disrupted dimerization (Fig. 6), suggesting that self-association is driven in a similar manner. Controls for chimeric protein expression levels and their correct membrane insertion were performed in all cases, and no mutation (except for a double mutation in ErbB4-TM) either reduced expression significantly or adversely affected the ability of the ToxRЈ(TM)MBP fusion protein to complement a malE deletion (Fig. 4). . Similar data were obtained for all other chimerae discussed in this paper. In each panel, the left-hand gel is a Western blot probed with an anti-MBP antibody, while the right-hand gel is stained with Coomassie Blue to monitor the extent to which proteinase K has proteolyzed E. coli proteins in general. The intact ToxRЈ(TM)MBP fusion is denoted at the left of each anti-MBP Western blot with an arrow, and MBP that has been liberated from the chimera by proteolysis is denoted with an asterisk. In whole cells (W) there is no cleavage of the chimerae whether the cells are left untreated (U), treated with proteinase K (P) or treated with 1% Nonidet P-40 plus proteinase K (DP). The Coomassiestained gel on the right of each panel shows no general cleavage of E. coli proteins. By contrast, while chimerae are not cleaved in spheroplasts (S) that are left untreated (U), substantial cleavage is seen when proteinase K is added (P), despite a lack of general proteolysis in the Coomassie-stained gels. This indicates that the ToxRЈ(TM)MBP chimerae are accessible to added proteinase K in spheroplasts but not in intact E. coli. Detergent solubilization of spheroplasts followed by proteinase K treatment (labeled DP) results in substantially more general proteolysis of E. coli proteins, but no significant increase in ToxRЈ(TM)MBP chimera cleavage. This indicates that nearly all of the chimeric protein is accessible to proteinase K in the spheroplast preparation. These results support the data presented in Fig. 4   In the ErbB1 TM domain, mutation (to valine) of the last residue in the COOH-terminal motif (1C) reduced dimerization (assessed in CAT assays) by ϳ40%, whereas mutation of neither motif 1N 1 nor 1N 2 influenced dimerization (Fig. 6A). Mutants with alterations in both the COOH-terminal and one of the NH 2 -terminal motifs (1N 1 /C and 1N 2 /C) dimerized to an extent that was indistinguishable from those with changes in only the COOH-terminal motif. These results argue that motif 1C (AXXXG) is involved in ErbB1-TM self-association, but that the NH 2 -terminal motifs (1N 1 and 1N 2 ) do not contribute sig-nificantly. In ErbB4-TM the situation appears to be reversed (Fig. 6B). Mutation of the NH 2 -terminal ErbB4-TM motif 4N 2 , but not motif 4C, reduced ErbB4-TM dimerization (to ϳ65% of wild-type levels). Interestingly, disruption of motif 4C in ErbB4-TM resulted in an apparent enhancement of dimerization that was seen consistently in all experiments. This enhancement appeared more robust when motif 4C was mutated together with motif 4N 2 (mutation 4N 2 /C). We do not have a satisfying explanation for this observation, but it does suggest that an alternate mode of TM domain association occurs in these mutants. In ErbB3-TM, mutation of the single (NH 2 -terminal) motif (3N) had no effect upon its dimerization (Fig. 6B).
Results obtained with the ErbB2 TM domain were arguably the most clear (Fig. 6C). Alteration of either the NH 2terminal (2N) or COOH-terminal (2C) GXXXG motif (Table I) reduced ErbB2-TM dimerization to ϳ35% of the wild-type level. Simultaneous mutation of both these motifs reduced dimerization to less than 15% of wild-type levels (2N/C), to a level similar to that seen with the G83I mutant of GpA-TM. It therefore appears that both motifs contribute to dimerization of ErbB2-TM. For ErbB2-TM we also used alanineinsertion to disrupt the GXXXG dimerization motifs. Insertion of alanine in the middle of a GXXXG motif was previously shown to be highly disruptive for GpA-TM dimerization (36). Similarly, insertion of alanine in the COOHterminal GXXXG motif of ErbB2-TM (ins C ) reduced CAT activity to approximately just 45% of wild-type levels (Fig.  6C). Alanine insertion into the motif 2N (ins N ) had only a small effect on ErbB2-TM dimerization, possibly because the insertion results in a change from the sequence vvSavvGi to vvsAavvGi, which leaves an intact alternative GXXXG-like motif in this region that could drive ErbB2-TM dimerization.
The Neu Mutation Reduces Dimerization by the ErbB2 TM Domain in the TOXCAT Assay-The rat form of ErbB2 (the Neu oncogene product) was first isolated from ethylnitrosourea-induced rat neuro/glio-blastomas and was shown to have been activated by mutation of valine 664 in its TM domain to glutamic acid (14,37). This Val 3 Glu mutation in the TM domain induces constitutive activation and dimerization of the Neu oncogene product (14,15,37), and it has been proposed that the glutamic acid side chain makes hydrogen bonds across the interface between two TM ␣-helices and thus stabilizes the Neu dimer (16). Substitution of the equivalent valine (Val 659 ) in human ErbB2 with glutamic acid has similarly been shown to cause constitutive activation of this receptor (38), and we were interested to determine the effect of this V659E mutation on dimerization of ErbB2-TM in TOXCAT. By contrast with our expectations, introduction of the Neu mutation actually reduced ErbB2-TM dimerization to about 50% of the level seen with wild-type ErbB2-TM domain (compare WT and neu in Fig.  6C). In the context of the V659E mutation, alteration of the second GXXXG-like motif (motif 2C) further disrupted dimerization (neuC). Thus, we did not observe the enhanced TM domain dimerization that was anticipated when the Val 3 Glu Neu mutation was made in the context of human ErbB2-TM. Expression levels of the chimerae containing the mutation were indistinguishable from those of other proteins (Fig. 4A), and both malE complementation (Fig. 4B) and proteinase K digestion (Fig. 5) indicated proper membrane insertion by each chimera.

DISCUSSION
Several reports suggest a role for TM domain self-association in ligand-dependent (13,19) and/or ligand-independent (20) dimerization of bitopic cytokine and growth factor receptors. Although biochemical studies using SDS-PAGE and gel-filtra-FIG. 6. Mutation of GXXXG motifs disrupts dimerization of ErbB receptor TM domains. The ability of different mutations to disrupt ErbB-TM dimerization was assessed using the 3 H-based CAT assay, as for Fig. 3. Mutations are marked for the motif (or motifs) in which the P 4 position (or second G of GXXXG) was replaced with a valine (see Table I). A shows results for ErbB1-TM, B shows results for ErbB3-TM and ErbB4-TM, and C shows results for ErbB2-TM. For ErbB2, in addition to valine substitutions in motifs 2N and 2C, alanine insertions were individually made into both motifs (ins N and ins C , respectively). Neu refers to ErbB2 with the V659E mutation, and neuC refers to the V659E mutant with an additional valine substitution in motif 2C. CAT activities for each mutant are reported as a percentage of the values obtained for the wild-type TM domain. tion in detergents have not provided support for this hypothesis (39,40), solid-state NMR studies indicate that the isolated ErbB1 and ErbB2 TM domains form oligomers in lipid bilayers in vitro (41)(42)(43). Using the TOXCAT system we show that the ErbB1, ErbB2, and ErbB4 TM domains all dimerize robustly within membranes in vivo, with the caveat that the lipid composition of the bacterial inner membrane differs from that of the mammalian membranes where these receptors are found. The TM domains of ErbB receptors gave signals in the TOX-CAT dimerization assay that reached 60 -70% of levels seen with the well characterized and strongly dimerizing GpA TM domain. Furthermore, since ErbB-TM dimerization could be diminished significantly with point mutations in identifiable sequence motifs, our findings suggest that dimerization occurs through specific interactions that resemble those of GpA-TM.
By analogy with the known structure of the GpA-TM dimer (28), we envision that dimerization of the ErbB1 and ErbB4 TM domains will involve right-handed crossing of two straight ␣-helices, centered on a single GXXXG-related motif. For ErbB1-TM, the two helices will cross close to their COOH termini, with motif 1C at the center of the interface. For ErbB4, the center of the dimer interface will be close to the NH 2 terminus, since motif 4N 2 appears to be most important. While the likely configuration of the dimer is easy to envision in these cases, it is more difficult for ErbB2-TM, since our mutagenesis studies show that two GXXXG motifs contribute to its dimerization (Fig. 6C). When modeled based on GpA-TM, an ErbB2-TM dimer driven by only the NH 2 -terminal motif (2N) appears as shown in Fig. 7A. A dimer driven only by the COOH-terminal motif (2C) has a quite different configuration, as shown in Fig. 7C. For both GXXXG motifs to contribute to a single helix-helix interface, significant distortion of the ␣-helices would be required, with the helices coiling around one another in a right-handed sense. Another possibility is that the two motifs could cooperate in forming higher order ErbB2-TM oligomers, as reported for GpA-TM with interleaved dimerization motifs (44). We consider this possibility unlikely, since we were unable to dock a third helix onto either of the ErbB2-TM dimer models shown in Fig. 7 (using the exposed GXXXG motif) without significant steric clashes. A final possibility, which we consider most likely, is that ErbB2-TM has two alternative dimerization modes, represented in Fig. 7, A and C. Simultaneous mutation of the two GXXXG motifs would abolish both of the dimerization modes, while mutation of a single motif should cause a relatively small statistical reduction in overall dimerization.
The precise role of TM domain dimerization in signaling by bitopic receptors is unclear. Despite the report that TM domains contribute to EGF-induced ErbB1 dimerization (13), several studies have shown that overexpressed ErbB1 mutants with alterations in the TM domain (including those in which the GXXXG motifs are disrupted) respond to EGF in a way that is indistinguishable from the wild-type receptor (45,46). This argues that TM domain dimerization is not critical for ErbB1 function at the expression levels usually studied in such experiments. Recent studies of the EPO receptor have indicated that the effects of TM domain interactions may only be seen when the receptor is expressed or activated at very low levels. The single TM domain of the EPO receptor dimerizes in a TOXCATrelated assay (18), and Constantinescu et al. (20) have shown that TM domain interactions are important in ligand-independent dimerization of this receptor. As with ErbB1, signaling is affected very little by loss of TM domain interactions at the relatively high receptor and ligand levels typically employed (19,20). However, a significant effect on signaling is seen when the EPO receptor is activated at very low levels. Constanti- FIG. 7. Models of possible ErbB2-TM dimers. Models were generated of the possible configurations of ErbB2-TM dimers, with the assumption that, like GpA-TM, ErbB2-TM dimerizes by right-handed crossing of straight ␣-helices with a GXXXG motif central in the interface. If the NH 2 -terminal GXXXG motif (2N) were employed, the configuration in A would be predicted. If the COOH-terminal motif were employed (2C), the configuration in C is predicted. For each dimer model, one monomer is shown in Corey-Pauling-Koltun representation, and the other is shown as a stick representation of the helix backbone. The first and last positions of the two GXXXG motifs are colored black and are labeled in the view of the ErbB2-TM monomer (B). The two motifs are separated by almost 90°around the helix axis, and significant mutual coiling of the two helices in a dimer would be required if both motifs were to contribute simultaneously to the same helix-helix interface. Furthermore, the disposition of the two motifs is such that higher order oligomerization appears unlikely (see text). Dimer models were generated by superimposing the 2N (A) or 2C (C) motif of an ErbB2-TM helix on to each GXXXG motif in the GpA-TM dimer structure (PDB entry 1afo) determined by NMR (28) and then extending the helices using canonical geometry and ErbB2-TM sequence. Energy minimization was then performed using Insight II and Discover (Biosym Technologies). The side chain of Phe 671 was removed for clarity. The figure was generated using MOLSCRIPT (54) and Raster3D (55). nescu et al. (20) have argued from these findings that TM domain interactions may allow rapid activation of the receptor by aiding its association into a "predimerized" inactive state, thus enhancing the ability of the bivalent ligand to bind and stabilize the active dimeric state. A similar situation is possible for ErbB1. Indeed, ErbB1 has been reported to form ligand-independent, inactive oligomers under certain conditions (47,48), and TM domain interactions may contribute to this self-association. One study has provided evidence that the relative orientation of the two receptors is different in the inactive predimerized state and the active ligand-stabilized dimeric state (48).
One of the most surprising of our findings is that the activating Val 3 Glu mutation in ErbB2-TM does not enhance its dimerization, but actually reduces the tendency of the TM domain to self-associate (Fig. 6C). The Val 3 Glu mutation certainly leads to enhanced dimerization and activation of fulllength ErbB2 (14,15,38), but the initial proposal that this results from a direct stabilizing effect on ErbB2/Neu TM domain interactions (16) has never been directly confirmed. An alternative hypothesis, consistent with our findings and with other literature results, is that the Val 3 Glu mutation does not enhance dimer stability per se, but instead alters the configuration of existing TM domain interactions driven by the two GXXXG motifs. For example, the glutamic acid side chain introduced by the mutation could participate in interactions that alter the relative orientation of the two ␣-helices, perhaps to promote adoption of an "active" configuration. Several studies have shown that the Val 3 Glu mutation can only activate ErbB2/Neu (without major alterations in the TM domain) when the GXXXG motif is intact (49,50), arguing that it must occur within a motif that drives TM domain dimerization. Moreover, it has been shown that activation of ErbB2/Neu requires a precise relative orientation of protomers within a receptor dimer (51), and that the Val 3 Glu mutation induces structural alterations in the region of the NH 2 -terminal GXXXG-like motif of ErbB2-TM when studied in phospholipid bilayers (52). A further piece of evidence indicating that the wild-type ErbB2/ Neu TM domain drives dimerization just as well as the Val 3 Glu mutant was provided by Lofts et al. (53), who studied the ability of short ErbB2/Neu TM proteins to inhibit transformation of mouse fibroblasts by the Neu oncogene. Both wild-type and Val 3 Glu mutant TM domain proteins could associate with full-length ErbB2/Neu and inhibit cell transformation to similar extents, while a TM domain protein in which the GXXXG motif was disrupted proved much less effective. Taken together, these results indicate that the difference in dimerization propensity between the wild-type and Val 3 Glu mutated ErbB2-TM may be less relevant than are the structural details of the dimerization interface and the resulting relative orientation of the interacting protomers.
To summarize, we have shown that the TM domains of ErbB1, ErbB2, and ErbB4 all form homodimers. However, a variety of literature reports suggest that these TM domain interactions are not critical for ligand-induced receptor activation under conditions usually employed for their study. Instead, by analogy with studies of the EPO receptor (20), we argue that ErbB-TM interactions may play an important role in enhancing the efficiency of ligand-induced dimerization when receptor levels are very low (or when ligand is limiting). In this case, ErbB receptor dimers stabilized by TM domain interactions would not be expected to induce receptor activation. In fact, if they did, constitutive activation of the receptor would result. Our finding that the Neu mutation in ErbB2-TM does not enhance dimerization per se suggests that it may instead activate the receptor by altering the relative orientation of receptors in a predimerized state.