Dimerization of the Transmembrane Domain of Integrin αIIb Subunit in Cell Membranes*

Homo- and hetero-oligomeric interactions between the transmembrane (TM) helices of integrin α and β subunits may play an important role in integrin activation and clustering. As a first step to understanding these interactions, we used the TOXCAT assay to measure oligomerization of the wild-type αIIb TM helix and single-site TM domain mutants. TOXCAT measures the oligomerization of a chimeric protein containing a TM helix in the Escherichia coli inner membrane via the transcriptional activation of the gene for chloramphenicol acetyltransferase. We found the amount of chloramphenicol acetyltransferase induced by the wild-type αIIb TM helix was approximately half that induced by the strongly dimerizing TM helix of glycophorin A, confirming that the αIIb TM domain oligomerizes in biological membranes. Mutating each of the αIIb TM domain residues to either Ala, Leu, Ile, or Val revealed that a GXXXG motif mediates oligomerization. Further, we found that the residue preceding each glycine contributed to the oligomerization interface, as did the residue at position i + 4 after the second Gly of GXXXG. Thus, the sequence XXVGXXGGXXXLXX is critical for oligomerization of αIIb TM helix. These data were used to generate an atomic model of the αIIb homodimer, revealing a family of structures with right-handed crossing angles of 40° to 60°, consistent with a 4.0-residue periodicity, and with an interface rotated by 50° relative to glycophorin A. Thus, although the αIIb TM helix makes use of the GXXXG framework, neighboring residues have evolved to engineer its dimerization interface, enabling it to subserve specific and specialized functions.

By interacting with macromolecular extracellular ligands, integrins mediate essential cell-cell and cell-matrix interactions (1). Further, integrin occupancy transduces information into the cell interior that regulates processes as diverse as cytoskeletal organization, cell migration, cell proliferation, and cellular differentiation, whereas signals initiated within the cell regulate the ability of integrins to interact with ligands. The structural basis for integrin regulation is an area of intense study. For several integrins, there is a correlation between their activation state and the relative positions of the cytoplasmic (CYTO) 1 domains of their ␣ and ␤ subunits (2)(3)(4)(5). Thus, these integrins are inactive when their CYTO domains are in proximity and are active when the domains are far apart (6). Binding of cytoplasmic proteins such as talin to the cytoplasmic domain of one subunit or the other may influence this equilibrium, providing additional opportunities for regulating the activation state of the integrin (7).
The role of transmembrane (TM) domains in the integrin activation process is currently not well understood. On the basis of in vacuo molecular modeling, Gottschalk et al. (8) proposed a model in which the ␣ and ␤ subunit TM domains interact extensively in both the active and inactive states, with the interaction pattern changing during activation. Other models based on electron microscopy and protein engineering propose that TM domains interact only in the inactive state and move far apart upon activation (4,6).
The ability of the ␣ and ␤ TM domains to undergo homooligomeric interactions potentially adds a second level of regulation to the system. We have found that the ␣ IIb TM helix has a greater tendency to form a homodimer than to interact with the ␤ 3 TM helix in micelles. Schneider and Engelman (9, 10) reached a similar conclusion using the GALLEX assay. GAL-LEX is designed to determine the avidity of TM helix association via the suppression of bacterial ␤-galactosidase synthesis by two separately expressed LexA-TM-containing chimeric proteins. Depending on how it is configured, GALLEX can detect the hetero-or homo-dimerization of chimeric proteins. Using this construct, Schneider and Engelman (10) found that the ␣ IIb TM helix had a higher propensity to form homo-dimers than to dimerize with the ␤ 3 TM helix.
The avidity of interaction between the TM helices of ␣ IIb is relatively weak compared with that of constitutive homodimers such as glycophorin A (GpA). This observation is consistent with the hypothesis that ␣ IIb TM helix interaction is involved in integrin regulation. Thus, if ␣/␣ or ␣/␤ TM helix interactions were too strong, the system might be locked into a single activation state. Indeed, mutations that increase the propen-* This work was supported by National Institutes of Health Grants HL40387 and HL54500. 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 U.S.C. Section 1734 solely to indicate this fact. sity of the ␤ 3 TM helix to form oligomers lead to ␣ IIb ␤ 3 variants that are constitutively active (11).
The residues that determine the association of the ␣ IIb and ␤ 3 TM helices have not been thoroughly examined. It is noteworthy that integrin TM domains contain a GXXXG or a similar small residue-XXX-small residue motif. These motifs are a well-established framework for TM helix-helix interaction (12,13), and Schneider and Engelman (10) have demonstrated that they are involved in mediating homomeric and heteromeric associations of integrin TM helices. Here, we examine the structural basis for ␣ IIb TM domain dimerization. Using the TOXCAT system to measure helix-helix interaction in a lipid bilayer (14), our results confirm previous in vitro measurements of the dimerization of the ␣ IIb TM helix in micelles. Furthermore, we have identified the residues that constitute the dimerization interface and used this information to create a model for the structure of the dimer. These data should provide a powerful starting point for future investigations of the role of homo-and hetero-oligomerization of TM helices in integrin activation and clustering.

MATERIALS AND METHODS
Vectors and Strains-The expression vectors pccKAN, pccgpA-wt, and pccgpA-G83I, along with Escherichia coli strain MM39 (14), were kindly provided by Dr. Donald M. Engelman. The EcoRV restriction site between the TM region and the malB gene in pccKAN was changed to a BamHI site. The vector fragment resulting from NheI-BamHI digestion was purified by gel electrophoresis. The gene encoding the ␣ IIb TM domain, amplified from a full-length ␣ IIb cDNA and flanked by NheI and BamHI sites, was ligated in-frame with the vector fragment. Subsequent Leu scanning and other point mutants were generated using a QuikChange mutagenesis kit (Stratagene). The sequences of the wildtype and mutant ␣ IIb TM regions were confirmed by DNA sequencing. The resulting plasmids were transformed into E. coli MM39 cells for further analysis.
Expression of the Chimeric Protein in MM39 -A freshly streaked single colony was inoculated into 5 ml of LB broth containing 100 g/ml ampicillin and grown to A 600 of 0.6 at 37°C with vigorous shaking. One ml of the culture was chilled on ice, pelleted, and resuspended in 100 l of LDS sample buffer (Invitrogen). After an incubation at 4°C overnight, 18 l of bacterial lysate in LDS buffer, with the addition of 2 l of reducing agent (Invitrogen), was vortexed, boiled, and loaded onto a 10% NuPAGE Bis-Tris pre-cast gel in MOPS buffer system (Invitrogen). The separated proteins were transferred to nitrocellulose paper and immunoblotted with an anti-maltose-binding protein (MBP) monoclonal antibody (Sigma).
MalE Complementation Test-The glucose in M9 minimal medium plates was replaced by 0.4% maltose as the only carbon source (14). Transformed MM39 cells were streaked onto these plates containing ampicillin and incubated for 2 days at 37°C.
Protease Digestion of Spheroplasts-Preparation of spheroplasts was carried out following the protocol of Mendrola et al. (15). Proteinase K (Fisher Biotech) was then added to a final concentration of 50 g/ml, with or without 1% Nonidet P-40 (Sigma).
Chloramphenicol Acetyltransferase (CAT) Enzyme-linked Immunosorbent Assay (ELISA)-One ml of bacterial culture, freshly grown to an A 600 of 0.6, was chilled on ice before being pelleted at 4°C. The pelleted cells were resuspended and washed once using 0.6 ml of Trisbuffered saline buffer (20 mM Tris, pH 8, containing 100 mM NaCl and 2 mM EDTA). Upon resuspension in 0.5 ml of Tris-buffered saline on ice, the cells were treated with lysozyme at a final concentration of 100 g/ml for 20 min. Triton X-100 (to a final concentration of 0.4%), dithiothreitol (to a final concentration of 10 mM), and bacterial protease inhibitor mixture (Sigma) were then added to the mixture. After a 30-min incubation, the mixture was sonicated briefly at 4°C. On the following day, the supernatant was assayed for CAT concentration using a CAT ELISA kit (Roche Applied Science). Calibration of the assay was carried out using standards provided by the manufacturer. In each experiment, the strongly dimerizing GpA-wild type (GPA-WT), the weakly-dimerizing GpA mutant G83I, and ␣ IIb 968 -989 were included for comparison. CAT ELISA results were expressed as a percentage of CAT induced by GpA-WT in the same experiment. Chimeric protein expression was quantified from immunoblots using a Personal Densi-tometer SI (Molecular Dynamics) and used to normalize CAT expression by various constructs.
Protein Expression and Purification-Single-site mutations were introduced into the expression vector of wild-type protein using the QuikChange kit (Stratagene) and confirmed by DNA sequencing. The wild-type and mutant TM-CYTO proteins were expressed in E. coli and purified as described previously (16). The purity of each protein was confirmed by reverse phase high pressure liquid chromatography.
Characterization of ␣ IIb TM-CYTO Proteins-SDS-PAGE and analytical ultracentrifugation of the ␣ IIb TM-CYTO proteins were carried out as described previously (16). For SDS-PAGE, methanol was removed from 5 g of protein from a stock solution, and the protein was dissolved in 10 l of SDS sample buffer. After vortexing for 1 min, the protein sample was heated to 80°C for 10 min and loaded onto a 10% NuPAGE Bis-Tris pre-cast gel in MES running buffer (Invitrogen). The sample was electrophoresed at 4°C for 3 h at a constant current of 27 mA before being stained by Gelcode (Pierce).
For equilibrium sedimentation experiments in a Beckman XL-I analytical ultracentrifuge at 25°C, D 2 O was added to the buffer (10 mM dodecylphosphocholine, 20 mM MOPS, 100 mM KCl, 1 mM MgCl 2 , pH 7.4) to 50.34% to match the density of dodecylphosphocholine. The extinction coefficients were calculated as 16,500 M Ϫ1 ⅐cm Ϫ1 for all the proteins characterized in this report (17). The molecular mass and partial specific volume of wild-type and three mutant TM-CYTO proteins were calculated as described previously (16): ␣ IIb WT, 6017 Da and 0.7595 cm 3 /g; G972L, 6073 Da and 0.7633 cm 3 /g; G976L, 6073 Da and 0.7633 cm 3 /g; and L980A, 5975 Da and 0.7566 cm 3 /g. The obtained data sets were analyzed as described previously (16).
Perturbation Index Analysis-The perturbation index (P i ) for GpA was calculated based on the formalism of Treutlein et al. (18) using the exhaustive mutagenesis of the dimer interface in Lemmon et al. (19). A similar index was generated for the ␣ IIb TOXCAT data, where the perturbation index reflects the mean fold change in observed activity relative to the 968 -989 construct. The fold change was calculated using Equation 1, where n is the number of mutations at a given position i, and x j is the TOXCAT activity at position i for mutation j. P i values for GpA and ␣ IIb were then renormalized from 0 to 1, with 1 representing the maximum sensitivity to mutation. The helical phases for both molecules were determined by fitting P i as a function of position to a continuous periodic function with a period of 4.0 residue/turn (20).
␣ IIb Structural Model-The atomic model of the ␣ IIb dimer was constructed by searching an exhaustive grid of C 2 -symmetric parallel dimers. The grid is populated with 10 4 states representing rigid body transformations of idealized helices. Structural fitness was evaluated using a novel scoring protocol that automatically incorporates mutagenesis data into the energy function. This protocol has been shown to effectively eliminate false positive predictions in other TM proteins. The details of this approach will be described elsewhere. 2 Final predictions from the grid search were minimized in the constant valence force field using the Insight97.0/DISCOVER molecular modeling package (Biosym/MSI, San Diego, CA).

TM Domain of ␣ IIb Subunit Dimerizes in Biological
Membranes-Although the ␣ IIb TM domain dimerizes in zwitterionic and acidic micelles (16), it has not been established that the domain also dimerizes in membrane bilayers. We used the TOXCAT system to address this question. In TOXCAT, a chimeric protein consisting of an amino-terminal ToxRЈ DNA binding domain, an MBP domain, and an intervening TM domain of interest is expressed in the inner membrane of E. coli. Intramembranous dimerization of the TM domain results in ToxRЈ dimerization, thereby driving transcriptional activation of a CAT reporter gene, making the E. coli resistant to chloramphenicol. In addition, there is a positive correlation between the level of CAT expression and activity and the extent of TM helix-mediated chimera dimerization. For example, inserting the strongly dimerizing TM domain of GpA induces a high level CAT activity, whereas a G83I mutation (GpA-G83I) in the TM domain impairs both dimer formation and CAT activity (12,19).
The length of the inserted TM helix and the location of the junctions between the helix and MBP and ToxRЈ are important in defining the degree of CAT expression in the TOXCAT assay. Each defines the positioning of the TM helix in the membrane and influences the orientation of the fused domains with respect to the TM dimer interface. Accordingly, we identified the optimal length for an ␣ IIb TM helix in this assay by incrementally deleting single residues from its amino-terminal end (Fig.  1). As shown by the immunoblot for MBP in whole cell lysates in Fig. 2A, decreasing the length of the TM helix had essentially no effect on the expression of the chimeric protein for six of the seven constructs, whereas the level of expression of the chimeric protein with the shortest TM helix containing residues 969 -989 was decreased by Ϸ50% compared with the other chimeras. Two methods were used to confirm that the chimeric proteins were actually inserted into the E. coli inner membrane. First, the ability of each construct to facilitate the growth of an MBP-deficient strain of E. coli when maltose was the sole carbon source (MalE complementation) indicated that the MBP domain of the chimeric protein was located in the bacterial periplasmic space (Fig. 2B). Second, the ability of proteinase K to digest the chimeric proteins in E. coli spheroplasts indicated that the proteins were not sequestered in the bacterial cytoplasm and therefore protected from protease digestion (Fig. 2C).
Whereas the length of the ␣ IIb TM helix had essentially no effect on the extent of chimeric protein expression in E. coli, there were substantial differences in CAT expression as measured by CAT ELISA. As shown in Fig. 3, CAT expression induced by each ␣ IIb TM helix was significantly greater than that induced by the poorly dimerizing GpA mutant G83I with p values ranging from Ͻ0.025 to Ͻ 0.004 (t test for unpaired samples). Nonetheless, the extent of CAT expression was very sensitive to the length of the ␣ IIb TM insert, varying by more than 2-to 3-fold, depending on the length of the insert. The chimera containing ␣ IIb 968 -989 generated the highest ELISA signal and was used for subsequent experiments. Interestingly, this construct placed GXXXG in the same registry as in the originally reported GpA-TOXCAT construct (14). Thus, it is reasonable to compare the ELISA signals from the two TM helices. The signal from the ␣ IIb TM was approximately half of that generated by GpA-WT, indicating that the integrin TM domain was a somewhat weaker dimerization motif.
Identification of a Dimerization Motif in the ␣ IIb TM Helix- The experiments described above establish that the ␣ IIb TM helix can dimerize in biological membranes. To identify the sequence motif responsible for dimerization, we performed sitedirected mutagenesis, scanning the ␣ IIb TM helix sequentially with Leu, Ala, Val, and Ile residues. The ability of these mutants to complement MBP deficiency (Fig. 4A) indicates that the mutations did not alter the topology of the chimeric proteins in the bacterial inner membrane. Furthermore, randomly selected mutants remained susceptible to protease digestion in spheroplasts, confirming that they were correctly oriented in the bacterial membrane (data not shown). Lastly, as shown in Fig. 4B, the expression levels of the mutants were comparable with that of ␣ IIb 968 -989. The effect of single-site mutations on the dimerization of the ␣ IIb TM helix, as measured by CAT ELISA, is summarized in Fig. 5. Mutation of 12 of 21 residues in the ␣ IIb TM helix decreased CAT expression to some extent, indicating that the assay is highly sensitive to the native structure of the helix. In particular, mutation of Gly 972 and Gly 976 to Ala decreased dimerization, reducing CAT expression to nearly that of nondimerizing GpA-G83I. These results are consistent with the established role of GXXXG motifs in mediating TM helix interactions (19,21). However, mutations at sites flanking the critical GXXXG motif indicated that the interaction pattern is considerably different from that found in GpA. In GpA, the residues succeeding the critical glycines in GXXXG are important for dimerization, whereas the preceding residues are nonessential (12). Just the opposite was observed in ␣ IIb ; the residues preceding the critical glycines were important for dimerization, whereas the succeeding residues were found to be substantially more mutable. Thus, mutation of the residue preceding Gly 972 in ␣ IIb to Ala, Leu, or Ile resulted in substantial reductions in CAT activity. By contrast, the succeeding residue, Val 973 , was less sensitive to the same mutations. Similarly, the position preceding Gly 976 required a small side chain, either Gly or Ala, for efficient dimerization. Again, mutation of the succeeding residue had no or little effect.
Differences in the interaction of the GpA and ␣ IIb TM helices are further evident when one considers the effect of mutations at a position 4 residues distant from the second glycine of the GXXXG motif. In GpA, this residue is Thr, and mutation of this residue to Ala or other side chains disrupts dimer formation (19). The corresponding residue in the ␣ IIb TM helix is Leu 980 . Surprisingly, mutating Leu 980 to Ala resulted in a dramatic 2.5-fold increase in CAT expression. A smaller increase was observed when Leu 980 was mutated to Val, whose side chain is intermediate in size between Leu and Ala. This finding suggests that Leu 980 is also located at the dimerization interface, supporting the approximate 4-residue periodicity observed in other GXXXG-bearing TM sequences. In summary, the residues critical for dimerization of the ␣ IIb TM helix are XX-VGXXGGXXXLXX, whereas the critical residues for GpA are XXXGVXXGVXXTXX.
Quantitative Analysis of the TOXCAT Data-To place the data from TOXCAT on a more quantitative footing, we defined a mean P i , which defines the mean change in CAT activity for the various mutants. Fig. 6A illustrates the variation in P i versus the position of the mutations in comparison with that observed for the GpA TM helix. Both are qualitatively similar and show a pronounced 4-residue periodicity. The deviation from ideal ␣-helical periodicity (3.6 residues/turn) reflects the ϳ40°right-handed crossing angle between the two helices (19,21). Fig. 6, B and C, shows fits of the data to a sine function with 4.0-residue periodicity (Equation 2), in which x is the residue number, is the phase (in residues), A is the amplitude, and B and C are the constants describing the linear drift as the chain moves away from the point of closest approach. The quality of fit deteriorated for periodicities greater or less than 4.0. Two features are noteworthy: first, the phase is offset by about 47 Ϯ 8°for ␣ IIb relative to GpA, suggesting that the helices have been rotated by approximately this factor. Second, the degree of perturbation decreases beyond the first Gly residue in the GXXXG motif in ␣ IIb , but not in GpA. This suggests that the region of maximal contact is located closer to the first Gly residue in ␣ IIb than in GpA. We have also compared the P i profile for ␣ IIb with the kPROT scale (22), which describes the extent to which a given residue type prefers to engage in helix-helix interactions based on a sequence analysis of single spanning and polytopic membrane proteins (Fig. 6D). There is very good qualitative agreement between these two measures of interaction, particularly in the area surrounding the GXXXG motif.
Characterization of Mutant ␣ IIb TM-CYTO Proteins-In TOXCAT, the orientation of the ␣ IIb TM helix in the bacterial inner membrane is opposite to its orientation in eukaryotic membranes (14). Moreover, MBP and ToxRЈ are appended to its carboxyl and amino termini, respectively. Thus, to confirm that neither feature affected our results, we directly studied the oligomerization of a protein encompassing the wild-type ␣ IIb TM and CYTO domains and the same protein containing the mutations G972L, G976L, and L980A using SDS-PAGE and analytical ultracentrifugation.
SDS-PAGE has been used extensively as a simple, direct way to visualize the oligomerization of membrane proteins. In the  Fig. 7A, wild-type ␣ IIb TM-CYTO protein migrated as two discrete bands corresponding to monomer and the dimer species (16). The G972L mutant also migrated as two bands, but it was not apparent using this technique that the mutation affected dimerization. In contrast, the amount of G976L dimer was substantially decreased, and the amount of L980A dimer was substantially increased, as would be predicted from the TOXCAT assay.

SDS gel shown in
Analytical ultracentrifugation provides a quantitative measurement of ␣ IIb TM helix oligomerization. When dispersed into dodecylphosphocholine micelles at pH 7.4, the ␣ IIb TM-CYTO protein is in a monomer-dimer equilibrium (16). The equilibrium radial concentration profiles for mutants G972L and G976L also fit best with a monomer-dimer equilibrium; consistent with the data shown in Fig. 7A, the data for the L980A mutant could be fit equally well to either a single dimeric species or to a monomer-dimer-trimer equilibrium with a very small trimer contribution, consistent with the faint trimer band observed in Fig. 7A. Fig. 7B displays the calculated fraction of monomers for WT and the mutant proteins as a function of protein/detergent ratio, based on dissociation constants derived from data spanning a peptide/detergent ratio from 0.2/ 1,000 to 70/1,000. These data show that the G976L mutation substantially decreased ␣ IIb TM helix oligomerization and that the L980A mutation substantially increased ␣ IIb TM helix oligomerization. In contrast to SDS-PAGE, we were also able to observe a significant disruptive effect of the G972L mutation.
Modeling of ␣ IIb TM Dimer-The mutagenesis data were then used to construct an atomic model for the ␣ IIb dimer (Fig.  8, A and B), using a novel method that exhaustively searches conformational space and explicitly considers the computed energies of the wild-type, disruptive, and permissive mutations. A family of structures was found that satisfies these mutational restraints. All have right-handed crossing angles, as was observed in GpA, with a value ranging from ϳ40°to 60°. This finding is consistent with the 4.0-residue periodicity found in the P i profile. The face of the helix that is buried in the helix-helix interface of the model structures is constant and consistent with the analysis of the P i profile; examination of Fig. 8C shows that the interface for ␣ IIb is rotated by 50°r elative to the GpA dimer interface.

DISCUSSION
The TOXCAT system is an optimized method for overcoming the technical difficulty of studying the homomeric associations of TM helices in biological membranes (14,23) and has been used to investigate the homomeric interactions of the GpA TM domain (14), other naturally occurring TM helices such as ErbB4 (15) and VacA (24), and randomized recombinant peptides (12). Previously, we found that the homomeric association of proteins corresponding to the TM and CYTO domains of the integrin ␣ IIb and ␤ 3 subunits in micelles involved their TM segments (16). Inspection of the ␣ IIb TM domain sequence revealed that it contains the sequence GVLGG, reminiscent of the GXXXG motif that is present at the interface of GpA homodimers (21), which could account for its ability to oligomerize. On the basis of these observations, we used the TOXCAT system to confirm that the ␣ IIb TM domain can dimerize in biological membranes and to identify the residues that constitute its dimerization interface.
The TOXCAT assay exploits a dimerization-dependent activator of transcription (ToxR) to report the ability of transmembrane sequences to oligomerize in the E. coli inner membrane (14). An important feature of this assay is that the precise location of the fusion of the TM segment with the ToxR subunit has a very significant effect on the magnitude of CAT expression (14). Thus, to compare the dimerization efficiencies of TM segments from two different proteins, it is essential to first determine that an optimal junction has been identified for both targets. To identify the optimal length of the ␣ IIb TM segment to be inserted into TOXCAT, we expressed a series of ␣ IIb TM domain truncation mutants and found that a 22-amino acid peptide encompassing ␣ IIb residues 968 -989 yielded maximum CAT activity. The maximum amount of CAT induced by ␣ IIb 968 -989 was 40 -50% of that induced by the wild-type GpA construct. In this construct, the registry of the GXXXG sequences is the same as in the GpA construct. If a non-optimal junction had been used for the ␣ IIb TM helix, a significantly lower degree of dimerization would have been inferred. It is noteworthy that in the optimized construct, Trp residues are present at or near both ends of the 968 -989 peptide; this amino acid is generally located at the ends of a TM helix (25).
Schneider and Engelman (9, 10) have also examined the association of the ␣ IIb and ␤ 3 helices using the GALLEX assay. This method allows investigation of both homomeric and het- eromeric association of TM helices. Consistent with our previous report (16), Schneider and Engelman (10) detected both ␣ IIb and ␤ 3 oligomers. The strength of the ␣ IIb signal was about one-fourth to one-third that observed for GpA. Given that the optimal location for fusion of the TM domain with the LexA domain was not investigated in this study, we consider this result to be in good agreement with our own finding using the TOXCAT system. Also, consistent with our earlier observations, Schneider and Engleman (10) confirmed that the ␣ IIb /␤ 3 interaction is weaker than the corresponding ␣ IIb /␣ IIb interaction.
We used scanning mutagenesis to identify residues that are present at the ␣ IIb TM helix dimer interface. CAT expression was decreased to a variable extent, and the extent of the reduction correlated well with the distance from the predicted helix-helix interface. Thus, residues that were on the opposite face of the helix from the interaction site had little or no effect on dimerization. We found that replacing Gly 972 or Gly 976 , the first and last residues of a potential GXXXG motif, either with a residue having a large hydrophobic side chain or with Ala, markedly disrupted ␣ IIb TM dimerization. The GXXXG motif, first recognized as a framework for helix-helix association in studies of the GpA TM helix (19), has been consistently identified as the most overrepresented sequence motif in TM domain databases (13). In the NMR structure of the GpA TM helix dimer, GXXXG permits extensive backbone-backbone contacts at the dimer interface because a groove created by Gly 79 and Gly 83 of one GpA monomer packs against a ridge created by the side chains of Val 80 and Val 84 of the other (26). The sequence arrangement GXXXG with Val, Ile, or Leu at positions i Ϯ 1 is also overrepresented in TM domain databases (13) and is a frequent packing motif in high affinity homooligomers selected from a randomized GpA TM helix sequence library (12). In the ␣ IIb TM helix, Gly 972 is preceded by an immutable Val, whereas Gly 976 is preceded by Gly, which can be replaced by Ala, but not Val or Leu, without affecting ␣ IIb TM helix dimerization. Thus, although GXXXG is a framework for helix-helix association, the diversity of residues neighboring GXXXG, exemplified by the ␣ IIb and GpA helices, likely imposes specificity on this interaction.
Although the structural model shown in Fig. 8 explains the most perturbing mutations, it does not account for some aspects of the mutagenesis data shown in Fig. 5. For example, residues 981-984 are sensitive to mutations, even though the helices have diverged and are no longer in contact at this position. This may reflect a limitation of our method that uses straight helices in the modeling process. It is possible that gradual bending or a kink downstream from the second Gly in the GXXXG motif may extend the interfacial region toward the carboxyl terminus of the TM helix.
In addition, mutation of Leu 980 to Ala increased TOXCAT activity severalfold, an effect detected by SDS-PAGE and analytical ultracentrifugation as well. Mutation of Leu 980 to Val also increased TOXCAT activity, but to a lesser extent. It is noteworthy that Val is present in the corresponding position in the TM helix of ␣ 7 , which is known to have a very high tendency to form homodimers (9). The analogous residue in the GpA TM helix is Thr 87 , and mutation of this Thr severely disrupts GpA TM dimerization (19). It is difficult to explain these observations by considering the ensemble of structures alone. L980A failed to significantly improve the computed energy of interaction between the helices. We considered the possibility that this mutation induces an alternate conformation. Indeed, calculations showed that the minimum energy conformation for this mutation was much more similar to the GpA structure. Thus, the nature of the residue at position 980 may be important for defining the rotation of the helices.
Previously, we reported that the reversible homomeric association of the ␤ 3 TM domain may play a role in the regulation of both the ␣ IIb ␤ 3 activation state and ␣ IIb ␤ 3 clustering (11). Because the ␣ IIb TM domain undergoes reversible dimerization, it is also possible that the homomeric association of this domain could regulate ␣ IIb ␤ 3 function. Vinogradova et al. (5) have proposed that a clasp between the membrane-proximal regions of the ␣ IIb and ␤ 3 CYTO domains maintains ␣ IIb ␤ 3 in an inactive state and that disruption of the clasp enables ␣ IIb ␤ 3 activation. Our previously reported data (11,16) raise the possibility that disruption of the clasp is accompanied by the homomeric interaction of the ␣ IIb and ␤ 3 TM domains, thereby providing a physiologic context for movement of the ␣ IIb and ␤ 3 stalks (6). Consistent with this hypothesis, we have found that the affinity of the ␣ IIb TM helix for dimerization is modest (16), making it amenable to regulation. In this study, we have provided a structural basis for this decreased affinity. Thus, although the ␣ IIb TM helix makes use of the common GXXXG framework for helix-helix interaction, neighboring residues have evolved to re-engineer its dimerization interface, enabling it to subserve a specific and specialized function.