GxxxG Motifs, Phenylalanine, and Cholesterol Guide the Self-Association of Transmembrane Domains of ErbB2 Receptors

Abstract

GxxxG motifs are common in transmembrane domains of membrane proteins and are often introduced to artificial peptides to inhibit or promote association to stable structures. The transmembrane domain of ErbB2 presents two separate such motifs that are proposed to be connected to stability and activity of the dimer. Using molecular simulations, we show that these sequences play a critical role during the recognition stage, forming transient complexes that lead to stable dimers. In pure phospholipid bilayers association occurs by contacts formed at the C-terminus promoted by the presence of phenylalanine residues. Helices subsequently rotate to eventually pack at short separations favored by lipid entropic contributions. In contrast, at intermediate cholesterol concentrations, a different pathway is followed that involves dimers with a weaker interface toward the N-terminus. However, at high cholesterol content, a switch toward the C-terminus is observed with an overall nonmonotonic change of the dimerization affinity. This conformational switch modulated by cholesterol has important implications on the thermodynamic, structural, and kinetic characteristics of helix-helix association in lipid membranes.

Introduction

Stability and activity of transmembrane (TM) proteins depend on the successful association of individual fragments to higher complexes (1,2). Association events are critical to biochemical processes as signal transduction mechanisms as well as to several applications based on designing artificial peptides that display a desired function (e.g., inhibitors of association) (2–7). Despite extensive research during the last decades, connecting stability to structure and function remains a formidable task that requires detailed characterization of factors such as the amino acid sequence and the physicochemical characteristics of the membrane environment.

GxxxG motifs are amino acid sequences with two small hydrophobic residues on the same face of a helix that are common in several TM proteins and considered a primary factor for successful association (8,9); in fact, such sequences are often employed when engineering artificial peptides (7). The absence of side chains, which would penalize association through loss of entropy, and the increase of the area of the interface (and thus protein-protein interactions) are main factors that assist association along the GxxxG sequence (8). Glycophorin A (GpA) is a prototype of this type of dimerization (8,10–14), although it has been argued that association is not solely controlled by the interface formed between the two helices (15). The members of the family of epidermal growth factor receptors (ErbBs) contain GxxxG sequences in their TM domain and it is suggested that this facilitates their dimerization (16). ErbB2, which does not require a ligand for activation (5), contains two separate GxxxG motifs in the TM sequence, one closer to the N-terminus and one toward the C-terminus. Findings that a specific Val → Glu mutation within the TM domain of the Neu oncogene product (homologous position 659 in ErbB2) induces dimerization- and activation (17,18)-initiated extensive experimental studies whose aim was to connect dimer structure to thermodynamics and activity of the receptor. However, experiments with mutated ErbB2 TM domains introduced in chimera proteins in Escherichia coli resulted in lower dimerization affinities, supporting the idea that structural changes apart from proximity are required for activity (16); for the Neu-TM domain, there is evidence that such changes are connected to a rotational coupling (19).

Theory and simulations have also aimed to establish the connection between structure, thermodynamics, and activity of TM proteins. A measure of helix packing is the crossing angle Ω (20), and early studies reported a left-handed coiled coil structure (Ω ≈ 20–40°) for mutated ErbB2 (21–24). Fleishman et al. performed a conformational search, fixing Ω at −35° (right-handed, as in GpA) to propose an activation switch that explains many of the observed effects of ErbB2 activity (25). Results supported two favorable packings, the most stable with an interface at the C-terminus GxxxG motif (proposed to correspond to an inactive state) and a second at the N-terminus (active state induced by mutations) (25). The concept of a conformational switch for activation is also consistent with several experimental data that support the existence of inactive preformed ErbB dimers in cell membranes (26–28). A different conformational search resulted in three different accessible structures: a left-handed at the N-terminus and one left- and one right-handed at the C-terminus (29). Beevers and Kukol performed experiments and molecular dynamics (MD) simulations in a lipid bilayer to find that ErbB2 forms a right-handed dimer, in contrast to earlier simulations (30); however, their TM domain was the Neu product, which has small but significant differences, as discussed further in this study. We note, though, that the TM domains remained helical, no direct Glu-Glu interactions (point mutations) were observed, and phenylalanines close to the C-terminus were located away from the interface. Recently, Bocharov et al. performed NMR experiments with ErbB2 TM domains in bicelles together with short MD simulations in bilayers to propose a right-handed dimer structure that, due to contacts toward the N-terminus, presumably corresponds to the activated state, as suggested by Fleishmann (31). However, the hydrophobic environment plays a role in the dimerization of ErbB2 (32), and the structure proposed develops deviations within 20 ns in an ensemble appropriate for a tensionless lipid bilayer (33). It is also clear that atomistic simulations remain close to the initial configuration, since rotational diffusion in membranes occurs on timescales of 10⁻⁴–10⁻⁵ s, with even longer times for aggregated oligomers (34,35). To overcome such limitations, simpler coarse-grained (CG) models that retain a level of amino acid detail were employed successfully to describe the association of peptides in lipid bilayers (36,37). Using such models and parallel Monte Carlo (MC) simulations we found that packing along GxxxG motifs for GpA is promoted by favorable lipid-entropic contributions (38), a result supported by other research groups (39). Furthermore, we showed that such membrane-mediated effects depend on amino acid sequence, with ErbB1 forming a nonspecific packing in a cholesterol-free lipid bilayer due to residues that promote low tilting, in contrast to ErbB2, where a clear interface was identified close to the C-terminus (40).

The role of lipid-mediated contributions in observed protein activity can be quite diverse when lateral heterogeneity within a cell membrane is considered. Extensive data support that receptors localize in lipid domains rich in cholesterol and signaling is modulated by the concentration of the sterol molecules (41–43). Kinetic parameters such as mobility (44), as well as changes in thermodynamics (45), could be significantly different in these lipid domains and potentially alter the observed activity. Recent experiments with GpA support that the association affinity increases, although in plasma membranes (higher cholesterol content), a lower amount of dimers was reported (46,47). Using a mesoscopic model that accounts for hydrophobic mismatch, de Meyer et al. found that cholesterol is enriched in the proximity of the proteins and this effect can reduce a repulsive barrier present in the free-energy profile as a function of separation (48). In this study, we examine cholesterol-induced effects on the self-association of TM domains of ErbB2, and we present evidence that helices follow disparate association pathways with implications on the dimerization and signaling mechanisms of this important family of receptors.

Models and Methodology

The model TM sequence of ErbB2 employed is: L₆₅₁TSIISAVVGILLVVVLGVVFGILIKR₆₇₇, as in past work (40), with an α-helical secondary structure imposed between residues 653 and 677 according to the MARTINI forcefield (49) and based on input from atomistic simulations performed with single TM domains in dipalmitoylphosphatidylcholine (DPPC). The charged amino acids at the C-terminus interface (residues 676 and 677) are neutralized by two chloride ions. Simulations are performed at 10%, 20%, and 30% cholesterol concentration and 323 K with phospholipids, cholesterol, and water models described as in MARTINI (49). We performed two different series simulations: MD using Gromacs v3.3.3 (50) and MC free-energy calculations using our (MW)²-XDOS algorithm (38,51). Initial configurations for MD were constructed by embedding ErbB2 helices in lipid bilayers and performing subsequent equilibration with MD for 1 μs using weak Berendsen coupling (τ_t = 1 ps, τ_p = 0.2 ps semi-isotropic). A similar procedure was used to construct a configuration for a pair of helices. This structure was then subjected to an MC simulation with a large value of the modification factor, f (51), to create 128 pairs (512 total for all cholesterol concentrations) distributed over 16 windows covering all separations. These configurations served as initial structures for (MW)²-XDOS (with different random number seeds).

Critical features of the free-energy calculations are the implementation of a parallel MC scheme with multiple unconstrained walkers (see Fig. S1 in the Supporting Material (38)) over extended protein lateral center-of-mass (COM) separations, ξ (up to ∼5 nm), and flat histogram sampling resulting to an equal number of pairs over all ξ. Since diffusion over ξ depends implicitly on sampling collective variables, such as the position of lipid and cholesterol molecules, every MC attempt to displace proteins is followed by ∼99 moves on lipids and water, out of which ∼98 are focused on molecules and beads in immediate proximity to the proteins. In this study, we performed such preferential sampling using six points as centers for the selection probability (51), three on each helix: a middle point (COM of residues 662–664) and two at the interfacial residues (COMs of 651–654 and 674–677). Preferential sampling accelerates convergence of the underlying Wang-Landau scheme (52,53) extensively, given the large box sizes required to avoid finite-system size effects (lateral areas ∼109.5 nm² for 10%, ∼112 nm² for 20%, and ∼155 nm² for 30% cholesterol—a total of 400–500 membrane molecules). This local update was also supplemented with a global hybrid NVE MD (0.1% attempted ratio) that accumulated ∼500 ns of accepted trajectories for each replica. Despite all these improvements, our potential of mean force calculations converged only after exhaustive sampling over time periods up to six months on 128 CPU cores for each system. This requirement has a physical origin in the extent of helix conformational sampling required, shown further in the analysis.

Results

Single ErbB2 TM domain in membranes with cholesterol

We first examine the configuration of a single TM domain as a function of cholesterol content. As expected, cholesterol increases bilayer thickness with the helix orienting more parallel to the membrane normal. Fig. 1 A presents the free energy as a function of tilt angle, τ, with a broad minimum moving from ∼30° to 26°, 24°, and 17° by decreasing cholesterol content and mean values somewhat higher depending on the profile. The change is not proportional to cholesterol content: a significant decrease is found at 30%, associated with higher lipid tail ordering and an increase in the glycerol-glycerol separation between opposing leaflets (Fig. 1 A, inset). It is clear that the change of tilting and the increased lateral pressure at content >20% (54) will have an impact on dimerization. However, we first analyze the distribution of lipids and cholesterol in proximity to a single TM domain, which holds a critical role.

Figure 1.

Configuration of a single ErbB2 TM domain in membrane as a function of cholesterol content. (A) Free energy as a function of tilt angle, τ, and cholesterol content. (Inset) Separation between the two bilayer leaflets defined by lipid phosphate or glycerol beads. (B) Labeling of lipid molecules as proximal or bulk based on a simple distance criterion of individual beads. (C) Cholesterol COM packs at closer separations than phospholipids, with a fraction of proximal cholesterol lower than overall content. (D) Cholesterol distribution along the membrane normal together with the distribution of amino acids of the two GxxxG motifs (shown for 20% cholesterol content). (E) At left is a snapshot of a single ErbB2 TM domain, in which the location of the GxxxG motifs is indicated in blue at the N-terminus (Ser⁶⁵⁶ and Gly⁶⁶⁰) and red at the C-terminus (Gly⁶⁶⁸ and Gly⁶⁷²). Additional colors indicate Val⁶⁶⁴ (yellow) and Phe⁶⁷¹ (violet). Cholesterol molecules are rendered in green. At right is an image of an isosurface (green) for cholesterol calculated based on the spatial distribution function extracted from 1 μs molecular dynamics of a single helix in a bilayer with 20% cholesterol content. (F) Lateral distribution of COM of cholesterol molecules at the N- and C-termini, with black lines denoting the average location of COMs of the corresponding GxxxG motif.

Recent simulations reported that cholesterol molecules form a layer around TM proteins, which favors dimerization by reducing repulsive barriers (48). To examine such effects, we performed MD simulations with a single helix and we calculated lateral radial distribution functions between the COM of the protein and the COM of phospholipids (DPPC) and cholesterol (CHOL). Selected profiles for 20% cholesterol content are presented in Fig. 1 B. As in other studies (48), we found a pronounced peak for cholesterol at close distances; for phospholipids, a weak peak at longer separations was observed. To perform a simple separation of proximal molecules from distant, we labeled any DPPC or CHOL molecule as proximal if it has a bead within 0.6 nm of any bead of the protein (Fig. 1 B (40)). Fig. 1 C presents the concentration profiles of proximal phospholipids and cholesterol as a function of lateral separation from the helix COM for all systems studied. As observed, the decrease of the protein tilt angle, the introduction of cholesterol, and the ordering of lipids lead to a narrower distribution for ∼22.5, 20.9, 18.6, and 16.8 phospholipids (for 0–30% CHOL, calculated by integrating along cylindrical coordinates) centered at ∼1 nm. In contrast, rigid sterol molecules (∼0, 1.9, 4, and 6, respectively) pack at closer separations of ∼0.75 nm. To compare the extent of such preferential packing to bulk concentration, we calculated the fraction of each component within the population of all proximal molecules (approximated by the sum of the two populations). As shown in the inset, the resulting fractions vary linearly with overall cholesterol concentration in the system considered. However, the exact amount of cholesterol appears to be somewhat lower than the bulk concentration (i.e., 26% for bulk 30% content). Although the difference is small, it is justified by further spatial analysis.

A simple density calculation along the membrane normal reveals an intriguing finding (Fig. 1 D): the sterol molecules configure in a bilayer at positions that coincide with the location of the GxxxG motifs for ErbB2. Additional insight is gained by g_spatial, distributed by Gromacs (55), to extract the spatial distribution function (SDF) of cholesterol and visualize the resulting isosurfaces with VMD (56). Fig. 1 E presents two snapshots. The left is a single configuration of an ErbB2 TM domain in a membrane, emphasizing the two ends of each motif along the helix (notice that the motifs are not along the same face) together with the location of the middle Val⁶⁶⁴ and Phe⁶⁷¹ residues, which play an important role. An isosurface for the cholesterol SDF obtained by analyzing the whole trajectory is shown in the right image (system with 20% overall cholesterol concentration). It is apparent that the sterol preferentially packs around the GxxxG motifs; however, close to the C-terminus, a depletion is observed around Phe⁶⁷¹. Side-chain entropy of this residue and orientation of the ring relative to cholesterol could play a role; the Phe ring is normal to the helix axis, whereas cholesterol tilts 20–30° with respect to the membrane normal (Fig. S2), as in atomistic studies in the literature (57–59). Further evidence of this depletion is provided in Fig. 1 F by orienting the helix with the N-terminus GxxxG motif along the x axis and calculating the lateral distribution of cholesterol molecules averaged over the whole trajectory. This depletion of cholesterol is in agreement with the deviation for the fraction of proximal sterol molecules from the bulk content we calculated based on our simple distance criterion earlier. We conclude that cholesterol preferentially packs in proximity to the helix; however, this effect is absent for the region around residue Phe⁶⁷¹.

Thermodynamics of association in the presence of cholesterol

We extracted the free-energy profile (or potential of mean force (PMF)) as a function of separation of two ErbB2 TM domains in bilayers with cholesterol content and compared our past results in pure DPPC (40). As shown in Fig. 2, association in all environments is favorable, with a nonmonotonic effect observed for the free-energy minimum. The standardized free-energy differences, ΔG, scale to −7.64 ± 0.16, −6.99 ± 0.14, −6.06 ± 0.14 and −6.93 ± 0.22 kcal/mol for 0%, 10%, 20%, and 30% cholesterol, respectively (38,40). We note that since no extramembrane domains are considered, the thermodynamics presented are only qualitative in the context of full receptor dimerization. For FGFR3 receptor, extracellular domains can inhibit dimerization (1 kcal/mol recently reported (60)); for ErbB2, such domains do not provide significant inhibition, although a detailed estimate of contributions by each domain remains challenging (61,62). Despite such limitations, given the magnitude of the above contributions, the low inhibition by extracellular domains, and the abundance of data expressing TM domains in chimera proteins (16), our study can offer insight into membrane-mediated effects even with the present restriction to the TM sequence.

Figure 2.

Free energy (PMF) of dimerization of ErbB2 TM domains and decomposition to separate contributions based on projected average forces exerted on the two helices. Lipid-mediated interactions were computed based on forces exerted by both cholesterol and phospholipid molecules. Errors of mean forces obtained by block-averaging were used to estimate PMF errors by subsequent boot-strapping with randomly sampled forces up to 2 SD at each ξ (75,76).

We anticipated an increase in lipid-induced interactions at low separation values (ξ = 0.5–0.8 nm) as a result of higher lateral pressure (54) that could lead to an increase of the association affinity. However, lipid-mediated interactions are coupled to the arrangement of the TM domains in the hydrophobic environment (40) and, as shown in Fig. 3 A and Fig. 1 A, the tilt angle, τ, is highly affected by the properties of the membrane. A lower tilt decreases the lipid-excluded volume within the hydrophobic domain, therefore reducing exposure to the hydrophobic lipid tails. At large separations, the values agree with mean values extracted from Fig. 1 A, except in the case of 10% cholesterol, where small deviations were observed and attributed to limited sampling of the few cholesterol molecules. An intriguing result is that at 20% cholesterol content (weaker dimer), an average lower tilting is observed at the free-energy minimum (ξ ≈ 0.8 nm (Fig. 3)). As shown in Fig. 3 B, by the joint distribution for a pair of helices, this difference is a result of both helices in a dimerized pair experiencing lower tilt angles. The origin of this behavior, which will be analyzed in the following section, can only be explained by examining collective conformational changes beyond the description offered by a single helix in a membrane of specific thickness.

Figure 3.

(A) Average tilt angle, τ, of helices as a function of lateral COM separation and cholesterol content, with errors of the mean extracted by block-averaging. (B) Joint distribution for tilt angles of a pair of helices at 0 and 20% cholesterol in the dimerized state (states 2 and 4 in Fig. 4).

Before examining the configurations probed in our simulations, we note the presence of a weak repulsive barrier extending from ∼1.6 nm to larger separations (1.5, 1.9, 1.4, and 1.9 kcal/mol for 0, 10, 20, and 30% cholesterol, respectively). This is an indication that association could be an activated process, with transition states sampled before the TM domains fall into the low free-energy minimum formed by protein-protein and favorable lipid-induced interactions (63). Identification of such states by equilibrium simulations is challenging due to their limited lifetime. In our study, sampling is biased along the lateral separation which is an order parameter rather than a true reaction coordinate. Another order parameter is the minimum interhelical distance, with low values at first contact, ξ = 1.6–2.4 nm, depending on tilting (Fig. 2 B in Janosi et al. (38)). It is clear that to examine kinetics of the association process, it is necessary to examine the change of the true reaction coordinate, defined by following the minimum-free energy path. Rigorous simulation techniques that probe such paths exist (64,65), although their application is hindered by the slow dynamics in membranes. The minimum free-energy path could be defined not only by separation and helix-helix contacts but also by collective variables such as the position of surrounding lipid molecules. To provide insight beyond Fig. 2, we resort to additional measures (interface, crossing angle, Ω, and concentration of components) to determine pathways and potentially activated complexes that drive the dimerization of TM domains of ErbB2.

Association pathways

The free-energy calculations provided us with numerous structures equally sampled along the lateral COM separation, ξ. To identify free-energy minima with respect to other variables, we plot the probability $\mathcal{P}\left(\Omega ,\xi \right)$ of a specific crossing angle, Ω, as a function of ξ, extracted by forming a two-dimensional histogram based on the configurations probed. However, in contrast to our past studies (38,40), we extend the range up to ξ = 2.5 nm (Fig. 4). We remind the reader that Ω depends on the extent of tilting, with helices sampling values closer to zero for low τ (and forming contacts at closer distances). Furthermore, a pair of TM domains will sample separations with a probability prescribed by the free-energy profiles provided in Fig. 2. We examined $\mathcal{P}\left(\Omega ,\xi \right)$ in our past study for pure DPPC membranes (0% cholesterol), focusing on ξ < 1.2 nm, since at 1.5 nm no significant preference for a specific configuration was found. Upon reevaluation of our data we observed a population of preferred Ω ≈ −62° at 1.6–1.8 nm (labeled state 1). Contact maps extracted at these values of ξ reveal that a complex is formed with Phe⁶⁷¹ residues of the two helices interacting directly (Fig. 5). Formation of this complex is further assisted by the lack of side chains for Gly in the C-terminus GxxxG motif. A subsequent approach of helices at shorter separations requires a rotation to remove the rings out of the interface and pack the helices one turn closer to the middle at V⁶⁶⁴xxxG⁶⁶⁸ (state 2, Figs. 4 and 5). Therefore, the association process in pure phospholipid bilayers presents multiple saddle points; first, lipid molecules in proximity to the C-terminus (at ξ > 1.8 nm) need to be removed, and helices will then roll over toward the center at ξ ≈ 1.2–1.6 nm. Further support of this mechanism is provided by a horizontal alignment of the pair of helices and calculation of the lateral density of Phe residues for ξ = 1.6–1.8 nm and ξ = 0.6–0.8 nm in Fig. 5. As can be observed, Phe residues are originally at the interface but then rotate outward so that helices pack efficiently, releasing volume to lipids and experiencing favorable lipid-mediated interactions.

Figure 4.

Association for different cholesterol contents as described by the probability of a crossing angle, Ω, for a specific lateral COM separation ξ. Representative configurations shown in Fig. 5 (states 1–5) are sampled in the ranges indicated.

Figure 5.

Representative configurations with top and front views (states 1–5) sampled in the ranges indicated in Fig. 4. Along the snapshots, the average distribution of Phe⁶⁷¹ residues, given a horizontal orientation of the separation vector and residue contact maps, are provided.

It is reasonable to question the accuracy of the model, given the coarse description of the amino acid sequence. However, an evaluation of literature experimental data provides significant support of this pathway. For example, Melnyk et al. found that Phe residues far from the packing interface contribute to the dimerization propensity of helices (66). Stronger support is provided by the extensive experiments of Unterreitmeier et al. in bacterial inner membranes; the authors proposed explicitly that Phe residues located less than a full helical turn away from a packing interface contribute to the association process (67). Additional evidence has recently been reported based on experiments of Beevers et al. with the Neu TM domain; the authors postulated that this sequence undergoes a conformational switch during association that is blocked by a Val → Glu mutation toward the N-terminus (68). The Neu TM domain is homologous to ErbB2, with the important difference that a Phe residue is present at the N-terminus. It is possible that a similar rotation mechanism is present that is altered for the oncogene sequence.

The next step of our analysis is to examine the effect of cholesterol. At 10% cholesterol (Fig. 4), an additional population of left-handed dimers appears at somewhat longer distances (ξ ≈ 2 nm) and becomes predominant for 20%. Structural analysis shows that during recognition, helices now form transient states packing along the N-terminus (Ser⁶⁵⁶xxxGly⁶⁶⁰), with the Phe residues at the C-terminus already far from the interface (state 3 (Figs. 4 and 5)). Since we defined ξ using all beads, a rotation taking Phe residues out of the interface leads to slightly larger values of separation. There are multiple contributions that drive the system to such a switch in recognition stage, including membrane thickness and enthalpic interactions between Ser-Ser residues. As discussed earlier, overlap of cholesterol layers is also favorable (48); therefore, helices prefer to associate over the N-terminus and not penalize the entropy of Phe rings. By extracting cholesterol profiles for the recognition stage and the upper membrane leaflet (N-terminus), we found increased populations of sterol molecules parallel to the Ser-Ser contact (Fig. 6 A); these molecules were removed from the layer surrounding nonassociated helices. However, another interesting feature is revealed for the lower leaflet with a layer of cholesterol molecules extending between the helices (Fig. 6 B). Association proceeds to lower separations by forming a dimer mostly toward the N-terminus (but further down the sequence), with the domains exhibiting low individual tilting (state 4). We note that a decrease in tilting is an additional mechanism to release volume to lipid molecules and increase entropy of the system (40). We also add that the N-terminus interface for 20% cholesterol is the least well defined of all associated dimers, a feature that appears consistent with the higher free-energy minimum shown in Fig. 2 (despite the latter being extracted as a function of separation).

Figure 6.

Lateral cholesterol COM number density/nm² drawn separately for upper (N-terminus) (A) and lower leaflets (C-terminus) (B) of the membrane for ξ = 1.6–1.8 nm and 20% overall cholesterol content (state 3 in Fig. 4C).

A different character of the process is observed at high cholesterol content. Helices do not tilt extensively and recognition occurs at close separations, with the Phe groups protruding throughout the cholesterol layer and making first contacts. They approach further, forming a stable dimer toward the C-terminus, again rotating the rings out of the buried area and increasing the tilt angle. However, as shown in Fig. 5, state 5 is different from state 2. The Phe rings are now located on the same side, interacting directly and being surrounded by cholesterol molecules. This packing minimizes exposure to the high lateral pressure and suggests an overall different arrangement of the dimer in highly ordered lipid membranes.

Conclusions

Our results provide evidence that amino acids not in the packing interface of dimers of TM domains of ErbB2 can still contribute to the dimerization process, forming critical transient complexes. Past studies have shown that flanking residues, as well as residues that alter tilting of the helices, modulate the association affinity (40,69). Herein, we show that even hydrophobic phenylalanine residues can affect dimerization of ErbB2, acting as anchors over a pathway involving the C-terminus in pure phospholipid bilayers. Association proceeds with a conformational switch that is modulated by the concentration of cholesterol molecules in the membrane. This is particularly important given that ErbB2 (with no known ligand binding to this protein) is the preferred member for heterodimerization in the family of epidermal growth receptors (5). Our study would be incomplete without discussing the implications of our findings for the mechanisms of activation proposed in the literature for the full receptors: a conformational switch of preformed dimers or a change in a monomer-dimer equilibrium (6).

If activity is promoted by packing of TM domains along different interfaces, then we can assume that the pair formed with contacts toward the N-terminus (state 4 in Fig. 5) corresponds to an active configuration (25). However, our structure (state 4) does not agree with the structure proposed by Bocharov et al. (31) in bicelles, since our predictions favor a left-handed dimer, as in older atomistic simulations (21–24). Nevertheless, we believe that the most critical aspect found is that the dimer at 20% cholesterol allows for significant rotational flexibility, as evidenced by the weaker (less well defined) interface relative to the one formed at the C-terminus. We also note the higher free-energy minimum relative to the single helix state (Fig. 2); experiments support that sequences with mutations that promote activity can actually present lower dimerization affinities (16,68). Therefore, rotational flexibility of the TM domain (and not solely association affinity) could be critical for signaling mechanisms, especially in the context of a conformational switch.

However, proximity of the TM domains is necessary, potentially affected by mutations, and exploited by designing inhibitors to reduce activity (3,4). Recently, Lu et al. performed mutagenesis studies with ErbB1 receptors to report that there is a loose linkage between the configuration of TM domains and ligand-induced activity (70). This is in agreement with our past study with ErbB1 TM domains that do not appear to promote a specific dimer (40). Our results herein suggest that cholesterol and GxxxG motifs can play a role in transition states formed that could affect the kinetics of the process without necessarily being part of a packing interface. Such a mechanism would explain observed changes in the dimerization of TM domains of ErbB1 upon mutation (16,71).

We conclude by referring the reader to the recent review by Schreiber et al. (72) on the subject of activated protein-protein association. Association of proteins is not only limited by diffusion of molecules but also highly dependent on rates of conformational change required to form a stable dimer; in fact, recent experiments support that helix-helix association in membranes is not diffusion-limited but rather controlled by the assembling of complexes along favorable interfaces (73). In this scheme, association of receptors in membranes could be largely affected by the requirement of rotation before the helices fall into the free-energy minimum favored by lipid-induced attraction (at low and high cholesterol content). The conformational limits of such a mechanism could be exacerbated by rearrangements of large extracellular domains, supporting the idea that kinetic (and not necessarily thermodynamically) inactive dimers could be key to the activation process (28,74). Although our knowledge on protein association mechanisms in solutions is rapidly increasing (72), there is a clear need for further research to understand how these processes are mediated in cell membranes.