Dynamically varying interactions between heregulin and ErbB proteins detected by single-molecule analysis in living cells

Abstract

Heregulin (HRG) belongs to the family of EGFs and activates the receptor proteins ErbB3 and ErbB4 in a variety of cell types to regulate cell fate. The interactions between HRG and ErbB3/B4 are important to the pathological mechanisms underlying schizophrenia and some cancers. Here, we observed the reaction kinetics between fluorescently labeled single HRG molecules and ErbB3/B4 on the surfaces of MCF-7 human breast cancer cells. The equilibrium association and the dissociation from equilibrium were also measured using single-molecule imaging techniques. The unitary association processes mirrored the EGF and ErbB1 interactions in HeLa cells [Teramura Y, et al. (2006) EMBO J 25:4215–4222], suggesting that the predimerization of the receptors, followed by intermediate formation (between the first and second ligand-binding events to a receptor dimer), accelerated the formation of doubly liganded signaling dimers of the receptor molecules. However, the dissociation analysis suggested that the first HRG dissociation from the doubly liganded dimer was rapid, but the second dissociation from the singly liganded dimer was slow. The dissociation rate constant from the liganded monomer was intermediate. The dynamic changes in the association and dissociation kinetics in relation to the dimerization of ErbB displayed negative cooperativity, which resulted in apparent low- and high-affinity sites of HRG association on the cell surface.

The ErbB family includes four member proteins, which localize to the plasma membrane of various types of cells and mediate signal transduction involved in cell differentiation and proliferation (1, 2). Among the ErbB family members, ErbB3 and ErbB4 are receptors for several extracellular protein ligands, including heregulin (HRG, also called neuregulin). ErbB3 is involved in the development of the heart and nervous system, and ErbB4 plays a role in lactation (3, 4). The dysregulation of ErbB3 and B4 (referred to hereafter as “HRG receptors”) is often related to malignancy in human cancers (2). The long-term exposure of MCF-7 cells, a cultured cell line derived from a human breast carcinoma, to HRG induces the production of differentiation markers, such as milk proteins or lipid droplets, whereas similar exposure to epidermal growth factor induces proliferation (5).

ErbB molecules must dimerize to transduce signals across the plasma membrane (3, 4, 6). Single ErbB molecules associate with single ligand molecules, such as HRG, and two liganded receptor molecules form a doubly liganded dimer on the plasma membrane (7). Dimerization is essential for the activation of ErbB molecules. This activation involves the phosphorylation of multiple sites along the cytoplasmic tail domain, which is catalyzed by the cytoplasmic kinase domain of the dimerization partner. In addition to homodimerization, some combinations of heterodimerization between members of the ErbB family have been described (1, 2). Several forms of unliganded and liganded homo- and heterodimers of HRG receptors have been reported (8–15). However, discrepancies among these studies have not yet been resolved, and the full model of ErbB heterodimerization remains unclear.

The kinetic processes involved in the association of HRG with its receptors and the dimerization of the receptors are also not yet fully clarified. Several studies have described the association and dissociation kinetics of HRG with receptors on the cell surface (16) or in vitro (11, 14), but how HRG association is kinetically related to receptor dimerization is unknown. Single-molecule imaging (17, 18) is a suitable technique for addressing this question because the observation of individual ligand molecules on the cell surface allows the simultaneous detection of the ligand-receptor interactions and the dimerization of the liganded receptor molecules. Using this technique, we studied the association between EGF and ErbB1 in HeLa cells (19). The results suggest that a small fraction of ErbB1 molecules form predimers, which rapidly associate with EGF in contrast to the slower rate of association observed for the monomers. The association of the first EGF molecule with the predimer increases the second EGF association rate constant, thereby accelerating the formation of doubly liganded dimers. Therefore, the association rate between EGF and ErbB1 changes dynamically, depending on the molecular interactions. It is unclear whether this is a common feature of the ErbB family proteins, and the change in the dissociation rate constant as a function of the formation of signaling dimers has not been characterized.

Biochemical analyses of the equilibrium association between ErbBs and their ligands suggested the presence of low- and high-affinity association sites, which produce a concave-up Scatchard plot (20, 21). This biphasic property has been observed for HRG associations at the cell surface, even when only ErbB3 was expressed (8, 16). In vitro measurements using purified monomeric extracellular receptor domains revealed only a low-affinity association constant (10, 11, 14). Therefore, the biphasic association could not be attributed to the presence of two types of HRG receptors (ErbB3 and B4), but rather to the functional divergence of each receptor species on the cell surface. To explain the concave-up shape of the Scatchard plot, negative cooperativity (22–24), a heterogeneous density of receptors (25), or the presence of external binding sites (26, 27) have been proposed. These hypotheses assume that the structural differences among the receptor molecules affect their affinity for the ligand. Although the first crystal structures of human ErbB receptor extracellular domains appeared to predict low- and high-affinity monomeric and dimeric states, respectively, this interpretation could not explain the observed concave-up Scatchard plots (7, 26, 27). A more recent asymmetric dimer structure of the Drosophila EGF receptor, however, has suggested a structural model for negative cooperativity that can explain these plots (24).

In this study, we observed the association between HRG and its receptors and the dimerization process of the liganded receptor molecules with single-molecule imaging. The reaction rate constants for the unitary HRG-receptor interaction steps were determined to provide unique information about the mechanism underlying the formation of signaling dimers and the various ligand association sites in the ErbB system.

Results

Synthesis and Biological Activities of Tetramethylrhodamine (TMR)–HRG.

Human heregulin1 β1 (HRG) containing an artificial amino acid conjugated to the fluorophore (TMR) at the N-terminus side (TMR–HRG) was prepared by in vitro protein synthesis. The biological activities of TMR–HRG were examined in MCF-7 cells, which express ErbB3 and B4. Unlabeled HRG and TMR–HRG induced the phosphorylation of ErbB3, B4, and ERK, a downstream effector of the ErbB system, to similar extents (Fig. S1).

Single-Molecule Detection of the Associations Between TMR–HRG and ErbBs.

Using an oblique illumination fluorescence microscope (19) (Fig. 1A and Movie S1), we observed the associations between TMR–HRG and its receptors on the apical surfaces of MCF-7 cells in culture as single molecules. All experiments were performed at 4 °C. Because TMR–HRG molecules in solution do not appear as distinct spots (because of their rapid Brownian movement), the sudden appearance of fluorescent spots on the cell surface represents the associations between TMR–HRG in the culture medium with ErbBs in the plasma membrane. The waiting times for the individual association events after the introduction of TMR–HRG to the medium were measured to analyze the association kinetics. Each association site on the plasma membrane showed either a one- or two-step increase in fluorescence intensity (Fig. 1B and Fig. S2). The latter should indicate association events to sites involving more than one ErbB molecule.

Fig. 1.

Single-molecule associations between TMR–HRG and ErbBs. (A) A snapshot from a single-molecule movie showing the association of TMR–HRG on the apical surface of a living MCF-7 cell. Bar: 10 μm. (B) Typical stepwise increases in the fluorescence intensity at the association sites. TMR–HRG was added to the medium at time 0. Stepwise increases in the fluorescence intensity (the appearance of fluorescent spots) indicated the association of single soluble TMR–HRG molecules with ErbB molecules on the plasma membrane. Each reaction site showed one- (Left) or two-step (Right) association. The waiting times for the first (τ₁, τ₂) and second (τ₃) associations were measured at individual association sites. The red lines indicate the fluorescence intensity changes detected statistically. (C–E) Cumulative frequency distributions of τ₁ (C), τ₂ (D), and τ₃ (E) in the presence of 2 nM (Upper) or 4 nM (Lower) TMR–HRG in solution. The lines indicate the fits with the reaction models. The number of observed events (n) and the multiple correlation coefficients between the data and the model fits (R) are indicated. The best-fit values for the reaction rate constants were k₁ = 0.00206(2 nM) and 0.00175(4 nM) nM^-1s^-1, k₂ = 0.00239(2 nM) and 0.00268(4 nM) nM^-1s^-1, k_i = 6.11 (2 nM) and 3.28 (4 nM) s^-1, d_i = 0.467 (2 nM) and 0.754 (4 nM)s^-1, and k₃ = 3.16 (2 nM) and 4.81 (4 nM) nM^-1 s¹. The average values weighted by the event numbers were k₁ = 0.00193 nM^-1 s^-1, k₂ = 0.00255 nM^-1 s^-1, k_i = 4.51 s^-1, d_i = 0.629 s^-1, and k₃ = 4.09 nM^-1 s^-1. (The value for k₂ corresponds to the dimeric unit, not to the single molecules in the dimer.)

The waiting time distributions for the association events (τ₁) at the one-step association sites can be described by a simple stochastic reaction, Here, r, R, and L represent the vacant receptor, the liganded receptor, and the concentration of HRG in the solution, respectively. In the general reaction kinetics of single-step ligand-receptor interactions, association and dissociation proceed in parallel, . Here, d₁ is the dissociation rate constant. In a single-molecule analysis based on the waiting time for each unitary reaction, the association and dissociation reactions are separable. For example, when the incremental steps are selectively analyzed (as in the case of Fig. 1B), the reaction equation contains only the association term. The value estimated for the association rate constant k₁ was 0.0019 nM^-1 s^-1 (Fig. 1C). This reaction can be attributed to the association between soluble HRG and either ErbB3 or B4 molecules on the plasma membrane. Because the distribution of τ₁ could be fitted with a single-component exponential function, it is plausible that the ErbB3 and B4 molecules yielded similar HRG association rate constants. These association sites could be monomers or dimers between active and inactive states (see Discussion and SI Text for details). In the following, we call the one-step sites as monomers for simplicity.

The distributions of τ₂, which is the waiting time for the first association event at the two-step association site, can also be described by a single-component exponential function, with a rate constant k₂ = 0.0026 nM^-1 s^-1 (Fig. 1D). It is possible that the reaction sites showing one- or two-step associations comprise a homogeneous population, at least at the stage of the first HRG molecule association event, because the estimated values of k₁ and k₂ are similar. However, the probability of observing a two-step association event (16% of the total association events observed) was extremely high, assuming a random distribution of receptors (SI Text, Figs. S3 and S4), suggesting the presence of predimers and/or preclusters of ErbB molecules before the association of the first HRG molecule. The presence of ErbB3 homodimers and/or oligomers has been reported (14, 28), supporting the latter possibility. The monomers and predimers of HRG receptors should be in equilibrium even in the absence of HRG.

The waiting time distributions for the second HRG association events (τ₃) at the two-step association sites cannot be described by a single-component exponential function (Fig. 1E and Fig. S2). Instead, a reaction scheme that includes an intermediate state is required: . For simplification, in this scheme the receptor precomplex is assumed to contain one vacant and one liganded receptor molecule (rR). As in the case in Fig. 1B, the dissociation of HRG from the R₂ state is separable from this scheme. The presence of a reaction intermediate (ρR) suggests structural changes in the vacant receptor after the association of the first HRG molecule. The estimated values for the rate constants are k_i = 4.5 s^-1, d_i = 0.63 s^-1, and k₃ = 4.1 nM^-1 s^-1. The association rate constant increased after the state transition (k₂ = 0.0026 < k₃ = 4.1 nM^-1 s^-1).

Single-Molecule Detection of the Dissociation of HRG from the ErbB Dimers.

Stepwise reductions in the fluorescence intensity at the TMR–HRG association sites were observed on the cell surface (Fig. 2 and Fig. S2). Each reduction may result from a dissociation or photobleaching event. The first steps in the two-step reductions yielded a lifetime (τ₄) distribution that can be described by a single exponential function with a decay rate of 2.7 s^-1. This value represents the sum of the dissociation (d₃) and photobleaching (k_b) rate constants. From single-molecule photobleaching events among TMR molecules fixed on coverslips, k_b was determined to be 0.44 s^-1. Therefore, the best-fit value for d₃ is 2.3 s^-1, suggesting the rapid dissociation of HRG from doubly liganded ErbB dimers. The rate constant for the second dissociation event (d₂) cannot be determined using this method because the lifetime of the second step of the two-step intensity reduction was similar to k_b.

Fig. 2.

Single-molecule dissociations of TMR–HRG from the ErbB dimers. (A) A typical builddown of the fluorescence intensity after the two-step association of TMR–HRG. The builddowns resulted from TMR–HRG dissociation or photobleaching events. The waiting times for the first builddown events (τ₄) were measured at individual association sites. The red line indicates the fluorescence intensity changes detected statistically. (B) Cumulative frequency distribution of τ₄. The line indicates the fit with the model of a single-step stochastic builddown; the decay rate is the sum of the dissociation (d₃) and photobleaching (k_b) rate constants. The number of observed events (n) and the multiple correlation coefficients between the data and the model fits (R) are indicated. The best-fit value for d₃ + k_b = 2.70 s^-1; k_b for TMR was 0.44 s^-1 under the same observation conditions. Therefore, the best-fit value for d₃ = 2.26 s^-1. (The value for d₃ corresponds to the dimeric unit, not to the single molecules in the dimer.)

Ensemble HRG and ErbB Interaction Kinetics.

The affinities between HRG and its receptors were examined by counting the numbers of TMR–HRG molecules that associated with the cell surface under equilibrium conditions in the presence of various concentrations of TMR–HRG in the extracellular medium (Fig. 3). The equilibrium association between HRG and the ErbBs was biphasic (Fig. 3B). The association curve was fit with a phenomenological association equilibrium model (SI Text). The density of HRG receptors on the cell surface was determined to be 1.70 μm^-2.

Fig. 3.

Ensemble molecule interactions between TMR–HRG and ErbBs. (A) At the indicated times during the incubation of the cells with TMR–HRG, the densities and fluorescence intensities of the TMR–HRG spots on the cell surface were measured, and the total number of molecules associated per single cell was calculated based on the single-molecule fluorescence intensity and the average cell surface area. Typical results in the presence of 1 and 2 nM TMR–HRG are shown. (B) The cells were incubated with the indicated concentrations of TMR–HRG for 60 min. The numbers of TMR–HRG molecules associated with the surfaces of single cells were plotted as a function of the TMR–HRG concentration. The averages values over 16–55 cells are shown with the standard deviations. The line indicates the fit with a two-component association curve (SI Text). The best-fit parameters for the dissociation constants and the number of reaction sites were K₁ = 0.330 nM (2,860 cell^-1) and K₂ = 0.0107 nM (429 cell^-1); R = 0.992. (C) Cells equilibrated with 1 nM TMR–HRG were washed twice with ice-cold medium without TMR–HRG. The numbers of TMR–HRG molecules remaining on the cell surfaces were then measured over time. The averages of the measurements collected for five cells are shown with the SEMs. The red line indicates the result of fitting to a two-component exponential function (R = 0.973). The best-fit values for the dissociation rate constants were d_a = 0.0109 s^-1 and d_b = 3.68·10^-5 s^-1. The blue line indicates the result of fitting with a three-component exponential function in which one of the dissociation rate constants was fixed at 2.3 s^-1(R = 0.972). The best-fit values for the dissociation rate constants were d_a = 7.00·10^-3 s^-1 and d_b = 3.95·10^-5 s^-1.

Cells equilibrated in 1 nM TMR–HRG (Fig. 3A) were washed under a microscope, and the numbers of TMR–HRG molecules remaining on the cell surfaces were measured over time (Fig. 3C). No endocytosis of HRG molecules was observed at 4 °C. The rebinding of the dissociated molecules was negligible because the concentration of TMR–HRG would have been less than 0.1 pM, even if all the molecules were dissociated. The dissociation curve can be explained using a two-component exponential function (red line in Fig. 3C) with dissociation rate constants of 0.011 and 3.7·10^-5 s^-1. The dissociation rate constant d₃, determined from the single-molecule measurements, was too large to be detected in this experiment. When the dissociation curve was fitted with a three-component exponential function in which the largest dissociation rate constant was fixed at d₃ = 2.3 s^-1 (blue line in Fig. 3C), the best-fit values for the other two rate constants were d_a = 7.0·10^-3 s^-1 and d_b = 4.0·10^-5 s^-1.

A Simple Model of the Equilibrium Interaction Between HRG and ErbBs.

The kinetic analyses described above yielded several rate constants associated with the interactions between HRG and its receptors: k₁ - k₃ for their association, and d₃, d_a, and d_b for their dissociation. We also determined the rate constants for the formation and deconstruction of a reaction intermediate (k_i and d_i). Assuming that both monomers and (apparent) dimers of unliganded receptors were present, the following reaction schema describing the interactions between HRG and its receptors can be proposed:

This schema does not consider any differences between the ErbB3 and B4 proteins because most of the elementary reaction steps analyzed above (k₁ - k₃, k_i, d₃, and d_i) can be described by single-component functions. Heterodimerization with other ErbBs was not considered because we do not have sufficient information for reliable modeling. The constants d₁ and d₂ could be d_a and d_b, although the assignment of these two parameters is unknown.

A simple reaction network describing the equilibrium associations between HRG and its receptors was constructed involving the schema shown above (Fig. 4A). The reversible formation of EGF receptor dimers has been observed experimentally (29, 30). This type of reaction network yields biphasic association curves as a function of the ligand concentration (L) with a concave-up Scatchard plot under the appropriate conditions (31). The single-molecule experiments and the experimental HRG association curve provided the equilibrium constants K₁ - K₆. As determined from the single-molecule experiments, K₃( = d₃/k₃ = 0.55 nM) and K_i( = d_i/k_i = 0.14); K₄( = 9.0 μm^-2) was calculated from the fractional ratio (α = 0.84) of the monomeric association events, the total density of the association sites (T = 1.7 μm^-2), and the association rate constants k₁ and k₂ (SI Text); and K₁ = d₁/k₁ = 3.6 nM and K₂ = d₂/k₂ = 0.016 nM, where (d₁,d₂) = (d_a = 7.0·10^-3 s^-1,d_b = 4.0·10^-5 s^-1). Two other equilibrium constants (K₅ and K₆) were determined from the others by assuming detailed balances in the reaction network (K₅ = 2K₂K₄/K₁, K₆ = K₃K₅K_i/K₁/2). The initial densities [r₍₀₎ and r₂₍₀₎] were calculated from α, k₁, k₂, and T (SI Text). Thus, all the reaction parameters in this model were determined experimentally, except for the combination of d₁ and d₂.

Fig. 4.

A model of the equilibrium association between HRG and ErbBs. (A) The simplest reaction network that includes information from the single-molecule kinetic analysis. See the text for details. (B) Association equilibrium curves (solid lines) calculated using the parameters determined from the experiments. The circles indicate the experimental results (Fig. 3B); R = 0.969. The black, red, and blue lines indicate the total, monomeric, and dimeric associations of HRG, respectively. (The dimeric association is indicated as the density of sites, not molecules.) After optimizing the parameters by assuming maximum errors of 50% (50%·0.5–1.5) in the experiments, the fit results improved slightly (R = 0.980, dotted lines). The parameter values before and after optimization are listed in Tables S1 and S2. (C) Scatchard plots of the HRG association calculated using the parameters given in B. The goodness of fit to the experimental results (circles) increased after optimization (dotted lines).

The association curve was simulated (Fig. 4B and SI Text). The assignment of d₁ and d₂ as 7.0·10^-3 and 4.0·10^-5 s^-1, respectively, yielded a good fit to the association curve (Fig. 4B). Although the fit to the Scatchard plot displayed a systematic error at lower numbers of bound molecules (Fig. 4C), the concave-up shape of the plot was reproduced. Another combination of d₁ and d₂ did not fit the association curve at all. The reaction parameters were optimized by assuming errors of < 50% in the single-molecule experiments (dotted lines in Fig. 4 B and C). The model’s goodness of fit relative to the Scatchard plot improved significantly after optimization (Fig. 4C). The best-fit parameter values before and after optimization are listed in Tables S1 and S2.

Discussion

In this study, we observed the associations between single fluorescently labeled HRG molecules and the ErbB receptors on the surfaces of living MCF-7 cells. The measurement results supported kinetic models in which three different forms of HRG association sites were assumed: monomers, predimers, and singly liganded dimers of ErbB molecules. These three forms were kinetically distinguishable in their association with HRG. If receptor molecules are distributed randomly without predimerization, 1% or less of the association sites should show two-step association in our experimental condition. This probability is far smaller than the fraction of two-step association sites (16%, Table S1), suggesting some types of receptor aggregation preformed on the cell surface. Direct predimerization, at least among ErbB3 molecules, has been reported previously (11, 14). Predimerization has also been detected for ErbB1 (32). Therefore, we assumed that the two-step association sites were predimers of HRG receptors in the kinetic models. No association events of more than two steps were observed in our experiments. On the contrary, it is possible that the one-step association sites also formed predimers. In that case, because association to the second sites was not observed before photobleach of the first HRG molecule (> 2 s) in spite of the 0.3-s half-time of second association at the two-step sites (Fig. 1E), the one-step sites should be predimers between active and inactive states of receptor molecules. The equilibrium model shown in Fig. 4A is applicable even in this case, assuming that transitions occur between the active and inactive states (SI Text presents a discussion of monomers and predimers of HRG receptors).

The HRG association rate constants were small for monomers and predimers of receptors during the first association, but increased significantly for the singly liganded dimers (Fig. 1). In this sense, the HRG receptors behaved similarly to ErbB1 (19). The association rate constant for the interaction between the ligand and the receptor monomers (k₁) was 0.004 (EGF) or 0.002(HRG) nM^-1 s^-1. For the singly liganded dimers (k₃), the association rate constants increased to 2 (EGF) and 4(HRG) nM^-1 s^-1. Between the first and second ligand association events, a kinetic intermediate was detected for both the ErbB1 and HRG receptors, with similar time constants. Large association rate constants (approximately k₃) were not observed during the first association, suggesting that the intermediate form (ρ) was unstable as a monomer and was stabilized in a dimer by the adjoining receptor molecule in the R state. The ligand association kinetics of the ErbB1 and HRG receptors differed in the values of the association rate constants of the predimers (k₂): 0.4 nM^-1 s^-1 for EGF and 0.003 nM^-1s^-1 for HRG. We consider that the difference between these rate constants was not caused by differences in the receptor properties but by differences in the experimental temperature. Previous EGF measurements (19) were made at 25 °C, whereas the present HRG measurements were made at 4 °C. At 25 °C, we observed an association rate constant of 0.1–0.2 nM^-1 s^-1 for the first HRG association. The results of a single-molecule kinetic study performed at 25 °C are shown in Figs. S5 and S6. The HRG association rate constant for the extracellular domain of ErbB3 in solution was reported to be 0.004 nM^-1 s^-1 (11), which was the rate constant for the low-affinity site that was probably monomeric ErbB3. These results suggest the presence of at least two ErbB predimer conformations (i.e., fast- and slow-association states). In a possibly related phenomenon, the fraction of predimers among the ErbB1 molecules on HeLa cells was smaller (2% at 25 °C) (19) than the fraction of HRG receptor predimers on MCF-7 cells (13% at 4 °C), even though the density of the receptor molecules was much higher for ErbB1 on HeLa cells (29,000 cell^-1) (19). The conformational changes in the ErbB dimers after the association of the first ligand, which increased the association rate constant of the second ligand, appear to be a common feature of the ErbB family. This feature increases the probability of forming doubly liganded ErbB dimers, thereby facilitating cell signaling (33). MCF-7 induces prolonged or transient signals in response to HRG or EGF, respectively (5, 34), and these distinct signal kinetics ultimately lead to different cell fates. This study indicates that the kinetic schema and parameters that describe ligand binding to the EGF and HRG receptors are similar, suggesting that their distinct biochemical characteristics may not originate from the ligand-receptor association itself.

Three HRG dissociation rate constants were observed, each on a different time scale (Figs. 2 and 3). Dissociation from the doubly liganded dimer (d₃) was the fastest dissociation process, and our kinetic analysis suggests that the dissociation from the singly liganded dimer (d₂) was the slowest process. A dissociation rate constant of 0.007 s^-1 has been reported for the extracellular domain of ErbB3 (11). This value is equal to that observed in our experiments (d₁ = 0.0070 s^-1), which was attributed to the dissociation of HRG from the monomeric association site. In combination with the corresponding association rate constants, the vacant monomer (K₁ = 3.6 nM), the predimer (K₂ = 0.015 nM), and the singly liganded dimer (K₃ = 0.55 nM) are suggested to form low-, high-, and intermediate-affinity association sites, respectively. In other words, our kinetic analysis suggests that two sequential changes in the state or conformation of the HRG receptors dynamically affect the ligand-receptor interaction: (i) Predimerization increases the ligand affinity, and (ii) the intermediate formation induced by the first HRG association increases the reaction rate constants for both the association and dissociation of the second HRG molecule. The juxtamembrane domain of the HRG receptor may be responsible for these affinity changes, as suggested for ErbB1 (35). Asymmetric dimerization caused by ligand binding has been suggested to induce low-affinity association sites in the Drosophila EGF receptor (24).

The ligand association with the HRG receptors showed a biphasic equilibrium that resulted in a concave-up Scatchard plot (Fig. 4C), as observed previously for the HRG receptors (8, 16) and ErbB1 (20, 21). The reaction model shown in Fig. 4 predicts the emergence of differential HRG association via the molecular interactions between ErbBs and HRG. This model is essentially identical to the reaction network described by Wofsy et al. for the EGF/ErbB1 system (31). In that network, negative cooperativity in ligand binding (higher ligand affinity for the predimer than for the singly liganded dimer; K₂ < K₃) is a necessary (but not sufficient) condition for producing a concave-up Scatchard plot. Another type of model proposes a heterogeneity in the receptor density (25), such as the surface aggregation of the receptors (36) or external binding sites on the receptor dimers (26, 27), to explain the concave-up shape of the Scatchard plot. A recent biochemical study in which the cell surface density of ErbB1 was varied suggested that K₂ < K₃, consistent with the negative cooperativity model (21). In the present study, we directly determined the association and dissociation rate constants for the interaction between HRG and its receptors and found negative cooperativity in the context of the HRG association. A heterogeneous receptor density, as would result from the microclustering of the HRG receptors, may be involved as an additional mechanism, but the inclusion of heterogeneity in the model is unnecessary.

Heterodimerization occurs among the ErbBs. In our measurements, the two types of HRG receptor (ErbB3 and B4) were indistinguishable because the unitary reaction steps could be described using single-component models (Figs. 2 and 3). The equilibrium dissociation constants previously reported for the low-affinity forms of ErbB3 and B4 are similar (10, 37). However, it has also been reported that liganded ErbB3 does not form homodimers (10, 13). ErbB3 homo-predimers (11, 14), ErbB4 homo-liganded dimers (10, 13), and ErbB3/B4 hetero-liganded dimers (9) have been detected. ErbB2 has also been shown to form heterodimers with ErbB3 and B4 (2). Therefore, the model illustrated in Fig. 4A requires some modifications to describe the real reaction network in cells. However, the network in Fig. 4A is essential to any modified model that emerges diverged associations. Additional information is required to construct models that include heterodimerization.

As shown here, the HRG-ErbB system is considered to change both the association and dissociation kinetics dynamically as a function of the ligand-receptor and receptor-receptor interactions. These dynamics permit the ErbB system to detect rapidly and sensitively HRG signals in the extracellular medium and to allow sites with varying affinities for HRG to emerge.

Materials and Methods

Synthesis of TMR–HRG.

The NRG1/NRG-β1A (HRG) gene (number GC08P030042) was a gift from Genentech Inc. The synthesis and purification of carboxytetramethylrhodamine-labeled HRG (TMR–HRG) using the “Pin-point” in vitro protein synthesis system was performed by ProteinExpress. In brief, the N terminus of HRG was extended by 13 amino acid residues, including a nonnatural amino acid conjugated to TMR at position nine relative to the N terminus. The C terminus of HRG was also extended with a hexameric His-tag for affinity purification. The product was purified on a TALON metal affinity column and analyzed by SDS- and native-PAGE. The dye/protein ratio of TMR–HRG exceeded 0.99.

Cell Culture.

A human breast carcinoma cell, MCF-7, was obtained from the American Type Culture Collection and was maintained in DMEM supplemented with 10% FBS at 37 °C under 5% CO2.

Single-Molecule Imaging.

Cells were cultured on glass coverslips. Overnight before the experiments, the culture medium was substituted with DMEM without FBS or phenol red but supplemented with 1% BSA. Immediately before the experimental observations, the culture medium was substituted with the medium containing 5 mM Pipes (pH 7.2). The coverslip was mounted on a metal culture chamber (Invitrogen) and the cells were observed with an oblique illumination microscope based on a Nikon TE2000 inverted fluorescence microscope. The specimen and the objective were cooled to 4 °C with a microscope incubator (Peltier-4; Taisei Denki). The cells were illuminated with the 532-nm wavelength of a solid-state green laser (Compass 315M; Coherent) through the objective (PlanApo 60×, NA 1.49; NIKON) and were observed using the same objective. On the microscope, 600 μL of a 0.001–4 nM TMR–HRG solution was added to 300 μL of the cell-incubation medium. The solutions were mixed well within 0.5 s. Images of single TMR–HRG molecules on the cell surfaces were acquired using an EM-CCD camera (ImagEM; Hamamatsu) and were analyzed using custom-made software.

Kinetic Analyses.

The experimental data were fitted with the reaction models using KaleidaGraph (Synergy Software) and MATLAB and Simulink (The MathWorks). The fluorescence intensity changes (Figs. 1B and 2A) were detected statistically using a hidden Markov model, analyzed with the variational Bayes method (38).