The extracellular domain of fibroblast growth factor receptor 3 inhibits ligand-independent dimerization

Abstract

Dysregulation of ligand-independent receptor tyrosine kinase (RTK) dimerization, which is the first step in RTK activation, leads to pathologies. A mechanistic understanding of the dimerization process is lacking, and this lack of basic knowledge is one bottleneck in developing effective RTK-targeted therapies. For instance, the roles and the relative contributions of the different RTK domains to RTK dimerization are unknown. Here we use quantitative imaging Förster resonance energy transfer (QI-FRET) to determine the contribution of the extracellular (EC) domain of fibroblast growth factor receptor 3 (FGFR3) to FGFR3 dimerization. We provide the first direct experimental evidence that the contribution of FGFR3 EC domain to dimerization is repulsive in the absence of ligand, and on the order of 1 kcal/mole. The magnitude of this repulsive contribution is similar to the dimer over-stabilization that can occur due to pathogenic single amino acid mutations, and therefore significant for biological function.

INTRODUCTION

RTKs are the second largest family of membrane receptors, after the G protein-coupled receptor family. They are single-pass transmembrane (TM) proteins, with an N-terminal extracellular (EC) domain usually composed of several motifs and involved in ligand binding, a single TM domain followed by a juxtamembrane region, and an intracellular tyrosine kinase domain ^1-3. RTK activation is a two-step process involving (1) lateral dimerization in the membrane plane and (2) ligand binding, which is believed to stabilize the RTK dimers and likely alter their structure and activity. RTK monomers are inactive, while RTK dimers are active because the interactions between the two catalytic domains in the dimer stimulate catalytic activity ^4-6. Thus, RTK dimerization regulates RTK activity.

Ligand-induced dimerization and activation has been long recognized as critical in signaling by RTKs. In many cases, however, ligands are not necessary for RTK dimerization and activation, and stable dimers have been shown to exist in cellular membranes in the absence of ligand ^7;8. Furthermore, dysregulation of ligand-independent dimerization, due to RTK overexpression or due to RTK mutations, had been linked to human pathologies ^9-13. For example, mutations in FGFR3, an RTK which is critical for skeletal development ^14-16, are believed to cause skeletal dysplasias and cancer by enhancing ligand-independent dimerization ^16;17. Yet, a mechanistic understanding of ligand-independent FGFR3 dimerization is lacking.

The extracellular (EC) domain of FGFR3, which consists of three Immunoglobulin-like (Ig-like) domains (D1-D3), has been crystallized in the presence of the ligand fgf1 ¹⁸. The structure has revealed that D2 and D3 mediate ligand binding, while D1 inhibits it. The behavior of FGFR3 EC domain in the process of ligand-independent dimerization, however, is unknown.

While dimerization studies have been carried out for isolated RTK transmembrane (TM) domains, and FGFR3 TM domain in particular has been shown to possess a propensity for sequence-specific dimerization ^19-22, the dimerization energetics of RTK EC domains has not been characterized thus far due to experimental limitations. Tanner and Kyte have argued that the role of ErbB1 extracellular domain is inhibitory in the absence of ligand ²³. However, their experiments have been scrutinized, and their conclusions questioned ²⁴. The effect of the reduced dimensionality on interactions in membranes is believed to be large ²⁵, but is unknown. Thus, measurements of interactions between isolated soluble extracellular domains in 3D aqueous solutions^26;27, and measurements of interactions between isolated TM domains in 2D membranes^28-31 cannot be compared.

Recently, we introduced a methodology that yields dimerization free energies of proteins in mammalian membranes ³². This method allows us to measure membrane protein interactions in two dimensions. The measurements utilize the quantitative imaging FRET (QI-FRET) method which yields the FRET efficiency E, as well as the concentration of donors and acceptors, C_D and C_A (the three parameters needed to calculate association constants) ³³. Experiments are carried out in single plasma membrane-derived vesicles, which bud off from cells upon treatments that disrupt the cytoskeleton ^34;35. Their membranes exhibit a homogeneous distribution of the fluorescently-tagged proteins ³⁶, which allows us to compare the fluorescence intensity of the vesicles with that of purified fluorescent protein solutions of known concentration, and implement the QI-FRET method ³².

Here we use QI-FRET to determine the contribution of FGFR3 EC domain to FGFR3 dimerization in plasma membrane vesicles derived from CHO cells. We measure this contribution by comparing the dimerization of two FGFR3 constructs, with and without the three Ig-like motifs of FGFR3 EC domain, in plasma membrane-derived vesicles. We provide the first direct experimental demonstration that FGFR3 EC domain inhibits ligand-independent dimerization.

RESULTS

The two gene constructs used in this study are shown in Figure 1, and the design of the plasmids is described in Supplementary Material. The two constructs used were (i) a EC+TM construct, consisting of the signal peptide, the EC domain and TM domain of FGFR3, a 15 amino acid flexible linker (GGS)₅ ³⁷, and the fluorescent protein, and (ii) a TM construct lacking the three immunoglobulin-like motifs, consisting of the signal peptide, a short 24 amino acid C-terminal segment from the EC domain, the TM domain of FGFR3, a 15 amino acid flexible linker (GGS)₅, and the fluorescent protein. The attachment of the fluorescent proteins via the 15 amino acid flexible linkers has been used in a previous study ³², and thus the distance between the fluorescent proteins in the dimers is known, ~ 48.5 Å.

Figure 1.

Overview of measurements of dimerization free energy in plasma membrane-derived vesicles. Two different constructs were used (i) A EC+TM construct consisting of the signal peptide, the EC domain, the TM domain, a 15 amino acid-long (GGS)₅ flexible linker, and the fluorescent proteins and (ii) a TM construct lacking the three Ig-like motifs of FGFR3 extracellular domain, consisting of the signal peptide, a 24 amino acid segment of the EC domain, the TM domain, a 15 amino acid (GGS)₅ flexible linker, and the fluorescent proteins. We determine the contribution of the EC domain to FGFR3 dimerization by measuring the difference in dimerization free energy for these two constructs.

The homodimerization of these two constructs was assessed in the following two sets of experiments. In the first set of experiments, EC+TM_EYFP and EC+TM_mCherry were co-transfected into CHO cells. In the second set of experiments, TM_EYFP and TM_mCherry were co-transfected into CHO cells. Twenty-four hours after transfection, vesiculation was carried out as described in Li et al ³⁶. Vesicles loaded with EC+TM_EYFP/EC+TM_mCherry and TM_EYFP/TM_mCherry were collected and imaged in a Nikon confocal microscope as described ³². Briefly, the intensity of the donor emission was collected from 500 to 530 nm in the donor scan by exciting with a 488 nm laser. The sensitized emission was obtained from the FRET scan, which used the 488 nm laser to excite EYFP, and recorded the emission of mCherry in the interval 565 – 615 nm. The acceptor scan utilized the 543 nm laser to excite mCherry and collected photons with wavelengths >650 nm to yield the emission intensity of the acceptor mCherry.

Typical images are seen in Figure 2. The fluorescence is uniform, indicative of homogeneous protein distribution, and is concentrated on the membrane, as expected for membrane proteins. The homogeneous protein distribution in the vesicles is expected, since the cytoskeleton, which plays a role in sustaining lateral membrane heterogeneity, is not present in the vesicles. As discussed previously, the uniform fluorescence allows the calibration of the donor and acceptor concentrations in the vesicles, and is critical for the successful implementation of the QI-FRET method ^32;36.

Figure 2.

One vesicle loaded with FGFR3 TM_EYFP and TM_mCherry. Intensities per unit membrane area were obtained by integrating the Gaussian intensity profiles across the membrane after correcting for the background.

The images were processed with a Matlab program written in the lab, which automatically recognized the vesicles and integrated the membrane intensity ³². The signal that originates from the membrane was fitted to a Gaussian function (solid lines in Figure 2), while the background fluorescence inside the vesicle was modeled by an error function (dashed lines in Figure 2). The integration of the Gaussian yielded the three intensities, $I_D^m{I}_{FRET}^m{I}_A^m$, per unit membrane area ³². The concentrations of donor-labeled and acceptor-labeled RTK constructs were determined from these intensities as described in detail in Chen et al ³². Briefly, the concentration of acceptor-labeled proteins per unit membrane area (normal to the focal plane) was calculated for each vesicle as:

$$C_A^m=\frac{I_A^m}{i_A},$$ (1)

where i_A is a calibration constant obtained by imaging purified mCherry solution of known concentration using the acceptor scan ³⁶. On the other hand, the concentration of donor-labeled proteins per unit membrane area ($C_D^m$) was determined as ³²:

$$C_D^m=\frac{I_{D,corr}^m}{i_D}=\frac{I_D^m+G_F(I_{FRET}^m-{\beta }_D{I}_D^m-{\beta }_A{I}_A^m)}{i_D}$$ (2)

Next, the FRET efficiency (E) was calculated for each vesicle according to ³⁶:

$$E=1-\frac{I_D^m}{I_D^m+G_F(I_{FRET}^m-{\beta }_D{I}_D^m-{\beta }_A{I}_A^m)}$$ (3)

In these equations, β_D and β_A are the bleedthrough coefficients, determined by imaging vesicles expressing only donor or only acceptor ³⁶. The Gauge factor, G_F, which correlates the sensitized acceptor emission with donor quenching, was determined by analyzing vesicles expressing a construct of the donor fused to the acceptor ³⁶.

The FRET efficiency E is plotted as a function of the acceptor concentration in the vesicles in Figure 3 for both the EC+TM and TM constructs. Each data point represents a single vesicle. The acceptor concentration in Figure 3 varies over more than an order of magnitude, yielding a functional dependence of the measured FRET efficiency as a function of the concentration. The FRET efficiency increases with the acceptor concentration, as expected for dimerization ^20;38;39. The data scatter can be explained by the random noise associated with imaging ³², and is reducible by acquiring a large number of data points (the approach taken in single-molecule experiments). Despite data scatter, we see that the FRET efficiency measured for the TM construct is higher than the EC+TM efficiency, demonstrating that the EC domain inhibits dimerization.

Figure 3.

FRET data and proximity contribution for FGFR3 EC+TM and TM in CHO plasma membrane-derived vesicles. Each data point represents a single vesicle, for which E, C_A and C_D are determined using the QI-FRET method. Filled squares: FRET efficiencies measured for the FGFR3 EC+TM construct. Hollow circles: FRET efficiencies for the FGFR3 TM construct. QI-FRET data analysis has demonstrated that the scatter is largely due to random noise in image acquisition, and is reducible by collecting a large number of data points ³².

The FRET efficiencies E given by equation (3) contain two contributions: E_D, the FRET efficiency due to sequence-specific dimerization, and E_proximity, the FRET efficiency due to random co-localization of donor and acceptor ^28;38. The proximity contribution, arising from random co-localization of donors and acceptors has been discussed in detail previously ^19;40;41, and is plotted in Figure 3 with the solid line. Next, it is subtracted from the measured FRET to obtain FRET due to sequence-specific dimerization, E_D:

$$E_D=E-E_{proximity}$$ (4)

The dimeric fraction f in each vesicle is determined from the corrected FRET efficiency E_D as ⁴²:

$$f=\frac{E_D}{x_A\tilde{E}}$$ (5)

where x_A is the acceptor fraction, a known parameter since the donor and acceptor concentrations are determined with the QI-FRET method. The parameter Ẽ is the FRET efficiency in a dimer with a donor and an acceptor ³⁶. Ẽ depends on the distance between the donor and the acceptor in the dimer, i.e. on the structure of the dimer. In both constructs, the fluorescent proteins are attached to the TM domains via a 15 aa flexible linker ⁴³, the same attachment used in previous studies of GpA dimerization ³². Ẽ for this attachment was determined as Ẽ = 0.63 ± 0.04, corresponding to 48.5 Å separation distance between the two fluorophores in the dimer. With x_A and Ẽ known, the dimeric fractions in each vesicle can be calculated from equation (5).

Next, the dimeric fraction 2[D]/[T] was calculated as a function of the total concentration [T] for various values of K_D, based on the following dimerization model :

$$M+M\stackrel{K_D}{\leftrightarrow }D,$$ (6)

where the dimerization constant is:

$$K_D=\left[D\right]∕{\left[M\right]}^2,$$ (7)

and the total concentration is given by:

$$\left[T\right]=\left[M\right]+2\left[D\right].$$ (8)

We fitted the calculated prediction to the experimentally measured single-vesicle dimeric fractions, while optimizing for the dimerization constant K_D. The optimal values for the two constructs determined in the fit were K_D(EC+TM) = 227 ± 45 nm² and K_D(TM) = 1170 ± 260 nm² (see curves in Figure 4). These optimal K_D values were independent of the initial guesses used in the fit.

Figure 4.

Filled squares: Averaged dimeric fractions measured for the FGFR3 EC+TM construct. Hollow circles: averaged dimeric fractions measured for the FGFR3 TM construct. The data were fitted to the dimerization model given by equations (6)-(8), yielding the dimerization constants K_D. The contribution of FGFR3 extracellular domain to dimerization is inhibitory, ΔΔG = 0.9 ± 0.2 kcal/mole.

Since the probability of having exactly the same total RTK concentration in several vesicles is very low, the average dimeric fractions and the respective standard errors were calculated by averaging data with similar concentrations, within bins of size 0.0005 molecules/nm² that had 3 to 20 data points. The average values of the calculated dimeric fractions are shown in Figure 4, as open circles for the TM construct and filled squares for the EC+TM construct. Note that the curves in Figure 4 were obtained by fitting all the single-vesicle data points, and are shown together with the experimental averages to facilitate comparison between the fits and the data. We see that the averaged data and the fits are in agreement. We also see that the dimeric fraction is higher for the TM construct, indicating that the propensity for dimerization is reduced in the presence of the EC domain.

As discussed previously ³², the units of the dimerization constants are area (units of dissociation constants are 1/area), because the lateral interactions between the RTK constructs are interactions in two dimensions. Defining the standard state as K_D = 1 nm², we obtained a free energy of dimerization ΔG(TM) = -RTlnK_D(TM) = -4.2 ± 0.2 kcal/mole and ΔG(EC+TM) = -3.3 ± 0.1 kcal/mole. Thus, the contribution of FGFR3 extracellular domain to dimerization is positive, ΔΔG = 0.9 ± 0.2 kcal/mole, and therefore inhibitory.

DISCUSSION

The role of the TM domains in the RTK dimerization process has been controversial. For many years, they were believed to be passive anchors. These views were based on experiments in cells which showed that alterations in TM domains had negligible effects on signaling ^44;45. In other cases, however, changes in the sequences of the TM domains affected signaling ^46;47. Furthermore, pathogenic mutations in TM domains were discovered and proposed to promote ligand-independent dimerization ^48;49. Studies of the isolated TM domains in lipid bilayers or bacterial membranes invariably showed that these domains can dimerize by themselves ^{19;29;31;50-52}, and they were proposed to be a major driving force for RTK dimerization in the absence of ligand ^22;53. However, a formal proof for such a role has been missing due to experimental challenges in measuring dimerization energetics of RTK constructs in mammalian membranes.

In the absence of ligand, RTK extracellular domains are believed to explore two different conformations, a dimerization-competent conformation with the dimerization interface exposed, and a dimerization-incompetent conformation with the dimerization interface hidden ^54-56. The crystal structures of RTKs in the absence of ligand are locked in the inactive conformation ^57-59. In cellular membranes, the time spent in these two conformations determines the propensity for dimerization of the extracellular domain. Based on the crystal structures and biochemical evidence, Ferguson and colleagues have argued that an RTK EC domain (other than ErbB2) spends most of the time in the dimerization-incompetent conformation⁵⁴. A question remains, however, whether the dimerization-incompetent EC conformation is repulsive, attractive, or neutral in the dimerization process.

Here we present the first direct experimental evidence that the contribution of FGFR3 EC domain to dimerization is repulsive in the absence of ligand. We accomplish this by measuring directly the dimerization energetics of two FGFR3 constructs, with and without the three Ig-like motifs of the EC domain, using a new methodology.

The dimerization of the isolated FGFR3 TM domain (residues 367 to 399) has been characterized previously in lipid bilayers ^19;38. Its free energy of dimerization is about -3 kcal/mole, and can now be compared to the value reported here for the TM construct, ~ -4 kcal/mole. We see that the TM construct used here dimerizes more strongly than the shorter TM domain studied in lipid bilayers. There may be different reasons for the observed difference. First, the native membrane environment is complex and crowded with membrane proteins, unlike the model lipid bilayer, and crowding is known to have a strong effect on macromolecular interactions ⁶⁰. Second, the short EC sequence (residues 343 to 366) that is a part of the TM construct also may be contributing to FGFR3 dimerization.

Studies of the effect of pathogenic mutations on FGFR3 TM domain dimerization have provided new insights into the mechanism behind human pathologies ^22;61. One mutation that has been studied using this approach is the Ala391→Glu mutation, associated with Crouzon syndrome with acanthosis nigricans, an autosomal dominant disorder characterized by premature ossification of the skull (craniosynostosis), as well as hyperpigmentation and hyperkeratosis of the skin ⁶². The change in the free energy of dimerization of the isolated FGFR3 TM domain in lipid bilayers due to this mutation is -1.3 kcal/mole ¹⁹. Thus, the Ala391→Glu mutation over-stabilizes FGFR3 TM domain dimers¹⁹. Since dimerization and activation are tightly coupled, it can be expected that dimer over-stabilization leads to FGFR3 overactivation and the observed pathology.

Interestingly, the magnitude of the EC contribution measured here, about 1 kcal/mole, is similar in magnitude and opposite in sign to the effect of the Ala391→Glu mutations on dimerization (i.e. it is repulsive instead of stabilizing). It therefore appears that changes in dimerization propensities on the order of 1 kcal/mole can have a significant effect on RTK signaling. Therefore, the repulsive contribution of the EC domain in the absence of ligand is likely critical for maintaining signaling at normal levels.

MATERIALS AND METHODS

Plasmid constructs

pRSET_mCherry was obtained from the laboratory of Roger Tsien (University of California, San Diego), and pEYFP was obtained from the laboratory of Michael Betenbaugh (Johns Hopkins University, Baltimore). PcDNA3.1_FGFR3 was a kind gift from Dr. Daniel Donoghue from the University of California, San Diego. All plasmids used for mammalian expression were engineered using the pcDNA3.1(+) vector (Invitrogen). All primers were purchased from Invitrogen. For a detailed description of plasmid design, see Supplemental Material.

The FGFR3 TM construct consisted of the FGFR3 signal peptide (MGAPACALALCVAVAIVAGA), a 24 amino acid-long C-terminal sequence of the EC domain, the TM domain (DEAGSVYAGILSYGVGFFLFILVVAAVTLCRLR), a flexible 15 amino acid-long linker (GGS)₅ and fluorescent proteins at the C terminus. The gene was constructed between the multicloning sites HindIII and XbaI of pcDNA3.1(+), as described in Supplemental Material.

FGFR3 EC+TM construct contained the signal peptide, the extracellular and transmembrane domains of FGFR3, a linker and fluorescent proteins at the C terminus. The linker was same as the linker used in the FGFR3 TM construct.

CHO cell growth, transfection and vesiculation

Chinese hamster ovary (CHO) cells were grown in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1 mM non-essential amino acids, 1.8 g/L D-glucose, and 1.5 g/L sodium bicarbonate. 2×10⁴ cells/well were seeded in a 6-well plate one day before transfection.

Transfection was carried out using Fugene® HD transfection reagent (Roche Applied Science), following the manufacturer's protocol. Vesiculation was carried out using a previously described protocol ⁶³. Cells were rinsed three times with PBS (pH 7.4) containing 0.75 mM calcium and 0.5 mM magnesium (CM-PBS), and incubated with 1 ml of vesiculation buffer at 37 °C. To quench the formaldehyde, a glycine solution in PBS was added to the vesiculation buffer to a final concentration of 0.125 M. The vesiculation buffer consisted of CM-PBS with 25 mM formaldehyde and 0.5 mM 1,4-dithiothreitol (DTT).

Image acquisition

Vesicles were imaged using a Nikon Eclipse confocal laser scanning microscope with a 60X water immersion objective. All the images were collected and stored at a 512×512 resolution. Three different scans were performed for each vesicle: (1) excitation at 488 nm, with a 500-530 nm emission filter (donor scan); (2) excitation at 488 nm, with a 565-615 nm emission filter (FRET scan); and (3) excitation at 543 nm, with a 650 nm long-pass filter (acceptor scan). Gains of 8.0 were used for all the three scans. To minimize the bleaching of fluorescent proteins, ND8 filters were used during excitation with the 488 nm laser, and pixel dwell time was set to the minimum (1.68 μs).

Image analysis

The fluorescence intensities of vesicles were analyzed with a Matlab program developed in the lab, which automatically recognizes the vesicles and calculates the intensities in the three channels.