Conformational transition of FGFR kinase activation revealed by site-specific unnatural amino acid reporter and single molecule FRET

Protein kinases share significant structural similarity; however, structural features alone are insufficient to explain their diverse functions. Thus, bridging the gap between static structure and function requires a more detailed understanding of their dynamic properties. For example, kinase activation may occur via a switch-like mechanism or by shifting a dynamic equilibrium between inactive and active states. Here, we utilize a combination of FRET and molecular dynamics (MD) simulations to probe the activation mechanism of the kinase domain of Fibroblast Growth Factor Receptor (FGFR). Using genetically-encoded, site-specific incorporation of unnatural amino acids in regions essential for activation, followed by specific labeling with fluorescent moieties, we generated a novel class of FRET-based reporter to monitor conformational differences corresponding to states sampled by non phosphorylated/inactive and phosphorylated/active forms of the kinase. Single molecule FRET analysis in vitro, combined with MD simulations, shows that for FGFR kinase, there are populations of inactive and active states separated by a high free energy barrier resulting in switch-like activation. Compared to recent studies, these findings support diversity in features of kinases that impact on their activation mechanisms. The properties of these FRET-based constructs will also allow further studies of kinase dynamics as well as applications in vivo.


Purification of recombinant proteins
Pellets derived from 200 mL cultures were thawed on ice and resuspended in 8 mL of chilled lysis buffer (25 mM Tris-HCl, 250 mM NaCl, 40 mM imidazole, 100 µg/ml lysozyme, pH 8.0) and supplemented with EDTA-free protease inhibitor cocktail (Roche). Once resuspended, 2 mL of 10% (v/v) Triton-X-100 and 1 Ku of bovine pancreatic DNAse I was added and the lysate was placed on the orbital shaker at 150 rpm at 4 °C for 1 hour. Cell lysates were clarified by centrifugation at 13,000 rpm at 4 °C for 1 hour in an SS34 rotor (Sorvall). The clarified lysate was loaded onto a 1-mL HisTrap Fractions were collected, combined and concentrated in spin concentrators, snap frozen in liquid nitrogen and stored at -80 °C. Purified proteins were analyzed by 10% SDS-PAGE.

Kinase assays
In vitro kinase assays of FGFR1 synthetic variant proteins were carried out using the ADP-Glo™ Kinase Assay (Promega) as described previously [4]. The assays were carried out at 25 °C in 40 mM Tris-HCl pH 8.0, 20 mM NaCl, 20 mM MgCl 2 , 1 mM MnCl 2 , 1 mM TCEP, and 100 µM Na 3 VO 4 , in a total volume of 60 µL (15 µL kinase reaction; 15 µL ADP-Glo™ Reagent; 30 µL Kinase Detection Reagent) in solid white, flat-bottom 96-well plates. In some assays, Poly (E 4 Y 1 ) peptide (Sigma-Aldrich) was used as kinase substrate. Kinase reactions were stopped after 90 min of incubation for autophosphorylation reactions and 45 min of incubation for time-course with substrate. ATP-to-ADP standard conversion curves, Z-values and signal-to-background ratio were calculated according to the manufacturer's published procedure. For determinations of the K m for ATP (K m,ATP ), reactions contained a peptide concentration of 0.5 mg/mL and ATP at varying concentrations from 10 to 240 µM. Data was then fitted using non-linear regression to the Michaelis-Menten equation. The optimal kinase amount was determined when 50% of substrate conversion was achieved (half maximal reaction velocity, ½V max ), and assays were carried out at the apparent K m,ATP , at 150 µM ATP.
Negative control experiments were performed in the absence of ATP, substrate, enzyme and positive controls were performed with FGFR1 WT (data not shown). Luminescence output was recorded using BMG Labtech FLUOstar Optima plate reader at 520nm. Curve fitting for was performed using GraphPad Prism® software.

SDS-PAGE and Western blotting
Proteins were separated using standard procedures and transferred onto polyvinylidene fluoride (PVDF) membrane for 16 hours at 30 V at 4 °C. For immunoblotting, the membranes were blocked with 5% non-fat dry milk (Sigma) in TBS pH 7.5 containing 0.1% Tween-20 (Sigma-Aldrich) (TBS-T) for 30 -60 min with agitation. Membranes were incubated with the indicated primary antibodies for 1 hour at 25 °C or overnight incubation at 4 °C. Membranes were washed three times for 5 min in TBS-T followed by incubation with the required conjugate diluted in 5% milk/TBS-T for 1 hour with agitation. Blots were subject to three washes in TBS-T and developed using with ECL prime detection kit (Pierce) and exposed to Hyperfilm ECL (Amersham Biosciences). Protein transfer to the membrane was then visualised by Amido Black staining.

Imaging instrumentation
All single-molecule imaging was conducted using a custom-built microscope, based on a fully motorised Olympus IX81 base. Four lasers (100 mW 405 nm Coherent Obis, 100 mW 488 nm Coherent Sapphire, 150 mW 561 nm Coherent Sapphire and a 150 mW Toptica iBeam Smart), each with their own shutter control, were expanded to the same diameter and combined using a series of dichroic mirrors into a single free-space beam. Half-wave plates were used to adjust the polarisation before passing the beams through an Acousto-Optical Tunable Filter (AA Optoelectronics) to quickly modulate laser power. The combined beams were again expanded and launched into a single-mode optical fibre (Thorlabs PM-S405-XP) using an Olympus (0.1 N.A. 10x Air) objective lens. The output of the optical fibre was collimated using an achromatic parabolic mirror collimator (Thorlabs) and passed through a quarter-wave plate to circularly polarise the beam. The free beam then passed through the "TIRF" lens (L1, Thorlabs 200 mm achromatic doublet) and was focused directly onto the back focal plane of the objective lens. This entire subsystem was mounted on a micrometer translation stage to adjust the TIRF angle. The beam was reflected to the objective lens via a multi-edge dichroic filter (Semrock Di01-R405/488/561/635-25x36). The objective lens was an oil-immersion Olympus TIRF apochromatic objective lens (1.49 N.A. 100x Oil). Actively cooled electron-multiplying charged coupled device (EMCCD) cameras (Andor iXon Ultra) were coupled to the camera port of the microscope via an additional 1.5x magnifying relay. An additional dichroic mirror (Semrock FF560-FDi01-25x36) in the 4f relay was used to simultaneously image the second colour. Bandpass filters (BP1: Semrock FF01-520/35-25 and BP2: Semrock FF01-617/73-25) in front of the two cameras selected the appropriate dye emission. Typically, camera acquisition was at 20 Hz. Laser shutter, acousto-optical tunable filter (AOTF) and camera firing were synchronised using a Data Translation DT9834 data acquisition (DAQ) module, using the internal clock to provide synchronised TTL pulses.
Sample positioning was controlled via a motorised micrometer stage (Physik Instrumente) with a XYZ-Nanopositioning stage (Physik Instrumente). All software for microscope control was written in C++ and Python.

Camera registration
Chromatic aberrations cause a slight misalignment in the images returned from the two cameras.
Therefore, we need to register the images to create a mapping from one space to the other. Before every experiment, we imaged immobilised MultiTetraSpec beads (Invitrogen) using both cameras, and used the image data to calculate the correction field using the following method: 1. Crudely align the channels using a normalised cross-correlation 2. Use the non-overlapping regions of the images to discard particles which are not present in both Once we have registered the channels ( Figure S4A), we end up with a set of particles, with positions ! and alignment vectors, ! . We can determine the expected position in the second channel for any new particle ( !"# ) using the weighted (based on distance) sum of the vectors in the vector field:

Data acquisition
We utilised an Alternating-Laser EXcitation (ALEX) illumination scheme [6], coupled with objective type TIRF microscopy to image the individual proteins, and verify their dye labeling stoichiometry.
Utilising the external D/A converter, an alternating sequence of laser pulses (12.5 mW 488 nm laser, 50 mW for the 561 nm laser) was generated, and synchronized with the both camera's exposure (typically 50 ms, at 20Hz) using TTL pulses. The cameras were operated at a "real-gain" of 750, and cooled to a temperature of -80 o C. Movies were recorded using the camera's full-frame (512x512 pixels) and streamed for up to 90 s.

Data analysis
Data analysis was performed according to Hohlbein et al. [7]. Here we briefly describe the methods.
The process proceeds as follows: 1. Acquire ALEX data and separate into Donor excitation and Acceptor excitation 2. Align the two image sets to account for chromatic aberration 3. Find candidate molecules in the direct excitation of Acceptor channel 4. Find their centroids and extract a window around the molecules 5. Correct for background fluorescence 6. Correct for dye cross-talk 7. Calculate the proximity ratio FRET (E PR ) and stoichiometry (S) Utilising alternating laser excitation, we generate a movie containing an alternating sequence of frames, such that in the odd frames we illuminate the sample with the 488 nm laser (Donor excitation, D ex ) and gather information about ! !" ! !" and ! !" ! !" and in the even frames we illuminate the sample with the 561 nm laser (Acceptor excitation, A ex ) and gather information about ! !" ! !" . These movies are then aligned to account for chromatic aberrations.
We initially select molecules for further analysis, by localising bright spots (corresponding to the Point Spread Function, PSF, of a single molecule) found in the ! !" ! !" channel, which possess brightness above a threshold value and corresponding to molecules containing a single Tet1-TAMRA-X dye.
Next we calculate the centroid of the molecules by fitting the PSF to a symmetrical two-dimensional Gaussian function. For a sub-wavelength diameter fluorescent molecule, fitting of the PSF to a Gaussian function yields the highest accuracy and precision of localisation, thus: Finally, we correct for "dye-crosstalk". In this experiment, there are two types of dye-crosstalk correction factor that we need to take in to account, (i) Leakage of donor emission captured in the acceptor channel, and (ii) Direct excitation of the acceptor by the donor excitation laser. To calculate the correction factor for leakage (L), we measure the fluorescence signal in the donor and acceptor channels using protein labeled with only the donor fluorophore. We can then calculate correction factor L as: spectrofluorometer [8]. The purified proteins (0.5 µM) were tagged with FlAsH-EDT2 (100 µM) or Tet1-TAMRA-X (50 µM) in 40 mM Tris-HCl pH 8.0, 20 mM NaCl, 20 mM MgCl2, 1 mM MnCl2, 1 mM TCEP, and 100 µM Na3VO4 and were incubated in the presence or absence of ATP (150 µM) at 25 oC for 1 hour. Excitation of the FRET pair was performed at 490 nm using the picosecond supercontinuum laser excitation source, and fluorescence emission was detected at 545 nm.

Molecular dynamics calculations
Molecular dynamics simulations were performed using GROMACS 4.5 [11] with the PLUMED plug-in and the Charmm22* force field [12]. Phosphorylated and non-phosphorylated FGFR systems were solvated with TIP3P water molecules in a dodecahedral box. Both systems were minimized with 10000 steps conjugated gradient and equilibrated for 10 ns in the NPT ensemble. Production runs were carried out in the NVT ensemble using the velocity-rescale thermostat [13] with a timestep of 2 fs. The PME algorithm was used for electrostatic interactions with a cut-off of 1.2 nm while a single cut-off of 1.2 was used for Van der Waals interactions.

Enhanced Sampling
Parallel Tempering Metadynamics (PT-metaD) [14] was performed for FGFR systems using 20 replicas in the 298.0 K -337.5 K temperature range. The well-tempered metadynamics algorithm was used, so that gaussian height is progressively decreased throughout the run [15]. The initial gaussian height was set at 10 kJ/mol with a bias factor of 10. The two collective variables used were the distance in contact map space to the inactive A-loop conformation and the distance in contact map space to the active conformation, as in previous simulations [4,16], with a sigma of 0.5 for both CVs.
Note that the CVs are adimensional. A total of 450 ns was needed to fully converge the free energy surfaces of the two systems.    An Olympus IX81 base with high NA objective lens is used as the body of the microscope. Laser illumination is coupled into the system using a single-mode polarisation maintaining fibre. Two EMCCD cameras are attached to the camera port via a relay, enabling simultaneous two-colour imaging. Cameras and lasers are synchronised using an external signal generator. An infrared laser autofocus system is coupled via the camera port to maintain a focus lock during image acquisition.