Ligand modulation of the conformational dynamics of the A2A adenosine receptor revealed by single-molecule fluorescence

G protein-coupled receptors (GPCRs) are the largest class of transmembrane proteins, making them an important target for therapeutics. Activation of these receptors is modulated by orthosteric ligands, which stabilize one or several states within a complex conformational ensemble. The intra- and inter-state dynamics, however, is not well documented. Here, we used single-molecule fluorescence to measure ligand-modulated conformational dynamics of the adenosine A2A receptor (A2AR) on nanosecond to millisecond timescales. Experiments were performed on detergent-purified A2R in either the ligand-free (apo) state, or when bound to an inverse, partial or full agonist ligand. Single-molecule Förster resonance energy transfer (smFRET) was performed on detergent-solubilized A2AR to resolve active and inactive states via the separation between transmembrane (TM) helices 4 and 6. The ligand-dependent changes of the smFRET distributions are consistent with conformational selection and with inter-state exchange lifetimes ≥ 3 ms. Local conformational dynamics around residue 2296.31 on TM6 was measured using fluorescence correlation spectroscopy (FCS), which captures dynamic quenching due to photoinduced electron transfer (PET) between a covalently-attached dye and proximal aromatic residues. Global analysis of PET-FCS data revealed fast (150–350 ns), intermediate (50–60 μs) and slow (200–300 μs) conformational dynamics in A2AR, with lifetimes and amplitudes modulated by ligands and a G-protein mimetic (mini-Gs). Most notably, the agonist binding and the coupling to mini-Gs accelerates and increases the relative contribution of the sub-microsecond phase. Molecular dynamics simulations identified three tyrosine residues (Y112, Y2887.53, and Y2907.55) as being responsible for the dynamic quenching observed by PET-FCS and revealed associated helical motions around residue 2296.31 on TM6. This study provides a quantitative description of conformational dynamics in A2AR and supports the idea that ligands bias not only GPCR conformations but also the dynamics within and between distinct conformational states of the receptor.


Rationale for the Design of the T119-Q226 A2AR Mutant for smFRET:
To track inward and outward movements from TM6, several published crystal structures of the A2AR were evaluated to find the positions which provided most dynamic changes for smFRET. Overlapping the crystal structures of the inverse agonist-, full agonist-, and mini-G-bound crystal structures of A2AR (i.e., 3EML, 2YDO, and 5G53, respectively) revealed that position Q226C on TM6 showed a pronounced displacement and rotated outwards whereas position T119C located on the alpha helix in intracellular domain II (ICL2) near TM4, remained relatively immobile ( Figure S1). The displacement of the between these two points corresponded to ~10 Å (Figure S1), which is similar with the results suggested by our smFRET measurements.

Double-stranded DNA: smFRET Corrections and Static Limit Controls
To obtain the γ correction factor for smFRET measurements (Eq. Donor-and acceptor-labelled oligonucleotides were annealed to obtain dsDNA with desired donor-acceptor basepair (bp) separations. The annealing buffer contained 10 mM Tris pH 7.5, 50 mM NaCl and 1 mM EDTA.
Equimolar amounts (100 μM) of the two strands were heated to 95 o C for 2 minutes using a thermocycler (Biometra TPersonal Thermocycler), followed by slow cooling to 25 o C for 45 minutes, after which the sample was aliquoted and frozen at -20 o C.
The smFRET control measurements were performed on two dsDNA samples with separations of 10bp and 17bp between the donor (fluorescein) and the acceptor (AF647) fluorophores, which were attached via a six-, and a five-carbon linker, respectively. The experiments were conducted using the same excitation power as was used for A2AR (donor and acceptor lasers intensities of ~25 kW/cm 2 and ~12 kW/cm 2 , respectively), and in a buffer solution (50 mM TRIS, 1 mM EDTA, pH 8.5, 150 mM NaCl) supplemented with 10 mM cysteamine. Singlepeak FRET histograms were observed for both 10 bp and 17 bp dsDNA samples, having mean efficiencies E of 80 ± 1 % and 29 ± 1 %, and widths ΔEFWHM of 19 ± 1 % and 38 ± 1 %, respectively. In addition to γ correction for smFRET, the width of the FRET histogram for the 10 bp dsDNA (pink band in Figure 1b) serves as a quasistatic limit for the exchange broadening in A2AR, i.e. the minimum FRET width in the absence of structural dynamics of the labelled biomolecule.
An upper limit for the smFRET width due to shot noise, ΔEshot-noise, can be computed using the expression: where E is the FRET efficiency and N is the average number of photons per burst. Four values were calculated using mean efficiencies E and photon numbers for the A2AR samples. The blue band in Figure 1b represents a spline (polynomial) interpolation of these points, with the width obtained by propagating errors in Eq. S1.

Förster Radius in A2AR: Limitations due to the Donor-Acceptor Orientation Factor κ 2
The efficiency of the Förster resonance energy transfer E between a donor and an acceptor fluorophore depends on the distance R between the fluorophores according to the expression: The Förster radius Ro is calculated according to the expression: where J is the normalized spectral overlap integral , ΦD is the fluorescence quantum yield of the donor , N is Avogadro's number , κ 2 is the orientation factor, n is the refractive index, εA is the extinction coefficient of the acceptor, and λ is the wavelength. For the AF488 and AF647 dyes conjugated to A2AR, absorption and emission spectra were measured and used to calculate the spectral overlap integral using open source software (UV-Vis-IR Spectral Software 1.2, FluorTools). We estimated the quantum yield of AF488 as ΦD = 0.72 ± 0.02 using the comparative method 1 .
To assess whether the isotropic value κ 2 = 2/3 is a good approximation for the A2AR samples, we measured the steady-state anisotropy or donor-and acceptor-labelled receptors. The values obtained from averaging over hundreds of single-molecule bursts were rAF488 = 0.34 ± 0.02 for the donor and rAF647 = 0.024 ± 0.005 for the acceptor 2 . Using these values, we calculated the lower and the upper limit of the orientation factor, κmin 2 and κmax 2 , respectively 3 :

Eq. S5
These limits correspond to a minimum and maximum Ro of 45 and 56 Å, respectively. This range exceeds the typical error margin for Ro, which depends on the error in measuring the quantum yield of the donor, ΦD, and in estimating the donor-acceptor spectral overlap integral.

SUPPORTING INFORMATION SECTION 2:
Expression, Purification, and Labelling of A2AR and Mini-Gs

Expression and Purification of R373C-ΔCys-mini-Gs
R373C-ΔCys-miniGs was expressed and purified according to the methodology described by Carpenter & Tate

Simulations of A2AR in Different Extracellular Ligand States
To build simulation systems, extracellular ligands, water molecules, and other solvents resolved in crystal structures were removed. Missing wild-type backbone atoms were modeled with the program Loopy. 5,6 Missing side chain atoms and side chain reversions to wild-type were modeled with the program SCWRL4. 7 Each repeat used a different model of missing receptor atoms. Disulfide bonds (C71-C159, C74-C146, C77-C166, and C259-C262) and all hydrogen atoms were placed with the GROMACS tool pdb2gmx. 8 A unique bilayer conformation was constructed for each simulation by extracting a randomly selected snapshot from a 300 ns simulation of a neat POPC bilayer with 166 lipids per leaflet, which was initially constructed with the CHARMM-GUI 9 membrane builder. 10 The receptor was oriented for insertion using the program LAMBADA, 11 and embedded in the bilayer using 20 cycles of the InflateGRO2 routine 11 with double-precision GROMACS steepest descent energy minimization.
During this procedure, 11-17 lipids were removed from each leaflet, allowing the final numbers to be asymmetric.
Each system was hydrated with ~85 waters per lipid and 100 mM excess KCl, disallowing placement of water or ions in the bilayer's hydrophobic core. Subsequently, water, lipids, and ions were relaxed over 30 ns by sequential 5-ns simulations using position restraints on receptor heavy atoms with force constants of 10 4 and 10 3 kJ/mol/nm 2 , followed by restraints on receptor Cα atoms with force constants of 10 3 , 10 2 , 10, and 1 kJ/mol/nm 2 . Pressureequilibrated systems had average Cartesian x × y × z dimensions of 7.33 ± 0.08 nm × 7.33 ± 0.08 nm × 10.73 ± 0.43 nm, an average volume of 576 ± 24 nm 3 , and an average of 59,000 ± 2,000 atoms.
All restraints were removed for production simulation. Simulations were conducted with mixed-precision (SPFP 12 ) AMBER 16 software. 13 Protein was modeled by the CHARMM36m force field 14 and lipids were modeled by CHARMM36. 15 The water model was TIP3P 16 with CHARMM modifications. 17  Lennard-Jones (LJ) interactions were evaluated using an atom-based cutoff with forces switched smoothly to zero between 1.0 and 1.2 nm. This is the recommended LJ cutoff for the CHARMM36 protein force field. 21,22 Note that the CHARMM36 lipid force field was parameterized with LJ switching between 0.8 and 1.2 nm. 15 Coulomb interactions were calculated using the smooth particle-mesh Ewald method 23,24 with Fourier grid spacing of 0.08 to 0.10 nm and fourth order interpolation. Temperature and pressure were controlled by velocity Langevin dynamics 25 at 310 K with a coupling constant of 1 ps and semi-isotropic coupling to Monte Carlo barostats 13 at 1.01325 bar with compressibilities of 4.5×10 −5 bar −1 , respectively. The integration time step was 4 fs with hydrogen mass repartitioning. 26 Non-bonded neighbor-lists were built to 1.4 nm and updated heuristically.
To estimate the precision of our results, the standard error of the mean is obtained using Eq.S6, for N repeat simulations with mean values μi and overall mean μ.
To quantify relaxation, we estimated exponential autocorrelation times, τacorr, for values of the distance between the BODIPY modified cysteine Cβ atom and the center of geometry of non-hydrogen atoms in the dye's aromatic group, l. The normalized autocorrelation function was evaluated according to Eq. S7, where the mean and standard deviation of the sample of l are represented by μ and σ, respectively, Δt is the time interval, and angular brackets indicate averaging over all possible initial times t (i.e., t ≤ tMAX -Δt, for a simulation of tMAX).

Eq. S7
The autocorrelation function, C(Δt), was fit to an exponential decay of the form    Table S1.     reports on the TM3-TM6 displacement, whereas the latter term, shown on the ordinate, reports on formation of the ionic lock. Data shown reports on ionic lock stability for simulations initiated using crystal structures in which the A2AR was crystalized bound to (a) an inverse agonist, (b) a full agonist, or (c) a full agonist plus mini-G, as shown in Table S2. Shaded dashed yellow region (a-c) indicate distances were the ionic lock is broken, whereas shaded dashed green regions (a-c) indicate the regime for salt bridge formation with the ionic lock intact. Probability scales in (a-c) are indicted by the spectrum of colors from white to dark red and follow a logarithmic scale.

6GDG
A agonist mini G (trimer) 6 5 a,b Indicates crystallization conditions; co-crystallized extracellular ligands and intracellular proteins were not included in simulations.