Structural assemblies of the di- and oligomeric G-protein coupled receptor TGR5 in live cells: an MFIS-FRET and integrative modelling study

TGR5 is the first identified bile acid-sensing G-protein coupled receptor, which has emerged as a potential therapeutic target for metabolic disorders. So far, structural and multimerization properties are largely unknown for TGR5. We used a combined strategy applying cellular biology, Multiparameter Image Fluorescence Spectroscopy (MFIS) for quantitative FRET analysis, and integrative modelling to obtain structural information about dimerization and higher-order oligomerization assemblies of TGR5 wildtype (wt) and Y111 variants fused to fluorescent proteins. Residue 111 is located in transmembrane helix 3 within the highly conserved ERY motif. Co-immunoprecipitation and MFIS-FRET measurements with gradually increasing acceptor to donor concentrations showed that TGR5 wt forms higher-order oligomers, a process disrupted in TGR5 Y111A variants. From the concentration dependence of the MFIS-FRET data we conclude that higher-order oligomers – likely with a tetramer organization - are formed from dimers, the smallest unit suggested for TGR5 Y111A variants. Higher-order oligomers likely have a linear arrangement with interaction sites involving transmembrane helix 1 and helix 8 as well as transmembrane helix 5. The latter interaction is suggested to be disrupted by the Y111A mutation. The proposed model of TGR5 oligomer assembly broadens our view of possible oligomer patterns and affinities of class A GPCRs.


Supporting Tables
Supporting Table 1: Parameters for determination of the corrected green to yellow fluorescence intensity ratio F D /F A necessary for the 2D histograms. The background B was determined from untransfected cells. The green to yellow fluorescence intensity ratio (F D /F A ) was corrected for crosstalk (characterized by the crosstalk factor α), background B, detection efficiencies of D (g G ) and A (g Y ). The acceptor fluorescence used for 2D-FRET must also be corrected for additional direct acceptor excitation DE and relative concentration dependent brightness DE rel .
All samples were corrected for distinct fluorescence quantum yields Φ and a spectral shift factor γ (especially for TGR5 Y111A) which is considered in the corrected green detection efficiency (g G *). Supporting Table 2: Parameters for ε(t) diagram in Fig. 4 for each TGR5 variant. The parameters b0-b4 are obtained from the fit equation diagrams.          Table 5). The corresponding dimers are colored in light grey or dark grey.

Proximity FRET
Pixel-wise analysis of the fluorescence data in TGR5 Y111A compared to wt and Y111F showed strong differences in the FRET properties, which were only detectable in an acceptor concentration-dependent manner (Figure 4, main text). Thus, we tested whether the observed FRET could simply be caused by a very high local concentration of acceptor proteins in the membrane, so that donor and acceptor are in proximity even though they do not interact. This phenomenon is called "proximity FRET".
Due to the single-molecule sensitivity of our confocal microscope, we could perform FRET experiments with acceptor concentrations of ~1-6 µM in 1.23 fl, which corresponds to a molecule density of less than ~0.02 acceptor molecules/nm 2 . According to King et al. 3 , proximity FRET is negligible (E < 0.1) at these concentrations.
The pixel-integrated, time-resolved FRET analysis ε(t) supported the pixel-wise analysis and clearly demonstrated the presence of different FRET species in TGR5 wt and Y111F and therefore the formation of higher-order oligomers as compared to Y111A.

Cloning of TGR5
Human TGR5 was cloned as previously described 4 . Constructs were cloned into the pcDNA3.1+   to 48 h before analysis. Cell vitality and successful transfection was visually inspected before MFIS measurements.

Microscope calibration
Calibration measurements with Rhodamine 110 delivered the G-factor G = Sg  /Sg  for the GFP emission wavelength range (green channels). The G-factor accounts for the detection efficiency difference between detectors of both polarizations (g  and g  ). The instrument response function (IRF) was measured with the back-reflection of the laser beam using a mirror and was used for iterative re-convolution in the fitting process. Furthermore, untransfected cells and water were measured at 488 nm and 559 nm for background determination.

Time series experiments of TGR5 stimulation by Taurocholic acid (TC)
To study the effect of bile acid agonists on the FRET parameters we used the water-soluble ligand TC, because addition of DMSO (necessary to dissolve TLC) affects the fluorescence signal significantly. For the time series experiments the time laps viewer function supplied by Olympus LSM was used. The motorized table was calibrated, and three cells were selected and monitored over a 40 minutes time period. FRET measurements were taken every 10 minutes: before the addition of TC immediately after addition (t = 0 min), and after ten and twenty minutes (t = 10 min; t = 20 min). Cells were excited with 488 nm and 559 nm laser light as described above. Where necessary, changes in focus and system drift were corrected.

Pixel-wise analysis
To determine fluorescence-weighted lifetimes in a pixel-wise analysis, the histograms presenting

MFIS-FRET 2D histograms
For oligomerization analysis, we plotted the 2D histograms of donor lifetime  D(A)  f vs the green to yellow fluorescence intensity ratio (F D /F A ) (see equations (2) and (3)) corrected for crosstalk (characterized by the crosstalk factor α ), background B, detection efficiencies of D (g G ) and A (g Y ). The acceptor fluorescence used for 2D-FRET must also be corrected for additional direct acceptor excitation DE and relative concentration dependent brightness DE rel . Furthermore all samples were corrected for distinct fluorescence quantum yields Φ and a spectral shift factor γ (especially for TGR5 Y111A) which is considered in the corrected green detection efficiency (g G *).
The crosstalk factor α is determined as the ratio between donor photons detected in the yellow channels and those detected in the green channels for the Donor only (Donly) labeled sample.
The corrected detection efficiency g G * is determined as the ratio of the spectral shift influenced by green detection (0.69) and expected green detection (1.12) multiplied with the quantum yield Φ Y111A obtained from a self-made detection efficiency software. The F D /F A parameters for each variant are provided in Supporting Table 1 The simultaneous reduction in both The average GFP fluorescence intensity of an image with GFP excitation was also corrected for detector dead time, and then the obtained intensity ( m G G S , ) was further corrected for the quenching effect due to FRET: Assuming the concentration of the FPs reflects the concentration of their host proteins, the TGR5 concentration (without non-fluorescent molecules) in µM was determined as:

Estimation of the association constants for oligomerization
The total protein concentration and the protein association constants have to be considered to determine the oligomerization state or the chemical speciation. To calculate the transferefficiency for a given oligomerization the spatial organization of the molecules within the oligomers and the concentration of donor, acceptor and non-fluorescent molecules has to be considered. The total protein concentration (equation (6)) is given by the sum of the acceptor, the donor and the unlabeled protein concentration: Here the unlabeled protein concentration c u equals the concentration of immature mCherry. The protein concentrations were calculated using the brightness of free GFP and free mCherry as reference Even though higher-order oligomerization is anticipated we used a simple dimer/tetramer model to describe our data as this allows for a quantitative description. In this model we assume that a tetramer is constituted of a dimer of dimers (Supporting Fig. 6) Here o is a monomer, oo a dimer and (oo)(oo) is a tetramer. We use the monomer o as a master species. Then the total protein concentration is given by: Now, the concentrations of the three species o, oo and (oo)(oo) for any given the total protein concentration is obtained by solving the three equations above.
To calculate the transfer efficiency we assume that donor, acceptor and unlabeled molecules behave biochemically identical. Hence, the probability of an oligomer composition is given by the probability of finding a donor, acceptor or unlabeled molecule and the counting statistics.
The probabilities of finding a donor, acceptor or unlabeled molecule depend on their respective concentrations. For instance the probability of an acceptor molecule is given by the respective species and total protein concentration: In a tetramer the sum of donor, acceptor and unlabeled molecules is constant. The probability of a certain tetramer composition is obtained by the multinomial distribution: FRET-rate constants are additive. Therefore in case of multiple acceptors the total FRET-rate constant experienced by a donor (i) is given by the sum of all FRET-rate constants of all acceptors (j): Here ) (ij DA R is the donor acceptor distance between the donor (i) and the acceptor (j) which is determined by the spatial arrangement of the oligomer. For instance, in the case as illustrated in Supporting Fig. 6 the two FRET-rates experienced by the donor at position 1 and the donor at position 4 are given by: It has to be considered that the contribution to the fluorescence signal depends on the number of donor molecules. For instance a tetramer constituted out of three donors and one acceptor molecule contributes three times more to the total signal as compared to a tetramer only constituted out of one donor, one acceptor and two unlabeled molecules.
The predicted transfer efficiency for each data point depends now only on the equilibrium association constants K 1 , K 2 and the spatial arrangement of the fluorophores in the dimer and the tetramer. To reduce the number of free parameters we assumed that the tetramer can be described by a rectangular geometry where one edge is approximately 100 Å long while the second edge is between 40-50 Å long (Supporting Fig. 6). This assumption is in line with the homology models (Supporting Table 5 and Supporting Fig. 7). Furthermore, only FRET molecules have been selected. Therefore, the first equilibrium from monomer to dimer is not monitored and only the equilibrium constant of the tetramer formation is probed. Thus, only K 2 and the dimer distance in the range of 40-60 Å is reflected in the data. For the measurements we find that a short distance of approximately 45 Å describes the data best. For the TGR5 wt and Y111F variant we find predominately a tetrameric or higher-order oligomer configuration while in case of the Y111A mutant the molecules are predominately in a dimeric configuration.

Statistical analysis
Experiments were performed independently at least three times. For MFIS-FRET at least nine cells per series in three independent experiments were measured. Results are expressed as mean ± standard error of the mean (SEM) and analysed using the two-sided student t-test. A p ≤ 0.01 was considered statistically significant.

Structural models of TGR5 dimers and tetramers
Dimer models with the interface TM1 and H8 (1/8) were generated by structurally aligning two homology models of TGR5 8 onto the dimeric crystal structure of the human κ-opioid receptor (PDB ID: 4DJH 9 ) via the 'cealign' command in Pymol 2 . For dimer models with the 4/5 interface and the 5/6 interface the same procedure was applied using the human CXCR4 receptor (PDB ID: 3ODU 10 ) and the murine μ-opioid receptor (PDB ID: 4DKL 11 ) as alignment templates, respectively.
Tetramer models were built in a similar fashion. Here, two TGR5 dimers with the same dimer interface, e.g. (1/8), were aligned on another TGR5 dimer with a different interface, e.g. (4/5).

Explicit linker simulations: Molecular dynamics simulations of GFP bound to a linker
For computing a thermodynamic ensemble (TE) of GFP positions with an explicit linker/GFP construct, initially, the structure of the TGR5 C-terminal residues 296-330, for which no experimental structural information is available, and the nine residues that connect the Cterminus to GFP (total sequence: QRCLQGLWGRASRDS PGPSIAYHPSSQSSVDLDLN YGSTGRHVS) was generated with the 'Protein building' approach in Maestro. Phi and psi angles of zero were chosen, resulting in a straight peptide conformation and, hence, a structurally unbiased starting structure for the molecular dynamics (MD) simulations. This linker was subsequently fused to enhanced GFP (PDB ID: 4EUL 16 ), and the resulting structure was capped with acetyl and N-methyl amide groups at the N-and C-termini, respectively, and protonated with PROPKA 17 according to pH 7.4. We assumed the thermodynamic ensemble (TE) of mCherry to be identical to that of GFP.
Then, the linker/GFP construct was neutralized by adding counter ions and solvated in an octahedral box of TIP3P water 18 with a minimal water shell of 12 Å around the solute. The Amber14 package of molecular simulation software 19,20 and the ff14SB and GAFF 21 force fields were used to perform an all-atom MD simulations. To cope with long-range interactions, the "Particle Mesh Ewald" method 22 was used, and the SHAKE algorithm 23 was applied to bonds involving hydrogen atoms. The time step for all MD simulations was 2 fs with a directspace, non-bonded cut-off of 8 Å. The first linker residue was fixed with positional harmonic restraints with a force constant of 100 kcal mol -1 Å -2 throughout the simulations to emulate that this residue would be bound to TGR5 embedded in a membrane. At the beginning, 17500 steps of steepest decent and conjugate gradient minimization were performed; during 2500, 10000, and 5000 steps positional harmonic restraints with force constants of 25 kcal mol -1 Å -2 , 5 kcal mol -1 Å -2 , and zero, respectively, were applied to the solute atoms. Thereafter, 50 ps of NVT-MD (MD simulations with a constant number of particles, volume, and temperature) were conducted to heat up the system to 100 K, followed by 300 ps of NPT-MD (MD simulations with a constant number of particles, pressure, and temperature) to adjust the density of the simulation box to a pressure of 1 atm and to heat the system to 300 K. During these steps, a harmonic potential with a force constant of 10 kcal mol -1 Å -2 was applied to the solute atoms. As the final step in thermalization, 300 ps of NVT-MD simulations were performed while gradually reducing the restraint forces on the solute atoms to zero within the first 100 ps of this step. Afterwards, six independent production runs of NVT-MD simulations with 150 ns length each were performed.
For this, the starting temperatures of the simulations at the beginning of the thermalization were varied by a fraction of a Kelvin. The conformations obtained in these simulations were pooled for further analyses.

Implicit linker simulations
Inter-dye distance distributions for all TGR5 dimer and tetramer models were calculated using an modified Accessible Volume (AV) approach 24 . Firstly, the different protein models (see 5.14) were embedded in an explicit membrane via the CHARMM-GUI membrane builder 25 Table 2) with a length of 3.7 Å each 26 . A dye radius of 25 Å was used as an approximation for the GFP size, resulting in a total length of 229 Å for the linker/GFP construct. The distance between linker attachment points in most of the screened oligomer models was shorter than the effective size of the AVs resulting in AV overlap. The AVs were constructed considering geometric factors in terms of steric exclusion effects caused by the TGR5 oligomer and the membrane. To account for clashes between the dyes, which are not addressed in the AV simulations, the inter-dye distance probability was set to zero for all distances below 25 Å. To account also for entropic effects, we introduced position weights for the implicitly modelled linker according to the Gaussian chain model, so that the non-uniform dye position probability distribution in the AV was scaled (Supporting Fig. 7) 27