Measuring G Protein Activation with Spectrally Resolved Fluorescence Fluctuation Spectroscopy

G protein-coupled receptor signaling has been posited to occur through either collision coupling or pre-assembled complexes with G protein transducers. To investigate the dynamics of G protein signaling, we introduce fluorescence covariance matrix analysis (FCMA), a novel implementation of fluorescence cumulant analysis applied to spectrally resolved fluorescence images. We labeled the GPCR, Gα, and Gβγ units with distinct fluorescent protein labels and we applied FCMA to measure directly the complex formation during stimulation of dopamine and adrenergic receptors. To determine the prevalence of hetero-oligomers, we compared the GPCR data to those from control samples expressing three fluorescent protein labels with known stoichiometries. Interactions between Gα and Gβγ subunits determined by FCMA were sensitive to stimulation with GPCR ligands. However, GPCR/G protein interactions were too weak to be distinguished from background. These findings support a collision coupling mechanism rather than pre-assembled complexes for the two GPCRs studied.


Introduction 23
Fluorescence fluctuation spectroscopy (FFS) is a set of statistical techniques used to extract physical 24 parameters from fluorescence signals by using physical models of fluorescence detection (1). In 25 biological applications, the most frequently measured parameters are diffusion coefficients, 26 concentrations, and the molecular brightness of fluorescently labeled biomolecules (2-4). Imaging-FFS, 27 by analyzing multidimensional fluorescence images, has become increasingly popular due to its potential 28 to provide spatially resolved information (5-8). FFS is also used to analyze samples containing multiple 29 chromophores and provide information about heteromeric molecular interactions (9, 10). Towards 30 measuring an increasing number of molecular components, multicolor FFS has been expanded to utilize 31 spectrally resolved detection. In these systems, a prism or diffraction grating is used to redirect photons 32 onto an array of detectors so that the energy of incident photoelectrons is known more precisely than in 33 systems based on dichroic mirrors and multiple independent detectors (11-13). Recent developments 34 have paired spectral detection with spectral unmixing techniques for better signal-to-noise ratios (SNRs) 35 (14-16), and as many as four chromophores have been used simultaneously in live cell experiments (17). (26-28). Pre-assembly models assert that GPCRs and G proteins interact constitutively in stable 48 complexes (29-31). Kinetic models have been developed to accommodate both signaling mechanisms 49 (32), but experimental evidence to date has been contradictory regarding which mechanism is dominant 50 (24). Multicolor FFS approaches are well-suited to address this problem because they enable us to 51 observe the principal components of GPCR signal transduction simultaneously and in real time. 52 To investigate the mechanisms underlying GPCR activation, we introduce fluorescence 53 covariance matrix analysis (FCMA), which is a new approach to analyzing spectrally resolved imaging 54 data that provides detection and quantification of multi-chromic complexes. Additionally, FCMA 55 provides a simple graphical interpretation of complex formation. FCMA is an extension of cumulant-56 based approaches applied to spectrally resolved imaging data (8, 33-35). We applied FCMA to 23 57 channel images of tri-chromic samples expressing permutations of green, yellow, and red fluorescent 58 protein monomers and dimers expressed on the plasma membrane. Using FCMA, we can detect and 59 measure trimeric interactions without invoking higher order correlations (17). 60 We applied FCMA to the analysis of G protein signaling mechanisms for two Gαi-coupled GPCRs, 61 DRD2 and ADRA2A, whose exogenous ligands are dopamine and epinephrine, respectively. The Gαi 62 subset of G proteins are involved in the inhibition of cAMP production through interactions with 63 adenylyl cyclase (36). In co-expression experiments with labeled Gαi and Gβγ functional signaling units, 64 we monitored these three components during the signaling event. The Gα/Gβγ interactions were 65 sensitive to stimulation with GPCR agonists. However, we did not observe ternary interactions between 66 GPCRs, Gαi, and Gβγ units, which is consistent with the absence of pre-assembled complexes. 67 68

69
Fluorescence Covariance Matrix Analysis of CD86 Controls 70 To calibrate FCMA as a tool for analyzing samples expressing multiple chromophores, we used the 71 monomeric plasma membrane protein CD86 as a scaffold for permutations of fluorescent heteromers. 72 Three fluorescent proteins were used, mEGFP (G), mEYFP (Y), and mCherry2 (R). Three fluorescent 73 monomers were analyzed, CD86-G, CD86-Y, and CD86-R, as well as five monomer/heteromer 74 combinations, CD86-G + CD86-Y + CD86-R (three monomers), R-CD86-Y-G (trimer), CD86-Y-G + CD86-R 75 (GY dimer with R monomer), CD86-G + R-CD86-Y (G monomer with YR dimer), and R-CD86-G + CD86-Y 76 (GR dimer with Y monomer) (Fig. 1). These combinations were expressed in HEK 293 cells and the 77 plasma membrane adjacent to the cover glass was imaged using confocal microscopy. 78 Samples containing a single chromophore feature a single peak on the main diagonal of the 79 covariance matrix (Fig. 1 A-C). The location of this peak matches the position of the maximum signal in 80 the detection spectrum (Fig. S1 C). When noninteracting species are present in a sample their 81 contributions to the covariance matrix are additive ( Fig. 1 D, F-H). When two chromophores interact 82 and act as a single species this has a multiplicative effect on their contribution to the covariance matrix. 83 Interacting chromophores with high spectral overlap produce a broadened peak, as in the case for G and 84 Y ( Fig. 1 E-F). If interacting chromophores are well separated spectrally, their contribution to the 85 covariance matrix produces lobes away from the main axis, as is the case for GR and YR interactions ( Fig.  86 1 E, G-H). In the case of ternary interactions, there is both the more prominent, broad peak from the 87 highly overlapping GY interactions and broader off axis lobes from concomitant GR and YR interactions 88 ( Fig. 1 E). 89 90

Distribution of Fluorescent Oligomer States by Fluorescence Covariance Matrix Analysis 91
The covariance matrices and corresponding detection spectra from images of cells expressing three 92 chromophores on CD86 were fit using a seven-component model accounting for the following species: 93 G, Y, R, GY, GR, YR, and GYR. For each component, an apparent number density is determined and the 94 fractional distribution of each chromophore across different oligomer states is found (Fig. 2). The sum 95 of fractional densities for each chromophore is unity. For the triple expression of monomeric proteins, 96 CD86-G, CD86-Y, and CD86-R the largest fractions for each chromophore are observed in monomeric 97 states (Fig. 2 A). We did observe some apparent GY and YR interactions which we take as the 98 background levels for further experiments. GYR fractions were negligible. 99 In cells where the heterotrimer, R-CD86-Y-G was expressed, all three chromophores are found 100 to be spread across several oligomer states (Fig. 2 B). This phenomenon can be understood by 101 considering the existence of dark state fractions for each chromophore. A single protein will only 102 contribute to the GYR (trimer) fraction if all of its constituent chromophores are correctly folded into the 103 fluorescent state. Due to this effect, only small fractions (~20-30% per chromophore) appear in the 104 trimeric state. Correspondingly, the dimeric states are populated by contributions from proteins with 105 one chromophore in a dark state and the monomeric states are populated by contributions from 106 proteins with two chromophores in dark states. 107 Data from cotransfections of dimer and monomer constructs produce similar results. For CD86-108 Y-G and CD86-R cotransfections, G is distributed predominantly between G (~65%) and GY (~35%) states 109 ( Fig. 2 C). Similarly, Y is split between Y (~50%) and GY (~50%) states. R is found almost exclusively in its 110 monomeric state. For CD86-G expressed with R-CD86-Y, YR interaction is detected along with smaller 111 fractions in Y and R monomeric states (Fig. 2 D). G is found almost exclusively in its monomeric state. 112 For R-CD86-G expressed with CD86-Y, G and R are split between their respective monomeric states and 113 GR interacting state, whereas Y almost exclusively found in its monomeric state (Fig. 2 E). To compare with an established method, we processed the same datasets used for FCMA with spectrally 117 resolved raster image correlation spectroscopy (RICS). In contrast with FCMA, the 23 channel spectrally 118 resolved images must be unmixed prior to analysis into three single-color images, each corresponding to 119 a single chromophore (Fig. S1 B-D). Fitting model spatial auto-and cross-correlation functions allows for 120 the calculation of relative cross-correlation amplitudes, a readout of the interaction between pairs of 121 chromophores. Results from RICS analysis (Fig. 2 F) are in good agreement with those found with FCMA. 122 The data show very little interaction between chromophores in triple transfections of G, Y, and R 123 monomers. Images of the R-CD86-Y-G trimer exhibit strong, but not ideal, relative cross-correlations 124 among the three possible pairings. For monomer/dimer coexpressions, CD86-Y-G with CD86-R displays 125 strong GY interactions, R-CD86-Y with CD86-G displays strong YR interaction, and R-CD86-G with CD86-Y 126 displays strong GR interactions (Fig. 2 F). Conventional RICS analysis is limited to concomitant 127 measurements of binary interactions while ternary interactions cannot be detected directly and must be 128 inferred. In contrast, ternary interactions can be quantified directly from covariance matrices ( Fig. 1  To express all three G proteins at physiologically appropriate and experimentally pragmatic levels, we 133 modified a polycistronic construct introduced by Unen et al (37) to carry mCherry2 tags on GNB1 and 134 GNG2 and an mEYFP(Q69K) tag on GNAI1 (Fig. S2). With this configuration, we labeled the two 135 functional G protein components with Y and R. This construct was expressed in tandem with a G labeled 136 GPCR so that all three components of the GPCR signaling triplet could be monitored simultaneously. 137 The two Gαi-coupled GPCRs studied here were DRD2 (dopamine) and ADRA2A (epinephrine). 138 GNAI1-Y is primarily a fluorescent monomer, as determined by FCMA, with ~50-70% of Y 139 chromophores appearing monomeric before the addition of a GPCR stimulating ligand (Fig. 3). The weak 140 interactions detected between GNAI1-Y and GPCRs G-DRD2 (Fig. 3 A-B) and G-ADRA2A (Fig. 3 C-D) are 141 comparable to background interaction levels found in control experiments featuring CD86 (Fig. 2 A). 142 There is significant interaction between GNAI1-Y and R-GNB1/R-GNG2 with ~20% of Y chromophores 143 being found in these interactions. The Y and YR fractions are sensitive to GPCR stimulation by its native 144 ligand. When expressed with G-DRD2, the fraction of Y participating in YR interactions decreases from 145 ~20% to ~10% after stimulation with 100 μM dopamine (Fig. 3 B). Similarly, when expressed with G-146 ADRA2A, the YR fraction decreases from ~20% to ~10% after stimulation with 30 μM epinephrine ( interactions and the fraction of R participating YR interactions comes from the relative expression levels 155 of these two chromophores. Because GNAI1-Y expression is dictated by an internal ribosome entry site 156 ( Fig. S2), its expression is approximately three times lower than that of R-GNB1 and R-GNG2 (37). 157 We observe small relative changes in oligomer state distributions for G-GPCRs and R-GNB1/R-158 GNG2 in response to GPCR stimulation. The GR fraction of R chromophores increases after dopamine 159 stimulation ( Fig. S3 D), the GYR fraction of G chromophores decreases after epinephrine stimulation (Fig.  160 S4 B), and the GYR fraction of R chromophores also decreases after epinephrine stimulation (Fig. S4 D). 161 In each of these cases, the absolute oligomer fractions do not differ significantly from background levels 162 ( Fig. 2) and are unlikely to represent biologically relevant findings. 163 164

Spectrally Resolved RICS Analysis of GPCR Stimulation Experiments 165
Datasets for FCMA of G protein activation were also processed using spectrally resolved RICS. The 166 results from these analyses are in good agreement with those determined by FCMA. We observe 167 negligible relative cross-correlations for GY and GR pairings (Fig. 4). Like FCMA, we observe YR relative 168 cross-correlations that are sensitive to GPCR stimulation (Fig. 4 B, D). We have demonstrated FCMA as a fluorescence fluctuation analysis tool suitable for multicolor imaging 182 experiments in live cells. In this analysis, heteromeric combinations leave unique fingerprints on the 183 covariance matrices calculated from spectrally resolved images (Fig. 1). The relative contributions of 184 different oligomer states can be determined from fitting model functions and the resulting information 185 tells us how chromophores are distributed across these states (Fig. 2). 186 FCMA is complementary to recent developments in spectrally resolved image correlation 187 spectroscopy. FCMA and spectrally resolved RICS achieve many of the same goals. Both quantify the 188 degree of interaction between two or more chromophores (Figs. 2-4). RICS, and other correlation 189 function-based approaches, have the advantage of providing information about transport properties by 190 outputting fitted diffusion coefficients (Fig. 5). However, as we show in this work, FCMA detects ternary 191 interactions directly (Fig. 2 B), which offers a more robust and simpler computational procedure. 192 Ternary complex detection has been achieved with triple correlation analysis (TRICS), but that relies on 193 higher order correlation functions which greatly increase the signal-to-noise requirements for the data 194 and computational complexity (17, 38). Additionally, visual inspection of covariance matrices allows for 195 the straightforward observation of complex formation that has an intuitive connection to the emission 196 spectra (Fig. 1, Fig. S1 C). In practice, both analyses can be implemented in parallel with the same 197 fluorescence imaging data. 198 The FMCA approach allows for simultaneous measurements of the three major components of 199 the canonical GPCR/G-protein signaling pathway directly with fluorescence in live cells (Fig. 3, Fig. S3-200   S4). These data are highly relevant to our mechanistic understanding of the signal propagation through 201 GPCR/G protein pathways (23). The two predominant models of G protein activation are collisional 202 coupling (26-28) and pre-assembled complexes (29-31). In collisional coupling models, GPCRs and G 203 proteins have independent Brownian motions aside from their brief interactions when the GPCR is 204 activated. Conversely, pre-assembly models posit that stable GPCR/G protein complexes are present 205 with the components maintaining contact throughout the signaling processes. Biochemically, the 206 distinction between collisional coupling and pre-assembly models arises from the affinities of GPCR/G 207 protein interactions (32). In this work, we did not find interactions between GPCRs (DRD2 and ADRA2A) 208 and G proteins (Gαi1/Gβ1/Gγ2) (Fig. 3-4, Fig. S3-S4) that were distinguishable from background (Fig. 2). 209 These data suggest that these GPCR-G protein components interact through weak, transient 210 associations below what is detectable with our current experimental sensitivity. Additionally, we 211 observed that the GPCR and G protein components have distinct apparent diffusion coefficients 212 suggesting that they are not constitutively coupled as a pre-assembly model would suggest (Fig. 5). 213 Although these data are consistent with a collisional coupling mechanism of GPCR signaling, we are 214 limited by the fidelity of the chromophores to radiative states (Fig. 2 B-E) Table S1.

Fluorescence Covariance Matrix Analysis 284
FCMA is an extension of multicolor fluorescence cumulant analysis applied to spectrally resolved 285 imaging data (8, 34, 35). For spectrally resolved imaging data the first order cumulants, [1] ( ), are 286 equal to the average detection spectrum of the pixels within the region of interest, R. The second order 287 cumulants, [1,1] ( , ), are equal to the covariance matrix for all pairs of channels, ( , ). 288 performed on the spectrally unmixed images (Fig. S2 B-D) found for spectrally resolved RICS described 294 below. To avoid complications from crosstalk in spectral detection we used a single pixel offset in the 295 scanning axis (i.e. ( + 1)) discussed in depth in previous work (8). 296 Apparent number densities ( ) and molecular brightnesses ( ) of species within the detection 297 volume of the confocal microscope are related to the first and second order cumulants by: 298 300 2 is a shape factor depending on the geometry of the detection volume. For this work we used 2 = 301 0.5 corresponding to a two-dimensional Gaussian detection volume (47). {1} ( ) for a species 302 consisting of a single chromophore is equal to that chromophore's detection spectrum normalized such 303 that its sum is unity. {2}  For the three chromophore experiments performed in this work, we fit a seven species model to 311 the detection spectra and covariance matrices. The number density was allowed to vary for each 312 species and the molecular brightness was linked across species so that 10 variables were determined 313 when fitting [ 1] ( ) and [ 1,1] ( , ): , , ,  ,  ,  , , , , and . Fitting was 314 performed using Levenberg-Marquardt least squares minimization comparing experimentally 315 determined [ 1] and [ 1,1] to theoretical values [1] and [1,1] (48, 49). 316 In this work, we focus on the distribution of number densities. Data are presented as fractional 317 number densities. For example, for G containing species the fractional number density for species is: 318