G protein diffusion states reveal receptor binding dynamics at single-molecule scale

Summary

G protein-coupled receptors (GPCRs) signal through heterotrimeric G proteins, yet the kinetics of receptor-G protein interactions in cells remain poorly understood. Using single-molecule imaging, we tracked individual Gα_i1 proteins diffusing in the plasma membrane, expecting that receptor binding should slow these rapidly diffusing lipid-anchored proteins to match the substantially slower transmembrane GPCRs. Gα_i1-Halo showed no detectable slowing upon dopamine D2 receptor activation. In contrast, the nucleotide-decoupled Gα_i1 4A-Halo mutant exhibited receptor-dependent slowing with a lifetime of ~0.8 seconds. Agonist increased the Gα_i1 4A-Halo association rate without affecting the lifetime of the receptor-engaged complex, consistent with the idea that agonist efficacy is reflected in an increased probability of productive G protein engagement. These findings suggest that native GPCR-G protein interactions are not resolved at 15 ms time resolution and establish a diffusion-based approach for studying receptor-G protein kinetics in living cells.

Graphical Abstract

eTOC blurb

Mathiasen et al. provide a diffusion-based single-molecule approach for studying how G protein-coupled receptors engage G proteins in living cells. Native interactions are not resolved at 15 ms, but a nucleotide-decoupled Gαi1 4A mutant shows that agonist increases the rate of receptor engagement without altering complex lifetime, consistent with efficacy being reflected in engagement probability.

Introduction

One-third of all FDA-approved drugs target G protein-coupled receptors (GPCRs), and efforts to discover additional compounds that modulate the activity of these receptors remain intense¹. Agonists stabilize receptor conformations that recruit and activate G proteins, and the capacity of a given agonist to evoke a downstream cellular response is quantified as efficacy. The molecular basis of efficacy, including how ligand binding relates to receptor-G protein association and dissociation, remains poorly understood. Cryo-electron microscopy has revolutionized structural biology by providing static snapshots of ligand-bound GPCR-G protein complexes, yet these reveal similar receptor conformations regardless of ligand efficacy, suggesting that efficacy is dictated by dynamic parameters: binding kinetics and complex lifetime.

Upon association with a GPCR, the heterotrimeric G protein is rapidly destabilized by exchange of guanosine diphosphate (GDP) for guanosine-5'-triphosphate (GTP) at the α subunit. This leads to dissociation of both the GPCR-G protein complex and the G protein heterotrimer. Electrophysiological methods show that effector responses, despite being several steps downstream of GPCR activation and G protein coupling, can begin tens of milliseconds after receptor activation^2-5, highlighting the high temporal resolution needed to capture the dynamics of GPCR-G protein complexes in their native environment. From ensemble energy-transfer-based assays, receptor-G protein association has been inferred to occur within 45-450 ms. However, these measurements reflect the time required for GPCR-G protein association to reach a new steady state and average signals from hundreds of thousands of receptors^6,7. Thus, the microscopic association rate constant (k_on) and particularly the lifetime of receptor-G protein complexes cannot readily be extracted from such data.

Single-molecule imaging combined with single-particle tracking (SPT) offers the advantage of directly revealing the dynamic behavior of individual proteins diffusing in the plasma membrane of living cells⁸. A prior single-molecule investigation used two-color intensity colocalization to estimate receptor-G protein complex lifetimes of approximately 1-2 seconds⁹. This is difficult to reconcile with the rapid electrophysiological timescales and may reflect limitations of colocalization-based approaches: the diffraction limit makes it difficult to distinguish true molecular interactions from random proximity, particularly if both species independently enter shared membrane microdomains.

Here, we report a single-molecule imaging approach that takes advantage of the difference in diffusion rates between GPCRs and G proteins at the plasma membrane to investigate the kinetics of agonist-induced GPCR-G protein association in living cells, and thus circumvents both the complications of colocalizing two intensity signals given the diffraction limit and the resulting problem of differentiating random colocalization from true interactions. If productive interactions between GPCRs and G proteins persist for as long as 1 second, it should be possible to directly observe rapidly diffusing lipid-anchored G proteins slow to the diffusion rate of the transmembrane receptor. We studied the dopamine D2 receptor (D2R), a prototypical class A GPCR that couples to the inhibitory Gi/o/z family. Using imaging conditions that allow us to track individual Gα_i1 subunits for seconds with a time-resolution of 15 ms, we failed to observe a change in Gα_i1-Halo diffusion state lifetimes in the presence of agonist-activated D2R, suggesting that the interactions are either too short or too infrequent to be captured. In contrast, the nucleotide-decoupled mutant Gα_i1 4A-Halo^10,11, which exhibits prolonged receptor binding in the presence of nucleotides, showed receptor- and activation-dependent slowing. We demonstrated that Gα_i1 4A-Halo-receptor interactions persist for nearly 1 second, and that agonist enhances the rate of G protein association without affecting the rate of complex dissociation, providing insight into the molecular basis of efficacy. Our findings also reveal that membrane-associated proteins can enter an intrinsic, background slow diffusion state independent of receptor binding, highlighting confounds in single-molecule analysis and indicating that improved methods will be necessary to capture wild-type G protein kinetics.

Results

Single-molecule imaging detects differential diffusion of GPCRs and G proteins

We generated a Gα_i1-HaloTag fusion protein (Gα_i1-Halo), using the well-validated position 91 for insertion^12,13, and verified that its function was indistinguishable from unlabeled wild-type Gα_i1 (Figure S1A-B). Using total internal reflection fluorescence microscopy (TIRFM) and the cell-permeant HaloTag ligand JF-646¹⁴, we imaged individual Gα_i1-Halo subunits diffusing in the plasma membrane of Chinese hamster ovary (CHO) cells (Figure 1A-B). For all conditions tested in this study we achieved optimal labeling density for single particle tracking (≤1 molecule per μm²) (Figure S1C-H), and used u-track¹⁵, an established multiple-particle-tracking algorithm, to generate trajectories from one-minute movies (Figure 1C). The divide-and-conquer moment-scaling spectrum (DC-MSS) analysis¹⁶ assigned segments of each trajectory into four motion types: immobile, confined, Brownian (hereafter denoted free) and directed diffusion. In the absence of D2R co-transfection, Gα_i1-Halo spent ~80% of the time freely diffusing with a mean diffusion coefficient of ~0.35 μm²/s (Figure S1I-J), consistent with previous reports and Gα_i1 being lipid-anchored to the inner membrane leaflet^17-21. In contrast, D2R-Halo diffused much more slowly (~0.15 μm²/s) (Figure S1K-L), as expected for a multi-pass transmembrane protein^22-24.

Figure 1. Live-cell Single-molecule imaging of Gα_i1.

(A) Schematic outlining the experimental approach for obtaining single-molecule diffusion trajectories of Gα_i1-Halo proteins labelled with the JF-646-Halo fluorophore. (B) Representative micrographs recorded after labelling with JF-646-Halo for a mock cell (not expressing Gα_i1-Halo) (left), and a cell expressing Gα_i1-Halo (right), scale-bars, 10 μm. Inset: corresponding bright field images. (C) Exemplary Gα_i1-Halo segment classified as Brownian (free). Orange circle and square mark the start and end of the segment. (D-E) Total fraction of time spent (occupancy) (D) and average diffusion coefficients (E) in the free motion state for Gα_i1-Halo co-transfected with the dopamine D₂ receptor (D2R), in the absence (light grey bars, n=11 cells) or presence of quinpirole (10 μM, dark grey bars, n=12 cells). In (E) the diffusion coefficient of freely diffusing D2R-Halo (darkest gray) is plotted for comparison. Dots represent individual cell means. Box plots indicate the median (central line), overall mean (central +), and interquartile range (lower and upper lines represent the 25th and 75th percentiles, respectively), while the whiskers are defined according to the Tukey criterion. Statistics: (D) Unpaired two-tailed t-tests. (E) Ordinary one-way ANOVA, Tukey’s multiple comparisons test with a single pooled variance. Only significant differences are indicated; all p values are reported in Table S1.

If receptor activation leads to long-lived G protein-receptor complexes, we would expect G protein diffusion rates to approach those of receptor; however, neither co-expressing high amounts of the D2R (1200 ng) nor saturating amounts of the full agonist quinpirole affected Gα_i1-Halo’s occupancy in the free diffusion state (Figure 1D) or its diffusion rate (Figure 1E). This suggests that Gα_i1-Halo-receptor interactions are too rapid or too rare to detectably slow G protein diffusion at our current imaging rate.

Membrane complexity leads to receptor-independent slow states

The DC-MSS analysis only reports a single average diffusion coefficient over the entire collection of free segments, potentially obscuring transient slowing events within individual trajectories. We therefore applied a hidden Markov modeling approach (HMM-Bayes)²⁵ to test whether Gα_i1-Halo shifts between a fast unbound state and a slower receptor-bound state and whether such shifts depend on receptor activation (Figure 2A). In this analysis we included the condition with high D2R co-transfected (1200 ng), with or without co-expression of the S1 catalytic subunit of pertussis toxin (PTX) to disrupt coupling of Gα_i1-Halo.

Figure 2. HMM-Bayesian analysis reveals two distinct free diffusion modes for Gα_i1.

(A) Schematic of data processing workflow for the HMM-Bayes analysis of freely diffusing segments. (B) Diffusion coefficients of fast and slow states for Gα_i1-Halo with and without co-transfection of PTX, in the presence of vehicle (light blue/orange, −PTX n=8, +PTX n=7) or quinpirole (10 μM, dark blue/orange, −PTX n=9, +PTX n=9). (C-H) Probabilities of transitioning from fast to slow (C) and from slow to fast (D), state occupancy of fast (E) and slow (F) states, fast (G) and slow (H) state lifetimes for Gα_i1-Halo co-transfected with D2R, with or without PTX, in the presence of vehicle (light blue/orange) or 10 μM quinpirole (dark blue/orange). Number of cells as in (B). (I-L) State occupancy of fast (I) and slow (J) states and fast (K) and slow (L) state lifetimes for the myr-Halo control in the presence of vehicle (light blue/orange, n=9) or quinpirole (10 μM, dark blue/orange, n=10). Dots represent individual cell means. Box plot details are described in Figure 1D-E. Statistics: All panels, ordinary two-way ANOVA, Tukey’s multiple comparisons test (single pooled variance). Only significant differences are indicated; all p values are reported in Table S1. In B, the main effect comparing fast vs. slow is significant with p < 0.0001 (****).

This procedure identified a two-state Brownian model as the most probable description of the freely diffusing Gα_i1-Halo segments, with two distinct diffusion coefficients: slow (~0.1 μm²/s) and fast (~0.45 μm²/s) (Figure 2B). Independent re-runs of the HMM routine yielded less than 1% variance in diffusion coefficients and state occupancy, attesting to the stability of the HMM-Bayes procedure and model selection (Figure S2A-B).

For the freely diffusing segments, the HMM-Bayes routine returned segments with a) no transitions between slow and fast states, b) one transition between slow and fast states, or c) two or more transitions (Figure S2C). For Gα_i1-Halo the number of segments with ≥2 transitions was ~7% (Figure S2D). Because HMM-Bayes identifies transitions between fast and slow states along every freely diffusing single-molecule trajectory, this analysis can be used to quantify the probabilities of transitioning from fast-to-slow and slow-to-fast (Figure 2C-D). We failed to observe any significant differences in transition probabilities related to receptor activation or PTX. We also examined whether the overall fraction of time spent in the fast or slow states (occupancy) changes with activation of D2R and observed a slight drop in fast state occupancy with quinpirole treatment that was reversed by PTX (Figure 2E-F).

We quantitated lifetimes using only segments that display ≥2 transitions such that the fast or slow states contributing to the lifetimes are always flanked by the other state. For example, in a slow-fast-slow segment, only the fast state would contribute to the fast state lifetime and vice versa, eliminating both left-censoring (unknown state duration before tracking began) and right-censoring (tracking termination during a state). Although HMM-Bayes analysis can detect slowing for even 1-2 frames²⁵, we failed to observe significant activation-dependent differences in lifetimes for Gα_i1-Halo, or sensitivity to PTX treatment (Figure 2G-H), suggesting that receptor-dependent effects on Gα_i1-Halo diffusion state lifetimes are below the detection threshold of our current method.

As receptor-dependent effects on Gα_i1-Halo diffusion state lifetimes were too small to be reliably detected, we tested whether the discrete fast and slow diffusion states we observed reflect a general property of membrane-associated proteins rather than G protein-specific behavior such as receptor binding. To test this, we expressed a palmitoylated and myristoylated HaloTag (myr-Halo) engineered to have the same membrane anchoring properties as Gα_i1. Free segments (Figure S2E-F) recorded for this construct were analyzed by the HMM routine, which also returned a two-state model with two diffusion coefficients (Figure S2G). The number of segments with ≥ 2 transitions and the probability of transitioning were also on par with Gα_i1-Halo (Figure S2H), and as expected, the myr-Halo state occupancies and lifetimes were insensitive to quinpirole (Figure 2I-L). The lifetime of the slow state was very similar for myr-Halo and Gα_i1-Halo, and we therefore conclude that palmitoylated and myristoylated proteins diffusing within the plasma membrane can transition into an intrinsic, background slow diffusion state with a ~0.5 second lifetime that is unrelated to GPCR signaling. This is consistent with heterogeneous diffusion behavior previously reported for membrane-associated proteins^9,21.

Single-molecule imaging captures receptor-dependent slowing of nucleotide-decoupled G protein

GPCR-G protein complexes dissociate when guanine nucleotides bind to the Gα subunit. This suggests that nucleotide-induced D2R-Gα_i1-Halo dissociation is too rapid to detectably slow diffusion of the G protein. To test this idea, we turned to a previously described Gα mutant (Gα_i1 4A) that exhibits prolonged binding to GPCRs even in the presence of guanine nucleotides, effectively uncoupling receptor binding from nucleotide exchange¹¹. The Gα_i1 4A mutant carries a four-alanine insertion in the α5 helix, extending it by one helical turn. This rotates residue F^H5.8 out of the hydrophobic core that stabilizes the GDP-bound state, mimicking the core disruption that accompanies receptor-induced activation in wild-type Gα^10,11. The 4A mutant therefore adopts a receptor-bound-like conformation without releasing GDP, decoupling receptor engagement from the conformational changes that normally drive nucleotide exchange (Figure S3A-B). Consequently, Gα_i1 4A binds receptors in the presence of guanine nucleotides and shows little or no nucleotide-induced displacement from activated receptor. The receptor-engaged 4A complex thus persists on a timescale governed by a nucleotide-independent dissociation process that is much slower than nucleotide-dependent dissociation.

Using the same single-molecule imaging approach described above, we imaged individual Gα_i1 4A-Halo proteins under conditions identical to the experiments with Gα_i1-Halo (Figure S1D and H). The fraction of trajectory time spent in the freely diffusing state remained unchanged at ~80% for Gα_i1 4A-Halo, regardless of D2R co-expression or agonist stimulation (Figure 3A), indicating that, like Gα_i1-Halo, Gα_i1 4A-Halo is predominantly freely diffusing in the plasma membrane. However, in contrast to Gα_i1-Halo, the Gα_i1 4A-Halo diffusion rate slowed significantly in the presence of high D2R expression (Figure 3B, compare with Figure 1E). This slowing occurred without agonist but was enhanced by saturating amounts of quinpirole. Indeed, at the highest level of D2R expression tested (1200 ng), agonist reduced the average diffusion rate to that of freely diffusing D2R (Figure 3B), consistent with tracked Gα_i1 4A-Halo being associated with activated receptor. We also imaged Gα_i1 4A-Halo in the presence of increasing amounts of co-transfected D2R, to experimentally test the effect of increasing receptor surface expression. We observed that diffusion rate decreased monotonically with increasing amounts of D2R cDNA transfected (Figure 3C).

Figure 3. Agonist drives Gα_i1 4A-Halo to a slow diffusion state.

(A-B) Occupancy (A) and diffusion coefficients (B) in the free motion state from the DC-MSS analysis for Gα_i1 4A-Halo co-transfected with D2R in the absence (light grey bars, n=9 cells) or presence of quinpirole (10 μM, darker grey bars, n=8 cells). In (B) the diffusion coefficient of freely diffusing D2R-Halo (darkest gray) is repeated from Figure 1E for comparison. (C) DC-MSS diffusion coefficients for Gα_i1 4A-Halo in the presence of increasing amount of co-transfected D2R, before (light gray) and after addition of quinpirole (10 μM, black). Data are shown as mean, +/− SEM, for n= 7-13 cells. (D) Average diffusion coefficients for Gα_i1 4A-Halo in the presence of increasing amounts of co-transfected D2R cDNA, as determined by the HMM-Bayes analysis for the fast (light and darker blue) and slow (light and darker orange) state, for vehicle (light colors) and quinpirole (10 μM, dark colors). Number of cells in each condition: Vehicle (100 ng n=7, 300 ng n=7, 1200 ng n=18), quinpirole (100 ng n=8, 300 ng n=8, 1200 ng n=8). (E-F) Occupancy of the fast-diffusing (E) and slow-diffusing (F) state for Gα_i1 4A-Halo in the presence of increasing amounts of transfected D2R for vehicle or quinpirole (10 μM), Number of cells/experiments identical to D. (G) Occupancy of the slow-diffusing state for Gα_i1 4A-Halo co-expressed with D2R (1200 ng), in the presence of vehicle, quinpirole (10 μM), or of inverse agonist sulpiride (10 μM, red, n=12). Conditions for vehicle and quinpirole are repeated from F for comparison. (H) Occupancy of the slow-diffusing state for Gα_i1 4A-Halo in the presence of D2R (1200 ng) transfected in the absence (−PTX) or presence (+PTX) of co-expressed PTX (vehicle n= 5, quinpirole n=7). Conditions for vehicle and quinpirole −PTX are repeated from F for comparison. In all panels except (C) dots represent individual cell means. Box plot details are described in Figure 1D-E. Statistics: A, Unpaired two-tailed t-test. B, Ordinary one-way ANOVA, C-H Ordinary two-way ANOVA, all ANOVA performed with Tukey’s multiple comparisons test (single pooled variance). Only significant differences are indicated in the figures; all p values are reported in Table S2. In D, the main effect comparing fast vs. slow is significant with p <0.0001 (****).

We expected the progressive slowing of the mean diffusion rate observed in the DC-MSS analysis to be reflected in the two-state Brownian diffusion model, such that Gα_i1 4A-Halo would shift between a fast unbound state and a slower receptor-bound state in a receptor- and agonist-dependent manner. We therefore analyzed these data with the HMM-Bayes routine. As for Gα_i1-Halo, HMM analysis returned a two-state model as the most probable, and increasing amounts of transfected D2R did not affect either the slow state (D~ 0.1 μm²/s) or fast state (D~ 0.45 μm²/s) diffusion coefficients of Gα_i1 4A-Halo prior to agonist stimulation (Figure 3D and Figure S3C). Only quinpirole at the highest transfection level (1200 ng) slowed the fast state (Figure 3D). We speculate that with a high density of agonist-bound receptor, Gα_i1 4A-Halo may rebind to active receptors faster than we can capture, and thus the fast state becomes contaminated with sub-frame slow states.

Increasing D2R transfection and activation by quinpirole decreased the fraction of time (occupancy) spent in the fast state (Figure 3E), and correspondingly increased the occupancy in the slow state (Figure 3F). The impact of receptor alone is consistent with D2R constitutive activity and ensemble data showing that the 4A mutant can bind to receptors in the absence of agonist stimulation. The selective D2R inverse agonist sulpiride reversed this redistribution, as expected for a ligand that blocks D2R activity and prevents G protein coupling (Figure 3G). We also co-expressed PTX to prevent coupling of the Gα_i1 4A-Halo. As expected, PTX disrupted Gα_i1 4A-receptor association in a parallel ensemble assay (Figure S3F-G). At the single-molecule level, PTX markedly reduced the occupancy in the slow state and the effect of quinpirole was diminished (Figure 3H). With high D2R transfection (1200 ng), the quinpirole-induced slowing of the fast state diffusion (shown in Figure 3D) disappeared with PTX co-expression (Figure S3H), and the variance reduced significantly, consistent with a block of G protein engagement with receptor. In summary, agonist shifts the occupancy of Gα_i1 4A-Halo towards a slow state, and inverse agonist or PTX shift occupancy towards the fast state, demonstrating that our diffusion-based approach can detect bona fide receptor-Gα_i1 4A-Halo association and dissociation events in living cells.

Agonist efficacy is related to the rate of G protein association

Having verified that we could detect interactions between D2R and Gα_i1 4A-Halo, we asked whether agonist activation alters the rates of receptor-G protein association, dissociation, or both. For the highest D2R transfection level, the fast-to-slow transition probability increased significantly with quinpirole, and this effect was abolished with PTX co-expression (Figure 4A), consistent with an increase in the rate of receptor-G protein association. Conversely, the slow-to-fast transition probability, and thus escaping the receptor-bound state, was unaffected by quinpirole stimulation (Figure 4B).

Figure 4. Agonist activation enhances the rate of G protein association.

(A-B) Transition probabilities for transitions from fast to slow state (A) and from slow to fast state (B), for Gα_i1 4A-Halo co-transfected with D2R (1200 ng), in the absence (−PTX) or presence (+PTX) of PTX and treated with either vehicle or quinpirole (10 μM). Number of cells in each condition: Light blue/orange, n=18 (−PTX), n=5 (+PTX), dark blue/orange, n=8 (−PTX), n =7 (+PTX). Box plot details are described in Figure 1. (C-D) Fast state lifetimes (C) and slow state lifetimes (D) using only free segments that display two or more transitions between fast and slow states, for Gα_i1 4A-Halo co-transfected with D2R (1200 ng), in the absence (−PTX) or presence of PTX (+PTX), and treated with either vehicle (light blue/light orange), quinpirole (10 μM) (dark blue/orange), or sulpiride (10 μM) (purple/red). Number of cells and number of experiments as in (A) and n=12 for sulpiride. (E-F) Fast (E) and slow (F) state lifetimes for Gα_i1 4A-Halo co-transfected with increasing amounts of D2R and treated with vehicle (light blue/orange) or quinpirole (10 μM, dark blue/orange). Lifetimes are computed as described in (C-D). Number of cells/experiments as in Figure 3D. (G-H) Data from E and F shown as mean +/− SEM, for each cell population. In (A-F) dots represent individual cell means. Box plot details are described in Figure 1D-E. Statistics: A-F, Ordinary two-way ANOVA with Tukey’s multiple comparisons test (single pooled variance). Only significant differences are indicated in the figures; all p values are reported in Table S3.

As for Gα_i1-Halo, we also examined the lifetimes, computed from segments with ≥2 transitions. For Gα_i1 4A-Halo, ~20% of segments showed ≥2 transitions, three times more than for Gα_i1-Halo, yielding on average >100 lifetimes per cell (Figure S4A-C). For high D2R transfection, quinpirole decreased the fast state lifetime, consistent with faster G protein engagement with receptor (Figure 4C). In contrast, the inverse agonist sulpiride increased the fast state lifetime, consistent with a block of D2R constitutive activity. PTX also extended the lifetime to a level similar to that of sulpiride. PTX attenuated but did not abolish quinpirole's effect: the fast state lifetime with quinpirole plus PTX was longer than with quinpirole alone, yet remained shorter than vehicle plus PTX (Figure 4C). This might reflect incomplete ADP-ribosylation of Gα_i1 4A-Halo or short-lived interactions between ADP-ribosylated Gα_i1 4A-Halo and D2R.

Quinpirole had no effect on the slow state lifetime (Figure 4D), suggesting that, once Gα_i1 4A is engaged with the receptor, the lifetime of engagement does not depend on agonist. However, treatment with sulpiride or PTX reduced the slow state lifetime from ~0.8 seconds to ~0.5 seconds, similar to the background observed for Gα_i1-Halo and myr-Halo, and as expected for treatments that prevent receptor-G protein association (Figure 4D). We also found that progressively lowering D2R transfection shifted the slow state lifetimes of Gα_i1 4A-Halo towards this ~0.5 second background (Figure 4F, H) and abolished quinpirole-induced shortening of fast state lifetimes (Figure 4E, G). This shortening of fast state lifetimes by quinpirole, without any change in slow state lifetimes, indicates that the full agonist increases the on-rate of Gα_i1 4A-Halo binding to receptor rather than affecting the off-rate (Figure 4G-H). Thus, although membrane-associated proteins exhibit a background ~0.5 second slow state independent of receptor binding, and Gα_i1-Halo binding events cannot be observed under our current imaging conditions, we are able to detect receptor-dependent slowing and extract a slow state lifetime of ~0.8 seconds for active D2R bound to Gα_i1 4A-Halo.

Discussion

Here, we report a single-molecule diffusion-based method in living cells that measures receptor-dependent slowing of G proteins to provide insight into the timing of receptor-G protein coupling. In native systems, effector activation downstream of G protein coupling occurs within 5-55 ms, as measured by electrophysiological approaches. For example, D2R-dependent activation of GIRK channels controls inhibitory postsynaptic currents in brain slices within this timeframe^2,4. A change in channel activity requires multiple steps after the Gα_i1-D2R interaction, including Gβγ release, diffusion to, and activation of GIRK channels, suggesting that the actual time required for receptor-G protein interactions must be substantially shorter. Furthermore, presynaptic inhibition downstream of the serotonin receptor, another Gi-coupled receptor, occurs in less than 20 ms in vivo³, and rhodopsin can activate more than 50 transducin G proteins per second⁵. These reported timescales (tens of milliseconds) for the physiological response are consistent with receptor-G protein interactions that form with high on rates and last only for very short times, and thus also have high off rates.

This kinetic profile is difficult to reconcile with studies suggesting that receptor-G protein complexes remain intact for seconds. For example, several approaches, including solution nuclear magnetic resonance spectroscopy²⁶, surface plasmon resonance²⁷, bio-layer interferometry²⁸, time-resolved cryo-EM and smFRET^29,30, have used purified preparations of G proteins and receptors in detergent or nanodiscs to estimate receptor-G protein complex lifetimes on the order of seconds. However, these approaches have used either non-physiological, nucleotide-free conditions or nucleotide-insensitive G protein dominant negative mutants and mini-G proteins²⁸, all of which artificially stabilize receptor-G protein complexes. Additionally, receptors and G proteins reconstituted in detergent solutions may not recapitulate the dynamics of interactions that occur in cell membranes.

Receptor-G protein association has been studied by ensemble energy-transfer-based assays that inferred the interaction to occur within 45-450 ms. It is important to note that the timescales from such ensemble techniques reflect the approach (k_app) of a population of receptors and G proteins towards a new steady-state and do not correspond to the lifetime (1/k_off) of a single receptor–G protein complex. Our failure to detect a change in Gα_i1-Halo lifetimes in the presence of activated D2R at 15 ms time resolution is consistent with both the physiological response-time being below 50 ms and with the expected distinction between ensemble and single molecule kinetics.

Sungkaworn and colleagues reported agonist-dependent receptor-G protein colocalization for 1-2 seconds in living cells using single-molecule methods, although such colocalization was rare, and often occurred during periods when the colocalized molecules slowed or stopped together, as if both were confined in the same location⁹. We did not detect such long-lived complexes of Gα_i1-Halo with D2R in living cells. Furthermore, using the Sungkaworn methodology, receptor-G protein interactions lasting less than 284 ms could not be distinguished from random colocalizations, and we failed to resolve a receptor-dependent lifetime for Gα_i1-Halo even with a 15 ms frame rate. The simplest explanation consistent with both the rapid electrophysiological timescales and our failure to resolve a receptor-specific Gα_i1-Halo lifetime is that the productive GPCR-G protein interactions responsible for rapid signaling are too short-lived to be captured by colocalization-based single-molecule methods or diffusion-based analysis at current temporal resolution. Notably, we found that membrane-associated proteins can transition into background slow diffusion states (~0.5 second lifetime) independent of receptor binding, a potential confound for colocalization-based approaches, particularly if receptors and G proteins share common spatial domains where slowing occurs.

In contrast, with the "sticky" nucleotide-decoupled Gα_i1 4A, the two distinct diffusion states became receptor-dependent. The time spent in the slow state (D~0.1 μm²/s) scaled positively with both receptor transfection and agonist activation, while the fast state (D~0.4 μm²/s) occupancy decreased correspondingly. This enhanced slow state occupancy was reversed by treatment with an inverse agonist or PTX, indicating that this state reflects the receptor-bound state of the G protein. Despite the background ~0.5 second slow state lifetime observed for all membrane-associated proteins, receptor binding extended the slow state lifetime to ~0.8 seconds, demonstrating that we can detect receptor-specific effects above this background. For Gα_i1 4A we found that the fast state lifetime was significantly decreased by saturating agonist treatment, while the slow state lifetimes remained unchanged, indicating that agonist increases the on-rate for productive receptor engagement rather than affecting the Gα_i1 4A off-rate. For the on-rate, these findings should translate to the wt nucleotide-coupled G protein, which we expect to be independent of nucleotide exchange. The off-rate, however, will need to be assessed with the wt nucleotide-coupled G protein directly, since it is likely to be regulated by nucleotide binding. The increase in on-rate with agonist treatment, indicates an increase in the probability of coupling, an observation that is consistent with the suggestion that ligand efficacy is related to enhanced conformational sampling of receptors prior to G protein binding^30-32.

Limitations of the study

A key limitation of this study is that our 15 ms time resolution is insufficient to resolve the duration of receptor engagement of wild-type Gα_i1, precluding direct measurement of native GPCR-G protein interaction kinetics. As a result, whether the lifetime of receptor engagement is agonist-dependent in the context of GDP release and GTP binding remains unresolved¹¹. Because the Gα_i1 4A is nucleotide-decoupled, the slow-state lifetime we measure reflects dissociation of the receptor-engaged complex via a nucleotide-independent process rather than the wt nucleotide-coupled G protein. Technical improvements in imaging rates and labeling technology would be required to detect wt G proteins and would also allow testing whether transient intermediate state complexes exist for partial agonists²⁹. Furthermore, the presence of endogenous Gi/o/z proteins in CHO cells introduces potential competition with Gα_i1-Halo for receptor binding, and we cannot fully exclude that this affects the measured association rates. Finally, this study is limited to Gα_i1, and whether the findings generalize to other G protein subtypes remains to be established.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines

1. Flp-in CHO cells (Thermo Fisher Scientific)

2. LEx-Flp-In T-REx (FITR) CHO cells

3. LEx-FITR CHO line stably expressing D2R-Halo upon tetracycline induction.

4. HEK-293-T cells (ATCC)

5. HEK-293 (ATCC)

All CHO lines (1-3) were maintained in Ham’s F12 medium (Corning), 10% fetal bovine serum (FBS) (Corning), 2 mM L-glutamine (Corning) and 1% penicillin–streptomycin (Gibco) at 37 °C in 5% CO₂.

For LEx-Flp-In T-REx (FITR) CHO cells (2), the medium contained 15 μg ml⁻¹ blasticidin (InvivoGen) and 50 μg ml⁻¹ zeocin (Invitrogen).

For the D2R-Halo stable LEx-FITR CHO line (3), the medium contained 500 μg ml⁻¹ hygromycin (Thermo Fisher Scientific) and 15 μg ml⁻¹ blasticidin. All CHO lines were grown to ~70% confluency in six-well tissue culture dishes before transfection, tetracycline (MilliporeSigma) induction and/or labeling for microscopy.

All HEK-293 cell lines (4-5) were maintained in DMEM (Gibco) supplemented with 10% FBS and 1% penicillin–streptomycin at 37 °C in a 5% CO₂ humidified incubator.

Sex of cell line 1-5: The CHO cells lines used in this study, which are all derived from CHO-K1, as well as the HEK293 lines, are female in origin.

Cell authentication of cell lines 1-5: Cell identity was authenticated by the vendor.

METHOD DETAILS

Reagents

Quinpirole and sulpiride (Tocris Biosciences) were dissolved immediately before use in DPBS at 10 mM final concentration.

Dopamine hydrochloride was dissolved immediately before use in ultrapure water at 22.22 mM with 40 mg/L sodium bisulfite (Thermo Fisher Scientific).

Forskolin (Merck) was dissolved immediately before use in ultrapure water at 133.33 mM with 40 mg/L sodium bisulfite. The HaloTag ligand JF-646 was synthesized, dissolved and characterized as described previously¹⁴.

Plasmid construction

Plasmid construction was done either by restriction enzyme digestion followed by ligation or by Gibson Assembly using NEBuilder HiFi DNA Assembly Master Mix (NEB). All sequences were confirmed with Macrogen’s DNA sequencing service. All constructs were propagated in Escherichia coli BL21-Gold competent cells (Agilent Technologies).

Four plasmids were cloned directly for this project:

1. pcDNA3.1 Gα_i1-Halo: HaloTag was inserted into Gα_i1 at amino acid position 91, flanked by linkers SGGGGS and SGGGSG using restriction sites BamHI and BspEI in the pcDNA3.1 (Thermo Fisher Scientific) expression vector, in the same position where luciferase was introduced previously⁶. This construct was shown not to significantly alter function as compared to wild-type Gα_i1 (Figure S1B).

2. pcDNA3.1 Gα_i1 4A-Halo: Four alanines were inserted at Gα_i1 amino acid position Q333¹¹ (numbering according to wt Gα_i1), extending the α5 helix by one helical turn, using Gibson Assembly and restriction sites KpnI and XbaI, in the Gα_i1-Halo plasmid.

3. pcDNA5/FRT/TO-IRES kozac −3T+4T D2R-Halo: Halo was inserted in the C-terminal of the short version of the human D2 receptor after linker ATGLRSRAQASNSAVDGTAGPVAT using Gibson Assembly and restriction sites AgeI and SpeI, in expression vector pcDNA5 FRT/TO IRES kozac −3T+4T, described previously⁸.

4. pcDNA3.1 myr-Halo: The first 10 amino acids from Gα_i1 was spaced with a hydrophilic linker TKLTEER, that was previously shown to ensure correct access to fatty acylation enzymes as well as anchoring to membranes^36,37, and assembled with HaloTag into pcDNA3.1+ using Gibson Assembly and restriction sites NheI and EcoRI.

Generation of stable CHO cell line

To generate stable CHO cell lines expressing D2R-Halo, we used a previously described FITR CHO line containing an FRT recombination target site³⁸ in the CHO genome that confers low expression levels and that also allows for tetracycline-inducible expression, referred to as LEx-FITR cells³⁹. Stable integration of the pcDNA5/FRT/TO-IRES kozac −3T+4T vector^8,39 encoding the D2R-Halo was achieved by co-transfecting 200 ng of the vector with 1,800 ng of the pOG44 Flp-In Recombinase vector (Thermo Fisher Scientific) using Lipofectamine LTX (Thermo Fisher Scientific) according to the manufacturer’s protocol and selecting for cells with 500 μg ml⁻¹ hygromycin in the presence of 15 μg ml⁻¹ blasticidin.

Sample preparation for TIRF imaging

Transfection and induction

Flp-in CHO cells were plated in six-well tissue culture dishes and cultured in Hams F12 containing 10% fetal bovine serum, 1% glutamine and 1% penicillin–streptomycin prior to transfection. Cells at ~70% confluence were transfected overnight with Gα_i1-Halo or Gα_i1 4A-Halo at 200 ng, D2R at either 100, 300 or 1200 ng, Gβ₁ at 300 ng and Gγ₂ at 300 ng, with pcDNA3.1 carrier plasmid added to bring the total DNA to 2000 ng per 2 mL medium using Lipofectamine LTX as directed by the manufacturer’s protocol, 18–24 h before labeling and imaging.

The LEx-FITR CHO D2R-Halo stable expression line was plated in six-well tissue culture dishes and cultured in Hams F12 containing 10% fetal bovine serum, 1% glutamine and 1% penicillin–streptomycin, 500 μg ml⁻¹ hygromycin and 15 μg ml⁻¹ blasticidin, prior to induction. Cells at ~70% confluence were induced with tetracycline at 200 ng per mL 18–24 h before labeling and imaging.

Labeling and preparation of TIRF chambers

For labeling of intracellular HaloTags, cells were incubated with 0.5 nM of JF-646-Halo¹⁴ in Hams F12 containing 10% fetal bovine serum and 1% glutamine for 15 min at 37 °C in 5% CO₂, and subsequently incubated in fresh media for 1 hour before being resuspended in enzyme-free cell dissociation buffer (Millipore). After resuspension, the cells were washed 3 times in DPBS supplemented with 0.1% BSA.

Cells were then seeded on fibronectin-coated (0.1 μg/μL) (Sigma-Aldrich) glass coverslips (dimensions 22 x 22 mm, thickness 0.17 mm, SCHOTT Nexterion) and incubated in FluoroBrite DMEM medium (Thermo Fisher Scientific) for 1.5-2.0 h at 37 °C in 5% CO₂. Immediately before imaging, coverslips with seeded cells were assembled into a microscopy chamber⁴⁰ and washed with excess DPBS x10.

Coverslips were chemically cleaned by incubation for 1.5 h in a 1:0.25:5 solution of hydrogen peroxide and ammonium hydroxide in deionized water at 75 °C. Coverslips were subsequently rinsed in deionized water, followed by 100% ethanol and finally rapidly dried under a filtered air flow. The chemically cleaned coverslips were subsequently cleaned under argon plasma for 5 min immediately before being coated with fibronectin. After incubating with fibronectin (0.1 μg/μL) for 1 h at room temperature, the fibronectin-coated coverslips were washed x5 with PBS prior to addition of cells.

TIRF-based single-molecule imaging

All movies were recorded at a frame rate of 66 Hz at 25°C and saved in a 16-bit file format. Each movie frame has the size of 256 x 256 pixels. Image sequences were collected as described⁴⁰ using an objective-based TIRF microscope (IX81 with CellTIRF illuminator, Olympus) equipped with a 100× oil-immersion objective (100xUAPON NA 1.49, Olympus). JF-Halo-646 was excited by a 640 nm laser line (100 mW, Olympus). Fluorescence emission was separated from excitation light using a dual band filter set suited for Cy3 and Cy5 fluorophores (ZET532/640x, ZET532/640m, ZT532/640rcp, Chroma). Emission was further passed through a dual emission image splitter (OptoSplit-II, CAIRN) using an Optosplit filter cube for separating Cy3 and Cy5 emission (ET585/65m, ET655lp, zt640rdc) and projected onto an EMCCD camera (Evolve 512, Photometrics).

Experimental design – TIRF imaging

The unit of replication is the individual cell. No formal sample-size estimation or power calculation was performed; the number of cells per condition follows established practice for single-molecule diffusion measurements, in which each cell contributes thousands of independent single-molecule trajectories. Acquisition was not randomized and investigators were not blinded to condition, as the downstream analysis pipeline is automated and applies identical parameters to every condition.

Cells were selected for imaging at the time of acquisition, based on appropriate labeling density (Figure S1). Vehicle and agonist conditions were prepared from the same transfection and imaged on the same day. Experimental conditions were imaged in a single microscopy chamber, except for the condition co-transfected with 1200 ng D2R and treated with vehicle, where data were combined from two chambers.

Individual cell movies were excluded post hoc on analysis-based criteria. For Gα_i1-Halo and Gα_i1 4A-Halo combined, a total of 164 cell movies were included in this study from an initial 175. Eleven were excluded: 9 because the HMM analysis did not return a slow state diffusion coefficient due to insufficient occupancy of this state, and 2 because their fast and slow state occupancies were reversed in comparison to the 8-9 other cells in the same experimental group and were >2.4 SD from the mean occupancy, most consistent with a misclassification in the HMM-Bayes analysis.

Ensemble BRET

CAMYEL sensor assay

HEK-293-T cells were cultured in DMEM medium supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin.

For transient transfection, cells were plated in six-well culture plates at a density of 9 × 10⁵ cells per well. The following day, cells were transfected with linear polyethylenimine (PEI; Polysciences, Inc.) at a 1:1 PEI:cDNA ratio. Cells were transfected with D2R, CAMYEL, Gα_i1 or Gα_i1-Halo, Gγ₂, Gβ₁ and pcDNA5/FRT in a (1:12:2:6:6:13) ratio for a total of 3.33 μg in each well of a six-well plate. The medium was changed 24 h after transfection, and after approximately 40 h, the cells were washed with PBS, and two wells were combined for each dose response curve, by resuspension in a total of 4 ml BRET buffer (PBS containing 5 mM glucose) and distributed into 96-well OptiPlates (Perkin Elmer Life Sciences) at 45 μL per well.

Cells were incubated 2 min with 30 μM forskolin (Merck) and subsequently 8 min with 5 μM coelenterazine H (NanoLight Technologies), before addition of Dopamine hydrochloride to reach a final well volume of 100 μL.

BRET intensities were measured on a Pherastar FS plate reader (BMG) at 10 min after ligand addition. The donor (Rluc) and acceptor (YFP) emission was collected at 485 nm and 525 nm, respectively. The BRET signal was calculated as the ratio of light emitted at 525 nm over that emitted at 485 nm. The net BRET ratio was obtained by subtracting baseline BRET. Dose-response curves evaluating the ligand-induced net BRET ratios as a function of ligand concentration were fit by nonlinear regression to a sigmoidal dose-response relationship using Prism.

Receptor to Gβγ sensor assay

HEK-293 Cells were transfected in growth medium using linear polyethyleneimine MAX (PEI MAX; molecular weight 40,000) at a nitrogen to phosphate ratio of 20 and were used for experiments 24 h later. Cells were transfected with D2-Rluc8, Gα_i1 4A, Venus-1-155-Gγ₂, Venus-155-239-Gβ₁ and pcDNA3.1(+) or PTX-S1 in a (1:10:5:5:2) ratio for a total of 2.3 μg of plasmid DNA in each well of a six-well plate. After a 24 h incubation, cells were washed twice in PBS, collected by trituration in DPBS, and distributed into 96-well plates (0.2 mL per well). Measurements were made from intact cells with final concentrations of 10 μM quinpirole, 1 μM haloperidol or DPBS control.

Steady-state BRET were made using a Mithras LB940 photon-counting plate reader (Berthold Technologies GmbH) running MicroWin2000 software. Coelenterazine h were added to all wells immediately before making measurements. Raw BRET signals were calculated as the emission intensity at 520–545 nm divided by the emission intensity at 475–495 nm. Net BRET signals were calculated as the raw BRET signal minus the raw BRET signal measured from cells expressing only the Rluc8 donor. Concentration–response curves were fitted to a four-parameter binding logistic model using Prism.

All BRET measurements were performed at room temperature.

Analysis of TIRF-based single-molecule data

Wide-field movies were processed using u-track¹⁵. For setting the tracking parameters, we empirically determined initial parameter estimates for spot detection and optimized these based on manual inspection of consistency and quality of spot detection. Frame-to-frame linking and additional parameters were chosen based on quality of tracking as determined by manual inspection of output tracks and are listed in Table S4.

All trajectories obtained from u-track were further processed for motion type classification using Divide-and-Conquer moment-scaling spectrum (DC-MSS)¹⁶. DC-MSS divides those tracks which are at least 20 frames long into segments based upon maximum pairwise distance computed over rolling windows of time to accurately identify switch points between mobility types of immobile, confined, free (Brownian) or directed motion.

From the output of DC-MSS, we isolated those segments which were freely diffusing and contained no gaps, that would have been filled in the prior tracking step by u-track. These freely diffusing segments were then used as inputs to HMM-Bayes software²⁵, a software package which efficiently computes a model of diffusion states, combined of Brownian and directed motion, for the population of segments, as well as state-labels for each timepoint of the input segments²⁵. Given our hypothesis that the biological system would lead to a fast and a slow population of diffusion steps (unbound versus receptor-engaged G proteins), see Figure S3C-E for validation, we set the input for the maximum number of diffusion states in our model to 2.

Extraction of occupancy

The occupancy of either motion types (immobile, confined, free (Brownian) or directed) were calculated as:

$$\frac{\sum framesinmotiontype}{\sum framesinallmotiontypes}$$

The occupancy of either slow or fast states were calculated as

$$\frac{\sum framesineithersloworfaststate}{\sum framesinslowandfaststates}$$

Extraction of diffusion coefficients

The DC-MSS analysis package generates a diffusion coefficient for each motion type segment longer than 20 frames, based on a mean square displacement analysis. For each motion type we generated a mean diffusion coefficient per cell, by averaging the individual segment diffusion coefficients.

HMM-Bayes uses the displacements, or steps, along a freely diffusion segment (see Figure S3C-E), to infer one global diffusion coefficient per diffusion state. We calculated the diffusion coefficient for each state using the direct output of emission standard deviation (σ) and localization error (σ_ε) according to the guidelines provided in the HMM-Bayes documentation²⁵. Since we ran HMM for a collection of free segments per cell, we obtained one set of (σ) and (σ_ε) per state per cell.

Extraction of lifetimes

To avoid the possibility that we would inflate our estimate of state lifetimes by including trajectories of particles unable to transition between a slow and fast state, we limited our lifetime analysis to only those segments that contained more than two states. This led to average lifetimes for fast and slow states in free segments capable of transitioning between states.

QUANTIFICATION AND STATISTICAL ANALYSIS

In single-particle tracking experiments, each cell provides thousands of independent trajectories, and variance is dominated by cell-to-cell biological heterogeneity rather than differences between experimental sessions. While pooling trajectories across cells is common practice, doing so would obscure this variability. We therefore treat individual cells as independent replicates (n), providing a more conservative statistical framework. Thus, for all experiments, each replicate (n) corresponds to one cell imaged within one microscopy chamber for the duration of a movie (60 s). The number of cells contributing to each condition is stated in the corresponding figure legend.

Individual cell movies were excluded post hoc on analysis-based criteria. For Gα_i1-Halo and Gα_i1 4A-Halo combined, a total of 164 cell movies were included in this study from an initial 175. Eleven were excluded: 9 because the HMM analysis did not return a slow state diffusion coefficient due to insufficient occupancy of this state, and 2 because their fast and slow state occupancies were reversed in comparison to the 8-9 other cells in the same experimental group and were >2.4 SD from the mean occupancy, most consistent with a misclassification in the HMM-Bayes analysis.

MATLAB (MathWorks, versions R2016b-R2023a) was used to run all data analysis packages, including u-track, DC-MSS, and HMM-Bayes, and subsequently used to extract diffusion state occupancies, mean diffusion coefficients and lifetimes from DC-MSS as well as HMM-Bayes output.

Plotting, distribution fitting and statistics for data were carried out using GraphPad Prism Software (GraphPad, version 10.6.1). For statistical testing either an ordinary one-way or two-way ANOVA with Tukey’s post hoc test was performed for multisample comparisons, while an unpaired two-tailed t-test was used for two-sample comparisons. Each figure legend reports which test was applied. Only significant differences are indicated in Figures 1-4; all p values are reported in Tables S1-S3.

Definitions of center, dispersion and precision are given in the legend to Figure 1. In all figures, box plots indicate the median (central line), overall mean (central +) and interquartile range (lower and upper lines represent the 25th and 75th percentiles), with whiskers defined according to the Tukey criterion.

Supplementary Material

Document S1. Figures S1-S4 and Tables S1-S4.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
Escherichia coli: BL21-Gold competent cells	Agilent Technologies	Cat#230130
Chemicals, peptides, and recombinant proteins
Quinpirole	Tocris	Cat#1061
Dopamine hydrochloride	Sigma	Cat#H8502
S(−)-Sulpiride	Tocris	Cat#0895
Sodium bisulfite	Thermo Fisher Scientific	Cat#419440025
Forskolin	Merck	Cat#F3917
NEBuilder HiFi DNA Assembly Master Mix	NEB	Cat#E2621S
Lipofectamine LTX	Thermo Fisher Scientific	Cat# A12621
JF-646-Halo	Grimm et al.14	N/A
Enzyme-free cell dissociation	Millipore / Merck	Cat # S-014-M
Fibronectin	Sigma	Cat#F4759
FluoroBrite DMEM medium	Thermo Fisher Scientific	Cat#A1896701
Coelenterazine h	NanoLight Technologies	Cat#20-0770
linear polyethyleneimine MAX (PEI MAX; molecular weight 40,000)	Polysciences	Cat#24761-1
Haloperidol	Thermo Fisher Scientific	Cat#0931
Experimental models: Cell lines
Flp-in CHO cells	Thermo Fisher Scientific	Cat#: R758-07; parent line CHO-K1, ATCC# CCL-61
LEx-Flp-In T-REx (FITR) CHO cells	Asher et al.8	generated from T-REx-CHO cells, Thermo Fisher Scientific, Cat. # R718-07; parent line CHO-K1, ATCC# CCL-61
LEx-FITR CHO line stably expressing D2R-Halo upon tetracycline induction	This paper	N/A
HEK-293-T cells	ATCC	CRL-3216
HEK-293	ATCC	CRL-1573
Recombinant DNA
pcDNA3.1 Gαi1-Halo	This paper	N/A
pcDNA3.1 Gαi1 4A-Halo	This paper	N/A
pcDNA5/FRT/TO-IRES kozac −3T+4T D2R-Halo	This paper	N/A
pcDNA3.1 myr-Halo	This paper	N/A
Plasmid: wt D2R short	Donthamsetti et al.33	N/A
Plasmid: wt Gαi1	cDNA.org	Cat#GNAI100000
Plasmid: PTX-S1	Jang et al.11	N/A
Plasmid: CAMYEL	Jiang et al.34	N/A
Plasmid: Gβ1	cDNA.org	Cat#GNB0100000
Plasmid: Gγ2	cDNA.org	Cat#GNG0200000
Plasmid: D2-Rluc8	Donthamsetti et al.33	N/A
Plasmid: Gαi1 4A	Jang et al.11	N/A
Plasmid: Venus-155-239-Gβ1	Hollins et al.35	N/A
Plasmid: Venus-1-155-Gγ2	Hollins et al.35	N/A
pcDNA3.1+	Thermo Fisher Scientific	Cat#V79020
pOG44 Flp-In Recombinase vector	Thermo Fisher Scientific	Cat # V600520
Software and algorithms
MATLAB 2023a	MathWorks	https://www.mathworks.com
u-track	Jaqaman et al.15	https://github.com/kjaqaman
DC-MSS	Vega et al.16	https://github.com/kjaqaman
HMM-Bayes	Monnier et al.25	https://github.com/lcbb/HMM-Bayes
GraphPad Prism	GraphPad Software	https://www.graphpad.com

Significance.

Receptor–G protein structures look similar regardless of ligand efficacy, suggesting that efficacy is set by dynamics rather than structure. How long an activated receptor remains bound to a G protein in a living cell is unknown, and reported values span milliseconds to seconds. Ensemble kinetics report how a population approaches steady state, not the lifetime of a single complex. Single-molecule colocalization only detects interactions long enough to outlast random proximity, biasing estimates upward and conflicting with the millisecond timescale of physiological responses. To get below this floor, we imaged individual G proteins and inferred engagement from their transitions between fast and slow diffusion states. Activated dopamine D2 receptors (D2R) did not detectably change Gα_i1-Halo state lifetimes. To observe engagement, we turned to Gα_i1 4A-Halo, a mutant that cannot dissociate through nucleotide exchange; co-expression of D2R increased the fraction of time it spent in the slow diffusion state with a lifetime of ~0.8 s. A full agonist increased how often the G protein entered this state without changing the duration of each engagement, while an inverse agonist or pertussis toxin shifted the balance back toward fast diffusion. Slowing alone is not evidence of binding: a lipid-anchored control that cannot engage receptors also entered a slow state, but with a ~0.5 s lifetime. Because receptors and G proteins can slow in the same membrane domains, this background can mimic binding in colocalization assays. Dissociation of the native G protein is likely promoted by nucleotide exchange, which Gα_i1 4A-Halo cannot undergo; association, which precedes nucleotide release, should be less affected. These findings are consistent with agonist efficacy being reflected in the probability of productive receptor engagement rather than in complex stability alone, a distinction ensemble measurement cannot make.

Highlights.

1. Diffusion-based single-molecule method reveals GPCR-G protein kinetics in live cells

2. Native GPCR-Gαi1 interactions are not resolved at 15 ms time resolution

3. Membrane anchoring alone drives slow diffusion states independent of receptor binding

4. Full agonist increases Gαi1 4A-GPCR association rate without affecting dissociation