Immobilization of cell-derived nanovesicles for single-molecule imaging of membrane proteins.

a Schematic illustrating both sample preparation and the imaging setup. Briefly, cells expressing the membrane protein of interest (e.g. GFP-TAX-4) are fragmented using N2 cavitation to spontaneously form nanoscale vesicles, some of which contain the protein of interest in a mixture of inward and outward facing orientations. Vesicles comprised of membrane from either the plasma membrane (PM) or endoplasmic reticulum (ER) are further separated by gradient ultracentrifugation. The fraction of PM vesicles is applied to a sample chamber for immobilization and imaging with micro-mirror total internal reflection (mmTIRF) microscopy. b Illustration of an individual immobilized vesicle in the sample chamber. The chamber consists of a glass coverslip coated with a layer of PEG doped with PEG-biotin to which a biotinylated anti-GFP nanobody (bait protein) is attached via streptavidin. Vesicles containing GFP-TAX-4 oriented such that GFP is exposed to the extravesicular solution are immobilized at nanobody locations on the optical surface. TIRF excitation, indicated by the blue gradient, ensures that bulk freely diffusing fluorescent ligand (e.g. fcGMP) above the vesicle layer are not appreciably excited. Fluorescence from the vesicle layer is imaged on an EMCCD as depicted to the left. Note that the vesicle is not drawn to scale as it would on average have a diameter about 20-fold larger than GFP-TAX-4.

Single-molecule imaging of fcGMP binding to GFP-TAX-4 in cell-derived nanovesicles.

a Time-averaged fluorescence for GFP (green) overlaid with fcGMP (magenta). Colocalized GFP and fcGMP signals appear white. From left to right depicts sequential epochs for the same field of view showing binding in 10 nM fcGMP, block of specific binding by coapplication with an excess of 500 μM non-fluorescent cGMP, and recovery upon removal of cGMP. The yellow box is expanded to the right of each image and arrows indicate locations exhibiting specific fcGMP binding to GFP-TAX-4. Below each image are time series for fcGMP fluorescence at a single colocalized spot during each of the epochs in the corresponding images above. Transient increases in fluorescence reflect individual fcGMP binding events in the vesicle layer at the optical surface. The shaded region is expanded below. b The frequency of binding events in 10 nM fcGMP is greatly diminished in the presence of 500 μM competing non-fluorescent cGMP.

Number of GFP-TAX-4 subunits per vesicle assessed by photobleaching of GFP.

a Stepwise photobleaching of GFP fluorescence from a single diffraction-limited spot. b The number of observed bleach steps for colocalized spots included in the analysis. The good fit to a binomial distribution with four sites suggests that most spots contain a single tetrameric channel where bleaching of each GFP is observed with a probability of 0.82 and 95% CI [0.74, 0.89].

Concentration-dependent fcGMP binding.

a Time series for fcGMP binding to single TAX-4 containing nanovesicles at increasing concentrations of fcGMP. Horizontal dashed lines indicate approximate fluorescence levels reflecting stepwise binding of one or two fcGMP molecules. b Idealization (black) of fluorescence (magenta) time series for the number of bound fcGMP at an individual molecule in 200 nM fcGMP (see Methods). c Bound probability distributions for all tested concentrations and fits to a binomial distribution assuming four identical and independent sites. Bound probability [95% CI] per site from the binomial fits: 10 nM = 0.08 [0.079, 0.081], 30 nM = 0.10 [0.09, 0.11], 60 nM = 0.12 [0.10, 0.14], 100 nM = 0.21 [0.20, 0.22], 200 nM = 0.26 [0.24, 0.28]. d Average fcGMP bound probability across all molecules and normalized ionic current46 as a function of fcGMP or cGMP concentration, respectively. Shaded area represents SEM (partially hidden by the line width). Current values are for fits of the Hill equation to inside-out patch clamp recordings in Komatsu et al. (1999)46.

Bound and unbound dwell time distributions and correlations.

a Bound and unbound fcGMP dwell time distributions across all molecules from idealized records (see Methods). b Dwell time distributions for data (gray) overlaid with mono-(blue dashed) and bi-exponential (red dashed) maximum likelihood fits (see Supplementary Table 1). c Weighted time constants from biexponential fits of bound (black) and unbound (gray) dwell times as a function of fcGMP concentration with associated 95% confidence intervals (shaded area). d Correlation between the duration of sequential singly bound events i and i+1 within individual molecules. The concentration of fcGMP is the same as for the dwell time distributions to the left. Color bar denotes number of events. If short and long events arise from distinct populations of molecules, we would expect to observe clusters of events primarily along the dashed diagonal, whereas events on the off-diagonal represent sequential short and long bound durations within individual molecules.

Models for the first and second binding steps.

a The preferred model for binding dynamics at individual CNBDs. This model depicts both a ligand association step (horizontal transition) as well as a conformational change of the CNBD in both unliganded and liganded states (vertical transitions). The dashed arrows indicate transitions that have a relatively smaller contribution to the observed dynamics. Postulated structures for the ligand-bound conformations are shown to the right. They are based on cryo-EM structures of TAX-4 in cGMP-bound (green) or unliganded (blue) conformations34, 35, where cGMP has been added to the unliganded structure in the position that it is found in the bound structure, and arrows denote the observed movement of the C helix between the two structures. Capping of the binding site by the C helix limits binding and unbinding, similar to dynamics observed at isolated CNBDs from HCN2 channels49. See Supplementary Fig. 9 and Supplementary Table 2 for all explored models of individual CNBDs and their optimized rate constants. b-c Two potential models depicting binding of the first two ligands and either a global conformational change of all four CNBDs in each ligation state (b) or individual conformational changes of each ligand-bound CNBD (c). We were unable to distinguish between these mechanisms. See Supplementary Fig. 10 and Supplementary Table 3 for all explored models of the first two binding steps and their optimized rate constants.

Transmission electron microscopy (TEM) images of cell-derived nanovesicles.

Vesicle diameters in TEM images of a standard vesicle preparation (see Methods) ranged from approximately 50–1000 nm in diameter similar to previous reports based on dynamic light scattering1. Arrow indicates a single vesicle with a diameter of ∼150 nm.

Western blot of plasma membrane fraction labeling.

The left-most lane corresponds to molecular weight markers used as a standard. The same amount of nanovesicle lysate from fractions 1-8 following density gradient ultracentrifugation was loaded into the corresponding wells. Fractions containing nanovesicles derived from plasma membrane were detecting using fluorescent antibodies specific to Na+/K+-ATPase, which were enriched in plasma membrane fractions 5-8. These data confirm an earlier report2.

Specific immobilization of nanovesicles containing a membrane protein of interest.

a Cartoon illustrating topology of a GFP-TAX-4 subunit with intracellular cyclic nucleotide binding domain and vesicles derived from cells expressing GFP-TAX-4 tetrameric channels. Below are mmTIRF images of GFP fluorescence (green) showing specific immobilization of GFP-TAX-4 containing vesicles at GFP-nanobodies deposited on the imaging surface (left) and a lack of appreciable nonspecific surface adsorption of vesicles in the absence of GFP-nanobodies (right). In both cases chambers were incubated in GFP-TAX-4 containing vesicles and then rinsed with buffer before imaging. b Cartoon illustrating topology of GABAA receptor α1, β2 and γ2L subunits and vesicles derived from cells expressing heteropentameric channels. For visualization and immobilization mScarlet was inserted in the intracellular M3-M4 linker of α1 (α1-mScarlet) and EGFP was inserted near the N-terminus of β2 (β2-GFP) on the same side of the membrane as the extracellular agonist and benzodiazepine binding domains. Below are mmTIRF images of GFP (green) and mScarlet (red) fluorescence showing colocalization (yellow) of β2-GFP and α1-mScarlet subunits at vesicles immobilized with GFP-nanobodies and a lack of appreciable nonspecific surface adsorption of vesicles in the absence of GFP-nanobodies.

Distinct properties between colocalized and non-colocalized fcGMP binding events.

a A comparison of the mean and standard deviation of the fluorescence signal from individual fcGMP binding events at spots that either do or do not colocalize with GFP-TAX-4 in a single field of view. The non-colocalized binding events are likely due to fluorescent contaminants in the PEG layer or non-specific adsorption of fcGMP to imperfections in the surface passivation. b Representative traces for low-intensity non-colocalized fcGMP signals, which make up a majority of the nonspecific signal. c Representative trace for high-intensity non-colocalized fcGMP signal likely due to non-specific adsorption to glass imaging surface. These high intensity traces make up a minority of observed non-colocalized signals.

Summary of GFP and fcGMP colocalization for vesicles with GFP-TAX-4 and controls for vesicles with TRPV1-GFP channels.

a Time averaged fluorescence for GFP (green) from nanovesicles containing either GFP-TAX-4 (left) or TRPV1 channels fused with an intracellular GFP (right) overlaid with 10 nM fcGMP (magenta) from the same field of view (two separate imaging locations shown for each construct). Colocalized GFP and fcGMP signals appear white. Note the relative absence of colocalization for TRPV1 channels as compared to TAX-4 channels. Also note that some of the fainter green spots are subthreshold as only the brighter more distinct spots were observed in the presence but not absence of GFP. b Histograms of the nearest neighbor distance from each identified GFP location to the closest identified fcGMP location. Two separate examples are shown for both GFP-TAX-4 and TRPV1-GFP vesicle preparations. The percentage (mean ± standard deviation) of GFP spots that colocalized with fcGMP spots (i.e. were within 3 pixels) across the dataset were for GFP-TAX-4 (at each fcGMP concentration): 33 ± 9% (10 nM), 32 ± 9% (30 nM), 31 ± 7% (60 nM), 33% (100 nM), and 39 ± 3% (200 nM). For TRPV1-GFP at 10 nM fcGMP a background colocalization of 7 ± 0.5% was observed.

No colocalization between immobilized GABAAR containing nanovesicles and fcGMP.

Time averaged fluorescence for GFP (green) from nanovesicles containing either GFP-TAX-4 (left) or GABAA receptor α1, β2 and γ2L subunits with GFP inserted in the N-terminus of the β2 subunit (β2-GFP) (right) overlaid with 100 nM fcGMP (magenta) from the same field of view. Colocalized GFP and fcGMP signals appear white. See Supplementary Fig. 3 for a description of the GABAAR subunits.

Evaluating the accuracy of idealization of fluorescence time series for ligand binding using simulated data.

a Distributions of baseline noise, event noise, and event amplitudes for experimental binding data from single, isolated events at 30 nM fcGMP. b Example of time series from independent site binding simulations, followed by addition of gaussian noise and event amplitude heterogeneity representing isolated event data (a), followed by the results of our idealization procedure. Comparison of the known simulated time series to the results of our idealization procedure after adding noise drawn from experimental observations provides a metric for testing the accuracy of our idealization procedure (see Methods in main text). c Example of binding data at 30 nM fcGMP, followed by the results of our idealization procedure. d-f Same as described for a-c, but for simulations and data at 200 nM fcGMP.

Missed events during idealization of simulated fluorescence time series.

a-b Simulated bound dwell time distributions of isolated singly bound events either including (a) or excluding (b) single frame events. Note that durations are shown in frames. Comparison of known simulated bound series (blue) to the idealized bound series obtained from idealization after adding experimentally relevant noise and event heterogeneity (red). The primary discrepancy is limited to events lasting only a single frame, which are sometimes missed during the idealization procedure. However, nearly all events lasting two or more frames are reliably detected. c-d Same as described for (a-b), but for dwell times in doubly bound states. All simulations were run with a frame rate of 20 Hz (50 ms per frame), equivalent to experimental recordings.

Single site dynamics for the first binding event.

a Schematic illustrating binding to the first of four sites per channel. CNBDs depicted in unbound (open circles) and bound (filled circle) conformations. b Evaluated single site models describing binding and conformational exchange between distinct bound or unbound states. State names indicate the number of bound ligands and an asterix denotes distinct states having the same number of bound sites. Rate constants are in μM−1s−1 for binding transitions (0 → 1, 0 → 1∗, 0∗ → 1∗) and s−1 for all other transitions. [L] indicates ligand concentration for binding steps. Rate constants were optimized in QuB for data segments comprised of isolated binding events (no stacked events) across all molecules and concentrations (see Methods in main text). For models M1.D and M1.F one of the rate constants was constrained to enforce microscopic reversibility in the loop. Rate constants and their estimated errors are given in Supplementary Table 2. c BIC scores for the models shown in C relative to the model with the best score (smaller score is better; ΔBIC = BIC – BICbest model).

Dynamics for the first two binding events.

a Schematic illustrating binding to the first two of four sites per channel. CNBDs depicted in unbound (open circles) and bound (filled circle) conformations. b Evaluated two site models describing binding and conformational exchange between distinct bound or unbound states. State names indicate the number of bound ligands and an asterisk denotes distinct states having the same number of occupied sites. Rate constants are in μM−1s−1 for binding transitions (0 → 1, 1 → 2, 1 → 2†, 0∗ → 1∗, 1∗ → 2∗) and s−1 for all other transitions. [L] indicates ligand concentration for binding steps. M2.A is the simplest possible scheme for sequential binding at two sites. M2.B allows for two distinct di-liganded states (e.g. adjacent and diagonally opposed occupied sites as depicted in a). M2.C and M2.D extend M2.A with a global conformational change of both sites (between state % and %*). M2.E describes binding followed by a conformational change at each site separately. Note that only model M2.B distinguishes between adjacent and diagonal doubly bound conformations in the channel tetramer. Each model is further subdivided into several models sharing the same schematic but differing in the applied constraints. For example, M2.Ai and M2.Ac both have the M2.A structure, but Ai constrains the two binding steps to be identical and independent, whereas Ac allows the dynamics for the two steps to differ (i.e. allows cooperativity between binding sites). Models Ci and Cc share the same constraints as Ai and Ac. M2.Eii assumes completely independent and identical sites, whereas M2.Eci allows for binding of the 2nd ligand to differ from that of the first (binding cooperativity), and M2.Eic allows the conformational change following binding at either site to depend on the number of bound ligands. See Supplementary Table 3 for a description of all applied constraints. Rate constants were optimized in QuB for the entire data set across all molecules and concentrations excluding events with more than two bound ligands (see Methods in main text). Rate constants and their estimated errors are given in Supplementary Table 3 along with model constraints. c BIC scores for each model relative to the model with the best score (smaller score is better; ΔBIC = BIC – BICbest model). See description above for model subscripts.

Comparison of observed dwell time distributions with predictions for two-site models.

Experimentally observed bound and unbound dwell time distributions (gray) overlaid with predicted distributions from simulated bound time series for each two-site model (colored; see Supplementary Fig. 10). For each ligand concentration, simulations matched the length of data collected.

Comparison of experimental data with simulations for independent CNBDs.

a Four simulated single-site bound time series were added together to generate bound seeries representative of tetramers comprised of independent CNBDs. b Gaussian noise and event amplitude heterogeneity were added to simulated bound series to reflect experimental observations, and the noisy simulations were idealized to obtain bound series. c Experimental fcGMP binding fluorescence traces idealized to obtain estimated bound series in exactly the same way as for the simulated data in (b). Dwell time distribution in singly bound (blue; one of four CNBDs occupied) and doubly-bound (red; two of four CNBDs occupied) states are shown for both known simulations (a), idealized simulations after adding noise (b), and experimental fcGMP fluorescence traces (c). The reduction in long-lived doubly bound events as compared to singly-bound events in the simulated data is a consequence of the fact that unbinding of either ligand will exit a doubly-bound state.

Latency between 1st and 2nd binding events.

Distribution of latencies to binding of the second ligand after binding the first ligand. Comparison between our experimental data (magenta) and simulations for four independent sites (blue, see Methods in main text

Photobleaching of single fcGMP molecules.

Lifetimes of individual fcGMP molecules at noncolocalized spots assumed to largely reflect dye adsorption to the surface and subsequent terminatin by photobleaching. The abscissa is limited to the first 100 seconds for visualization, although lifetimes up to 143 seconds were observed. Lifetimes of colocalized binding events at TAX-4 channels in 30 nM fcGMP are shown for comparison. The mean lifetime for noncolocalized events was 23.4 seconds, an order of magnitude longer than the time constant for the longest duration bound component (Supplementary Table 1).

Time constants (τ) and relative weights (A) for biexponential maximum likelihood fits to bound and unbound dwell time distributions across tested fcGMP concentrations.

Rate constants and relative BIC scores for single site models (Supplementary Fig. 9). Units are μM-1s-1 for binding transitions (k0→1, k0→1*∗, k0*→1*) and s-1 for all other transitions. Rate constants were optimized in QuB for data segments comprised of isolated binding events (no stacked events) across all molecules and concentrations (see Methods in main text). For models M1.D and M1.F one of the rate constants was constrained to enforce microscopic reversibility in the loop. Errors are standard deviations across optimized rate constants for five randomized folds of the data set where each fold contained ∼20% of the data for each concentration.

Rate constants and relative BIC scores for models of the first two binding steps (Supplementary Fig. 10). Units are μM-1s-1 for binding transitions (k0→1, k1→2, k0*→1*, k1*→2*) and s-1 for all other transitions. Gray shaded cells indicate constraints, and all other cells were free parameters. All constraints for k1→2 and k2→1 reflect the statistical factors associated with independent binding at each site in a tetramer, whereas otherwise the binding steps were allowed to differ (i.e. be cooperative). Constraints for model M2.C transitions ki→i* and ki*→i reflect the additive effect of ligand binding on the conformational change between states i and i* as depicted by the factors α and β in Supplementary Fig. 10. For all three M2.D models, constraints on transitions not involving ligand binding/unbinding imply that the ratio of the transition rate for the conformational change in doubly liganded states (2 → 2*) to that in singly liganded states (1 → 1*) is given by %, and the factor of 2 is a statistical factor given that either site could undergo the conformational change. Likewise, the ratio of the reverse transition rates for 2* → 2 and 1* → 1 is given by β. Constraints for all M2.D models not shown in the table are k1*→2* = αk1→2, k2*→1* = (β⁄2) k2→1, k2*→2** = (α⁄2) k2→2* and k2**→2* = 2β k2*→2. Rate constants were optimized in QuB for the entire data set across all molecules and concentrations excluding events with more than two bound ligands (see Methods in main text). Errors are standard deviations across optimized rate constants for five randomized folds of the data set where each fold contained ∼20% of the data for each concentration.