A single-molecule method for measuring fluorophore labeling yields for the study of membrane protein oligomerization in membranes

Abstract

Membrane proteins are often observed as higher-order oligomers, and in some cases in multiple stoichiometric forms, raising the question of whether dynamic oligomerization can be linked to modulation of function. To better understand this potential regulatory mechanism, there is an ongoing effort to quantify equilibrium reactions of membrane protein oligomerization directly in membranes. Single-molecule photobleaching analysis is particularly useful for this as it provides a binary readout of fluorophores attached to protein subunits at dilute conditions. However, any quantification of stoichiometry also critically requires knowing the probability that a subunit is fluorescently labeled. Since labeling uncertainty is often unavoidable, we developed an approach to estimate labeling yields using the photobleaching probability distribution of an intrinsic dimeric control. By iterative fitting of an experimental dimeric photobleaching probability distribution to an expected dimer model, we estimate the fluorophore labeling yields and find agreement with direct measurements of labeling of the purified protein by UV-VIS absorbance before reconstitution. Using this labeling prediction, similar estimation methods are applied to determine the dissociation constant of reactive CLC-ec1 dimerization constructs without prior knowledge of the fluorophore labeling yield. Finally, we estimate the operational range of subunit labeling yields that allows for discrimination of monomer and dimer populations across the reactive range of mole fraction densities. Thus, our study maps out a practical method for quantifying fluorophore labeling directly from single-molecule photobleaching data, improving the ability to quantify reactive membrane protein stoichiometry in membranes.

Fig 1. Expected photobleaching probabilities for monomers and dimers at the single-molecule limit.

(A) The equilibrium dimerization reaction of two CLC-ec1 monomers (2M) forming a dimer (D) in the cellular membrane. In the labeling model, we consider that there is a high probability of labeling at an exposed cysteine site (P_site) and a lower probability of non-specific labeling contributing to the background (P_bg), which is measured in the protein sample without the cysteine. Experimentally the total labeling yield per subunit is measured, P_fluor = P_site + P_bg for each protein sample. (B) Photobleaching of the fluorophores attached to the protein happens in a step wise fashion when imaged using a TIRF setup. (C) The monomeric and dimeric photobleaching probability distributions for labeling yields with P_bg = 0.1, and total P_fluor varied from 0.4 to 1.0, at a single-molecule density of χ_rec. = 1 x 10⁻⁶ subunits/lipid, i.e. << 1 protein species per liposome. Data reported as mean ± SE, for n = 3 simulation replicates. (D) Plot of P₁ vs. P₂ photobleaching probabilities showing the dependency of the dimer signal on the labeling yield and overall dynamic range between dimer and monomers.

Fig 2. Single-molecule photobleaching estimation of (P_fluor, P_bg) for covalently cross-linked CLC-ec1 dimers across a wide range of protein densities.

(A) Heatmaps of the inverse normalized sum of squared residuals (Norm. SSR^-1) over the fluorophore labeling parameter space of (P_fluor, P_bg). The experimental photobleaching data used in this analysis are from [25] for the covalently cross-linked R230C/L249C CLC-ec1 dimer, for a single sample (n₁) across a 5-magnitude range in mole fraction densities: χ₁ = 2 x 10⁻⁹, χ₂ = 2 x 10⁻⁸, χ₃ = 2 x 10⁻⁷, χ₄ = 2 x 10⁻⁶, χ₅ = 2 x 10⁻⁵ subunits/lipid. Maximum value, corresponding to the (P_fluor, P_bg) pair that best fits the experimental data is indicated by the red "x". (B) Heatmaps of the probability distribution of the sum of squared residuals, P_SSR. (C) Bootstrapping analysis from the P_SSR distribution. The sampling number, N, is set to 10⁷. The mode and standard deviation around the mode, σ, from each bootstrapped distribution are marked in panel (B) with the circle and error bars, respectively. (D) Mode ± σ (circle ± error bars) and max values (red "x") of P_fluor and P_bg from the bootstrapping analysis along with SSR values compared to the experimental data.

Fig 3. Single-molecule photobleaching estimation of fluorophore labeling recapitulates sample variability.

(A) Heatmap of P_SSR over the parameter space of (P_fluor, P_bg) for different experimental samples, n, of the covalently cross-linked R230C/L249C CLC-ec1 dimer. The circle and error bars reflect the mode ± σ from the bootstrapping analysis. (B) The maximum and mode values of P_fluor and P_bg, along with SSR compared to the experimental values, dotted line. (C) Global fit of (P_fluor, P_bg) obtained from pooling experimental data for samples n = 1–5. (D) Experimental photobleaching data (P₁, P₂, P₃₊), along with the modeled probability distribution (circles) using (P_fluor, P_bg) = (0.72,0.10) corresponding to the maximum value from the global fit in panel (C).

Fig 4. Estimating K_D values for CLC-ec1 dimerization in membranes.

(A) P_SSR for dimeric glutaraldehyde cross-linked WT, reactive WT, reactive I422W, ’W’, and monomeric I201W/I422W, ’WW’, as a function of the dissociation constant parameter K_D. Curves reflect the model using the experimental labeling parameters (P_fluor, P_bg)_expt.—black or the R230C/L249C fitted labeling parameters (P_fluor, P_bg)_max—cyan, (P_fluor, P_bg)_boot.−red. Dotted line represents the maximum P_SSR value and best-fit K_D using the (P_fluor, P_bg)_max labeling parameters, and the yellow box reflects the uncertainty based on the bootstrapping analysis. (B) Range of photobleaching probability distribution (P₁, P₂, P₃₊) simulated using (P_fluor, P_bg)_max and K_D,boot.—σ, K_D,boot., K_D,boot. + σ and agreement with experimental data (white circles). (C) ΔG⁰ for WT and W [20] based on least-squares estimation of F_Dimer from expected monomer and dimer photobleaching probability distributions compared to the direct fitting of the photobleaching probability distribution while iterating over K_D as a parameter. ΔG⁰ = −RTln(K_eqχ^°), where K_eq = 1/K_D and χ^° = 1 subunit/lipid represents the mole fraction standard state. Results shown for the best-fit, ’max’, value as well as the mode of the bootstrapping analysis for (P_fluor, P_bg)_expt.—grey, compared to the R230C/L249C fitted labeling parameters (P_fluor, P_bg)_max—cyan, (P_fluor, P_bg)_boot.—red. All data are shown as mean ± SE of fits of independent data sets and statistical significance is calculated via t-test.

Fig 5. The operational range for fluorophore labeling yields.

(A) Chi-squared (χ²) analysis between monomer and dimer model photobleaching probability distributions over (P_fluor, P_bg) fluorophore labeling parameter space for P_bg < P_site = P_fluor—P_bg. χ² heatmaps are shown for mole fraction densities χ₃ = 2 x 10⁻⁷, χ₄ = 2 x 10⁻⁶, χ₅ = 2 x 10⁻⁵ subunits/lipid. (B) Heatmaps of null hypothesis testing for the χ² values in (A) for alpha = 0.001 significance. The null hypothesis is that the monomer and dimer distributions are the same, where H = 1 (yellow) indicates a rejection of the null hypothesis, and H = 0 (purple) indicates that the null hypothesis cannot be rejected at the indicated significance level. (C) P₁ vs. P₂ for monomer (white) and dimer (orange) models for P_fluor = 0.3 to 0.9 and P_bg = 0.1. Symbol size increases with increasing P_fluor with endpoints labelled as shown. Error bars depict representative standard deviation values of the experimental data for RCLC, std = ± 0.06. (D) Maximal scalar distance, R_max, between (P₁, P₂) signals from monomer and dimer model distributions as a function of mole fraction density. The background labeling yield is set to P_bg = 0.1. The dotted line indicates R_max = 0.25 that corresponds to the significance testing cutoff in the χ² analysis in (B).