Stoichiometric analysis of protein complexes by cell fusion and single molecule imaging

The composition, stoichiometry and interactions of supramolecular protein complexes are a critical determinant of biological function. Several techniques have been developed to study molecular interactions and quantify subunit stoichiometry at the single molecule level. However, these typically require artificially low expression levels or detergent isolation to achieve the low fluorophore concentrations required for single molecule imaging, both of which may bias native subunit interactions. Here we present an alternative approach where protein complexes are assembled at physiological concentrations and subsequently diluted in situ for single-molecule level observations while preserving them in a near-native cellular environment. We show that coupling this dilution strategy with fluorescence correlation spectroscopy permits quantitative assessment of cytoplasmic oligomerization, while stepwise photobleaching and single molecule colocalization may be used to study the subunit stoichiometry of membrane receptors. Single protein recovery after dilution (SPReAD) is a simple and versatile means of extending the concentration range of single molecule measurements into the cellular regime while minimizing potential artifacts and perturbations of protein complex stoichiometry.


FCS/FCCS data analysis.
For non-PKA experiments, data was fit to a single component diffusion with triplet model: where τT and FT are the triplet time and fraction, respectively, τd is the diffusion time, S is the structure factor for the focal volume and G(0) is the correlation at τ = 0. The structure factor was set to 10 for all fits. From the two-color cross-correlation measurements, the average number of particles was determined using: where NG,R is the number of green or red particles, and NX/NG and NX/NR are the heterodimer fractions. Absolute concentrations for cytoplasmic mNeonGreen were obtained by calibrating the focal volume with known concentrations of Alexa488.
For Protein Kinase A experiments, PKA-transfected U2OS cells were mixed 1:10 with nonexpressing VSVG cells and incubated in doxycycline-supplemented Fluorobrite DMEM for 24 hours. Cells were then fused by a 5-minute incubation in fusion buffer and FCS was performed in syncytia one hour later. In order to maintain the same syncytial position for post-stimulation measurements, 2x cAMP-stim buffer (50 μM forskolin, 200 μM IBMX in Fluorobrite DMEM) was added directly to the imaging dish in equal volume to the residual media and a second FCS recording was initiated 5 minutes later. PKA data was fit to a two-component diffusion model: where τdi are the diffusion times for the respective fractions (i = 1, 2), α is the fractional contribution of the first component and all other parameters are as defined above.
Single particle tracking of membrane proteins in giant syncytia. Membrane proteins are typically free to diffuse in the plasma membrane unless tethered to larger intracellular structures.
To examine protein mobility in giant syncytia, U2OS cells expressing mNG-ADRβ2 were co-plated with non-expressing VSVG cells and immersed in fusion buffer for 5 minutes before returning them to Fluorobrite DMEM and incubating at 37°C for 75 minutes. Cells were imaged at 30 fps on a lab-built TIRF microscope at 37°C with an EMCCD (iXon 887, Andor).
Single molecule trajectories were analyzed using the ImarisTrack module in Imaris Bitplane. The raw data was temporally averaged by five frames to improve signal-to-noise, yielding a substack interval of 150 ms, and SPT parameters (PSF quality and size) were chosen based on the dataset. Spots were detected in each substack frame and trajectories were assembled with an autoregressive motion model. 4580 trajectories lasting longer than five substack frames were generated and included for further analysis. MSD plots were fit to extract single particle diffusion coefficients (MSD = 4Dt), which were distributed exponentially. Trajectories were sorted by their diffusion coefficient, heuristically classified into slow, intermediate and fast motions (bottom, middle and top thirds) and average MSD plots were calculated for each of these fractions.
mNeonGreen folding efficiency and true oligomer calculations. Our FCS data of mNeonGreen (mNG) monomers and covalent dimers permitted an estimation of the fluorescent protein maturation efficiency, denoted as f. In the case of monomers, misfolded proteins are not detected and thus, the detected brightness-per-particle is equal to the brightness of the monomeric species. For simplicity, we normalize this to 1.
In the case of dimers, we observe a mixture of monomers and dimers. The overall brightnessper-particle (count rate divided by average number of particles) is given by: where the prefix to each term denotes the probability of an mNG dimer having one or two fully mature fluorescent domains. Assuming B2 = 2B1 = 2, this simplifies to: Using our normalized value for Bmix (1.7, the observed count rate for dimers divided by count rate for monomers), we estimate that mNeonGreen folds to 80 -85% efficiency.
We can use this fraction to estimate the true dimer propensity of ADRβ2, which was largely present in monomeric and dimeric forms. If we denote the true monomer and dimer fractions as χ1 and χ2, then we obtain the relation 1 + 2 = 1 assuming no higher-order oligomers exist.
The "emitter fractions" χi ' are given by considering that each ADRβ2 oligomer can have properly folded or misfolded mNG domains. The fractions are given by: It is important to note that these emitter fractions are not the directly observed quantity (largely because χ0 ' cannot be detected). Instead, we detect For ADRβ2, φ = 0.75. From the above system of linear equations, we can uniquely determine the values of the "true" monomer and dimer fractions, χ1 and χ2 (0.64 and 0.36 for ADRβ2).
Generalized Dark Fraction Correction. Assuming a constant fluorescent protein dark fraction independent of oligomerization state, the observed fractions of each state can be corrected for the presence of dark protein components as an estimate of the actual fraction. p = fraction of fluorescent proteins (0 to 1.0)

-p = dark fraction
The fractional losses of each n-mer state are given by binomial coefficients. These form a series of linear equations that are solved to correct the observed fractions. The matrix form, written here for a maximum oligomer state of 6 (k): Where the upper triangular matrix (U) elements are: The vector b is the observed number of steps (oligomer subunits) and the vector a is the corrected number of subunits per complex. Ua = b can be solved for a by back-substitution since all of the diagonal elements are non-zero.
Analysis software for bleach step quantification.
ImageC: ImageC.exe is a lab-written image analysis program that has single molecule/centroid localization functions useful for single molecule analysis. The application is a Windows based program written in C/C++ using Microsoft Visual Studio 2017. For a portion of the bleach step analysis used in this work, ImageC was used to two modes: (1) user-determined number of steps based on observation of time traces, where the users tally the results within a program spreadsheet, or (2) an automated bleach step-counting algorithm that is described in Figure S7.
Spot location method. Single molecule spots are located by analysis of an image created by summation of a subset of the first 20-50% of image frames. A histogram of the summed image is calculated and pixels at the maximum pixel value located, analyzed either on a peak vs background levels basis for the NxN box (typically 5x5 pixels), or using a Gaussian mask-based merit function described below. Once the bright center pixel is analyzed, the pixels within the NxN box are set to zero. The level criteria determine whether the peak pixel is greater than the userset background value (BG) for all the pixels within the NxN box. The Gaussian difference mask method compares the NxN pixels surrounding the peak pixel to see if they conform to a Gaussian profile relative to the center of the spot. The extent of conformance is a user-selectable value that is the fractional ±amount the pixel can deviate from the expected value, based on what would be expected if the spot were a perfect Gaussian. If a spot has nearest neighbor pixels that do not meet this criterion, the spot is not used.
Gaussian Difference Mask (GDM) method. Assume a Gaussian PSF with a 1/e radius of σ centered at pixel i = j = 0. The camera pixel size is p, and p and σ are in the same units. The pixels at and near the center of the PSF would have values of: The above can be used to pre-calculate a mask for a given σ and pixel size, which can then be applied as a merit function to judge whether a spot on an image is a PSF: PIF 40 : In brief, spots are located by scanning a Laplacian of a Gaussian kernel across the surface. Fluorescent puncta above a certain threshold with a good Gaussian fit are selected as valid spots for analysis. The intensity vs time traces of these spots are analyzed using an algorithm which locates intensity drops larger than a predetermined threshold in the trace. A level is determined as the average value of the trace between two drops. The intensity difference between adjacent levels is then compared and they are combined if it is below the threshold. This is repeated in an iterative manner until all remaining levels are separated by an amount greater than the threshold. Each image contained ~600 spots. Figure S1: Time course of syncytial formation and protein diffusion in the cytosol and on the membrane.

SUPPLEMENTAL FIGURES
(a) The cell fusion process acquired using an incubator microscope. Initially each cell's bounding membranes are clearly discernible but this morphology disappears within 30 minutes. The resulting syncytium remains bound to the coverslip for 4-6 hours, at which point it begins to detach from the substrate. By about 16 hours, the syncytium is mostly detached and cells begin to die. Scale bar is 100 μm. (b) mNG-ADRβ2 diffusion in the VSVG-induced syncytium measured by the change in signal variance across the image as fusion precedes. Equilibrium is reached within 30-60 minutes for all conditions (we looked at a number of different co-plating ratios and temperatures). (c) Exponential decay times for image variance in syncytia. Equilibration takes longer for higher co-plating ratios and lower temperatures, as expected for diffusion of macromolecules (mNG-ADRβ2 in this case).      (a) -(c) Bleach step traces of mNG-Orai1 complexes: (a) 3 steps, (b) 5 steps, (c) an example of fluorophore blinking artifact (arrow). This trace could be scored as 2 or 3 steps depending on whether drop at around 20s is considered to be background level or the intensity of a single fluorophore. (d) -(f) mNG-EGFR complexes: (d) typical single step trace (EGFR monomer), (e) 2 bleach steps (EGFR dimer), (f) an example of a trace with decaying signal (arrow). Both software used (PIF and our own) corrected for this artifact by fitting the background decay (all pixels not identified as PSFs) and subtracting it from the data before analysis (traces a through e have been adjusted to remove any background decay signal). Intensities are the average of a 5x5 pixel box (500x500 nm) centered on the PSF. The algorithm begins by convoluting the time trace with the kernel defined in B. kdata(t) is effectively the derivative of the time trace. kdata(t) is squared to make it positive, normalized and displayed in the lower plot window of the Time Trace Display window (lower plot in A). The peaks in kdata(t) locate potential bleach steps in time trace as shown above in A. The peak detection threshold level determines how many steps to use in the processing algorithm. Once a peak threshold is found that finds no more than user-set maximum number of allowed "jogs" (convoluted time trace spikes), the time trace is broken up into #jogs + 1 segments defined by the threshold crossing points. The segments are fit to a linear function to obtain their slope. A characteristic deviation (noise level) is calculated for each segment and if the segments are of acceptable length (in time) and slope, they are further analyzed to identify whether they overlap in pixel value. If so, they are considered to be in the same level group (i.e. a fluorophore blinking, rather than a different molecule) rather than assigned as new level. The total number of levels are tallied and the number minus 1 (to account for the final fully bleached level) is reported by ImageC to be the number of photobleach steps in the trace. Figure S8: Effect of the short pH drop on intracellular pH using SNARF.

D
After a 24-hour induction period in doxycycline, Tet-VSVG U2OS cells fuse immediately after a brief pH drop from 7.4 to a pH < 6.0. We measured the effect on the intracellular pH during incubation at pH 5.5 and found that during the short time needed for activating VSVG (0-2 minutes) the pH remains above pH 7.0. Even during longer incubations (up to 5 minutes), the pH was never found to be lower than ~6.8. Data are mean ±SEM n = 2.  Frequency of bleach steps for each of the membrane protein oligomers studied. Adjusted percentages are adjusted assuming 20% dark fraction for mNeonGreen.