Measuring the co-localization and dynamics of mobile proteins in live cells undergoing signaling responses

Single molecule imaging in live cells enables the study of protein interactions and dynamics as they participate in signaling processes. When combined with fluorophores that stochastically transition between fluorescent and reversible dark states, as in super-resolution localization imaging, labeled molecules can be visualized in single cells over time. This improvement in sampling enables the study of extended cellular responses at the resolution of single molecule localization. This chapter provides optimized experimental and analytical methods used to quantify protein interactions and dynamics within the membranes of adhered live cells. Importantly, the use of pair-correlation functions resolved in both space and time allows researchers to probe interactions between proteins on biologically relevant distance and time-scales, even though fluorescence localization methods typically require long times to assemble well-sampled reconstructed images. We describe an application of this approach to measure protein interactions in B cell receptor signaling and include sample analysis code for post-processing of imaging data. These methods are quantitative, sensitive, and broadly applicable to a range of signaling systems.

information relevant to interactions between proteins in live cells undergoing signaling responses after the B cell receptor is engaged with a multivalent cross-linker [22,31]. Since then, we have further optimized and extended these approaches, enabling the detection of weak signals over time [24]. This protocol chapter describes optimized experimental approaches including sample preparation, fluorescent probe selection, and microscopy imaging methods. We also include protocols for single fluorophore localization in raw images, the evaluation of properly normalized pair correlation functions, and use of correlation functions to characterize colocalization and molecular motions in live cells. Analysis code is provided to facilitate the application of these methods by other groups.

Cell culture and transfection
1. Cells and culture medium. The imaging methods described here can be used to study interactions and dynamics of plasma membrane signaling proteins in a variety of cell types. The only requirements are a relatively flat cell membrane surface for imaging through cell adhesion to a bare glass or functionalized coverslip, and that proteins of interest can be labeled with SMLM-compatible fluorescent tags. For our studies of the B cell receptor, we use CH27 mouse B cells [36]. CH27 cells are maintained in culture in growth media at 37˚C and 5% CO2 as previously reported [31].

Imaging reagents
1. Fiducial markers: sub-diffraction registration of multiple color channels requires reference markers that fluoresce in both channels, such as small multiwavelength fluorescent beads (e.g. .1µm Tetraspeck beads, ThermoFisher T7279). We prepare a fiducial marker sample by first cleaning a #1.5 glass coverslip cell culture dish in an oxygen plasma cleaner at full power for 3 min. A 1:100 dilution of beads in PBS is immediately added to the dish and incubated for 30 min-2 hr at room temperature.
Unbound beads are removed with gentle rinsing. This sample can be stored in PBS at 4˚C and reused.
2. Imaging buffer reagents: live cells are imaged in buffer conditions that maintain the cellular signaling process under study and are also conducive to single molecule photoswitching necessary for localization imaging of organic dyes. Different formulations are possible and generally include a reducing agent and oxygen scavenging system to support dye photoswitching. We use the following reagents for SiR/mEos3.2 two-color imaging: 1. L-glutathione reduced (Sigma G6013) (see Note 1). Handling large amounts of image data and determining single-molecule localizations from raw data are computationally intensive. Numerous commercial and freely available software packages are available for single molecule localization. We find that the imageJ plugin ThunderSTORM [38] is flexible and accessible. We use MATLAB for all further analysis of single molecule localization to quantify protein co-localization and dynamics.

Methods
All steps are carried out at room temperature unless otherwise noted.

Preparation of fluorescent Fab fragments:
Labeling proteins such as antibodies, Fab fragments, toxins, or other binding proteins can be conjugated with organic dyes suitable for super-resolution imaging (see Note 2) via NHS ester chemistry. In these reactions, a reactive group commonly available on organic dyes is used to covalently attach the dye to amines on lysine residues or N-termini of purified proteins. After conjugation, unreacted dye is separated from labeled protein by size exclusion. For applications where the labeling protein is crosslinked during the imaging experiment, for example to study the effects of protein clustering, labeling proteins can be conjugated with both dye and biotin groups using the same succinimidyl ester chemistry. In our specific application, we label goat anti-Mouse IgM(µ) Fab fragments with biotin and silicon rhodamine (SiR) NHS esters.
1. Conjugation reactions should be carried out near pH 8.5 under buffer conditions where proteins are stable. Exchange purified proteins into the reaction buffer through dialysis. If the initial protein storage buffer contains free amines such as Tris or glycine, several rounds of dialysis may be necessary to remove them.
Final protein concentrations between .5 -2mg/ml are optimal for labeling. 3. Incubate the reaction for 1 hr at room temperature with rotation. Protect from light e.g. by enclosing the reaction tube in a larger foil-covered tube.
4. After incubation, pass the reaction solution through a NAP-5 gel filtration column using PBS with 1mM-EDTA as the equilibration and elution buffer.

Modified proteins can be further purified and concentrated by centrifugation in a
Vivaspin-500 Polyethersulfone concentration spin column. 6. Estimate the degree of labeling using absorbance measurements of the dyeconjugated protein at the dye's absorption maximum and 280nm. We aim for a dye-to-protein ratio around 2. If the protein is under-modified, the reaction can be repeated to increase the dye-to-protein ratio.
7. For proteins modified with both fluorescent dyes and biotin, we use a two-step procedure. First, react unlabeled proteins with a mixture of fluorophore and biotin NHS esters in 5x and 12x molar excess, respectively. Purify as described above.
Second, repeat the labeling procedure with only fluorophore NHS ester in 5x molar excess. 2. If needed, treat glass bottom wells or culture dishes with reagents to facilitate adhesion. For example, VCAM-1 coated dishes can be prepared as follows:

Cell preparation:
1. Plasma clean imaging chambers using an oxygen plasma cleaner at the highest power setting for 3 min.
2. Immediately incubate cleaned dishes with a solution of Fcγ-specific goat anti-human IgG antibodies in PBS at 100 μg/mL for 30 min at room temperature.
3. Rinse dishes with PBS and block for 30 min with a 5% BSA solution in PBS.

Rinse dishes with PBS and incubate dishes for 1 hour at room
temperature or overnight at 4˚C with recombinant VCAM-1/human Fc chimera protein at 10 μg/mL, 5. Rinse dishes thoroughly in PBS before use. 6. Coated dishes can be stored in PBS at 4˚C for up to 1 week.
3. Cells can be grown overnight on untreated glass, but should be plated soon before imaging if 1) cells secrete an expressed fluorescent protein which can become deposited on the glass (see Note 4) or 2) a surface treatment such as VCAM-1 coating is used. CH27 cells adhere to untreated glass within 2-4 hours at 37˚C and VCAM-1 coated dishes within 10 min at room temperature. 4. In both cases, cells should be plated at a density such that they remain isolated from one another for imaging.

5.
A day of imaging usually involves multiple replicates of a live cell imaging measurement. Prepare as many dishes of cells as replicates are desired.
6. Label cells with dye conjugated proteins at room temperature or on ice if needed to minimize probe internalization. We label BCR by incubation with 5 µg/ml SiRbiotin-Fab in culture medium for 10 min at room temperature. Cells should be labeled shortly (within 30 min) before imaging to prevent internalization. 7. At this stage and prior to imaging, cells can be pre-treated with drugs or other perturbations as needed for the experiment.
3.3 Imaging: Sparse, single molecules are simultaneously imaged in two color channels over time.
1. Configure microscope such that two fluorophores can be imaged simultaneously. This likely will involve illumination from several excitation lasers (e.g 405nm, 561nm, 647nm) paired with a multi-band filter cube that reflects laser lines and passes fluorophore emission. An emission splitter unit is used to separate emission from distinct fluorophores and direct them onto separate cameras or separate sides of the same EMCCD or sCMOS camera (see Note 5).
Lasers and emission optics should be aligned prior to sample preparation. We recommend using an automated correction for drift in the axial direction (e.g. ZDC from Olympus). Total internal reflection (TIR) illumination is commonly used for selective imaging of cell surface proteins and structures on the plasma membrane.
2. Acquire images of fiducial markers to register color channels: Take many (>100) images of the fiducial marker bead sample using the same illumination laser lines and filters to be used in the cellular imaging measurement. Between acquisition of individual bead images, translate the stage so that the field of view is uniformly sampled by bead positions. It is important that the splitting optics are stable throughout the course of the imaging experiment. We recommend acquiring bead images before and after a cell measurement, to ensure that the relative positions of the two color channels do not shift during a measurement, impairing channel registration.
3. Add oxygen-scavenging enzymes glucose oxidase and catalase to the imaging buffer (see Note 1) and replace buffer in cells with fresh imaging buffer.

2.
Generate a spatial mapping between image channels using fiducial markers, for example using the procedures described in [40]. To accomplish this, fiducial markers are first matched across image channels and used as control points to generate a transform. Once the positions of the control points are established, this map can be generated from control points using the function fitgeotrans() and it can be applied to points using the function transformPointsForward() in MATLAB.

Cull localizations to remove poor fits. This is typically done by accumulating
distributions of fit parameter values and removing the top and bottom 5% for parameters such as intensity or width.

4.
Localized positions can be corrected for stage drift or collective cellular motions using a drift correction algorithm, for example using the procedures described in [41]. Although drift correction is not strictly required for the cross-correlation or diffusion analysis because it applies at slower time-scales than are typically of interest, it can be important when establishing a region of interest (ROI) that is appropriate for localizations acquired over the entire dataset.

5.
Images representing the time-averaged positions of localized molecules can be rendered by generating two dimensional (2D) histograms from the localized positions. This can be accomplished using the histcounts2 function in MATLAB.
Histograms can be presented as is or blurred using a Gaussian filter for display purposes.

Post-processing: Cross-correlation analysis. Cross-correlation analysis is conducted
using the locations of molecules in both space and time. Cross-correlation functions are generated by tabulating pairwise distances between localizations in different color channels, assembling these distance into histograms, then normalizing these histograms to account for the expected number of pairs if probes are randomly co-distributed over the specified region of interest. We have assembled a collection of MATLAB functions to perform these calculations from lists of localizations, as well as sample analysis scripts, publically available at https://github.com/VeatchLab/smlm-analysis. Scripts and localizations used to generate  [42] and described previously [41], as well as boundary conditions for arbitrary spatial ROIs and temporal windows, as described [34]. In some cases, longrange gradients in images can obscure results when an analysis of short-range correlations are desired. To reduce the impact of long-range gradients, a local density correction can be applied, tabulated using spatial_gradient_correction() in the SMLM-analysis distribution, following an approach described previously [43].
We find that this correction is often necessary when imaging mobile (d)STORM probes such as SiR that diffuse into the illuminated area from the cell edges in an activated state. These probes tend to take some time to be converted into a reversible dark state, leading to a larger density of probes at the cell periphery (evident in Figure 1B for SiR). A properly normalized correlation function should decay to 1 when probed at large separation distance or large separation-times.
Examples of correlation functions tabulated from the same set of localizations but with different normalizations are shown in Figure 3.

3.
Correct for bleed-through and/or other fluorophore related photo-physical effects between image channels. Fluorophores have broad emission spectral characteristics, making it difficult to fully isolate emission with filters alone. Even low intensity emission in the incorrect color channel can bias the localizations to generate the appearance of shorter pairwise distances, when molecules appear in both color channels simultaneously and within a diffraction limited separation distance [31]. FRET or other proximity related photo-physical effects (e.g. [44]) can also give rise to correlations that do not reflect the actual density of labeled molecules. These effects can be detected and corrected by analyzing how crosscorrelations vary with the time-interval between observations. Bleed-through or other fluorophore related photo-physical effects are expected to be most apparent when probes are observed simultaneously within a diffraction limited separation distance (~200nm). An example is shown in Figure 4 in which one probe (SiR labeling BCR) is clustered and largely immobile and the second probe (mEos3.2 labeling a minimal GPI linked protein) is mobile. In this case, crosscorrelation amplitudes (Figure 4B) are highest for simultaneous observations (time-separation = 0) and decay over a fraction of a second for short separation distances (r<150nm). Since BCR clusters are largely immobile, the crosscorrelation amplitude should not depend on when exactly labels decorating those clusters are observed, therefore we expect the actual cross-correlation amplitude to not vary with time-interval. These artificial probe-related correlations can be removed by either averaging over a range of time-intervals excluding the effected points, or by fitting quickly varying contributions and reporting only the slowly varying component. The impact of these corrections can be seen in the spatial correlation functions of Figure 4C. 4. In some cases, we are most interested in the change in the magnitude of correlations upon stimulation, in which case we subtract correlation functions generated from localizations detected before stimulation from the correlation function generated from localizations obtained after stimulation (see Note 6).
One advantage of this approach is that it provides a means to exclude systematic contributions from membrane topography that are present over the entire measurement but are not removed by the ROI. We found this to be important when quantifying small signals [45].

Post-processing: Diffusion analysis.
Single molecule mobility can be characterized by quantifying the evolution of the spatial auto-correlation in time. As molecules move, they tend to explore larger distances over larger time-intervals, and this can be detected as a broadening of the initial peak in the auto-correlation function, over time-scales relevant to the off-rate of the fluorophore blinking. This analysis yields similar information as is obtained in single particle tracking, but has the advantage of not requiring that localizations be assigned to trajectories, which may be ambiguous for densely sampled data. An example showing GPI-mEos3.2 motion is shown in Figure 5. The script and localizations used to generate Figure 5 is included in the SMLM-analysis distribution.

1.
Using the same ROI constructed for the cross-correlation analysis, tabulate the spatial auto-correlation for the color channel corresponding to the molecule being characterized. This can be done using the same MATLAB function used to tabulate cross-correlations, spacetime_acor(), which resembles the function spacetime_xcor() but in this case only positions for one color channel are inputted. Auto-correlation functions should include normalizations for the ROI and spatial gradients.

2.
Auto-correlation functions can be fit to a superposition of Gaussian functions to extract the mean square displacements (MSDs) and relative populations associated with each state. The example of Figure 5 is fit to a superposition of two Gaussians of the form: is a normalization factor, MSD1 and MSD2 are the mean squared displacements in the two states, and α is the fraction of molecules in the 1 st state. This fitting approach makes the assumption that motion is normally distributed, which is typically appropriate as long as directed motion is not expected. This approach does not assume that motion is Brownian. where LP is the average localization precision of the measurement. For the example of Figure 5, the slow state is highly confined, in that its MSD does not increase with time-interval. In this case it is more appropriate to report the confinement radius, which is 2. For quantitative measurements of protein co-distributions by super-resolution imaging in two color channels, special considerations apply for dye selection. It is imperative that each orthogonal label is identified in one color channel only.

Best fit
Cyanine-based dyes commonly used in localization microscopy can undergo reactions that shift their absorbance spectra to shorter wavelengths, leading to detection of spurious signals in the wrong color channel from these dyes [47,48].
Two-color detection in the red and far-red emission channels is commonly used because of availability of dyes and fluorescent proteins optimal for localization microscopy. For this combination, we favor silicon-substituted rhodamine dyes [49] for far-red detection. 5. Quantitative analysis of multi-color SMLM data is very sensitive to even low levels of spectral bleed-through across color channels. Emission filters should be selected to minimize bleed-through, with more restrictive requirements than are typical for standard TIR or wide field fluorescence imaging. Small residual bleedthrough artifacts can be corrected in post-processing (see 3.5.3 and Figure 4).
6. Factors such as subtle membrane topography can contribute to cross-correlation functions in ways that can be difficult to identify or isolate (see 3.5.1 and Figure   2). These contributions add to cell-to-cell variation in cross-correlation functions and are not typically the focus of the imaging experiment. In situations where structures are expected to be present throughout the imaging experiment, it can be useful to quantify the difference in the cross-correlation function before and after the addition of a cell treatment or stimulus rather than the absolute amplitude of the cross-correlation after treatment. For this reason, it can be helpful to acquire sufficient image data before treatment to establish a baseline for the cross-correlation function.
7. Quality assessment for raw data requires special attention. It should be noted that fitting algorithms are capable of producing localizations from even very poor quality raw data and therefore the ability of processing procedures to reconstruct super-resolution images is not a sufficient data quality standard. Isolated single molecules should be identifiable by eye in raw images. Reconstructed images generated from raw data with a high density of single molecules often contain recognizable artifacts such as intensity bleeding between higher intensity objects or narrowing of linear structures. Localization algorithms return quality control metrics (e.g. localization error, spot width and intensity, etc.) that degrade when probe density is too high, indicating single molecules are not well resolved.
8. It can be useful to localize single molecules on images that have undergone a background subtraction to remove intensity from signals that do not vary in time.
The background is typically estimated by taking the median of images acquired over some finite time (e.g. 500 frames), subtracting this from the raw images, then adding back the average intensity value of the median image. The most commonly used localization algorithm [50] incorporates information about pixel intensity, camera gain and noise when localizing and estimating the localization error on single fits. These error estimates will contain systematic errors when applied to background subtracted image frames.