Oligomerization Dynamics of Cell Surface Receptors in Living Cells by Total Internal Reflection Fluorescence Microscopy Combined with Number and Brightness Analysis

Summary

We describe an imaging approach for the determination of the average oligomeric state of mEGFP-tagged-receptor oligomers induced by ligand binding in the plasma membrane of living cells. The protocol is based on Total Internal Reflection Fluorescence (TIRF) microscopy combined with Number and Brightness (N&B) analysis.

Abstract

Despite the importance and ubiquity of receptor oligomerization, few methods are applicable for detecting clustering events and measuring the degree of clustering. Here, we describe an imaging approach to determine the average oligomeric state of mEGFP-tagged-receptor homocomplexes in the membrane of living cells. The protocol is based on Total Internal Reflection Fluorescence (TIRF) microscopy combined with Number and Brightness (N&B) analysis. N&B is a method similar to fluorescence-correlation spectroscopy (FCS) and photon counting histogram (PCH), which are based on the statistical analysis of the fluctuations of the fluorescence intensity of fluorophores diffusing in and out of an illumination volume during an observation time. In particular, N&B is a simplification of PCH to obtain information on the average number of proteins in oligomeric mixtures. The intensity fluctuation amplitudes are described by the molecular brightness of the fluorophore and the average number of fluorophores within the illumination volume. Thus, N&B considers only the first and second moments of the amplitude distribution, namely, the mean intensity and the variance. This is, at the same time, the strength and the weakness of the method. Because only two moments are considered, N&B cannot determine the molar fraction of unknown oligomers in a mixture, but it only estimates the average oligomerization state of the mixture. Nevertheless, it can be applied to relatively small time series (compared to other moment methods) of images of live cells on a pixel-by-pixel basis, simply by monitoring the time fluctuations of the fluorescence intensity. It reduces the effective time-per-pixel to a few microseconds, allowing acquisition in the time range of seconds to milliseconds, which is necessary for fast oligomerization kinetics. Finally, large cell areas as well as sub-cellular compartments can be explored.

Introduction

We describe a Total Internal Reflection Fluorescence-Number and Brightness (TIRF-N&B) imaging approach for determining the average oligomeric state of receptor molecules at the plasma membrane of live cells, aiming at linking the receptor assembly dynamics to the biological function of the proteins (Figure 1).

Upon extracellular ligand binding, receptors initiate the intracellular signal transduction depending on their conformation, oligomerization, potential co-receptors and membrane composition. Despite the importance and ubiquity of receptor oligomerization, recognized as a key event in cellular signaling^{1,2,3,4,5,6,7}, few methods can detect clustering events and measure the degree of clustering experimentally^8,9. The confocal volume (x,y ≈ 300 nm, z ≈ 900 nm) is insufficiently resolved for proving molecular interaction and stoichiometry, even after optimization by image restoration algorithms¹⁰. The sub-unit composition of protein oligomers cannot be resolved on a purely spatial basis even by super-resolution methods at x,y resolution of 20-70 nm such as PALM¹¹, STORM¹², and STED¹³. Moreover, their temporal resolution (in the order of minutes per image) cannot follow kinetics in the range of seconds. Single molecule step-bleaching resolves the stoichiometry of protein oligomers only if they are immobile¹⁴.

One of the most versatile methods to measure density and oligomerization of fluorescently tagged proteins within single images is the spatial intensity distribution analysis (SpIDA), which relies on spatial sampling. It is applicable to both chemically fixed and live cells, and allows the analysis of several regions of interest of the cell simultaneously using standard fluorescence microscopy¹⁵. Alternatively, moment methods, such as fluorescence-correlation spectroscopy (FCS)¹⁶, photon counting histogram (PCH)¹⁷, and Number and Brightness (N&B)^18,19, are suitable for quantitative oligomer measurements. These methods analyze the fluorescence intensity fluctuations that can be observed in time when the fluorophores diffuse in and out of an illumination volume. The amplitudes of the intensity fluctuations can be uniquely described by the molecular brightness of the fluorophore (ε) and the average number of fluorophores (n) within the illumination volume¹⁷ (Figure 2). Typically, the diffusion coefficient of the fluorophores and the average number of molecules (inversely related to the G(0) value) within the illumination volume can be obtained by FCS²⁰. However, since the diffusion time only scales with the cubic root of the mass, FCS is not sufficiently sensitive to detect changes in molecular mass²¹. In practice, single color FCS cannot detect dimerization of membrane receptors. PCH resolves mixtures of different oligomers accurately. Using more than two moments of the amplitude distribution, it detects molecules of different brightness that occupy the same illumination volume. Scanning FCS²² and developments, such as the interesting pair-correlation of molecular brightness (pCOMB) approach²³, introduced to extend the range of applicability of fluorescence correlation methods in biological systems²⁴, remain single point methods lacking the capability of fast measurements in a large area of a cell, requiring many consecutive observations at each pixel and data acquisition in the order of seconds.

N&B is a simplified version of PCH that considers only the first and second moments of the amplitude of the fluorescence distribution, namely the mean intensity, <I>, and the variance, σ² (Figure 2)^18,19 and, because of that, it cannot determine the molar fraction of unknown oligomers in a mixture, but only estimates the average oligomerization state of the mixture. Nevertheless, N&B has the advantage of working with relatively smaller time series of images of live cells than PCH on a pixel-by-pixel basis, simply by monitoring the fluctuations on time of the fluorescence intensity. Because N&B reduces the time-per-pixel to a few microseconds, it can follow fast oligomerization kinetics over large cell areas, allowing image acquisition on a time scale of seconds in raster scanning microscopy (e.g., confocal, 2-photon) and milliseconds in camera-based microscopy (e.g., TIRFM).

Several reports have demonstrated the capability of N&B to quantify the number of subunits in protein clusters by imaging extended cell regions. Paxillin-EGFP clusters were detected at the adhesion sites in CHO-K1 cells²⁵, and the intracellular aggregation of the pathogenic Httex1p peptide was described in COS-7 cells²⁶. N&B was applied for following the ligand-driven oligomerization of the ErbB receptor²⁷, and the effect of the ligand FGF21 on Klothob (KLB) and FGFR1c in HeLa cells²⁸. The combination of TIRF imaging and N&B analysis was used to show that dynamin-2 is primarily tetrameric throughout the entire cell membrane²⁹. We applied N&B to both raster scanning and TIRF images to prove ligand-driven dimerization of uPAR and FGFR1 cell membrane receptors^30,31.

Fluorescence correlation methods, such as N&B, FCS and PCH, are based on the notion that in an open volume the occupation number of particles follows a Poisson distribution. Because only the photons that the fluorophores emit can be detected, the mean value for a measured fluorescence intensity versus time in a pixel of the image, , is the product of the average number of fluorophores in the illumination volume, n, and their molecular brightness, ε¹⁷:

where ε is expressed as the number of photons emitted per unit of time (conventionally per second) per molecule when the molecule is at the center of the illumination volume.

Brightness is a property of each fluorophore in a given acquisition set up, while intensity is the sum of all contributions from all fluorophores. In biological contests, brightness will increase with the increase of the number of fluorophores that fluctuate together, giving information on the oligomerization state of the fluorescently-tagged protein. The fluctuation amplitudes at a given pixel is measured from the variance of the fluorescence signal, σ²:

Where the mean of the square of intensity, , and the square of the mean of intensity, , are computed from the individual intensity values in each pixel of each frame:

where K is the number of total frames in the time series. Experimentally, it is necessary to compute for the entire image series the variance that describes the scatter of the individual intensity values at each pixel of a single image around the mean intensity value. The variance includes all fluctuations of different origins. In a first approximation, the variance due the diffusing particles in the illumination volume, σ²₀, can be separated from the variance due to the detector shot noise, σ²_d. The two variances are independent; thus, the total variance is given by their sum:

The variance, due to molecular fluctuations in and out of the detection volume, is linearly dependent on the molecular brightness and intensity:

Rearranging eq. 6 according to eq. 1:

According to the typical concept in fluorescence correlation spectroscopy, equation 7 states that the variance due to the number of fluctuations depends on the square of the particle brightness.

Then, the variance due to detector fluctuations is a linear function of the detected intensity, under the assumption that the detector is operated below its saturation limit¹⁹:

In the case of photon counting detectors a=1 and c=0, thus the detector variance is equal to the average intensity:

To apply these concepts to real measurements in live cells, Gratton and colleagues¹⁸ define the apparent brightness, B, for each pixel as the ratio of the variance over the average intensity:

B is the parameter that is measured experimentally. In this work, time series images of FGFR1 receptors at the plasma membrane of HeLa cells are captured by TIRF microscopy and the average apparent brightness, B, is determined by the N&B analysis. Then, after addition of FGF2, consecutive time series are captured to follow the changes in the self-assembly of the receptor molecules in the membrane surface after stimulation of the receptor with the canonical ligand.

However, since the detector of the TIRF microscope is a EMCCD camera, the expression for the apparent brightness needs to be modified as¹⁹:

where offset is the intensity offset of the detection electronics that is a characteristic of the detector settings. The variance and average intensity for an analog detector are respectively given by:

where G is the analog gain in digital levels (DL/photons), S, the digital levels per photon¹⁹, is given by the slope of an intensity versus variance plot for a light source with constant intensity (no temporal fluctuations). The γ factor is related to the shape of the pixel detection volume. According to Hassler et al.³², the γ factor is equal to 0.3 for TIRF imaging working at the maximum gain of the detection camera¹⁹. The offset, S and G parameters are characteristics of the camera and the microscope. The apparent brightness, B, is obtained by rearranging eq. 11 according to eq. 12 and 13:

Experimentally, ε is a complex function of laser intensity and the detection efficiency of the system. Nevertheless, since B/S is linearly dependent on ε, it is only important to determine the relative value of ε for a given detection mode:

where ε' is proportional to ε. Still, a calibration is performed using an internal reference.

Protocol

1. Sample Preparation

1. Day 1. Seed HeLa cells in complete medium at a concentration of 100,000-200,000 cell/mL in glass-bottom dishes. Seed 6-8 replicate dishes.
   NOTE: In this example, the medium is supplemented with 10% heat inactivated Fetal Bovine Serum (FBS), 1 mM sodium pyruvate, 100 U/100 µg penicillin/streptomycin. Several replicate dishes are prepared.
2. Day 2-3. When cells are at sub-confluency, transfect half of the dishes with the protein plasmid and the second half with reference plasmids (monomer and dimer), in serum-free medium.
   NOTE: Transfection is made in serum free medium supplemented with antibiotics, 0.1% Bovine Serum Albumin and 25 mM HEPES buffer, without Phenol Red.
3. Day 3-4. Check that the transfected cells are adherent to the bottom of the dishes and the cell membrane is fluorescent. Discard dishes with overgrown cells or with very low fluorescence.
   NOTE: Do not let cells overgrow. Cells must be well distributed and be adhered to the glass area of the dish (Figure 1A). Precoated glass bottom dishes can be used for favoring cell adhesion. The cell culture is tested for mycoplasma contamination before any experiment. In this example, cells are transfected with a (A207K)mEGFP-FGFR1 plasmid and the reference cells are transfected with a GPI-(A207K)mEGFP and a GPI-(A207K)mEGFP-(A207K)mEGFP plasmids using standard protocols. For live cell microscopy, an indicator-free medium is recommended; 25 mM HEPES buffer is added to prevent pH changes during imaging.

2. TIRF Imaging — Alignment of the Laser Line and Optimization of TIRF Illumination

1. Four hours before experiment, activate the temperature incubator of the microscope at 37 °C.
2. Turn on the microscope, computers and cameras and wait for the cameras to reach the proper working temperature.
   NOTE: The working temperature of the camera used in this study is -75 °C.
3. Place a little drop of oil on the objective. Put a sample dish in place. Close the doors of the incubator and let the temperature of the dish equilibrate (~10 min).
4. Turn on the epifluorescence lamp and the 488 nm laser.
5. Select the epifluorescence contrast mode to explore the sample, searching a cell to focus from the ocular.
   NOTE: The use of a fluorescent lamp for searching cells trough the ocular is not mandatory. A suitable laser line can be used instead.
6. Select the proper filter for collecting the green emission through the microscope camera (Band Pass Ex 490/20 (500) Band Pass Em 525/50, or similar.
7. Switch from the ocular to the camera port (camera #1 in Figure 1) in epifluorescence mode, refine the focus and change to TIRF mode. Epifluorescence and TIRF modes might be named with a different nomenclature depending on the brand of the microscope.
   NOTE: There may be issues focusing or aligning the laser if there are no fluorescent markers at the coverslip interface. To align the laser properly (essential for good TIRF), focus on the coverslip. It is often very difficult to determine whether the coverslip is in focus. As a suggestion, focus on the edges of the cells.
8. Activate the auto-alignment following the instructions of the TIRF microscope.
   NOTE: Briefly, for steps from 2.4 to 2.8, first find the cells through the ocular and focus on them, then send the emission to the camera port of the TIRF microscope, re-focus the cells on the microscope computer screen and activate the procedure for laser alignment. The alignment consists in finding the critical angle at which illumination becomes evanescent (Figure 3). Commercial microscopes might have slightly different alignment protocols and also be fully automated; others might have a small camera for facilitating the visualization of the critical angle conditions.
9. Choose a suitable illumination depth and optimize the direction of the evanescent field (Figure 3).
   NOTE: The penetration depth is kept constant for all controls and samples.

3. TIRF Imaging: Capture of the Time Series

1. Define a region of interest (ROI) of at least 256 x 256 pixels.
   NOTE: In this set up, the capture is done with camera #2 under software that directly controls only the camera (See Figure 1 legend).
2. Set the exposure to 1 ms and the EM gain to 1,000 (this is the G factor in eq. 12 and 13). At such a speed, it might be necessary to adjust or increase the laser power. Here laser power is 0.5 mW.
   NOTE: Depending on the type of the camera and the limits imposed by the diffusion coefficient of the protein, fluorescence intensity and background, the general criteria for setting the laser power are not to saturate the detector, minimize photobleaching, and capture as fast as possible at a reasonable S/N. The EM gain is always set at the maximum of the camera (see Introduction).
3. Run a first trial sequence under initial conditions and roughly estimate the S/N value. The conditions are acceptable at S/N = 2-3 or higher, as measured in the first frame of the first time series.
4. Use the slider of the emission splitting system connecting camera #2 to the microscope for masking a side of the image (Figure 1B, Figure 4A-B)
   NOTE: In this set up a multichannel imaging connector is installed on camera #2 to enable the acquisition of two spatially identical images simultaneously. The system is equipped with slides for mounting different emission filters. One of the sliders mounts a black mask to cover a side of the image. The masked area is used for the internal calibration of each time series, to determine the camera parameters (eq. 12 and eq. 13). In this way there is no need for an independent calibration step and, importantly, calibration is performed in parallel to the capture of each time series. In the absence of this system, the camera can be calibrated applying published protocols³³.
5. Select the camera file autosave option.
6. Start the acquisition of the image series. Acquire a minimum of 700 frames at a minimum S/N ratio of 2.
   NOTE: The number of frames that are necessary for analysis depends on the sample stability to photobleaching and on the dispersion of the data. Therefore, the quality of each time series is assessed during N&B analysis.
7. Without taking the dish out of the microscope, add the ligand.
8. Select a cell with a bright fluorescence membrane and quickly start the first time series of the kinetic run.
   NOTE: If the addition of the ligand is done quickly, this first capture sets the point = 0 time of the ligand kinetics. The software registers the exact time of the capturing.
9. Search a second cell and acquire the second time point of the kinetics.
   NOTE: Point-visiting routines are available in some microscopes equipped with x,y,z motorized stages. These allow the memorization of multiple positions on the cell dish, and can help in keeping a more constant interval of time between image-series on different cells.
10. Capture a new cell for each time point of the kinetic run.
    NOTE: After capturing, a cell is partially photobleached and it cannot be re-imaged. Because of that, the kinetics is obtained by combining time series of many cells, each captured at a different time point.
11. For each new dish, repeat the protocol from step 2.3 to 3.9.
    NOTE: For reference dishes, add a volume of the vehicle (PBS supplemented with 0.01% bovine serum albumin) equivalent to that used for the ligand.

4. Number & Brightness (N&B): Quality Check of the Time Series

1. Convert and save as .TIFF the files acquired with the camera software (.sif files in this example).
2. Import .TIFF files in the analysis software routine by activating the N&B graphical user interface (GUI) MATLAB.
   NOTE: A customized Matlab executable N&B routine is used here (N&B analysis at https://www.cnic.es/en/investigacion/2/1187/tecnologia). By opening an imported .TIFF file, the routine generates the average intensity image, the average intensity profile and it allows inspecting the series frame-by-frame (Supplemental Figure 1). Other software are available for N&B analysis (e.g., SimFCS software).
3. Discard series for which the average intensity profile shows more than 10% photobleaching, and series in which there has been an evident cell membrane distortion or translation during acquisition.
4. Crop frames that are evidently out-of-focus.
   NOTE: A cropping tool is implemented in the routine to discards single or multiple frames within the image series. This operation is allowed because frame-to-frame time is not critical whereas the pixel dwell time (exposure time) is (see Discussion).
5. Keep for the analysis only series with at least 500 time frames.

5. Number & Brightness (N&B): Determination of the Camera Parameters (Offset, σ and S)

1. Activate the routine Calibrate Camera.
2. Select an area of at least 20 x 50 pixels in the detector noise region (Figure 4).
   NOTE: The routine originates a histogram of the values (also defined Digital Level, DL) and it returns a logarithm plot of the Frequency versus Digital Levels.
3. In the log Frequency versus Digital level plot, move the linear red cursor to delimit the Gaussian and the linear part of the curve.
   NOTE: The red cursor divides the two sections of the curve, and activates the routine returning the offset, which is the center of the Gaussian function of the camera response, the σ of the Gaussian fit, and the S factor, which is the slope of the linear part of the camera response (Figure 4C-D).

6. Number & Brightness (N&B): Computation of the B-values in Selected Region-of-interest (ROI)

1. Activate the B key.
   NOTE: This action generates the average intensity image (Figure 5, first column) and the B-image in which each individual B-value is associated to the related pixel in the image (Supplemental Figure 1).
2. Apply a minimum binning (2 2) to reduce the dispersion of the data and to generate the B-I histogram (Figure 5, second column).
   NOTE: The B-I histogram represents the distribution of the B-values of all pixels of the image versus the pixel intensity. Y = B/S; X = ( - offset)/S (Supplemental Figure 1 and eq. 11 and 15).
3. Inspect the B-I histogram using the interactive square cursor.
4. Select a square ROI for the analysis (Figure 5, third column).
   NOTE: The cursor synchronizes a mobile mask on the average intensity image, highlighting the pixels that are selected inside the square cursor area (Supplemental Figure 1). By this inspection, it is possible to exclude from the analysis the background and areas with very low intensity.
5. Generate the B-map of the selected ROI (Figure 5, fourth column).
6. Save the ASCII file of the B-values associated to the selection.
7. Import the ASCII file in a graphic software to compute the frequency distribution of the data and obtain the average B-value ± S.E (Figure 5, fifth column).
   NOTE: If data are homogeneous, the frequency distribution of the B-values approximates a Gaussian distribution.
8. Apply eq. 15 to derive the average brightness = - 1 [(counts/molecule) per dwell time] for each cell at each time point of the kinetic run. Normalize the data according to:
   
   where is the average B-value measured at time "t" after ligand addition, and is the average B-value measured at the time t=0 (10-20 s after ligand addition).
   NOTE: The normalization of the results allows the direct comparison of experiments that are carried out in different days. It compensates for differences in the measured brightness due to laser power and technical fluctuations.
9. Plot the Normalized Average Brightness versus acquisition time to build the kinetic run (Figure 6).

Representative Results

The results for two representative HeLa-mEGFP-FGFR1 cells seeded in the same culture dish are shown in Figure 5 and Supplemental Table 1. The two cells were captured at time 0 min (Figure 5A, top) and 7 min (Figure 5A, bottom) after addition of the FGF2 ligand.

Figure 5 also shows the results of two representative HeLa cells expressing either the pure monomer, GPI-mEGFP (Figure 5B, top), or the covalently linked dimeric fluorophore, GPI-mEGFP-mEGFP (Figure 5B, bottom), exposed at the cell membrane and captured under the same experimental conditions.

The average apparent brightness, B, in the HeLa-mEGFP-FGFR1 cells increases from 1.070 ± 0.001 S.E. to 1.141 ± 0.001 S.E., whereas the reference monomeric (Figure 5B, top) and dimeric (Figure 5B, bottom) samples return B values of 1.070 ± 0.001 S.E. and 1.141 ± 0.001 S.E respectively. Thus, by comparison, the FGFR1 receptor is present mainly in the monomeric form at the cell membrane surface at start, but it progresses towards a predominant dimeric state upon stimulation with its canonical ligand FGF2. On average then, the prevalent state of the FGFR1 molecules in the two representative cells is clearly different.

By applying the analysis to several cells in the same dish, each one captured at a different time point, the average brightness as a function of time is obtained (Figure 6A). The kinetic run in Figure 6A describes a slow process of dimerization that persists for several minutes at the cell surface. The FGF2-induced FGFR1 dimerization and subsequent internalization of the receptor is a well-known mechanism³⁴; therefore, the results are in full agreement with the present notion on the FGFR1 signal transduction, and confirm the potentiality of the TIRF-N&B approach for studying the oligomerization of cell membrane proteins up to the determination of subtle monomer-dimer dynamics.

The normalized average brightness analysis of the results is a suitable tool for comparing the effect of different ligands on the same receptor. One example is given in Figure 6B. The protocol was repeated using the same standards and stimulating the cells with a non-canonical FGFR1 ligand, NCAM-Fc (50 µg/mL). In this case, the kinetic profile reveals fast and cyclic transitions of the receptor in oligomeric mixtures, which also reaches brightness values above that of the dimer. A normalized average value of 3 is repeatedly observed. However, the limitation of the N&B analysis (only two moments of the intensity fluctuation versus time are considered) does not allow to demonstrate undoubtedly the formation of a trimeric form. The same normalized average brightness could be the result of various combinations of larger oligomers and monomers of the receptor. Yet, the results clearly demonstrate the spatiotemporal differences of the effect of the two ligands on the same receptor.

Figure 1: Overview of the experimental protocol. (A) Cells are plated on glass bottom dishes and transfected with the fluorescently tagged receptor. (B) Time series images are captured on a commercial TIRF microscope equipped with a TIRF 100x 1.46 oil objective and incubator chamber. In this commercial setup, the software does not allow the built-in EMCCD camera #1 to work at very short exposure times necessary for acquiring N&B time series. This is an important point since the exposure time limits the range of molecular diffusions that can be captured. The shorter the exposure time, the faster the molecular diffusion that can be analyzed. Exposures ranging from 0.5 to 1 ms are sufficiently fast for membrane protein diffusion. Therefore, a second EMCCD camera (#2) is added to an additional port of the microscope to work directly under the camera software, bypassing the microscope software. In this adapted configuration, the microscope software and camera #1 are used only for TIRF alignment. TIRF time series are then acquired using camera #2 that runs at very short exposure times such as 1 ms and at the maximum EM gain. Camera #2 also has a pixel size of 124 nm that allows oversampling and binning of the images (see protocol section 6.2). Other configurations to gain imaging speed are possible, depending on the characteristics of different TIRF microscopes, whereas the use of sCMOS cameras is not advisable, because noise is not random in the image³⁵. (C) After capturing, time series are inspected as a quality check. Series are discarded if photobleaching is higher than 10% as it can be determined by plotting the average frame intensity versus frame number. Series are also discarded if there has been an evident distortion of the cell membrane or translation of the cell during acquisition. (D) The average intensity in each pixel is saved. (E) A B-I histogram representing the apparent brightness, B, in each pixel of the image is generated. (F) The B-I histogram is used for selecting a ROI that is above the background. (G) The frequency distribution of the B-values is analyzed to determine the average B-value ± S.E. Please click here to view a larger version of this figure.

Figure 2: N&B principle. N&B quantifies the average oligomerization state of fluorophores by measuring the fluorescence fluctuations that occur when they move in and out the illumination volume during the acquisition of a time stack series of "K" images. The amplitude of the fluctuations is characterized statistically by computing the ratio between variance of the fluctuating signal, σ², and mean intensity value, . In the simplest scenario (A), when the illumination volume is empty (i.e., no fluorophores), the ratio describes instrument noise. If the fluorescence signal fluctuates due to mobile fluorophores (B,C), the "extra" variance is directly proportional to the molecular brightness, ε (detected photon counts per molecule and per second), of the diffusing molecules. In (B), there are 8 monomeric diffusing fluorophores and in (C), the same fluorophores diffuse as 2 tetrameric oligomers. In these two cases, the average intensity is the same, but the standard deviation and brightness are different (1ε, 4ε), because the amplitudes of the fluctuations are different. ε = 0 when fluorophores are immobile or absent. Please click here to view a larger version of this figure.

Figure 3: TIRF microscopy. Although N&B runs equally well with raster scanning microscopes equipped with multiphoton, continuous wave or pulsed single photon lasers and analogue or photon counting detectors, TIRF microscopy is ideal for fast temporal imaging of molecular dynamic events occurring at or near the cell surface with speeds, resolution and signal/noise ratio (S/N) that are not possible to achieve by other imaging techniques. (A) TIRF microscopy employs the principle of total internal reflection by which an angled excitation laser light excites only fluorophores that are just underneath a glass-water interface of the coverslip. The laser irradiates the specimen at an angle of incidence greater than or equal to the critical angle of refraction whereas the excitation laser light is totally reflected. The reflection generates a very thin electromagnetic field in the specimen called the evanescent wave. The resulting fluorescent light emitted by the tiny illuminated section of the specimen is collected through a microscope objective placed perpendicularly below to the slide. (B) The panel shows a representative example at a given wavelength of the relative intensity of an evanescent field versus depth, which decreases exponentially with increasing distance from the surface, providing high axial resolution fluorescence images. The lateral resolution is set by the numerical aperture and magnification of the objective and by the pixel size of the detection camera. The cell interior is not illuminated and it does not contribute to the signal with intracellular autofluorescence. Multi-angle TIRF microscopes allow selecting various penetration depths. They provide a scale that depends on the excitation wavelength and usually goes from 70 nm up to 250 nm depth for single color TIRF image. The evanescent illumination depth chosen for this protocol is 110 nm, and it is the result of a compromise between the necessity of using low laser power and the intensity of the fluorescence signal that decreases sharply with the decrease of the penetration depth. It is important to avoid overly large evanescent fields that can illuminate many intracellular vesicles and intracellular population of fluorophores. Therefore, depending on the type of sample, several penetration depths should be explored, searching for the best combination: high signal-to-noise ratio, low excitation power, short exposure time, short penetration depth. Once this optimization is done, the penetration depth is kept constant for all controls and samples. (C) Representative epifluorescence and TIRF images of a HeLa-GPI-mEGFP cell after software-guided optimization of the depth and direction of the evanescent field. Multi angle TIRF microscopes also allow to optimize the direction of the evanescent field. This step is recommended for minimizing scattering (i.e., sharp increase of the intensity and less crisp image), and it can be carried out according to the instructions of the specific microscope used. In this protocol the microscope software includes an automatic optimization of the direction of the evanescent field. For manual optimization, refer to published protocols^36,37. (D) Representative cell expressing the mEGFP-FGFR1 construct in epifluorescence and after optimization of the TIRF illumination. Please click here to view a larger version of this figure.

Figure 4: Capture of the time series and calibration of the camera response to single photons. (A) Example of the first frame out of 700-frame time series captured on a HeLa-mEGFP-FGFR1 cell in the Cropped Sensor Mode. The camera chip is partially masked (red rectangles) using the dual view connector installed on the TIRF microscope (Figure 1B). The internal calibration regions are recognized in the average intensity image (B) and processed (C) for estimating the camera parameters by plotting the log-frequency versus Digital Levels (D). An analog detection system, such as an EMCCD camera, detects pulses of photocurrent instead of photon counts, and the photon pulse height distribution is quasi-exponential. The first part of the distribution is due to the amplifier and analog-to-digital converter and it contributes a Gaussian readout noise (the variance introduced by the signal recording). The most populated channel (i.e., the most frequent value) of the distribution is the offset (eq. 13). Using a vertical cursor in a log plot, the first part of the distribution is separated from the exponential second part, above 250 DL, which represents the average camera response to a single photon (the slope is the S factor in eq. 12). The measurement of these parameters allows estimating the density of photons that are recorded during the acquisition. Counts (DL) are in pseudo color scale. Pixel size = 124 nm; Image format = 256x256 pixels, penetration depth of the evanescent field ~ 110 nm; Calibration ROI #1 = 19x256 pixels; Calibration ROI #2 = 5x256 pixels. DL = digital levels. Exposure = 1 ms and EMGain (the G factor in eq. 12, 13 and 14) = 1000. Note that cameras working in photon counting mode with background are equivalent to analog detectors with S = 1 (eq. 12) and with σ²_d and offset (eq. 13) measured in the same way as for an analog system¹⁸. Please click here to view a larger version of this figure.

Figure 5: N&B analysis of FGFR1 oligomerization. (A) Representative results from two HeLa cells in the same dish expressing mEGFP-FGFR1 and captured at time 0 min (top) and 7 min (bottom) after addition of 20 ng/mL FGF2, and (B) two HeLa cells expressing the reference constructs GPI-mEGFP (top) and GPI-mEGFP-mEGFP (bottom). The entire analysis sequence shows (from left to right): Average intensity of the time series; Plot of all B-values as a function of fluorescence intensity in which the color code represents the number of pixels having the same B-value in the image and the rectangular cursors delimit the region of interest (ROI), the background (BG) and regions of very low intensity (LI). Notice that immobile fluorophores will give B-value = 1 since in the absence of fluctuations ε = 0; B-map of the chosen ROI and the associated B-distribution histogram. The B-value distribution is fitted with a Gaussian function (dashed red line) to compute the average apparent brightness, B (full red line) and the distribution statistics (Supplemental Table 1) B-value = B' = B/S (eq. 15); Intensity = ( - offset)/S; M = monomer; D = dimer; scale bars = 10 microns. Raw data time series in Supplemental Movie 1, Supplemental Movie 2, Supplemental Movie 3, and Supplemental Movie 4. Please click here to view a larger version of this figure.

Figure 6: Kinetics of FGFR1 oligomerization induced by ligand activation. Representative kinetic runs described by the Normalized Average Brightness (eq. 16) of HeLa-mEGFP-FGFR1 cells after stimulation with 20 ng/mL FGF2 (A) or 50 μg/mL NCAM-Fc (B). Cells in the same dish were captured at increasing time points and the apparent average brightness, B ± S.E., was computed from the B-distribution histograms. The figure is adapted with permission from Zamai et al. Journal of Cell Science (2019)³¹. Please click here to view a larger version of this figure.

Figure 7: Fast Kinetics and data interpretation. Scatter plot of the all Normalized Average Brightness values obtained from replicate kinetic runs in which HeLa-mEGFP-FGFR1 cells were stimulated either with NCAM-Fc (50 µg/mL) or FGF2 (20 ng/mL). Please click here to view a larger version of this figure.

Supplemental Figure 1: Screen snapshot of the analysis routine in MatLab. Screen snapshot of the N&B graphical user interface (GUI) MATLAB routine showing the initial analysis steps: upload time series; compute average intensity image, compute intensity profile, compute B-map and B-I histogram. Please click here to view a larger version of this figure.

Parameter	GPI-mEGFP	GPI-mEGFP-mEGFP	mEGFP-FGFR1 (time = 0')	mEGFP-FGFR1 (time = 10')
Gaussian mean B-value	1.070	1.141	1.070	1.141
S.E. on the mean B-value	0.001	0.001	0.001	0.001
S.D. of the B-vaue distribution	0.059	0.067	0.077	0.075
95% Confidence interval of the mean B-value	1.069 to 1.071	1.139 to 1.143	1.069 to 1.072	1.139 to 1.143
S.D. of the 95% confidence interval of the B-value	0.058 to 0.060	0.065 to 0.069	0.075 to 0.079	0.073 to 0.078
Fitting constrain	S.D. > 0	S.D. > 0	S.D. > 0	S.D. > 0
S.E. standard error; S.D. standard deviation

Supplemental Table 1: Apparent Brightness Analysis. Statistics and Gaussian fitting parameters of the B-distributions in Figure 5 were carried out. The least square Gaussian fits were obtained under the constrain of standard deviation > 0 with the X-range automatically chosen and maximum iterations = 1000. R square: mEGFP-FGFR1 monomer = 0.9957, mEGFP-FGFR1 dimer 0.9940; GPI-mEGFP = 0.9985; GPI-mEGFP-mEGFP = 0.9970.

Supplemental Movie 1: Time series of an HeLa-mEGFP-FGFR1 cell at time = 0 minutes after FGF2 stimulation (20 ng/mL in PBS supplemented with 0.01% BSA) shown in Figure 5. The series of 800 frames are reproduced uncompressed at 7 fps. Image format = 256 x 256 pixels; pixel size = 124 nm; The internal calibration area is identified as dark lateral bands. Please click here to view this video (Right click to download).

Supplemental Movie 2: Time series shown in Figure 5 of a HeLa-mEGFP-FGFR1 cell at time = 7 minutes after FGF2 stimulation (20 ng/mL in PBS supplemented with 0.01% BSA) The series of 800 frames are reproduced uncompressed at 7 fps. Image format = 256 x 256 pixels; pixel size = 124 nm; The internal calibration area is identified as dark lateral bands. Please click here to view this video (Right click to download).

Supplemental Movie 3: Time series shown in Figure 5 of a HeLa-GPI-mEGFP cell after adding the vehicle (PBS supplemented with 0.01% BSA). The series of 1,000 frames are reproduced uncompressed at 7 fps. Image format = 256 x 256 pixels; pixel size = 124 nm; The internal calibration area is identified as dark lateral bands. Please click here to view this video (Right click to download).

Supplemental Movie 4: Time series shown in Figure 5 of a HeLa-GPI-mEGFP-mEGFP cell after adding the vehicle (PBS supplemented with 0.01% BSA). The series of 1,000 frames are reproduced uncompressed at 7 fps. Image format = 256 x 256 pixels; pixel size = 124 nm; The internal calibration area is identified as dark lateral bands. Please click here to view this video (Right click to download).

Discussion

N&B requires several precautions in the choice of the cell model and labelling strategy. It can be applied only to live cells that remain stably adhered during the image capture time. Extra fluctuations due to the whole cell rigid displacement might be handled with appropriate image restoration approaches³⁸. However, generally when a cell moves, the cell membrane also deforms, and structure deformation, producing large extra variance, introduces serious limitation to the analysis of membrane proteins. In this work, the fluorescent construct is expressed in the HeLa cell line because the constitutive FGFR1 is negligible. The presence of the constitutive protein is a condition to avoid, otherwise mixed populations of fluorescent and non-fluorescent receptor oligomers would likely form. Consequently, the N&B analysis, which is based only on the brightness of the fluorescently-tagged protein population, would return an unreliable estimation of the average oligomeric state. This aspect is particularly important when applying N&B to detect receptor dimerization events in which the brightness only increases by a factor of 2, such as the FGFR1 dimerization in the presence of the FGF2 ligand. In such a case, the presence of mixed dimers can completely hide the dimerization events detectable by N&B. The contamination of fluorescent oligomers with non-fluorescent constitutive forms is one of the potential pitfalls of any approach based on the detection of fluorescence, unless the tagged protein is largely overexpressed. However, oligomerization studies in conditions of protein overexpression raise concerns about their functional significance. Transiently transfected cells will likely have variable levels of fluorescence (i.e. protein concentration) and are useful for testing the independence of the average brightness on the protein concentration, since N&B can be applied to a wide range of concentrations, under the condition of linear response of the detector (see the definition of the brightness, eq. 1).

Another constraint is the stoichiometric labelling of the receptor using a stable fluorophore in live cells. Halo-tags, fluorescent proteins (FP) or other covalent fluorescent tags are suitable labels to obtain an exact number of bound fluorophores per molecule of protein in cell. To reduce the complexity of the N&B analysis, we use either fluorescent proteins or halo-tags yielding a 1-to-1 fluorophore-to-protein labelling. The FP of choice must be monomeric in the mature form, having a negligible self-association tendency that, otherwise, might induce artificial oligomerization of the FP-tagged receptors. In this protocol, we use the (A207K)mEGFP because this single point mutation abolishes the mEGFP self-aggregating tendency as previously shown^31,39,40. Regardless of the labelling strategy, the fluorescently tagged protein should be checked for the retention of the biological activity in the chosen cell model, as the functionality of the tagged molecules is essential for linking oligomerization events to receptor signaling. In our model, we proved the autophosphorylation of the fluorescent (A207K)mEGFP-FGFR1 after stimulating the transfected cells with the receptor ligand FGF2³¹.

N&B is based on the diffusion of molecules in the illumination volume. In digital camera detection, the exposure time (i.e., the pixel dwell time in analog detection, t_dwell) must be short enough that one and only one configuration of particles in the focal volume is captured. This is to say that the probability of a fluorophore entering or exiting the illuminated volume during the exposure time is very small. The exposure time must be shorter than the fluorophore residence time (τ_D) which is the average time that a fluorophore spends in the illuminated volume. Therefore, before starting an N&B experiment, it is necessary to have some notion about the residence time (or the diffusion coefficient) of the target protein, which in a first approximation, depends on subcellular localization and molecular mass. Frame rate is the inverse of the time needed for the camera to acquire an image and then completely read that image out, and it is often, calculated approximately from the total number of pixels and the readout rate, combined with the exposure time. Even though frame rate does not need to be fast, cycle time (t_cycle), which is the sum of the frame rate and the time to prepare the capture of the next frame, might affect the analysis. These constraints can be generalized as: t_cycle >> τ_D >> t_dwell. If the cycle time is too fast, the molecules will not have moved appreciably from one frame to the next in that time, and will appear as immobile (B = 1). However, because hundreds of frames are needed for the analysis, cycling is set as fast as possible (increasing or decreasing the image format in number of pixels) to avoid cell displacements and distortion of the cell membrane that would add large extra variance during the series acquisition. In the example of this protocol, the slowest cycle time is 22 ms, so that 500 frames can be acquired in 11 s.

Molecular brightness is not an absolute quantity; It depends on the molecular properties of the fluorophore (cross section and quantum yield), on the experimental set up (detector and optics) as well as on the excitation conditions such as the laser power. Thus, the TIRF-N&B approach yields an apparent brightness and because of that it is fundamental to set up a "reference cell model" that expresses a reference construct with univocal oligomeric state. The reference construct carries the same fluorophore tag of the protein under study [here, the (A207K)mEGFP] and it localizes in the same subcellular compartment (here, the plasma membrane). In this work we use a glycosylphosphatidylinositol (GPI) anchored-(A207K)mEGFP construct that we have repeatedly demonstrated to be a reliable monomeric brightness standard for cell surface receptors labelled with the same fluorophore^30,31.

One major drawback is the occurrence of fluorophore photobleaching that very likely happens when the same cell is repeatedly exposed to light during a kinetic run, and leads to an underestimation of the oligomerization state. To prevent that, the brightness kinetics is obtained by combining the average brightness of different cells each captured at a different time point (Figure 6). This is to say that in a Petri dish each cell is a different time point of the kinetics and a petri dish represents an entire kinetic run.

When the oligomerization dynamics is fast and complex, such as the FGFR1 oligomerization induced by a non-canonical ligand, NCAM³¹, the profile of the normalized average brightness versus time can be variable and unstable (Figure 6B). In such a case, reproducibility cannot be determined by reading replicate dishes of cells at precise time points, due to the necessity of moving and focusing on a new cell each time, select a ROI, capture and inspect/discard time series. Thus, reproducibility can be evaluated in terms of kinetic profile similarity and amplitude of the brightness changes.

An overview of the results is presented in Figure 7. The scatter plot illustrates only the average brightness measured in cells of the same dish, neglecting the time of each capture. In this plot the major difference induced by the two ligands on the oligomerization state of the receptor remain evident, although all kinetic information is removed. However, for both set of experiments (FGF2- versus NCAM-induced oligomerization), the conventional approach of determining the mean ± S.D. of each run in Figure 7 would lead to misleading conclusions since the FGFR1 molecules stimulated by FGF2 clearly transit into two well defined states, whereas NCAM induces unstable and cyclic oligomeric FGFR1 mixtures that would not be well represented by a mean ± S.D. value.

In summary, from an experimental point of view, N&B requires only access to a microscope equipped with a fast acquisition module and a dedicated software. The protein of interest can be tagged with a variety of monomeric fluorescent proteins or organic fluorophores, but quantitative measurements require several conditions to be fulfilled: stoichiometric labelling, reference brightness standard, detector calibration and no photobleaching. In this context, the N&B approach is a powerful tool to decipher the spatiotemporal oligomerization of proteins in live cells. Furthermore, recent advances in resampling raw data to solve the statistical weighting⁴¹ and combining fluorescence cross-correlation spectroscopy with cross-correlation N&B⁴² are refining and improving the applicability of the N&B approach to protein oligomerization and interaction studies.

Materials

Name	Company	Catalog Number	Comments
3-Colour Fast TIRF Leica AM TIRF MC inverted microscope, with smi-automatic TIRF alignment. The microscope is equipped with a diode 488 nm laser, a 100x 1.46 oil TIRF objective, Ex/Em Bandpass filters at 490/20 and 525/50, temperature/CO2 incubator and Andor DU 8285 VP EMCCD camera. The microscope is operated by Leica LIF software.	Leica Microsystems, Wetzlar, Germany
Albumin from Bovine Serum 98% minimun	Sigma-Aldrich, St. Louis, MI, USA	A7906-100G
DMEM without Phenol Red with 25 mM HEPES	GIBCO Thermo Fisher Scientific,Waltham, MA, USA	21063029	Used serum free for microscopy
DMEM high-glucose GlutaMAX I	GIBCO Thermo Fisher Scientific,Waltham, MA, USA	10566-016	Used for complete medium
Dulbecco's Phosphate Buffered Saline 10x (PBS)	Biowest, Nuaillé, France	X0515-500
Emission splitting system Photometrics DV2	TeledynePhotometrics, Tucson, AZ, USA
Fetal Bovine Serum, qualified, Brazil	GIBCO Thermo Fisher Scientific,Waltham, MA, USA	10270106	10% inactivated supplement for complete medium
Glass bottom 35 mm sterile 1.5 dishes	MatTek, Ashland, MA, USA	P35G-0.170-14-C	uncoated, glass thickness 0.17 microns
GraphPad Prism	GraphPad Software Inc., San Diego, CA, USA
Human cervical carcinoma (HeLa), serum-free animal component (AC) cells	Millipore-Sigma ECACC, Darmstadt, Germany	CB_08011102
iXonEM+ 897 EMCCD (back-illuminated) ANDOR camera controlled by ANDOR Solis software	Oxford Instruments, Andor TM Technology, Abingdon-on-Thames, UK		This camera, installed in an additional port of the microscope, is used for acquiring the N&B time series
Matlab Executable N&B routine	Unit of Microscopy and Dynamic Imaging, CNIC, Madrid, Spain		download at https://www.cnic.es/en/investigacion/2/1187/tecnologia
MatLab v.2018b	The MathWorks, Inc. Natick, MA, USA		https://www.mathworks.com/products/matlab.html
Penicillin:Streptomycin for tissue culture 100x	Biowhittaker Inc. Walkersville, MD, USA	LONZA 17-602E	supplement for medium at Penicillin/Streptomycin 100 U/100µg.
pN1-mEGFP-FGFR1 expression vector	Unit of Gynecological Oncology Research, European Institute of Oncology IRCCS, Milan, Italy		Zamai et al., 2019
pN1-N-Gly-mEGFP-GPI expression vector	Unit of Microscopy and Dynamic Imaging, CNIC, Madrid, Spain		Hellriegel et al., 2011
pN1-N-Gly-mEGFP-mEGFP-GPI expression vector	Unit of Microscopy and Dynamic Imaging, CNIC, Madrid, Spain		Hellriegel et al., 2011
Recombinant FGF2	PeproTech EC, Ltd., London, UK		Ligand solution: 20 ng/mL of FGF2 in PBS supplemented with 0.01%BSA.
Sodium pyruvate GIBCO	ThermoFisher Scientific	11360070	1 mM supplement for medium
TransIt-LT1 Transfection Reagent	MirusBio LLC, Madison, WI, USA	MIR 2300
Trypsin-EDTA (0.25%), phenol red	GIBCO Thermo Fisher Scientific,Waltham, MA, USA	25200056
Type F Immersion liquid 10 mL	Leica Microsystems, Wetzlar, Germany	11513 859
UltraPure BSA (50 mg/mL)	ThermoFisher Scientific	AM2618	0.1% supplement for medium without phenol red used for transfections