This protocol describes a fluorescence fluctuation spectroscopy-based approach to investigate interactions among proteins mediating cell-cell interactions, i.e. proteins localized in cell junctions, directly in living cells. We provide detailed guidelines on instrument calibration, data acquisition and analysis, including corrections to possible artefact sources.
A variety of biological processes involves cell-cell interactions, typically mediated by proteins that interact at the interface between neighboring cells. Of interest, only few assays are capable of specifically probing such interactions directly in living cells. Here, we present an assay to measure the binding of proteins expressed at the surfaces of neighboring cells, at cell-cell contacts. This assay consists of two steps: mixing of cells expressing the proteins of interest fused to different fluorescent proteins, followed by fluorescence fluctuation spectroscopy measurements at cell-cell contacts using a confocal laser scanning microscope. We demonstrate the feasibility of this assay in a biologically relevant context by measuring the interactions of the amyloid precursor-like protein 1 (APLP1) across cell-cell junctions. We provide detailed protocols on the data acquisition using fluorescence-based techniques (scanning fluorescence cross-correlation spectroscopy, cross-correlation number and brightness analysis) and the required instrument calibrations. Further, we discuss critical steps in the data analysis and how to identify and correct external, spurious signal variations, such as those due to photobleaching or cell movement.
In general, the presented assay is applicable to any homo- or heterotypic protein-protein interaction at cell-cell contacts, between cells of the same or different types and can be implemented on a commercial confocal laser scanning microscope. An important requirement is the stability of the system, which needs to be sufficient to probe diffusive dynamics of the proteins of interest over several minutes.
Many biological processes occur at the sites of cell-cell interactions, e.g., cell-cell adhesion1,2,3, cell-cell fusion4 and cellular recognition5. Such events are particularly important during the development of multicellular organisms and for cell-cell communication, e.g., during immune responses. These processes are typically mediated by proteins that are localized at the surface, i.e., at the plasma membrane (PM) of neighboring cells and undergo specific interactions at the cell-cell contact that are precisely regulated in space and time. In many cases, these interactions are direct homo- or heterotypic protein-protein trans interactions, but may also involve ions or ligands acting as extracellular linkers1. Although of fundamental importance, there is a lack of assays probing these specific protein-protein interactions directly in the native environment of living cells. Many methods either require cell disruption (e.g., biochemical assays such as co-immunoprecipitation6), fixation (e.g., some of the super-resolution optical microscopy techniques and electron microscopy of cell-cell contacts7), or are non-specific, e.g., aggregation/ adhesion assays8,9. To overcome this issue, fluorescence techniques have been implemented based on fluorescence resonance energy transfer (FRET)10 or fluorescence complementation11. However, to achieve sufficiently small distances between fluorophores, these methods require fluorescent labels on the extracellular side of the proteins10, potentially interfering with trans interactions.
Here, we present an alternative fluorescence-based assay for protein-protein interactions at cell-cell contacts. This approach combines fluorescence cross-correlation approaches (scanning fluorescence cross-correlation spectroscopy (sFCCS), cross-correlation number and brightness (ccN&B)) and mixing of cells expressing a fusion construct of the protein of interest, e.g., an adhesion receptor. The investigated receptors in the two interacting cells are labeled with two spectrally separated fluorescent proteins (FPs), from the intracellular side (see Figure 1A).
The employed methods are based on the statistical analysis of fluorescence fluctuations induced by the diffusive motion of fluorescent fusion proteins through the focal volume of a confocal laser scanning microscope. More in detail, the assay probes the co-diffusion of the proteins of interest in both neighboring PMs at cell-cell contacts. If the proteins undergo trans interactions, these trans complexes will carry fluorescent proteins emitting in both spectral channels, causing correlated fluorescence fluctuations of both emitters. On the other hand, if no binding occurs, the number fluctuations of proteins in facing PMs will be independent, causing no correlated fluctuations. The acquisition can be performed in two ways: 1) sFCCS is based on a line-shaped scan across the cell-cell contact and effectively probes the interactions in a spot located in the contact region. Through a temporal analysis of fluorescence fluctuations, sFCCS provides also dynamics information, i.e., the diffusion coefficients of protein complexes; 2) ccN&B is based on a pixel-wise analysis of a sequence of images acquired at the cell-cell contact regions. It has capability to probe and map interactions along the whole contact region (in one focal plane), but does not provide information on dynamics. Both methods can be combined with an analysis of the molecular brightness, i.e., the average fluorescence signal emitted in the time unit by single diffusing protein complexes and, thus, provide estimates of the stoichiometry of protein complexes at cell-cell contacts.
In this article, we provide detailed protocols for sample preparation, instrument calibration, data acquisition and analysis to perform the presented assay on a commercial confocal laser scanning microscope. The experiments can be performed on any instrument equipped with photon counting or analog detectors and an objective with high numerical aperture. We further discuss critical steps of the protocol and provide correction schemes for several processes causing artefactual signal fluctuations, e.g., detector noise, photobleaching or cell movement. Originally developed to probe interactions between adherent cells, the assay may be modified for suspension cells, or adapted to model membrane systems, e.g., giant unilamellar vesicles (GUVs) or giant plasma membrane vesicles (GPMVs), allowing the quantification of interactions in different lipid environments or in the absence of an organized cytoskeleton12,13.
Scanning fluorescence cross-correlation spectroscopy is a modified version of fluorescence cross-correlation spectroscopy14 and was specifically designed to probe slow diffusive dynamics in lipid membranes15. It is based on a line scan acquisition perpendicular to the PM containing the fluorescent proteins of interest. To probe interactions of two differently labeled protein species, the acquisition is performed in two spectral channels using two laser lines and two detection windows for spectrally separated fluorophores. Due to the slow diffusion dynamics of proteins in the PM (D≤~1 µm2/s), a cross-talk-free measurement can be performed by alternating the excitation scheme from line to line15. The analysis starts with: 1) an alignment algorithm correcting for lateral cell movement based on block-wise averaging of ~1000 lines, 2) determination of the position with maximum fluorescence signal, i.e., the PM position, in each block and 3) shifting of all blocks to a common origin12,15, separately in each channel. Then, an automatic selection of pixels corresponding to the PM is performed by selecting the central region from a Gaussian fit of the sum of all aligned lines (i.e., center ± 2.5σ). Integration of the signal in each line yields the membrane fluorescence time series F(t) in each channel (g = green channel, r = red channel). Note that the pixel size has to be small enough, e.g., <200 nm, to reconstruct the shape of the point spread function and find its center, corresponding to the position of the PM. In the presence of substantial photobleaching, the fluorescence time series in each channel may be modeled with a double-exponential function and then corrected with the following formula:16
. (1)
It is important to note that this formula effectively corrects both the amplitudes and diffusion times obtained from correlation analysis of F(t)c, compared to parameter estimates that would be obtained from the uncorrected F(t). Then, the auto- and cross-correlation functions (ACFs/ CCFs) of the fluorescence signals are calculated:
, (2)
, (3)
where δFi = Fi(t) – Fi(t) and i = g,r.
A two-dimensional diffusion model is then fitted to all correlation functions (CFs):
. (4)
Here, N denotes the number of fluorescent proteins in the observation volume and τd the diffusion time for each channel. This model takes into account that in the described experimental setting, diffusion of proteins in the PM occurs in the x-z plane, in contrast to the commonly used configuration of fluorescence correlation spectroscopy (FCS) experiments on membranes probing diffusion in the x-y plane of the confocal volume17. The waist w0 and the structure factor S, describing the elongation wz of the focal volume in z, S = wz/w0, are obtained from a point FCS calibration measurement performed with spectrally similar dyes and same optical settings using already available values for the diffusion coefficient Ddye:
, (5)
where τd,dye is the measured average diffusion time of the dye molecules, obtained from fitting a model for three-dimensional diffusion to the data, taking into account transitions of a fraction T of all N molecules to a triplet state with a time constant ττ:
. (6)
Finally, diffusion coefficients (D), molecular brightness values (ε) and the relative cross-correlation of sFCCS data (rel.cc.) are calculated as follows:
, (7)
, (8)
, (9)
where Gcross(0) is the amplitude of the cross-correlation function and is the amplitude of the autocorrelation function in the i-th channel.
This definition of the relative cross-correlation, i.e. using max instead of mean in Equation 9, takes into account that the maximum number of complexes of two protein species present at different concentrations is limited by the species present in a lower number.
Cross-correlation number and brightness is based on a moment analysis of the fluorescence intensity for each pixel of an image stack acquired over time at a fixed position in the sample, typically consisting of ~100-200 frames, with two spectral channels (g = green channel, r = red channel). From the temporal mean Ii and variance , the molecular brightness εi and number ni are calculated in each pixel and spectral channel (i = g, r)18:
, (10)
. (11)
It is important to note that the given equations apply to the ideal case of a true photon-counting detector. For analog detection systems, the following equations apply19,20:
, (12)
. (13)
Here, S is the conversion factor between detected photons and the recorded digital counts, is the readout noise and offset refers to the detector intensity offset. Generally, these quantities should be calibrated, for any detector type, based on measuring the detector variance as a function of intensity for steady illumination19, e.g., a reflective metal surface or dried dye solution. The offset can be determined by measuring the count rate for a sample without excitation light. By performing a linear regression of the detector-associated variance versus intensity (I) plot, S and can be determined19:
. (14)
Finally, the cross-correlation brightness is calculated in each pixel and is defined in general as21
, (15)
where is the cross-variance .
In order to filter long-lived fluctuations, all ccN&B calculations are performed following a boxcar filtering, independently for each pixel22. Briefly, ni, εi (i = g, r) and Bcc are calculated in sliding segments of e.g., 8-15 frames. The values thus obtained can be then averaged to obtain the final pixel number and brightness values.
Stoichiometry analysis
In order to estimate the stoichiometry of protein complexes at cell-cell contacts, the molecular brightness can be separately analyzed in each spectral channel for the sFCCS or ccN&B data. In sFCCS, one brightness value is obtained per measurement in each channel. In ccN&B, a brightness histogram of all pixels corresponding to the cell-cell contact is obtained and the average (or median) value can be used as representative brightness for the measurement. By performing the same analysis on a monomeric reference, all brightness values can be normalized to directly obtain the average oligomeric state of the detected protein complexes. At this point, it is important to correct for the presence of non-fluorescent FPs that may result in an underestimation of the oligomeric state. This is typically performed by measuring the brightness of a homo-dimeric reference protein23,24 using one-color sFCS or number and brightness (N&B).
1. Sample Preparation: Cell-Cell Mixing Assay
NOTE: The following protocol describes the mixing procedure for adherent cells. It may be modified for cells cultured in suspension.
Figure 1. Experimental workflow and schematic representation of scanning fluorescence cross-correlation spectroscopy and cross-correlation number and brightness analysis at cell-cell contacts. (A) Scheme of sample preparation: Two cell populations transfected with the protein of interest (e.g., APLP1) fused to two spectrally distinct fluorescent proteins (e.g., mEYFP and mCardinal) are mixed after transfection. Contacts of differently transfected cells are selected in the microscopy experiments. To avoid interference with extracellular binding domains, the fluorescent protein should be fused to the intracellular terminus of the protein of interest. (B) Scanning FCCS (sFCCS) measurements are performed perpendicular to the cell-cell contact in two spectral channels (channel 1, green and channel 2, red). Scan lines (represented as kymographs) are aligned and membrane pixels summed. Then, ACFs and CCFs are calculated from the intensity traces Fi(t). ACFs are represented in red and green. CCF is represented in blue. (C) Cross-correlation N&B (ccN&B) acquisition results in a three-dimensional (x-y-time) image stack. A ROI is selected around the cell-cell contact. Then channel and cross-correlation brightness (ε1, ε2, and Bcc) values are calculated in each cell-cell contact pixel. The results are then visualized as histograms, pooling all selected pixels. Please click here to view a larger version of this figure.
2. Sample Preparation: Positive Control for Cross-Correlation Experiments and Homo-Dimer Construct for Brightness Analysis
3. Confocal Laser Scanning Microscopy: Setup and Focal Volume Calibration
NOTE: The following protocol is written for experiments performed with mEGFP/mEYFP and mCherry/mCardinal on the laser scanning confocal microscope used in this study. The optical setup, the software settings (laser lines, dichroic mirrors, filters) and choice of calibration dyes may be modified for other FPs and microscope setups.
4. Scanning Fluorescence Cross-Correlation Spectroscopy: Acquisition
NOTE: The following protocol is written for experiments performed with mEGFP/mEYFP ('green') and mCherry/mCardinal ('red') on the laser scanning confocal microscope used in this study. The optical setup and the software settings (laser lines, dichroic mirrors, filters) may be different for other FPs or microscope setups.
5. Scanning Fluorescence Cross-Correlation Spectroscopy: Data Analysis
NOTE: The following protocol follows an implementation of the analysis procedure described in detail in previous articles12,15. The software code is available upon request to the authors.
6. Cross-Correlation Number and Brightness: Detector Calibration
NOTE: The following protocol provides a general guideline regarding how to calibrate the detection system. This procedure is mandatory for analog detection systems, but is not strictly needed when true photon counting detectors are used.
7. Cross-Correlation Number and Brightness: Acquisition
8. Cross-correlation Number and Brightness: Data Analysis
NOTE: The following protocol follows a previously described analysis procedure12,31. The software code is available from the authors upon request.
A first test for the protein-protein interaction assay, i.e., mixing of cells expressing spectrally distinct fluorescent proteins followed by sFCCS/ccN&B measurements (Figure 1), should be performed on proteins that are not expected to interact at the cell-cell contact (i.e., a negative control). Therefore, HEK 293T cells expressing myristoylated-palmitoylated-mEYFP (myr-palm-mEYFP) or -mCardinal were mixed and sFCCS was performed across the cell-cell contact (Figure 2A). In an ideal case, the fluorescence signal in each channel is supposed to fluctuate around a stable mean, as a consequence of the diffusive motion of the proteins in the PM and the statistical variations of the number of proteins in the focal volume. For proteins that do not interact, the fluctuations in both channels are independent from each other and, thus, the spectral cross-correlation is expected to fluctuate around zero. Indeed, a relative cross-correlation close to zero was observed in typical measurements (Figure 2C). The ACFs show characteristic decay times of ~10-20 ms (corresponding to, on average, Dmyr-palm = 1.3 ± 0.3 µm2/s (mean ± SD, n = 20 cells)), as expected for the diffusion of myr-palm-mEYFP and -mCardinal in the PM, e.g., Dmyr-palm = 0.88 ± 0.11 µm2/s (mean ± SEM) based on fluorescence recovery after photobleaching (FRAP) experiments35. Notably, these rather slow dynamics allow the use of an alternating excitation scheme, i.e., switching between only green and only red excitation and detection every line, causing a ~0.5 ms delay between signals in both channels but suppressing spectral cross-talk. On average, a very low average cross-correlation of 0.08 ± 0.10 (mean ± SD, n = 17 cells) was obtained for the negative control (Figure 3E), as expected.
Next, a positive cross-correlation control was used to calibrate the maximum possible cross-correlation in the optical setup. Therefore, the membrane-anchored hetero-dimer myr-palm-mCardinal-mEYFP was expressed in HEK 293T cells and sFCCS measurements were performed on single cells (Figure 2B). The obtained CCFs had positive amplitudes and showed similar decay times as the ACFs, again ~10-20 ms (Figure 2D). On average, a relative cross-correlation of 0.96 ± 0.18 (mean ± SD, n = 14 cells) was measured for the positive control (Figure 3E).
Figure 2. Scanning fluorescence cross-correlation spectroscopy control measurements. (A) Representative images of mixed HEK 293T cells expressing myr-palm-mEYFP/-mCardinal as negative control for trans interactions. The yellow arrow indicates the sFCCS scan path. Scale bars are 5 µm. (B) Representative images of HEK 293T cells expressing myr-palm-mCardinal-mEYFP hetero-dimer (left: green channel, right: red channel) as positive cross-correlation control. The yellow arrow indicates the sFCCS scan path. Scale bars are 5 µm.(C) Representative CFs (green: ACF in green channel (mEYFP), red: ACF in red channel (mCardinal), blue: CCF) obtained in sFCCS measurements for negative control. Solid lines show fits of a two-dimensional diffusion model to the CFs. (D) Representative CFs (green: ACF in green channel (mEYFP), red: ACF in red channel (mCardinal), blue: CCF) obtained in sFCCS measurement of the positive control. Solid lines show fits of a two-dimensional diffusion model to the CFs. Please click here to view a larger version of this figure.
The suitability of this assay was then investigated in a biologically relevant context by probing the trans interactions of the amyloid precursor-like protein 1 (APLP1), a type I transmembrane protein that has been proposed to act as a neuronal adhesion receptor. To this aim, APLP1 fused to mEYFP or mCardinal was expressed in HEK 293T cells. To exclude interference with extracellular binding domains, the FPs were fused to the C-terminus of APLP1, i.e., at the intracellular domain (see Figure 1A). Then, sFCCS measurements were performed on cell-cell contacts between APLP1-mEYFP and APLP1-mCardinal expressing cells (Figure 3A), resulting in ACFs and CCFs (Figure 3C) that provide information about APLP1 diffusion and interactions. A positive relative cross-correlation of 0.45 ± 0.21 (mean ± SD, n = 17 cells) was observed, i.e., a value significantly larger than that of the negative control (Figure 3E). Interestingly, the average relative cross-correlation was lower than that of the positive control (Figure 3E), indicating only partial trans binding.
Finally, the assay was used to show that zinc ions facilitate enhanced APLP1 trans binding12,31. The sFCCS CFs obtained from measurements across APLP1 clusters at cell-cell contacts (characterized by a strong co-localization of APLP1-mEYFP and APLP1-mCardinal and forming rapidly in the presence of zinc ions, Figure 3B) showed strongly reduced dynamics, as evident from large decay times and the oscillations at large lag times (Figure 3D). Nevertheless, analysis of the amplitudes at short lag times revealed a significant increase of the relative cross-correlation to 0.8 ± 0.3 (mean ± SD, n = 17 cells), i.e., ~80% of the calibrated maximum (Figure 3E). It is to be expected that such slow dynamics induce severe distortions of the correlation curves (see Figure 3D) due the limited number of diffusive events that can be detected during the finite measurement time, inducing so-called particle noise30. For an accurate quantification, the maximum lag time should be at least 3 orders of magnitude above the diffusion time (see previous reviews30,17 for further details).
The molecular brightness from the sFCCS APLP1 data was further analyzed, using the myr-palm negative control as a monomeric reference in each channel and correcting for the amount of non-fluorescent proteins23. Upon zinc ion addition, the molecular brightness significantly increased from small oligomers (~dimers) to larger multimers consisting of ~10-50 monomers on each cell (Figure 3F). Thus, on average, up to ~100 APLP1 monomers are present in a whole protein cluster across the cell-cell junction.
Figure 3. Scanning fluorescence cross-correlation spectroscopy measurements of APLP1 interactions at cell-cell contacts. (A, B) Representative images of HEK 293T cells expressing APLP1-mEYFP (green)/ APLP1-mCardinal (red) before (A) and 30 min after zinc ion treatment (B, different cells). The yellow arrows indicate the sFCCS scan paths. Scale bars are 5 µm. (C, D) Representative CFs (green: ACFs in green channel (mEYFP), red: ACFs in red channel (mCardinal), blue: CCFs) obtained in sFCCS measurements for (C) APLP1 before zinc ion treatment and (D) after zinc ion treatment. Solid lines show fits of a two-dimensional diffusion model to the CFs. (E) Box plots of relative cross-correlation obtained from sFCCS analysis of negative control ("negative"), APLP1 in absence and presence of zinc ions, and positive cross-correlation control ("positive"). Plots show median values and whiskers ranging from minimum to maximum values. (F) Box plots of normalized molecular brightness in green channel (mEYFP) obtained from sFCCS analysis of APLP1 at cell-cell contacts in the absence and presence of zinc ions. Brightness values were corrected for non-fluorescent mEYFP based on sFCS measurements of myr-palm-mEYFP-mEYFP homo-dimers expressed in HEK 293T cells, measured under the same conditions23. Plots show median values and whiskers ranging from minimum to maximum values. Please click here to view a larger version of this figure.
All the measurements shown so far were corrected for additional, spurious fluctuations occurring in the cells that, if not taken into account, would make the sFCCS analysis challenging. For the negative control, for example, these may cause a false-positive cross-correlation, if no correction schemes are applied. Two major processes that can severely distort the CFs are: 1) instabilities in the recorded fluorescence signal due to intracellular vesicles that transiently enter the focal volume or slow membrane dynamics, e.g., drift in z-direction, and 2) photobleaching. To identify transient instabilities, it is recommended to divide the full measurements in 10-20 equally sized segments and to visually inspect the intensity time series and CFs in each segment. This procedure is illustrated in Figure 4, which gives an example of clearly distorted segments obtained in the analysis of a negative control measurement. Transient instabilities (Figure 4A, intensity traces of segments 1 and 2) may result in a CCF having negative values (Figure 4B, segment 1) or a high false-positive cross-correlation (Figure 4B, segment 2). Typically, such instabilities are visible in the intensity series as slow signal variations (Figure 4A). The corresponding CFs typically deviate strongly from the CFs of the majority of segments, displaying, e.g., higher amplitude and much slower decay times on the ~second scale (see Figure 4B–4D). It is recommended to remove such segments from the analysis and to calculate the final CFs by averaging all non-distorted segments, i.e., segments characterized by CFs not deviating from the majority of CFs. Typically, this is done in a graphical interface by 1) iteratively examining the segments, 2) removing clearly distorted segments from the average and 3) inspecting the CFs of remaining segments with respect to the updated average CFs of segments that were not removed. By applying this procedure, partially distorted long measurements (Figure 5A), showing slowly decaying (corrupted) CFs (Figure 5B) were successfully corrected and meaningful correlation curves were recovered (Figure 5C). Generally, this correction procedure can be automatized29, avoiding a visual inspection by the user, which may be prone to subjective bias. In comparison to intensity based filtering methods36, in which small bins of the measurement are evaluated based on their intensity compared to the average intensity of the full measurement and removed if exceeding a threshold parameter, the described procedure does not rely on external parameters and is sensitive also to minor instabilities.
In order to compensate for photobleaching, the described mathematical correction, Equation 1, was applied. Typically, photobleaching can be identified by an exponential decay of the fluorescence signals (Figure 5D), dominating the CFs, which then show a false-positive cross-correlation and decay times on the ~min scale (see the typical curve shape in Figure 5E). The correction procedure recovered the non-distorted CFs (Figure 5F). As already mentioned, similar intensity variations within single measurements may be also caused by PM movement, e.g., z-drift, or large slowly moving structures. However, if similar exponential decays appear in all measurements, photobleaching will be the most likely source. Generally, correction schemes can be combined, e.g., intensity filtering, CF based filtering and further methods such as Fourier transform based filtering of slow signal variations in frequency space27.
Figure 4. Segment-wise analysis of scanning fluorescence cross-correlation spectroscopy measurements of negative cross-correlation control. (A) Fluorescence intensity in green (F1) and red channel (F2) of two different time segments (each measurement was analyzed in 20 segments of ~20 s each), obtained from sFCCS measurement of negative control. (B) CCFs of each of the 20 segments. The CCFs for segments 1 and 2 are highlighted in red and orange, respectively. (C, D) ACFs of each segment in green (C) and red (D) channel. The ACFs for segments 1 and 2 are highlighted in red and orange, respectively. Please click here to view a larger version of this figure.
Figure 5. Perturbations in scanning fluorescence cross-correlation spectroscopy measurements at cell-cell contacts, exemplified for negative cross-correlation control. (A) Full fluorescence time series for an exemplar measurement in green (F1) and red channel (F2). The solid red lines represent a double-exponential fit of the time series in each channel. (B, C) CFs (green: ACFs in green channel, red: ACFs in red channel, blue: CCFs) of the fluorescence time series shown in A, calculated by (B) correlating the whole measurement or (C) correlating 20 segments separately and averaging the least distorted CFs of ~80% (green channel) and ~50% (red channel) of the segments. Solid lines represent fit of a two-dimensional diffusion model to the data. (D) Full fluorescence time series and double-exponential fit (solid red lines) of measurement characterized by substantial bleaching in green (F1) and red channel (F2). (E, F) CFs (green: ACFs in green channel, red: ACFs in red channel, blue: CCFs) of the fluorescence time series shown in D, calculated by (E) correlating the whole measurement or (F) applying the bleaching correction, Equation 1, correlating 20 segments separately and averaging the least distorted CFs of ~90% (both channels) of the segments. Solid lines represent fit of a two-dimensional diffusion model to the data. Please click here to view a larger version of this figure.
As a complementary approach to sFCCS, ccN&B (Figure 1C) can be used to detect protein-protein interactions after cell mixing. In contrast to sFCCS, ccN&B provides no information on protein dynamics, but allows measuring trans interactions along the whole cell-cell contact in one focal plane. Measurements on APLP1 samples were performed before and after treatment with 50 µM ZnCl2, as well as on the negative myr-palm control. These measurements are sensitive to cell movements, especially for zinc ion treated samples, requiring prolonged acquisition times to account for the slow dynamics of APLP1 clusters. Therefore, an image alignment algorithm was implemented to correct for lateral movement of the cells33. Also, a boxcar filter22 (8 frames box size, ~5 s) was applied to remove low frequency fluctuations of the measured signals. This procedure is very similar to other filters used in N&B analysis involving local averaging37 or detrending18,34, but keeps the original data unaltered, i.e., pixels are treated independently and no averaging or subtraction of signals is performed. This procedure effectively suppresses long-lived fluctuations on time scales longer than the box size22. Following such data analysis, the cross-correlation brightness values for all samples were compared by pooling all cell-cell contact pixels in a cross-correlation brightness histogram. For APLP1 in the absence of zinc ions (Figure 6A and 6D), a positive average Bcc of 0.068 ± 0.004 (mean ± SEM, n = 18 cells) was observed. After zinc ion addition (Figure 6B and 6E), the Bcc value increased to 0.266 ± 0.006 (mean ± SEM, n = 19 cells). For the negative control (Figure 6C and 6F), a lower average cross-correlation brightness was detected (Bcc =0.022 ± 0.002, mean ± SEM, n = 26 cells). To estimate the stoichiometry of APLP1 complexes at cell-cell contacts, the brightness of APLP1-mEYFP was normalized using the average value obtained for myr-palm-mEYFP (i.e., the negative control) and corrected for the amount of non-fluorescent proteins23. In agreement with sFCCS data, the brightness distribution was centered around a value corresponding to dimers (Figure 6G), in the absence of zinc ions, suggesting an average 2:2 stoichiometry. After zinc ion treatment, the normalized brightness strongly shifted to larger values, ranging from ~10 to ~60 (Figure 6H), i.e., stoichiometries of at least 10:10 or larger, again in good agreement with sFCCS data.
Figure 6. Cross-correlation Number and Brightness measurements of APLP1 interactions at cell-cell contacts. (A–C) Representative ccN&B image frames of cell-cell contacts between APLP1-mEYFP and APLP1-mCardinal expressing HEK 293T cells without (A) and with zinc ions (B) or myr-palm-mEYFP and myr-palm-mCardinal expressing cells as negative cross-correlation control (C). Scale bars are 5 µm. (D–F) Cross-correlation brightness (Bcc) histograms of all examined pixels and cells obtained from ccN&B analysis of cell-cell contacts in APLP1 samples (D), zinc-treated APLP1 samples (E) and samples containing myr-palm-mEYFP and myr-palm-mCardinal (F). (G, H) Normalized brightness histograms of APLP1 samples (G: without zinc ions, H: with zinc ions) for the green channel (mEYFP) obtained from brightness analysis of the same cells and ROIs used for the calculation of Bcc. Inset in G shows a magnification in the normalized brightness range of -2 to 10. Brightness values were corrected for non-fluorescent mEYFP based on N&B measurements of myr-palm-mEYFP-mEYFP homo-dimers expressed in HEK 293T cells, measured under the same conditions23. Please click here to view a larger version of this figure.
In order to perform the ccN&B analysis, a careful calibration of the detectors should be performed. For the experimental setup used in this study, the need for such a calibration became apparent when the molecular brightness was analyzed in fixed samples (i.e., in the absence of number fluctuations). In this case, the molecular brightness is expected to be zero according to Equation 10, since the variance should only originate from detector noise. However, when a ROI that contains only (immobile) background pixels was analyzed (Figure 7A and 7B), a positive brightness of ~0.1 cts./(MOL x dwell time) was determined. A similar value was obtained performing N&B measurements of HEK 293T cells expressing glycosylphosphatidylinositol-mCherry (GPI-mCherry, Figure 7C) with varying laser powers and extrapolating the measured molecular brightness to zero laser power. To correct for this effect, we performed a systematic calibration of the detector, as previously reported (see Equation 12)19,20. We therefore measured the variance as a function of detector count rate on a sample consisting of a dried fluorescent dye solution and determined the parameters S, and the dark count rate offset. The latter was obtained from measuring the intensity at zero laser power and was, in our case, negligible. From a linear fit of the variance versus intensity plot, we determined, according to Equation 14, a slope of S = 1.1, and a negligible readout noise. The determined values (S = 1.1, = 0, offset = 0) were then used to correctly calculate molecular brightness and number according to Equations 12 and 1319. Alternatively, a more empirical correction scheme can be applied based on the fact that a superpoissonian detector noise, i.e., ~10% larger noise than Poissonian shot noise, would also explain the observation of a positive molecular brightness in pixels without number fluctuations. Using this assumption, the determined brightness of ~0.1 cts./(MOL x dwell time) in such pixels can be considered as a constant brightness offset and should thus be subtracted from all brightness values calculated with Equation 10. Most importantly, both described approaches lead to the same results when calculating brightness ratios, as long as and offset are negligible. It is worth mentioning that among different microscopes of the same type (same vendor, same model, GaAsP detectors in photon counting mode) different S values were observed, highlighting the necessity for a careful detector calibration on each individual setup.
A further important parameter affecting the detector performance in brightness measurements is the detector dead time. As it has been previously shown, the detector dead time can substantially reduce the detected molecular brightness, even at medium count rates (above 102-103 kHz)38. To avoid this artefact, measurements should either be performed at lower count rates, or the dead time should be calibrated based on performing N&B or FCS in a dilution series of, e.g., EGFP in buffer solution. Then, measured count rates can be corrected using a calibrated dead time38. For the setup used in this study, such a calibration using N&B on diluted dye solutions revealed stable molecular brightness values up to count rates of ~0.5 MHz and a corresponding dead time of ~6 ns (Figure 7E). The decrease at higher count rates can be thus corrected by using a previously published correction formula38, resulting in a constant brightness value of ~8 kHz/MOL.
Figure 7. Detector calibration for Number and Brightness analysis. (A) Representative image from N&B measurements of HEK 293T cells expressing GPI-mCherry. A ROI (blue dashed rectangle) was selected in the background. (B) Pixel brightness histogram of all pixels corresponding to the ROI shown in A. The average pixel brightness, obtained from fitting the pixel brightness histogram with a Gaussian function, is ~0.1 cts./(MOL x dwell time). Data were acquired at 25 µs pixel dwell time. (C) Molecular brightness obtained from N&B analysis of GPI-mCherry expressed in HEK 293T cells, measured at three different laser powers (6 cells each, 561 nm excitation, 25 µs pixel dwell time). Data are displayed as mean ± SD. A linear regression (red line) provides an offset value of 0.11 ± 0.02 cts./(MOL x dwell time). (D) Plot of pixel variance as a function of pixel intensity from N&B measurements of a dried solution of a fluorescent dye (excited at 561 nm), pooled from all pixels of multiple measurements in different regions of the sample. The solid red line shows a linear fit of the data, resulting in a slope of 1.1, providing the S factor of the detector calibration. (E) Molecular brightness as a function of detector count rate, obtained from N&B measurements of diluted fluorophore solutions (excited at 488 nm). A previously published correction scheme38 was applied using different possible values for the detector dead time. Please click here to view a larger version of this figure.
The experimental procedure described here allows the investigation of protein-protein trans interactions at cell-cell contacts, employing fluorescence fluctuation spectroscopy techniques, namely sFCCS and ccN&B. These methods involve a statistical analysis of fluorescence fluctuations emitted by two spectrally separated FPs fused to the protein(s) of interest at a contact of two neighboring cells, each expressing one or the other fusion protein. The presence of trans complexes is quantified by probing the degree of co-diffusion of proteins in neighboring PMs. In addition to detailed protocols on sample preparation, data acquisition and analysis, this article provides experimental evidence of the successful application of the assay on the neuronal adhesion protein APLP1. We show that APLP1 undergoes specific, homotypic trans interactions at cell-cell contacts. Further, zinc ions promote the formation of APLP1 clusters at cell-cell contacts that provide a multivalent platform for trans interactions and, thus, induce enhanced trans binding.
In contrast to previous assays to detect such interactions based on disruptive biochemical methods6, the presented approach can be performed directly on living cells, with no need for fixation or isolation of protein complexes. Moreover, it provides molecular specificity and information by detection of fluorescent proteins genetically fused to the protein of interest, in contrast to previous qualitative assays8,9. Differently from other fluorescence based approaches such as FRET10 and fluorescence complementation11, there is no requirement for the fluorescent labels to be localized on the extracellular side (potentially interfering with protein-protein interactions). Nevertheless, it has to be noted that C-terminal FPs might still alter the binding to intracellular components, e.g., adaptor proteins that mediate interactions with the cytoskeleton. Notably, the assay is applicable to both homo- and heterotypic interactions.
A few requirements for a successful application of the presented assay are worth mentioning. The performed fluorescence fluctuation methods are based on temporal measurements that require bright and photostable monomeric FPs, which is a major constraint for many red FPs23. Even though the presented scanning scheme is particularly well suited for the investigation of the slow dynamics of transmembrane proteins, which are typically involved in cell-cell interactions, residual photobleaching may still occur. Therefore, we present a detailed description of correction schemes, i.e., a bleaching correction for sFCCS and boxcar filter for ccN&B, which minimize such perturbations. Further, we discuss numerical alignment algorithms that effectively correct for other systematic perturbations, such as lateral movement of cells, and provide guidelines to remove transient instabilities. A major source of such instabilities is vesicular transport, i.e., intracellular vesicles carrying the protein of interest that transiently enter the focal volume. Although varying from case to case, this phenomenon can in general severely disturb data acquisition and analysis. However, the application of correction algorithms improves the stability of the acquisition, allowing extended measurement times that are needed to probe the especially slow diffusive dynamics. In this regard, it has to be underlined that the presence of diffusive dynamics is an essential condition for the method to work. The example of zinc ion mediated APLP1 clustering shows a drastic example of large, very slow protein complexes (D ≈ 0.001 µm2/s) that push the technique to its limits and prevent an accurate quantification of the underlying dynamics, due to large noise induced by the limited acquisition time30,17.
In this context, recent advances on combing sFCS and super-resolution approaches, e.g., scanning stimulated emission depletion FCS (STED-FCS), may be beneficial by providing a smaller focal volume and thus smaller effective diffusion times39,40. This could also help to resolve protein clusters in case of an inhomogeneous localization of the protein of interest at the cell-cell contacts, as observed for APLP1 in the presence of zinc. Unfortunately, a cross-correlation implementation of scanning STED-FCS, i.e., STED-FCCS, that could be directly applied here, has not been successfully demonstrated yet due to the difficulty of finding compatible dyes for two-color STED-FCS. In the case of sufficiently fast dynamics (D ≥ ~0.05 µm2/s), sFCCS analysis allows to quantify the diffusion of proteins at cell-cell contacts or in other regions of the PM12.
Further, a brightness analysis can be performed for all data (sFCCS and ccN&B), providing an estimate of the stoichiometry of trans protein complexes. It should be mentioned that the accuracy of the stoichiometry quantification increases with the amount of proteins that bind in trans (i.e., if there are only few cis complexes which could also affect the brightness determination). Brightness analysis does not allow resolving mixtures of different oligomeric states, e.g.,cis monomers and trans tetramers. More in general, it should be noted that N&B cannot resolve mixtures of different oligomeric species homogeneously distributed in a sample. If multiple species are present within a pixel, one single value of brightness will be measured (i.e., a weighted average of the brightness of each oligomeric species). The statistical distribution of the observed brightness values (e.g., in different pixels) is not directly connected to the relative amounts of oligomeric species. To resolve such mixtures, other methods should be used (e.g., spatial intensity distribution analysis (SpIDA)41, photon counting histogram (PCH)42). Similarly, the quantification of the spectral cross-correlation in sFCCS provides an accurate quantification of the fraction of bound and unbound proteins only for simple, known stoichiometries, e.g., 1:1. For a more complex stoichiometry, further assumptions have to be made43. Overall, brightness analysis remains a powerful experimental tool, if the detector system and the fraction of non-fluorescent FPs are calibrated23 as described in the protocols.
Although the application presented here is focused on protein-protein interactions in adherent cells, the assay can be applied to a wider range of samples, e.g., suspension cells. For such systems, correction for lateral cell movement may be particularly crucial. In addition, it can be easily applied to model membrane systems such as GUVs or GPMVs, allowing the quantification of molecular interactions under well controlled conditions, e.g., different lipid compositions or membranes lacking an organized cytoskeleton. When performing sFCCS on such vesicles, vertical motion may cause additional signal fluctuations, but can be minimized when focusing on large vesicles. In this context, multiple combinations are promising, e.g., GUV-/ GPMV-cell mixing12. As shown in a recent publication, the formation of immune synapses can be successfully modeled by such systems44. Thus, the presented assay will certainly be useful for the investigation of a large variety of cell-cell interactions.
The authors have nothing to disclose.
This work was partially supported by the Deutsche Forschungsgemeinschaft (DFG) grant 254850309. The authors thank Madlen Luckner for critical reading of the manuscript.
DMEM growth medium | PAN-Biotech | P04-01548 | |
DPBS w/o: Ca2+ and Mg2+ | PAN-Biotech | P04-36500 | |
DPBS w: Ca2+ and Mg2+ | PAN-Biotech | P04-35500 | |
Trypsin EDTA | PAN-Biotech | P10-023100 | |
TurboFect Transfection Reagent | Thermo Fisher Scientific | R0531 | |
HEK 293T cells | DSMZ | ACC 635 | |
Alexa Fluor 488 NHS Ester | Thermo Fisher Scientific | A20000 | |
Rhodamine B | Sigma-Alderich | 83689-1G | |
Plasmid DNA | Addgene | NA | See reference 12 (Dunsing et. al., MBoC 2017),for a detailed description of all plasmids |
6-well plate | Starlab | CC7672-7506 | |
35-mm glass bottom dishes | CellVis | D35-14-1.5-N | |
Zeiss LSM780 confocal | Carl Zeiss | NA | |
MATLAB software package | MathWorks | 2015b | |
Neubauer cell counting chamber | Marienfeld | 640110 |