The Epidermal Growth Factor Receptor Forms Location-Dependent Complexes in Resting Cells

Abstract

The epidermal growth factor receptor (EGFR) is a prototypical receptor tyrosine kinase involved in cell growth and proliferation and associated with various cancers. It is commonly assumed that after activation by binding of epidermal growth factor to the extracellular domain it dimerizes, followed by autophosphorylation of tyrosine residues at the intracellular domain. However, its oligomerization state before activation is controversial. In the absence of ligands, EGFR has been found in various, inconsistent amounts of monomeric, inactive dimeric, and oligomeric forms. In addition, evidence suggests that the active conformation is not a simple dimer but contains higher oligomers. As experiments in the past have been conducted at different conditions, we investigate here the influence of cell lines (HEK293, COS-7, and CHO-K1), temperature (room temperature and 37°C), and membrane localization on the quantitation of preformed dimers using SW-FCCS, DC-FCCS, quasi PIE-FCCS, and imaging FCCS. While measurement modality, temperature, and localization on upper or lower membranes have only a limited influence on the dimerization amount observed, the cell line and location to periphery versus center of the cell can change dimerization results significantly. The observed dimerization amount is strongly dependent on the expression level of endogenous EGFR in a cell line and shows a strong cell-to-cell variability even within the same cell line. In addition, using imaging FCCS, we find that dimers have a tendency to be found at the periphery of cells compared to central positions.

Introduction

The intensively studied epidermal growth factor receptor (EGFR, also known as ErbB1 or HER1) is a transmembrane glycoprotein and belongs to the ErbB family of receptor tyrosine kinases together with three other homologous receptors (ErbB2-ErbB4) (1). The structure of all receptor tyrosine kinases consists of three parts: an extracellular ligand-binding domain, a transmembrane segment, and an intracellular tyrosine kinase domain with a carboxyl terminal segment. Ligand-driven signaling of EGFR plays an important role in cell growth as well as proliferation and cell division, and changes in EGFR properties such as overexpression and mutation have been implicated in different cancer types (2, 3). Therefore, understanding EGFR oligomerization and dynamics in the plasma membrane and its role in regulation of EGFR signaling is important to elucidate its role in cancer progression and for the development of new anti-cancer therapies.

According to the traditional model of EGFR activation, the receptor exists in a monomeric form in the membranes of resting cells. Upon binding of its ligand (epidermal growth factor, EGF) the receptor undergoes a conformational change into an open conformation allowing association of two EGFR molecules into a dimer. Dimerization then leads to autophosphorylation of tyrosine residues in the intracellular domain (4, 5, 6). This model has been supported by crystal structures of the extracellular domain of EGFR (7, 8) as well as by reports of EGFR dimerization after stimulation with EGF. However, numerous studies have also reported the presence of EGFR dimers or even larger oligomers in the membranes of resting cells (9, 10, 11, 12, 13). The presence of preformed EGFR dimers and oligomers indicates that regulation of EGFR signaling is more complex than implied by the model outlined above (Fig. S1 in the Supporting Material). The existence of preformed dimers, their relative amount in resting cells, and their role in EGFR signaling continue to present open questions and the findings of individual studies differ considerably (9, 14). While some authors report negligible amounts of preformed dimers (15), others have found most EGFR molecules in dimeric form (6, 9, 16, 17). Nagy et al. (18) observed preformed dimers only at high expression levels of EGFR of ∼600,000 receptors per cell. In contrast to that, in our previous study, we found dimerization to be independent of receptor expression levels in the range from 20,000 to 260,000 copies per cell (19). According to single particle tracking studies of EGFR, the receptor forms transient dimers in the absence of EGF (20, 21, 22); on the other hand, results of Tao and Maruyama (23) indicate that EGFR dimers are already formed in the endoplasmic reticulum, which would suggest a much more stable association. Among possible explanations of the contradictory findings are the intrinsic differences between cell lines as well as differences in the state of the cells used in the individual studies. It has been suggested, for example, that fixed cells frequently used in superresolution microscopy studies can display a higher abundance of receptor oligomers due to aggregation artificially promoted by fixation (24). Insufficient starvation of cells and the use of phosphatase inhibitors have been also proposed as potential sources of elevated levels of receptor dimers (18). EGFR dimerization has been found to depend strongly on temperature (11). Other factors that should be considered are the specific artifacts and limitations of the experimental techniques used. Techniques differ in the efficiency with which they can detect short-lived transient dimers as well as in the ability to distinguish true functional receptor dimers from groups of receptors in close proximity due to colocalization in membrane domains (25, 26, 27, 28). Partitioning of EGFR in cholesterol-rich microdomains has been reported in several studies (29, 30, 31, 32); we have shown recently that EGFR partitions into cholesterol-dependent as well as cholesterol-independent domains and interacts with the cytoskeleton (33).

In this work we probe EGFR dimerization by single wavelength fluorescence cross-correlation spectroscopy (SW-FCCS) (19, 34) with the aim to understand the potential role of selected experimental factors in the discrepancies in reported levels of EGFR dimerization in resting cells. In particular, we performed all measurements at room as well as at physiological temperature and on the apical as well as on the basal plasma membrane. To investigate the influence of endogenous receptors, we used three different cell lines. Experiments were performed in CHO-K1 cells, a cell line selected for its lack of endogenous EGFR, and on two lines of fibroblasts that express endogenously intermediate and low levels of EGFR (COS-7 ∼ 100,000 EGFR/cell (35) and HEK-293 ∼ 20,000 EGFR/cell (36)). To modulate EGFR dimerization, we performed experiments with high- and low-level stimulation by EGF, with the dimer-deficient EGFR mutant (37), and with transiently coexpressed unlabeled receptors.

Selected experiments were repeated with other fluorescence cross-correlation spectroscopy (FCCS) modalities, including dual-color FCCS (DC-FCCS), quasi pulsed interleaved excitation FCCS (PIE-FCCS) (38) and, for the first time to our knowledge, dual-color imaging total internal reflection FCCS (DC-ITIR-FCCS), or imaging FCCS in short, demonstrating that the measurement modalities have little influence on the results. Imaging fluorescence correlation spectroscopy (FCS) allowed us to image the dimer population at the basal cell membrane providing information on the distribution of dimers in live cells.

Materials and Methods

Cell culture conditions and EGFR plasmid construction are described in the Supporting Material.

SW-FCCS

The SW-FCCS instrumentation was described in Pan et al. (39). Briefly, the setup consists of an FV300 confocal microscope (Olympus, Melville, NY) equipped with two avalanche photodiodes (SPCM-AQR-14; PerkinElmer, Waltham, MA) as light detectors for FCS measurements. The Argon ion 514 nm laser line (Melles Griot, Albuquerque, NM) was focused by a 60×, NA 1.2 water immersion objective (UplanApo; Olympus) into a diffraction limited spot. Laser power (measured at the back-aperture of the objective) was 20 μW in all experiments. The emitted fluorescence passed a 488/514 major dichroic mirror (Omega Optical, Brattleboro, VT) through a 150 μm pinhole, was split by a 560 DCLP emission dichroic mirror (Omega Optical), passed through emission band-pass filters (545AF35 or 615DF45, respectively; Omega Optical) and reached the avalanche photodiodes. The output of the avalanche photodiodes was processed by a hardware correlator (Flex02-01D; Correlator.com, Bridgewater, NJ) to obtain the auto- and cross-correlation functions. The inbuilt photomultiplier tubes of the FV300 were used for confocal imaging of the cells. The effective detection volume was calibrated by FCS measurements in 20 nM aqueous solution of Rhodamine 6G (Sigma-Aldrich, St. Louis, MO), which has a known diffusion coefficient of 382 μm²/s at room temperature and 555 μm²/s at 37°C (as calculated from values given in Kapusta (40)).

The 37°C temperature during the measurements was maintained by an on-stage incubator (TempContro 37-2; Pecon, Erbach, Germany) and the corresponding temperature of the objective by an objective heating ring (TC-124A; Warner Instruments, Hamden, CT). The room temperature was 22°C, as maintained by the air-conditioning system in the laboratory. Three 30 s data acquisitions were run consecutively at each selected point.

SW-FCCS data analysis: curve fitting and dimer fraction calculation

The auto-and cross-correlation functions (ACFs and CCFs) were fitted by a model for two-dimensional (2D) diffusion with reversible switching of the fluorophores to a dark (triplet) state. The model (Eq. 1) contains five unknown parameters, including the apparent particle number N_app, the diffusion coefficient D, the fraction of fluorophores in the dark state F_trip, and the characteristic time of switching to the dark state τ_trip. The switching to the dark state is in this case most likely a combination of photophysical processes (triplet transitions, isomerization) and a protonation-deprotonation equilibrium (41, 42). While a single exponential may not be the correct description of the correlation function of such processes, it provides a reasonably good parameterization of the correlation curve and, thanks to a sufficient difference between τ_trip and the characteristic diffusion time, the actual choice of the parameterization of the initial decay of the correlation function has only a limited effect on the diffusional part. The asymptotic value for long correlation times G_inf is in the ideal case of a perfectly stationary process equal to 0; however, in practice small nonzero values are commonly encountered due to the finite length of the measurement or due to nonstationarity of the time trace caused by photobleaching. Therefore, we leave G_inf as a free fitting parameter. The radius of effective detection area ω₀ was determined by calibration with a dye with known value of D as mentioned above. The fitting was performed by a self-written module in Igor Pro (WaveMetrics, Portland, OR):

$$G\left(\tau \right)=\frac{1}{N_{\text{app}}}\frac{{\omega }_0^2}{{\omega }_0^2+4D\tau }\left[1+\frac{F_{\text{trip}}}{1-F_{\text{trip}}}\text{exp}\left(\frac{-\tau }{{\tau }_{\text{trip}}}\right)\right]+G_{\text{inf}}.$$ (1)

Background-corrected particle numbers N were obtained from N_app as described previously in Koppel (43) and Hess and Webb (44). The background was measured in cells not expressing any fluorescent proteins and illuminated by excitation light of the same intensity as used in the actual SW-FCCS recordings (19). The typical background levels were between 500 and 800 cps.

The amount of cross-correlation q (the ratio of the concentration of the double-labeled species to that of all the particles carrying the less abundant label) was calculated according to Eq. 2 and used as a measure of complex formation (45):

$$q=\frac{\max \left(N_g,N_r\right)}{N_x},$$ (2)

where N_g, N_r, and N_x are the background-corrected particle numbers extracted from the ACFs in the green and the red channels and from the CCF, respectively.

The mean values and standard deviations (SDs) or standard errors (SEs) of the mean are calculated from all measurements performed under the given conditions. We measured in at least four cells, with typically a maximum of nine measurements per cell. The numbers of individual cells and measurements are for each type of experiment provided in the table in Fig. 2 or in the text, together with the respective results.

Figure 2.

(A) ACF curves for EGFR-eGFP (dotted line); mRFP-EGFR (dashed line); and CCF curves (dash-dotted line) and their best fits (solid lines in corresponding colors) recorded by different FCCS modalities: SW-FCCS, quasi PIE-FCCS, DC-FCCS, and DC-ITIR-FCCS. The ACF curves from positive control mRFP-EGFR-eGFP are shown in the upper panels; the curves for mRFP-EGFR/EGFR-eGFP (the actual experiment) in the middle panels; and the negative control PMT-mRFP/EGFR-eGFP in the case of SW-FCCS and PMT-eGFP/PMT-mRFP for other FCCS modalities, in the lower panels. (B) The average cross-correlation amounts q from the different FCCS modalities are summarized in Table S1. To see this figure in color, go online.

Estimate of number of labeled receptors per cell

We used the sum of background-corrected particle numbers from both channels (N_s = N_g + N_r) to estimate the number of EGFR copies per cell. N_s is the number of labeled receptor molecules within the observation area π ω₀² defined by the focus of the confocal microscope. The radius of the observation area ω₀ is determined by the calibration of the confocal volume as described above. To obtain the number N_c of receptors per cell, we need to estimate the area of the whole cell membrane A_c and then N_c = N_s A_c/(π ω₀²). The area of the basal cell membrane A_b can be directly determined from the confocal images of the cells in which we performed the measurements. Approximating the shape of the apical membrane as a spherical surface, we can write A_c = 2A_b + π h², where the height of the cell h can be estimated as the typical difference in axial focus position between the basal and the apical membrane. The average calculated cell surfaces of CHO-K1, HEK293, and COS-7 cells were 691 ± 163 μm², 602 ± 153 μm², and 1501 ± 701 μm², respectively (the errors are based on the SD of A_b). The above surface areas can be overestimated for cells that are very flat in most of their area. A lower estimate of the cell surface can be obtained A_c = 2A_b. In this way we obtain 537 ± 163 μm², 448 ± 153 μm², and 1347 ± 701 μm² for CHO-K1, HEK293 and COS-7, respectively. Because the two estimates agree within the range of the error, we used the former estimate in calculating N_c.

DC-ITIR-FCCS

Measurements in CHO-K1 cells at 37°C were also performed by DC-ITIR-FCCS. The setup consisted of an inverted epi-fluorescence microscope (IX83; Olympus) equipped with a motorized TIRF illumination combiner (IX3-MITICO; Olympus), an image splitter (OptoSplit II; Cairn Research, Faversham, UK), and an electron multiplying charge-coupled device camera (Evolve 512; PhotoMetrics, Tucson, AZ). A 491 nm laser and a 561 nm laser (LAS/491/100 and LAS/561/100; Olympus) were connected to the TIRF illumination combiner in which the incidence angles for individual laser lines were adjusted to give 110 nm penetration depth of the evanescent field. The laser power measured at the back aperture of the objective was 0.6 mW for the 491 nm laser and 0.9 mW for the 561 nm laser. A 60×, NA 1.49 oil immersion objective (ApoN; Olympus) was used to illuminate the sample and collect the fluorescence image. The fluorescence light then passed through a major dichroic (Di01-R488/561; Semrock, Rochester, NY) and was split by the image splitter on two halves of the camera chip. The image splitter was fitted with an emission dichroic (FF560-FDi01; Semrock) and band-pass filters (510AF23 and 615DF45, respectively; Omega Optical). A bright-field image of a stage micrometer was used to align the image splitter following a procedure described in Krieger et al. (45). The camera was controlled by μManager 1.4 (46), and in each measurement a stack of 30,000 frames with 3.4 ms per frame was acquired. The captured region of interest consisted of 20 lines (to allow sufficiently fast readout) spanning the whole width of the camera chip (to include corresponding images in both the green and the red spectral regions). The chamber with imaged cells was placed in an on-stage incubator (Chamlide TC; Live Cell Instrument, Seoul, Korea) maintaining the temperature of 37°C and 5% CO₂ atmosphere.

The image stacks were then analyzed by a self-written FIJI (47) plug-in (48), which calculates the ACF from intensity fluctuations in each pixel as well as the CCF for each pair of corresponding pixels in the two halves of the image. To correct for gradual changes in fluorescence intensity caused by bleaching or membrane undulations, correlations were calculated in sliding windows of 2500 frames and then averaged to avoid distortion of correlation functions due to intensity changes caused by photobleaching or whole cell movement. The plug-in is analogous to our previously described program, ImFCS (49). The correlation functions were fitted with a model for 2D diffusion derived previously in Sankaran et al. (50) to extract values of apparent particle numbers N, diffusion coefficient D, and asymptotic correlation G_inf. The size of the microscope point spread function was calibrated by measurement in supported lipid bilayers as described in Bag et al. (51). The background was set at 500 counts based on recordings in samples of cells not expressing any fluorescent protein. The pixels were binned 2×2 for the analysis.

Because the electron multiplying charge-coupled device camera is not a true photon counting detector, recovering absolute particle numbers is not straightforward (52). We, therefore, did not attempt to determine the absolute dimer fractions but instead characterized the level of dimerization by cross-correlation amount q defined by Eq. 2, in line with a previous work on dual-color imaging FCCS (45). By performing the analysis for every pair of corresponding pixels, we obtain a map of distribution of q in the imaged cell.

DC-FCCS and quasi PIE-FCCS

Furthermore, DC-FCCS and quasi PIE-FCCS experiments were carried out in CHO-K1 cells at room temperature on a confocal microscope (FV1200; Olympus) equipped with a time-resolved FCS upgrade kit (PicoQuant, Berlin, Germany). For DC-FCCS, the cells were illuminated by two continuous wave laser lines of 488 nm and 543 nm (GLG 3135 and GLG 7000, respectively; Showa Optronics, Tokyo, Japan) through a 60×, NA 1.2 water immersion objective (UplanSApo; Olympus) to excite eGFP and mRFP, respectively. The same 543 nm laser was used for the quasi PIE-FCCS together with a pulsed 485 nm laser (LDH-D-C-488; PicoQuant) operated at 20 MHz repetition rate. 20 μW of laser power for each individual laser line was used in all measurements. The fluorescence emission passed through a 405/488/543/635 major dichroic mirror (Chroma Technology, Bellows Falls, VT), a 120 μm confocal pinhole and, after being split by a 560DCXR (Chroma Technology) emission dichroic, through a 600/50 (Chroma Technology) or 513/17 (Brightline; Semrock) band-pass emission filter, respectively, to be detected by avalanche photodiodes (SPCM-AQR-14; PerkinElmer). The photon counts from the detectors were registered by a TimeHarp 260 time-correlated single photon counting board (PicoQuant) and processed by the SymPhoTime 64 software (PicoQuant); the same software was also used to calculate the correlation functions. Those were then analyzed in the same manner as correlation function obtained by SW-FCCS. As in the case of SW-FCCS, the dimensions of the effective detection volumes were determined by calibration FCS measurements in solutions of reference dyes. Atto-488 (Atto-Tec, Siegen, Germany) was used for the 488 nm and 485 nm laser lines and Rhodamine 6G for the 543 nm line. The diffusion coefficient of Atto-488 was taken as 369 μm²/s at 22°C, based on values in Kapusta (40). Each data acquisition lasted 30 s in the case of DC-FCCS and 60 s in the case of quasi PIE-FCCS.

When analyzing data acquired with the pulsed 485 nm laser, we used statistical filtering (53) to eliminate detector after-pulses and spectral cross-talk between the two detection channels as described previously in Padilla-Parra et al. (38). This method of cross-talk elimination in FCCS is based on similar principles as PIE-FCCS (54); however, it requires only a single pulsed laser, the second laser working in continuous wave mode. We call it, therefore, “quasi PIE-FCCS”.

In the case of quasi PIE-FCCS, the cross-correlation amount q was calculated as q = N_g/N_x; the value N_g was preferred because the ACF in the green spectral channel was corrected for background and detector after-pulsing, unlike the ACF in the red channel (38). Based on the definition of q, using N_g in the calculation is correct if the concentration of red-labeled receptors is lower or comparable to that of the green-labeled ones. This assumption was satisfied in the majority of our measurements performed by the other FCCS modalities in cells prepared by the same protocols. In the case when the red-labeled receptors are predominant, the calculation using N_g would result in the underestimation of q. We can, therefore, regard the values obtained by quasi-PIE-FCCS as a lower estimate, if not the correct determination, of the actual q.

Results and Discussion

Diffusion coefficient D of EGFR in different cell lines

In the first part, we investigated the diffusion coefficient D of EGFR measured by SW-FCCS in cells, which coexpress mRFP-EGFR and EGFR-eGFP (Fig. S2). The average D ± SD of the transmembrane EGFR at the apical and the basal membrane was 0.3 ± 0.2 μm²/s at room temperature and showed no differences between these two membrane locations. At physiological temperature, D reaches a higher value of 0.7 ± 0.3 μm²/s. All correlation curves were successfully fitted with a model with a single diffusive component; this is in agreement with the Saffman-Delbrück model, which predicts only a negligible difference in D between a monomer and a dimer of a membrane protein (55). In general, the D of mRFP-EGFR and EGFR-eGFP was in the same range in the apical and the basal membranes of all studied cell lines. In all cases, D increases with temperature, as expected. In addition, we observed an increase of the SD of D with temperature, which is probably due to increased heterogeneity in local membrane organization at physiological temperature; similar effects have been observed by Ries et al. (56).

Consistently, D of EGFR measured by DC-FCCS and quasi PIE-FCCS at room temperature was 0.3 ± 0.2 μm²/s in agreement with SW-FCCS results. The average D at 37°C determined by DC-ITIR-FCCS in both the green and the red detection channels was 0.6 ± 0.3 μm²/s (mean ± SD).

EGFR complex fractions and receptor density determined by SW-FCCS

By using SW-FCCS, we found that EGFR forms complexes under all conditions investigated in the three different cell lines. Only cells with expression levels of the two fluorescent proteins differing by a factor of <2 were selected for data processing and evaluation (57). Positive and negative controls were performed for each cell line and each temperature to determine the dynamic range of our method. Using double-labeled receptor mRFP-EGFR-eGFP as the positive control gave a cross-correlation amount q of 58 ± 1% in CHO-K1 cells (Fig. 1 A); the value is the average from measurements on both membranes and at both temperatures, errors are given as the SE of the mean (mean ± SE) unless stated otherwise. The apparent complex fraction <100% can be explained by incomplete maturation of mRFP and its prolonged residence in dark or dim states, as discussed in detail earlier (57, 58). The apparent complex fraction of mRFP-EGFR-eGFP in COS-7 and HEK293 cells was 68 ± 2% and 58 ± 1%, respectively. In our earlier work, we found that the proteins PMT-eGFP/-mRFP do not interact with each other in the membrane in CHO-K1 cells (19), making them a suitable negative control for FCCS measurements in cell membranes. Surprisingly, different behavior of these proteins was observed in COS-7 and HEK293. In these cell lines, considerably higher amount of cross correlation was found. A possible explanation for the increased cross correlation is partitioning of PMT into membrane microdomains in those cell lines. Hence, we selected monomeric PMT-eGFP and mRFP-EGFR as the negative control for all three cell lines. We obtained the following cross-correlation amounts q: 10 ± 2% in CHO-K1 cells, 15 ± 0.5% in HEK293 cells, and 21 ± 1% in COS-7 cells. The significantly higher apparent complex fraction in COS-7 cells is possibly caused by partitioning of PMT and EGFR into the same domains, and makes detection of low levels of EGFR dimerization impossible in this cell line. The fact that different cell lines can show different organization on membranes is the goal of an ongoing study in our lab, but has also been shown in Kraft and Klitzing (59), Kreder et al. (60), and Bag et al. (61).

Figure 1.

SW-FCCS experiments under different conditions. (A) SW-FCCS for resting cells at two different temperatures and basal and apical membranes. The plots (A–C) show the cross-correlation amounts q in CHO-K1, COS-7, and HEK293 cells, respectively, for different expression levels of mRFP-EGFR/EGFR-eGFP. Measurements with <200 or >200 receptors per μm² are indicated by solid or open circles, respectively. The upper and lower gray zones represent the range of the positive and negative controls with their width given by the mean ± SD. (D) SW-FCCS measurements of CHO-K1 cells expressing mRFP-EGFR/EGFR-eGFP. The bars represent q (mean ± SE) of control (solid bars), LAT-A (shaded bars), and mβCD-treated cells (open bars). LAT-A experiments and mβCD experiments conducted at room temperature. (E) Time series of q (mean ± SE) values for high-dose (100 ng/mL) and low-dose (10 ng/mL) EGF stimulation on mRFP-EGFR/EGFR-eGFP on CHO-K1 cells. (F) Value of q (mean ± SE) of mRFP-EGFR/EGFR-eGFP (apical membrane, RT) compared with mRFP-EGFR/EGFR (1706Q, V948R)-eGFP (apical membrane, RT), EGFR interaction in COS-7 cells at receptor densities <200/μm² and >200/μm² (basal membrane, RT) in CHO-K1 cells at <200/μm² and >200/μm² (basal membrane, RT), mRFP-EGFR/EGFR-eGFP/wt-EGFR (1.5:1:2) (basal and apical membrane, RT), and mRFP-EGFR/EGFR/wt-EGFR (1.5:1:1.3) basal and apical membrane, RT).

Next, we investigated mRFP-EGFR/EGFR-eGFP interaction in resting CHO-K1 cells that were serum- starved for a minimum of 4 h (Fig. 1 A). Cells ranging from low to very high EGFR expression (receptor numbers in the observation area ranging from 3 to 462, which corresponds approximately to cell surface receptor density of 17–2222 per μm²) were examined to probe whether the complex fraction is concentration-dependent. Our results yielded an average apparent complex fraction of 32 ± 3% (n = 47, 20 cells) on the basal and 33 ± 5% (n = 46, 19 cells) on the apical membrane at room temperature. At physiological temperature, the apparent complex fraction in the basal membrane is 28 ± 2% (n = 43, 15 cells), which does not differ significantly from the value at room temperature. A slight increase to 38 ± 4% (n = 30, 17 cells) occurs on the apical membrane. The subpopulations of cells with receptor densities below and above 200 receptors per μm², respectively, do not differ in the apparent complex fractions (Fig. 1 A). The independence of the cross-correlation fraction on the expression level is even more evident from the plot of q versus the number of EGFR copies per detection area (Fig. S4, E and F).

The situation is different in HEK293 and COS-7 cells, which express endogenously ∼20,000 and 100,000 EGFR copies per cell, respectively. Formation of complexes between endogenous and fluorescently labeled EGFR molecules results in a decrease in cross-correlation amount q and, thus, in an underestimation of the complex fraction, unless the number of endogenous receptors is made negligible by overexpressing of the labeled ones. We examined the apparent EGFR complex fraction in HEK293 and in COS-7 cells for a range of labeled receptor densities of 34–6212 per μm² (N = 7–1242) and 62–1543 per μm² (N = 13–321), respectively (Fig. 1, B and C). In these cases, differences in q are observed between cells with labeled receptor densities below and above 200 per μm². The dependence (especially for COS-7 cells) of the apparent q factor on the labeled EGFR expression level is evident from Fig. S4, A–D. These results show that in the presence of endogenous receptor, there is an apparent dependence of dimerization on expression level caused by the competitive interaction of endogenous receptor with the labeled EGFR.

Effect of Latrunculin A treatment on complex fraction in CHO-K1 cells

To study the role of the actin cytoskeleton on EGFR dimerization in CHO-K1 cells, we disrupted the actin cytoskeleton by Latrunculin A (LAT-A) at a final concentration of 3 μM. The data were acquired in cells cotransfected with mRFP-EGFR/EGFR-eGFP 15 min after addition of LAT-A at room temperature. LAT-A-treated cells exhibit apparent complex fractions of 34 ± 4% (n = 34, 5 cells) and 32 ± 5% (n = 19, 4 cells) in the basal and the apical membrane, respectively. This agrees well with control cells having 32 ± 5% (n = 30, five cells) and 33 ± 5% (n = 25, 4 cells) complexes in the basal and the apical membrane, respectively (Fig. 1 D). These results indicate that the EGFR complex formation in the plasma membrane does not depend on the actin cytoskeleton, which is consistent with the study of Ariotti et al. (62).

Influence of cholesterol depletion on complex fraction in CHO-K1 cells

Next, we evaluated the influence of cholesterol-dependent domains in the plasma membrane on the observed EGFR complex formation in CHO-K1 cells. Cholesterol was depleted from the plasma membranes by treatment with mβCD (at 3 mM final concentration). EGFR-eGFP/mRFP-EGFR dynamics were monitored after 25 min mβCD incubation to maximize cholesterol depletion effect at room temperature (33) (Fig. 1 D). In untreated cells, the average apparent EGFR dimer fraction was 26 ± 3% (n = 25, 5 cells) in the basal and 33 ± 3% (n = 36, 5 cells) in the apical membrane. In mβCD-treated cells, the effect of cholesterol removal varies dependent on the membrane location. On average, cholesterol depletion showed a trend to increase the complex formation up to 40 ± 4% (n = 29, 5 cells) and 58 ± 5% (n = 40, 8 cells) on the basal and apical membrane, respectively (Fig. S3). These values are higher than the cross-correlation amounts measured in untreated cells, demonstrating that EGFR complex formation is to some extent inhibited by partitioning of the receptor into cholesterol-dependent domains. These results are in agreement with a previously published study (63) and are consistent with the findings that EGFR resides in cholesterol-dependent domains (33, 64).

EGFR complex fractions in CHO-K1 determined by DC-FCCS and quasi PIE-FCCS

The same sets of EGFR experiments in CHO-K1 cells at room temperature were also performed by DC-FCCS and quasi PIE-FCCS to test whether results obtained by these FCCS modalities are consistent with our SW-FCCS findings. The data in Fig. 2 illustrate the correlation functions from distinct experimental sets by different FCCS modalities and cross-correlation values are summarized in Table S1.

DC-FCCS measurements gave EGFR complex fractions of 44 ± 4% (n = 18, 9 cells) at the basal and 37 ± 4% (n = 19, 7 cells) at the apical membrane. The positive control mRFP-EGFR-eGFP yielded a cross-correlation amount of q = 60 ± 8% and 71 ± 6% at the apical and the basal membrane, respectively. The lower detection limit was determined by the negative control (PMT-eGFP/PMT-mRFP) to be 13 ± 2% and 9 ± 1% in the basal and the apical membrane, respectively (Fig. 2).

To obtain cross-talk free values of q, we repeated the measurements using quasi PIE-FCCS. Negative control measurements in cells expressing PMT-eGFR/PMT-mRFP give q = 0 ± 0%, thus, demonstrating the efficiency of spectral cross-talk elimination. For the positive control mRFP-EGFR-eGFP, we found q = 51 ± 11%. The EGFR apparent dimer fraction in the apical membrane is 33 ± 9% (n = 9, 7 cells). Collectively, all the FCCS modalities applied provided mutually consistent results, lending further support to the conclusions based on our SW-FCCS data.

EGFR dimer fractions in CHO-K1 determined by DC-ITIR-FCCS

Measurements in the basal membranes of CHO-K1 cells at 37°C were performed by DC-ITIR-FCCS. Cells expressing PMT-eGFP and PMT-mRFP and cells expressing mRFP-EGFR-eGFP were used as the negative and the positive controls, respectively. An analyzed region of interest consisted typically of >1000 pixels, not all of which contained useful information; there were pixels corresponding to areas outside of cells or to regions within the cells where the membrane is too far from the glass surface to be efficiently excited by the evanescent field or, possibly, regions of cell membrane inaccessible to the fluorescent tracer. To exclude pixels outside of the cell from the calculation and fitting of correlation functions, an intensity threshold was set for each stack and only pixels having an intensity larger than the threshold in the first frame were processed. Most stacks contained some pixels that, although having high intensity, gave very noisy ACFs, fitting of which was unreliable and likely to produce unrealistic parameter values. To exclude such pixels from further analysis, upper and lower limits were set on D and G_inf obtained from ACFs fits. Only the values from pixels for which the fitted D and G_inf in both autocorrelation channels lay between the respective lower and upper limits were included in further evaluation. The limits for G_inf were the ideal asymptotic value ± 50% of the correlation function amplitude G(0). The limits for D were set separately for EGFR and for PMT according to the actual distribution of the measured D values. Histograms of natural logarithms of measured D values were fitted with a normal distribution and the limits were set as the mean ± 3 times the SD. For PMT they were D_min = 0.2 and D_max = 2.7 μm²/s and for EGFR D_min = 0.1 and D_max = 1.4 μm²/s.

There were some pixels in which both ACFs were fitted successfully and the fit parameters satisfied the above described criteria, yet the fit of the CCF did not give realistic parameter values. This was especially common in the case of the negative control, where the CCFs consist mainly of noise correlation. The fits of such curves give sometimes extremely high value of D (e.g., on the order of 10¹⁶ μm²/s) or a very low value of D (e.g., on the order of 10⁻⁶ μm²/s) together with a very low G_inf (e.g. < 0). Both situations result in large overestimation of the amplitude (1/N; Fig. S5). To avoid such unrealistic values, the amplitudes of CCFs were set to 0 if the fitted D and G_inf were not within certain limits. The limits for G_inf were the same as in the case of ACFs (±50% of the amplitude from the ideal asymptotic value) and the limits for D were broadened to D_min/2 and 2 D_max, where D_min and D_max are the respective limits for ACFs. In other words, q is set to 0 in such pixels. The CCF amplitude was set to 0 also in pixels where the CCF fit gave a negative value of N. The value of q was, thus, set to 0 in almost 60% of pixels of the negative control, in 27% of pixels of the actual experiment, and only in ∼7% of pixels of the positive control (Fig. 3). It should be noted that the points in which CCF fitting fails, and which are clearly shown in DC-ITIR-FCCS, might be discarded in confocal FCCS analysis as failed measurements. Besides the additional information on frequency and location of those points, DC-ITIR-FCCS also prevents any artificial bias that could stem from rejecting them from evaluation.

Figure 3.

Histogram of pooled q values and examples of q maps obtained by DC-ITIR-FCCS. The q values obtained in all pixels of all investigated cells were pooled together for the three series of measurements: the negative control (10,053 pixels from 10 cells), the positive control (6696 pixels from eight cells), and the actual experiment with EGFR-eGFP and mRFP-EGFR (16,779 pixels from 16 cells). The frequencies plotted in the histogram (A) are numbers of pixels in each bin divided by the total number of pixels in the respective series. Examples of q maps are shown for a cell from the negative control (B), positive control (C), and the actual experiment (D). The color-scale of q and a 5 μm scale bar are shown next to the maps. Pixels shown in white were excluded from fitting (either because of having lower than threshold intensity or because the parameters obtained by fitting of the ACFs were not within the set limits). To see this figure in color, go online.

The final values of q, thus obtained, were 10 ± 14% (10 cells) for the negative control, 53 ± 23% (8 cells) for the positive control, and 34 ± 28% (16 cells) for the actual experiment (Fig. 2). The values are given as mean ± SD; the SEs of the mean are negligible because of the large number of pixels evaluated. It can be seen in Fig. 3 that the distribution of q measured in the positive control has a shoulder at ∼80%. Out of the eight cells investigated, the shoulder at ∼80% was dominant in three cells, while in the remaining cells the q distribution was monomodal with a maximum at ∼45% (Fig. S6). The very high amount of cross correlation observed in some pixels may be caused by elevated amounts of dimers of the EGFR tandem constructs or possibly even higher-order EGFR oligomers. The q values obtained in the actual experiment show an even broader distribution. Out of 16 cells investigated, five showed very low q (as low as 11%), indicating negligible complex fractions. Another five cells showed high q values similar to the positive control, indicating most EGFR molecules were in the form of complexes (Fig. S6). The remaining cells, which exhibited on average intermediate q values, contained regions of very high as well as negligible complex fractions. (Figs. 3 and S6). The regions of high q values are typically located closer to the cell periphery.

The effect of cholesterol-dependent domains on EGFR dimerization in the basal membrane of CHO-K1 cells at 37°C was investigated by performing DC-ITIR-FCCS measurements before and after cholesterol depletion by mβCD (at 3 mM final concentration). The measurements after cholesterol depletion were performed within the time interval of 25–35 min after mβCD addition. The average q values for each individual cell before and after cholesterol depletion are shown in Fig. S3 C. The average q from all the 20 cells increased from 20 ± 9% to 28 ± 11% after cholesterol depletion, which is consistent with the SW-FCCS results.

To conclude, the DC-ITIR-FCCS data are in agreement with the results of the confocal FCCS modalities; moreover, they provide further insight into the distribution of dimers and provide a comparison of the cell-to-cell variations with the variations in different regions of the membrane of each individual cell.

Effect of unlabeled receptor molecules on apparent dimer fraction

When unlabeled EGFR molecules (such as the endogenous EGFR expressed by COS-7 and HEK293 cells) are present in the cell membrane, the dimer fraction derived from our FCCS data is underestimated because some of the labeled EGFR molecules dimerize with unlabeled ones and appear as apparent monomers in FCCS. To test the sensitivity of our method in the presence of unlabeled EGFR, we cotransfected CHO-K1 cells with labeled mRFP-EGFR, EGFR-eGFP, and wild-type EGFR and performed SW-FCCS measurements at room temperature (Fig. 1 F). When the mRFP-EGFR, EGFR-eGFP, and wild-type EGFR plasmids were transfected in the ratio 1:1.5:2 (EGFR-eGFP/mRFP-EGFR/wt-EGFR), we obtained apparent complex amounts of 19 ± 2% (n = 25, 5 cells) in the basal and 22 ± 2% (n = 34, 5 cells) in the apical membrane. When the amount of transfected wild-type EGFR plasmid was reduced (1:1.5:1.3), the apparent complex fractions increased to 23 ± 2% (n = 16, 5 cells) and 34 ± 2% (n = 20, 5 cells) in the basal and apical membrane, respectively. The measurements with unlabeled EGFR demonstrate the sensitivity of our technique to changing complex fractions of the receptor; this is in line with our results in COS-7 and HEK293 cell lines where the competition between labeled and endogenous EGFR molecules causes an apparent dependence of the complex amount on the expression level.

Complex fraction of EGFR (1706Q, V948R) is reduced

EGFR (1706Q, V948R) is a dimer-deficient mutant (65). The mutations are located in the N-terminal lobe (1706Q) and in the C-terminal lobe (V948R). Their combination impairs activator and receiver function and demonstrates that the N- and C-lobes have an impact in the formation of EGFR dimers. In this work, we have tested the influence of double mutated EGFR (1706Q, V948R) on the presence of preformed dimers by using SW-FCCS. This construct was transfected together with mRFP-EGFR into CHO-K1 cells and measured in the apical membrane at room temperature (Fig. 1 F). Fitting of the correlation curves and data analysis showed in the presence of EGFR (1706Q, V948R) a cross-correlation amount of 16 ± 2% (n = 26, 5 cells). Compared to the control measurement of mRFP-EGFR/EGFR-eGFP, which exhibited ∼33% cross-correlation amount, the dimerization of the mutated EGFR (I706Q, V948R) is significantly reduced.

Low- and high-dose EGF stimulation

Cells expressing labeled EGFR were stimulated with EGF at low and high doses of 10 and 100 ng/mL, respectively. FCCS measurements were performed on serum-starved cells (4 h) before stimulation. The same cells were monitored at 3, 10, 15, and 20 min after EGF stimulation (Fig. 1 E; Table S2) (final concentration 100 ng/mL). Large receptor clusters could be seen in some measurements, which were manifested by high spikes in the intensity trace and by extremely broad ACF and CCF curves. The data sets affected by large oligomers were excluded from the analysis.

In resting cells, the average q value was 39 ± 3% (n = 22, 4 cells), and increased slightly to 44 ± 5% (n = 12, 4 cells) at 3 min and remained at 42 ± 6% (n = 10, 4 cells) at 5 min and 44 ± 6% (n = 8, 4 cells) at 10 min. The cross correlation then rose slightly to 52 ± 6% (n = 9, 4 cells) at 15 min and remained at 53 ± 6% (n = 8, 4 cells) at 20 min. Broadening of the correlation functions observed in some cells after 10 min of incubation with EGF is a sign of formation of large EGFR oligomers or clusters. At longer incubation times, the correlation function width decreases again without a decrease in the cross-correlation amount. This indicates disappearance of the large oligomers accompanied by an increase in the fraction of dimers and smaller oligomers (such as tetramers), which do not affect the width of the correlation functions. The loss of the large oligomers or clusters may be a result of their depletion from the plasma membrane by endocytosis.

The same set of experiments was carried out at a lower EGF concentration of 10 ng/mL. In this case, the cells were monitored over a longer time period at 10, 20, and 25 min after incubation with EGF as no strong internalization was expected at this concentration. As seen in Fig. 1 E, a stepwise increase in q factor was observed in the presence of EGF ligand. Unstimulated cells showed an average cross correlation of 41 ± 3% (n = 13, 3 cells), which increased to 58 ± 7% (n = 7, 3 cells) 10 min after stimulation. After 20 min of incubation with EGF, the cross-correlation reached its highest value of 77 ± 5% (n = 4, 3 cells) and then decreased to 59 ± 8% at 25 min (n = 7, 3 cells). The high cross correlation of 77 ± 5% observed after 20 min of incubation with EGF is higher than the cross-correlation for the positive control (positive control average ∼58%). This indicates that higher EGFR oligomers (such as tetramers or hexamers) are involved which are, however, not large enough to be manifested by broadening of the correlation functions as observed after high-dose EGF stimulation (Fig. S8 C).

Discussion

In this study, we have tested the influence of several experimental factors on the amount of EGFR complexes in resting cells as measured by SW-FCCS and other FCCS modalities. For easier comparison between the different FCCS measurements, in the following we use the normalized dimer fraction, which is the q value of an experiment divided by the q value of the positive control (Table S1). The normalized dimer fraction of EGFR was 57% determined by SW-FCCS and DC-FCCS (both at the apical surfaces), and 65% by quasi PIE-FCCS (apical) and imaging FCCS (basal). The normalized values are an estimate of the actual dimer fraction obtained and are comparable with our previous study on EGFR (19). On the other hand, we have observed significant cell-to-cell variability, which was especially obvious from the imaging FCCS experiments. Approximately one-third of all cells showed very low cross-correlation amounts indicating negligible complex fractions, while in approximately another one-third of cells, most EGFR molecules were in the form of complexes. The remaining cells, exhibiting on average intermediate cross-correlation, contained regions of very high as well as negligible complex fractions. This demonstrates the utility of DC-ITIR-FCCS (and imaging FCS modalities in general) for linking the cell-to-cell variability with the variability between regions within individual cells. The regions of high complex fractions were located predominantly at the cell periphery in agreement with previous reports (20, 33). It remains to be investigated further what is the nature of those regions and whether they contain only elevated amounts of dimers or whether they contain higher EGFR oligomers. Interestingly, we have found recently that EGFR forms microscopic clusters upon activation in cholesterol-depleted cells, those clusters being, in some cases, more frequent at the cell periphery (20).

Besides the brightness, the lifetime of the complexes also determines their contribution to the CCF. The contribution increases with increasing lifetime of the complexes and saturates when the lifetime is longer than the characteristic residence time of the molecules in the effective observation area. The characteristic residence times are ∼50 ms in our confocal measurements and close to 1 s in the case of DC-ITIR-FCCS. The good agreement of the cross-correlation amount between confocal FCCS and DC-ITIR-FCCS indicates that the EGFR complexes are stable on timescales of at least the order of seconds. This is consistent, for example, with the single-molecule study of Chung et al. (20) reporting transient EGFR dimers in resting cells with lifetimes ∼10 s.

We have observed no dependence of the cross-correlation amount on the EGFR expression level. The measurements were performed in in CHO-K1 cells expressing from ∼10,000 to 1,600,000 EGFR copies, values ranging from very low expression to overexpression (66). We can, therefore, conclude that the complex formation is not an artifact induced by high receptor densities. The CHO-K1 cells were selected because of their negligible endogenous EGFR expression. The presence of endogenous EGFR interferes with the FCCS detection of complexes because dimers between an endogenous and a labeled receptor carry only a single fluorescent label and contribute, therefore, to the monomer fraction. We used this effect in control experiments designed to show the sensitivity of our FCCS measurements to changes in EGFR complex amounts. Firstly, we reduced the cross-correlation amount by coexpressing wild-type EGFR alongside the labeled receptor. Secondly, we performed measurements in HEK293 and COS-7 cell lines, which express low to intermediate levels of endogenous EGFR (HEK293 ∼20,000 EGFR/cell (36) and COS-7 ∼100,000 EGFR/cell (35)). The cross-correlation amounts were low to negligible in cells expressing <200 labeled receptors per μm². Only at higher labeled receptor densities, which means at receptor densities much higher than those of the endogenous receptors (∼30/μm² in HEK-293 and (∼70/μm² in COS-7), the cross-correlation amounts are comparable to those obtained in CHO-K1 cells. Together, these results demonstrate the sensitivity of our FCCS approach to the changes in the EGFR complex fraction. Even more importantly, they prove that the observed complex formation is not an artifact of the artificially introduced labeled receptor, but that the same phenomenon involves also the EGFR in cells where it is endogenously expressed. To avoid interference of endogenous EGFR in specific cell lines one could use gene editing tools to create background free stable cell lines. This would be of particular interest to investigate EGFR dimerization within the same protein environment as different cell lines will have different protein expression profiles.

A third control experiment showing the sensitivity of our technique to the changes in EGFR complex fraction was based on measurements in CHO-K1 cells expressing an EGFR mutant (1706Q, V948R) with reduced propensity for dimerization (65). Unlike the previous two control experiments, which involved changes only in the apparent complex amounts (as observed by FCCS), this experiment involves an actual reduction in the receptor complex fraction. The concept of EGFR dimerization is based on the stabilization of asymmetric dimers between the kinases. The C-terminus from one kinase and the N-terminus from the other kinase are responsible for the formation of an asymmetric dimer (67, 68, 69). Using this EGFR mutant (1706Q, V948R) with labeled EGFR revealed a decrease in normalized dimer amount to 28%, which is significantly lower than the 57% dimer fraction in control experiments.

In contrast to our observations, Nagy at al. (18) observed pronounced EGFR complex formation only in cells with very high expression levels. This discrepancy is possibly explained by a dependence of the complex lifetimes on the receptor density. If the complex lifetime at low receptor densities is comparable with the temporal resolution of the number and brightness technique used in the study (on the order of seconds), the measured complex fraction would be likely underestimated and would increase with growing complex lifetime at higher receptor densities.

The involvement of cholesterol-dependent lipid rafts in EGFR activation and signaling has been reported earlier (70, 71). Our recent results show that EGFR partitions into cholesterol-dependent as well as cholesterol-independent domains and that its diffusion is affected by the actin cytoskeleton (33). Our data show that cholesterol depletion by mβCD leads to an increase in the EGFR complex fraction. This increase is more pronounced in the apical membrane. The result suggests that the observed cross-correlation does not stem directly from colocalization of multiple receptors into small plasma membrane domains, but results instead from the formation of EGFR complexes held together by receptor-receptor interactions. At the same time we may speculate that the complexes are of a transient nature; the transient trapping of receptor molecules into small cholesterol-dependent domains prevents them from diffusing freely and, thus, reduces the frequency of their encounters with other receptors, which lead to formation of the transient complexes. At the same time, there is no obvious reason why the fraction of long-lived stable dimers should be affected in any way by their membrane domain partitioning. Consistent with our results, Saffarian et al. (63) reported an increase of oligomeric EGFR fraction after cholesterol depletion. Other studies have reported decreased EGFR clustering upon cholesterol depletion (62, 64); however, the EGFR clusters described in those studies are much larger and most likely of a different nature than the mobile complexes investigated here. In addition, the disruption of the actin cytoskeleton by LAT-A treatment had no effect on the EGFR complex fraction. Similarly, Low-Nam et al. (21) reported that actin disruption did not change the EGFR dimer stability.

Activation of EGFR by its ligand EGF has been usually reported to enhance formation of receptor complexes, either dimers or higher oligomers (62, 72, 73). Our results agree with that. At 10 ng/mL, EGF induced a considerable increase in the EGFR complex fraction. The high values of cross-correlation (comparable to the positive control) suggest that the increase is not only caused by elevated dimerization, but probably also by higher oligomers (such as tetramers or hexamers) which are, however, not large enough to be manifested by broadening of the correlation functions. Different effects were observed after stimulation with a higher dose of EGF (100 ng/mL). The average increase in the cross-correlation amount was smaller in this case; on the other hand, broadening of correlation curves indicated presence of large EGFR complexes or clusters, which diffuse slower than the monomers and small oligomers. While oligomers of increasing size contribute more prominently to the CCF (the contribution scales with the square of the brightness), characterization of very large oligomers and clusters by FCCS is problematic. FCCS requires the complexes to be sufficiently mobile on the timescales of the measurement to ensure statistically significant number of passages of the complexes through the observation area within the measurement time. Besides, our recent findings show that such large complexes and clusters undergo rapid endocytosis, which depletes them from the plasma membrane (33).

Confocal FCCS modalities, due to the very limited number of measurements taken in each cell, are unable to separate the cell-to-cell variability from the variability in the membrane of a single cell. On the other hand, imaging FCCS, besides providing overall better statistics thanks to much higher numbers of measurement points, clearly separates the two sources of variations. That enabled us to classify the cells into subpopulations with high, low, and intermediate complex amounts as discussed above.

We also avoided all cells that showed already clustering before any stimulation. So the data are averages from the cells that do not show any clustering before stimulation. This would include cells that respond and cells that do not respond to activation and thus our dimerization is probably a lower limit. Only after stimulation can one detect which cells respond. We think that this problem is inherent in 2D cell cultures and it will be interesting to measure the same values in tissues or organisms.

Furthermore, we have shown that the fraction of receptors in complexes is insensitive to actin cytoskeleton disruption and increases upon cholesterol depletion. The amount of EGFR complexes also increases upon EGF stimulation at different doses, leading to the formation of larger oligomers, which are then endocytosed at different rates and possibly with different fates. As shown earlier, the low-dose EGF stimulation (10 ng/mL) leads to the recycling of the receptors while high-dose EGF (100 ng/mL) to receptor degradation (33, 74).

Future studies—taking into account receptor transport to the membrane and internalization—will have to elucidate how the receptor dimerization and its domain clustering are dynamically regulated and how an equilibrium is reached.

Conclusions

The dimer fraction of EGFR on unstimulated cells has been measured many times, however, with very different outcomes. Reported values range over almost all possible values from 0 to 100%. The aim of this work was, therefore, to investigate on one side the influence of experimental parameters on the quantification of the EGFR dimer fraction and on the other side the consistency of the results over a range of different FCCS modalities.

Our results show that temperature has little effect and the dimer fraction is fairly consistent between measurements at room temperature and at physiological temperature. Also the measurement location on the apical and basal membranes does not show strong differences. Although in some cases the apical membrane seems to react more to drug treatments, this effect can probably be explained by the better accessibility of the solvent-exposed apical membrane compared to the basal membrane, which is attached to the coverslip for the adherent cells investigated here. In contrast to these parameters, the cell line plays a strong role in the results. Cell lines that express endogenous EGFR (COS-7, HEK293) show low apparent dimerization at low expression of the labeled EGFR as the endogenous receptor will interfere and mainly dimers between endogenous and transiently expressed, labeled receptors will be observed. The apparent dimer fraction in these cases will increase with the expression level of the labeled EGFR. For cells that do not possess any endogenous EGFR (CHO-K1), the dimer fraction is always high, independent of the expression level of labeled EGFR.

Low- and high-dose stimulation by EGF leads to increased EGFR complex formation, clustering, and internalization at different rates influencing the FCS results. Higher EGF doses lead to faster clustering and endocytosis, while at low EGF dose more medium size clusters or oligomers are observed.

Another important factor is cell-to-cell variability, which is considerable. In confocal FCS measurements this is difficult to quantify, as one has only a limited number of measurements per cell. Imaging FCS allowed us to quantify with good statistical significance the dimer fraction over a large part of the cell membrane. This showed that we have approximately one-third of all cells that have a high dimer fraction, one-third that have low dimer fraction, and one-third in between. This alone can lead to large variability when averaging measurements over different cells and is possibly one of the factors leading to a variety of results. It will be interesting to investigate whether this variability in 2D cell cultures is an artifact or whether it also exists in 3D cell cultures, tissues, and live organisms. Other possible approaches are to create a CRISPR/CAS9 transgenic line that allows measuring of receptor interaction in a stable expressing cell line and as well as in synchronized cells.

All FCCS modalities used in this work (SW-FCCS, DC-FCCS, quasi PIE-FCCS, and imaging FCCS) show the same high dimer fraction but provide different advantages. SW-FCCS is the simplest in terms of setup but even the negative control will have some cross-correlation due to spectral cross-talk. Quasi PIE-FCCS is cross-talk free and provides perfect negative controls that should increase the sensitivity of the technique when low cross-correlation need to be quantified. Imaging FCCS has a lower time resolution but is well suited for membrane measurements. Its main advantage is the multiplexing of measurements and its spatial resolution. Here we were able to show that dimer fractions tend to be higher at the border compared to the center of the cell. This would not have been easily possible with single-point FCS measurements.

Overall, our results indicate that the largest factors in the variability of dimer measurements are cell line and cell-to-cell variability, and to some extent the location of the measurement while temperature plays a minor role.

Author Contributions

T.W. designed the studies; S.Y. and R.M. performed experiments and analyzed data; and S.Y., R.M., and T.W. wrote the article.

Editor: Kalina Hristova.