Quantifying protein oligomerization in living cells: A systematic comparison of fluorescent proteins

Fluorescence fluctuation spectroscopy has become a popular toolbox for non-disruptive studies of molecular interactions and dynamics in living cells. The quantification of e.g. protein oligomerization and absolute concentrations in the native cellular environment is highly relevant for a detailed understanding of complex signaling pathways and biochemical reaction networks. A parameter of particular relevance in this context is the molecular brightness, which serves as a direct measure of oligomerization and can be easily extracted from temporal or spatial fluorescence fluctuations. However, fluorescent proteins (FPs) typically used in such studies suffer from complex photophysical transitions and limited maturation, potentially inducing non-fluorescent states, which strongly affect molecular brightness measurements. Although these processes have been occasionally reported, a comprehensive study addressing this issue is missing. Here, we investigate the suitability of commonly used FPs (i.e. mEGFP, mEYFP and mCherry), as well as novel red FPs (i.e. mCherry2, mRuby3, mCardinal, mScarlet and mScarlet-I) for the quantification of oligomerization based on the molecular brightness, as obtained by Fluorescence Correlation Spectroscopy (FCS) and Number&Brightness (N&B) measurements in living cells. For all FPs, we measured a lower than expected brightness of FP homo-dimers, allowing us to estimate, for each fluorescent label, the probability of fluorescence emission in a simple two-state model. By analyzing higher FP homo-oligomers and the Influenza A virus Hemagglutinin (HA) protein, we show that the oligomeric state of protein complexes can only be accurately quantified if this probability is taken into account. Further, we provide strong evidence that mCherry2, an mCherry variant, possesses a superior apparent fluorescence probability, presumably due to its fast maturation. We finally conclude that this property leads to an improved quantification in fluorescence cross-correlation spectroscopy measurements and propose to use mEGFP and mCherry2 as the novel standard pair for studying biomolecular hetero-interactions.


INTRODUCTION
A large variety of biological processes rely on transport and interactions of biomolecules in living cells. For a detailed understanding of these events, minimally invasive techniques are needed, allowing the direct quantification of inter-molecular interactions in the native cellular environment. In recent years, fluorescence fluctuation spectroscopy (FFS) approaches have been often used to fulfill this task [1][2][3][4][5][6] . FFS is based on the statistical analysis of signal fluctuations emitted by fluorescently labeled molecules. While the temporal evolution of such fluctuations provides information about dynamics, the magnitude of the fluctuations contains information about molecule concentration and interactions (i.e. oligomeric state). In order to probe the oligomerization of a protein of interest directly in living cells, the molecular brightness (i.e. fluorescence count rate per molecule) of a genetically fused fluorescent protein (FP) is analyzed 4,5,7 . Comparison to a monomeric reference allows the quantification of the number of FPs within a protein complex, i.e. its oligomeric state. This analysis can be performed with different experimental methods, e.g. Fluorescence Correlation Spectroscopy (FCS) 1,8 , Photon Counting Histogram (PCH) 4,9 , Number&Brightness analysis (N&B) 2,7 or subunit counting 10,11 .
Measuring the oligomeric state from the number of fluorescent labels, it is often assumed that all FPs emit a fluorescence signal. However, various in vitro studies of FPs revealed complex photophysical properties such as: long-lived dark states of green FPs [12][13][14][15] , transitions between different brightness states (e.g. YFP 16 , mCherry 17 ) and flickering 18 . Additionally, limited maturation was reported for FPs expressed in cells 19 . All together, these observations challenge the suitability of FPs for quantitative brightness analysis 5 . In this context, partially contradicting results are reported: studies performing subunit counting typically indicate apparent fluorescence probability (pf) values of 50-80% 10,11,20,21 for GFPs. Very few investigations utilizing FFS approaches report similar values 22,23 , while very often it is simply assumed that all FPs are fluorescent. For commonly used red FPs (mainly RFP and mCherry), published results tend to agree, consistently reporting low pf values (ca. 20-40%) 24,25 , with only few exceptions 17 .
Notably, many investigations would profit from systematic controls testing the presence of nonfluorescent labels, but so far only few studies take explicitly into account the role of the pf in the exact quantification of protein-protein interaction 5,11,22,26 . Importantly, oligomerization data are prone to severe misinterpretations if non-fluorescent labels are not taken into consideration.
To our knowledge, this is the first report systematically comparing non-fluorescent states and associated pf for various FPs in one-photon excitation. We found significant amounts of nonfluorescent FPs in different cell types and compartments, and we determined the pf for each FP.
With appropriate corrections, we were able to correctly determine the oligomeric state of the homo-trimeric Influenza A virus Hemagglutinin glycoprotein, for the first time directly in living cells, as a proof of principle.
To investigate multiple interacting molecular species simultaneously, multicolor FFS analysis is often performed. For example, protein hetero-interactions can be quantified via fluorescence cross-correlation approaches 1,27 , even in living multicellular organisms 6,28 . Such methods require well-performing FPs with spectral properties distinguished from the typically used mEGFP. Therefore, current FP development focuses on red and far-red FPs 29 . Nevertheless, the pf for these proteins, although playing a fundamental role in brightness and cross-correlation analysis, has not been systematically investigated yet. We therefore screened different red FPs for the presence of non-fluorescent states, and found that mCherry2, a new mCherry variant, possesses superior properties compared to all other tested red FPs, i.e. mCherry, mCardinal, mRuby3, mScarlet and mScarlet-I. Additionally, by performing FCCS measurements of FP hetero-dimers, we show that mCherry2 improves the quantification of the spectral crosscorrelation compared to mCherry and propose to use mEGFP and mCherry2 as a novel standard FP pair for hetero-interaction studies.

Results
The brightness of homo-dimers of conventional FPs is lower than double the brightness of monomers. In an ideal case, i.e. if all fluorophores within an oligomer were fluorescent, a homo-dimer would emit twice as many photons as a monomer. We expressed several FPs in the cytoplasm of HEK 293T cells and performed FFS measurements. We found that the brightness of homo-dimers (normalized to the brightness of the corresponding monomer) for three widely used FPs, namely mEGFP (edimer=1.69 ± 0.05), mEYFP (edimer =1.63 ± 0.05) and mCherry (edimer =1.41 ± 0.04), are generally lower than two, indicating the presence of nonfluorescent proteins. The effect is particularly pronounced for mCherry ( Figure 1A) and does not depend on the specific FFS method used or cellular localization, as shown by comparing the results from N&B, pFCS (in cytoplasm and nucleus) and sFCS (for FPs associated to the plasma membrane (PM)) ( Figure 1A, B). Interestingly, we observed a 10% lower brightness for FP monomers within the nucleus compared to the cytoplasmic fraction ( Figure S1A). Furthermore, we measured homo-dimer brightness values of mEGFP and mCherry in different cell lines (HEK 293T, A549, CHO, HeLa) and obtained comparable values in all cell types for the same FP ( Figure S1B, C).
The maturation time of FPs might influence the fraction of non-fluorescent proteins and this, in turn, may be dependent on the temperature at which experiments are performed 30 . For this reason, we compared the homo-dimer brightness of mEGFP at 23°C and 37°C, but observed negligible differences ( Figure S1D).
The oligomeric state of mEGFP homo-oligomers is correctly determined by using a simple correction scheme for non-fluorescent states. Based on the observed non-fluorescent protein fractions for mEGFP, mEYFP and mCherry, we investigated whether it is possible to nevertheless correctly determine the oligomeric state of higher-order oligomers. To this aim, we expressed mEGFP-homo-oligomers of different sizes: 1xmEGFP, 2xmEGFP, 3xmEGFP and 4xmEGFP (i.e. monomers to tetramers). We then performed pFCS measurements in the cytoplasm of living A549 cells (Figure 2A).
We observed brightness values consistently lower than those expected. For example, the obtained tetramer brightness (etetramer=3.01 ± 0.08) is closer to the theoretical trimer brightness value ( Figure 2B, white boxes). Hence, we performed a brightness correction based on a simple two-state model 11,31 , taking into account the probability that each FP subunit emits a fluorescence signal. The pf values were determined from the brightness of 2xmEGFP (edimer=1.65 ± 0.06, pf=0.65). Thus, we were able to correctly determine the oligomeric state of all mEGFP-homo-oligomers investigated in this study ( Figure 2B, grey boxes). Consistent with the brightness data, pFCS analysis revealed an increase of the diffusion times with increasing homo-oligomer size (SI and Figure S2A-C).
Overall, these results highlight the importance of performing control experiments with suitable homo-oligomers for brightness-based oligomerization studies and demonstrate that the simple correction for non-fluorescent states presented here produces reliable results. Box plots of normalized molecular brightness obtained from pFCS analysis, pooled from at least three independent experiments (1xmEGFP: n=52 cells, 2xmEGFP: n=42 cells, 3xmEGFP: n=43 cells, 4xmEGFP: n=59 cells) before correction (white) and after correction (grey). First, a normalization of the uncorrected brightness data was performed using the brightness value of 1xmEGFP. Second, a correction was performed as described in the Methods section, using a pf of 0.65, as obtained from measurements on 2xmEGFP. C: Representative images of U2OS cells expressing 1xmEGFP, 2xmEGFP and GlnA-mEGFP (GlnA). Scale bars are 5 µm. D: Box plots of normalized molecular brightness obtained from N&B analysis, pooled from three independent experiments (1xmEGFP: n=34 cells, 2xmEGFP: n=35 cells, GlnA: n=41 cells) before correction (white) and after correction (grey). After normalization using the brightness value of 1xmEGFP, a correction was performed using a pf of 0.72, as obtained from measurements on 2xmEGFP. Influenza A virus hemagglutinin forms homo-trimers in the plasma membrane. We next verified whether the above-mentioned simple two-state brightness correction provides reliable quantitative results in a biologically relevant context. We analyzed an mEGFP-fused version of the Influenza A virus hemagglutinin (HA-wt-mEGFP), a biochemically well-characterized trimeric transmembrane protein 33,34 . To this aim, we expressed the fluorescent construct in living HEK 293T cells and performed sFCS measurements (Figs. 3A and S3) across the PM.
After correction for the non-fluorescent FPs contribution, we obtained an average normalized brightness of eHA=3.17±0.12 (Fig 3B), in line with the expected trimeric structure of HA-wt-mEGFP. We further investigated an HA-TMD mutant, in which the HA ectodomain is replaced by mEGFP on the extracellular side. This construct was shown to localize as HA-wt in the PM, but in a dimeric form 35 . The observed brightness of HA-TMD-mEGFP was significantly lower than that of HA-wt-mEGFP ( Figure 3B). After correcting for non-fluorescent FPs, we found an average normalized brightness of eHA-TMD=1.82 ± 0.07, compatible with the presence of a large dimer fraction.
In summary, these results clearly demonstrate that a simple two-state model for FFS-derived brightness data correction allows precise quantification of the oligomeric state of proteins in living cells. The mCherry variant "mCherry2" has a superior performance in FFS measurements, compared to other red fluorescent proteins. In order to extend brightness measurements to the investigation of hetero-interactions, FPs with spectral properties different from those of mEGFP are needed. Typically, red FPs are well suited for this task since spectral overlap with mEGFP is low, reducing the possibility of FRET or cross-talk. However, for the typically used mCherry, we and others 25 observed a high fraction of non-fluorescent states, i.e. only ca. 40% of the proteins were fluorescent. In order to identify red FPs with higher pf, we screened the recently developed FPs mCherry2 36 , mCardinal 37 , mRuby3 38 , mScarlet 39 and mScarlet-I 39 . We performed bleaching and N&B measurements of monomers and homo-dimers, expressed in HEK 293T cells. Notably, we observed strong photobleaching for mRuby3, mScarlet and mScarlet-I ( Figure 4A, Table 1) compared to the other three tested FPs. Therefore, N&B measurements on these proteins were conducted at lower excitation powers. This reduces their effective brightness, e.g. only 1 kHz for mRuby3, compared to the theoretically three-fold higher brightness when interpolated to the same laser powers used for mCherry, mCherry2 and mCardinal ( Figure 4B). All other FPs exhibit minor difference in the effective brightness ranging from 1.5 kHz (mCherry2) to 2.2 kHz (mCardinal, mScarlet) in our experimental conditions. However, when comparing the normalized homo-dimer brightness, we found strong differences between mCherry2 and the other FPs. We estimated a pf of 0.71 for mCherry2, which is ~1.8-fold higher than that of mCherry (pf=0.41) and mScarlet (pf=0.40), while mCardinal and mRuby3 show very low pf values of only 0.24 and 0.22, respectively. Notably, mScarlet-I also features a high pf (0.63), but still suffers from considerable photobleaching, even at lower excitation powers ( Figure 4C, Table 1).
The superior performance of mCherry2 was confirmed in other cell types, as we consistently observed a reproducible difference from mCherry ( Figure S4 A). Moreover, we compared the homo-dimer brightness of mCherry2 at 23°C and 37°C and, similarly to mEGFP, observed only negligible variations (Figure S4 B).
We therefore conclude that mCherry2 exhibits cell type-and temperature-independent, superior properties in the context of FFS measurements, compared to all the other tested red FPs.  are powerful methods for the investigation of protein hetero-interactions. These techniques are based on the analysis of simultaneous fluorescence fluctuations emitted by co-diffusing, spectrally distinct labeled molecules. In protein-protein interaction studies, fusion proteins with mEGFP and mCherry or RFP are typically used for this purpose 1,2,6,40 . However, due to the presence of non-fluorescent states, only a fraction of protein hetero-complexes simultaneously emits fluorescence in both channels (i.e. many complexes will contain fluorescent green proteins and non-fluorescent red proteins). This factor has to be taken into account when calculating e.g. dissociation constants from cross-correlation data 25 . Given the superior pf of mCherry2 compared to other red FPs, we hypothesized that mCherry2 would improve the quantification of cross-correlation data, since more complete fluorescent protein-complexes should be present. To test this hypothesis, we performed pFCCS experiments with mCherry-mEGFP and mCherry2-mEGFP hetero-dimers in the cytoplasm of living A549 cells. As presumed, we observed a higher auto-correlation function (ACF) amplitude G in the red than in the green channel (Gg/Gr=0.65±0.03, Figure 5A,C) for mCherry-mEGFP, indicating that the apparent concentration of mCherry is ca. 1.5-fold lower than that of mEGFP (i.e. in a significant fraction of hetero-dimers, only mEGFP is fluorescent). This is in agreement with the expected relative amount of hetero-dimers containing fluorescent mEGFP and/or mCherry, based on the above-mentioned pf values. Furthermore, we expect ⁓27% of hetero-dimers to carry both fluorescent mEGFP and mCherry (see SI related to Figure S5).
For mCherry2-mEGFP in contrast, the amplitudes of the ACFs in the red and green channel were comparable (Gg/Gr=0.97±0.05, Figure 5B,C), indicating, as expected, similar apparent concentrations of mCherry2 and mEGFP (see SI related to Figure S5). Also, the relative amount of hetero-dimers carrying fluorescent mEGFP and mCherry2 is estimated to be ⁓42%, i.e. 1.5fold more fully-fluorescent complexes than for mCherry-mEGFP. On the other hand, the expected cross-correlation values for mCherry-and mCherry2-mEGFP should be similar, which is confirmed by our data ( Figure S5). Nevertheless, the 1.5-fold higher relative fraction of fully-fluorescent hetero-dimers with mCherry2 should improve the quality of crosscorrelation data. We therefore compared the signal-to-noise (SNR) ratio of the measured CCFs for mCherry-and mCherry2-mEGFP, and observed a ⁓40% higher SNR of mCherry2-mEGFP CCFs (SNRmCh-mEGFP=1.39±0.10, SNRmCh2-mEGFP=1.93±0.10; Figure 5D).
These results demonstrate that using mCherry2 instead of mCherry in cross-correlation experiments leads to a more accurate quantification of the spectral cross-correlation, i.e. of the degree of binding in hetero-interactions or the mobility of hetero-complexes. shown that the latter proteins assemble as trimers and dimers, respectively 33 to a strong misinterpretation of the data, e.g. a tetramer being classified as a trimer (Fig 2).
These systematic errors are particularly pronounced for FPs with a low pf, as found e.g. for mCherry (⁓40%), a FP often used in the past to determine the stoichiometry of protein performed with two-photon excitation, which may influence the transition to non-fluorescent states 17 . In this context, our data provide the first complete and systematic comparison of pf for several FPs, in one-photon excitation setups.
In order to measure multiple species simultaneously, red FPs are required due to their spectral separation from mEGFP. Additionally, they are a preferential choice for tissue and animal imaging, due reduced light absorption and autofluorescence in the red and far-red spectral region 29 . Given the suboptimal pf we determined for mCherry, we screened several recently developed monomeric red FPs [36][37][38][39] . The pf is in fact an essential parameter that, until now, has not received appropriate attention in reports of new FPs. The suitability of these proteins for FFS studies depends on three important fluorophore characteristics: 1) a high photostability is required to enable temporal measurements under continuous illumination, 2) a high molecular brightness is needed to obtain a signal-to-noise ratio sufficient to detect single-molecule fluctuations, 3) a high pf is essential for a maximal dynamic range that allows reliable oligomerization measurements. Thus, red FPs which fulfill only one or two of these requirements are not recommended for FFS measurements. Among all red FPs investigated in this study, we found only one fulfilling all three important criteria: mCherry2, a rarely used mCherry variant 36 that has not been entirely characterized yet. However, for the remaining red FPs tested here, we found either low photostability albeit high monomer brightness (mRuby3, mScarlet, mScarlet-I; Fig. 4A, B), and/or low to medium pf of 20-45% (mCardinal, mCherry, mRuby3, mScarlet; Fig. 4C), very similar to previously published values for mRFP 24,25 and mCherry 25 . In contrast, mCherry2 possesses a high pf of ⁓70%. Very recent studies of FP maturation times report a faster maturation of mCherry2 and mScarlet-I compared to mCherry/mScarlet 30 . Together with our findings, this indicates that faster FP maturation could be the reason for the observed higher pf.
Finally, we demonstrate that quantification of hetero-interactions via cross-correlation approaches, so far typically performed with mCherry 1,2,40 , can be substantially improved by using mCherry2 instead. In agreement with the reported similar pf of mEGFP and mCherry2 (Figs. 1, 4), we observed that the amount of hetero-dimers containing both fluorescent mEGFP and mCherry2 increased significantly compared to those containing mCherry. For this reason, the CCF signal-to-noise ratio for mCherry2-mEGFP complexes increased by 40% compared to that measured for mCherry-mEGFP hetero-dimers (Fig. 5D). This could be particularly relevant for investigations of weak interactions, in which only a small number of hetero-complexes is present, compared to the vast amount of non-interacting molecules. Additionally, crosscorrelation techniques have been recently applied in living multicellular organisms 6,28 , which require low illumination to avoid phototoxicity and thus generally suffer from low signal-tonoise ratios. Therefore, we recommend using mCherry2 as the novel standard red FP in brightness and cross-correlation measurements.
In conclusion, this study provides a useful, comprehensive resource for applying FFS techniques to quantify protein oligomerization and interactions. We provide a clear, simple methodology to test and correct for the presence of non-fluorescent states, and argue that such controls should become a prerequisite in brightness-based FFS studies to avoid systematic errors in the quantification of protein oligomerization. Finally, our results suggest that the apparent fluorescence probability is an important fluorophore characteristic that should be considered and reported when developing new FPs and we provide a simple assay to determine this quantity. To obtain 2xmEYFP, mEYFP was amplified from mEYFP-N1 1 and inserted into mEYFP-C1 by digestion with KpnI and BamHI.

MATERIALS AND METHODS
The plasmids mCherry-C1 and mCherry2-C1/N1 (gifts from Michael Davidson, Addgene plasmids #54563 and #54517, respectively) were used to generate 2xmCherry and 2xmCherry2, respectively. First, mCherry-C1 and -N1 were generated by amplification of mCherry from mCherry-pLEXY plasmid (a gift from Barbara Di Ventura & Roland Eils, Addgene plasmid #72656) and inserted into a pBR322 empty vector. A second mCherry cassette was inserted into this vector by digestion with XhoI and BamHI to obtain 2xmCherry (i.e. mCherry dimer).
The 2xmCherry2 (mCherry2 dimer) plasmid was generated by amplification of mCherry2 and insertion of this construct into mCherry2-C1 through digestion with XhoI and BamHI. To clone mRuby3-C1, mRuby3 was amplified from the pKanCMV-mClover3-mRuby3 plasmid, a gift from Michael Lin (Addgene plasmid #74252). The obtained PCR product was digested with AgeI and XhoI and exchanged with mEYFP from digested mEYFP-C1 plasmid. For 2xmRuby3 (mRuby3 dimer), mRuby3 was again amplified by PCR and the product inserted into mRuby3- For two-color measurements, all ACFs were used to fit the diffusion model described above.
Relative cross-correlation values were calculated from the amplitudes of ACFs and CCFs: The pixels corresponding to the membrane were defined as pixels which are within ±2.5s of the peak. In each line, these pixels were integrated, providing the membrane fluorescence time series F(t). When needed, a background correction was applied by subtracting the average pixel fluorescence value on the inner side of the membrane multiplied by 2.5s (in pixel units) from the membrane fluorescence, in blocks of 1000 lines 12 . In order to correct for depletion due to photobleaching, the fluorescence time series was fitted with a two-component exponential function and a correction was applied 5 . Finally, the ACF was calculated as described above.
A model for two-dimensional diffusion in the membrane and a Gaussian focal volume geometry 11 was fitted to the ACF: .
Number&Brightness Analysis. N&B experiments were performed as previously described 13 , with a modified acquisition mode. Briefly, 200 images of 128x64-128 pixels were acquired per measurement, using a 300 nm pixel size and 25 µs pixel dwell time. Laser powers were maintained low enough to keep bleaching below 10% of the initial fluorescence signal (typically ~0.7 µW for 488 nm and ~4.9 µW for 561 nm) except for mRuby3 and mScarlet/ mScarlet-I. CZI image output files were imported in MATLAB using the Bioformats package 14 and analyzed using a custom-written script. Before further analysis, pixels corresponding to cell cytoplasm or nucleus were selected manually as region of interest. Brightness values were calculated as described 13 , applying a boxcar algorithm to filter extraneous long-lived fluctuations 15,16 . Pixels with count rates above 2 MHz were excluded from the analysis to avoid pile-up effects. To further calibrate the detector response, we measured the brightness on a reflective metal surface and dried dye solutions. The thus obtained brightness-versus-intensity plots (which should be constant and equal to 0 for all intensity values 13 ) were used to correct the actual experimental data. We applied this transformation to every brightness data point and obtained the "corrected" brightness. Notably, this transformation holds true also for fluorophores which have two brightness states rather than an on and off state 17 .
Statistical analysis. All data are displayed as box plots indicating the median values and whiskers ranging from minimum to maximum values. Statistical significance was tested using a two-tailed Mann-Whitney test or one-way ANOVA for multiple comparisons. Quantities in the main text are given as mean±S.E.M.
Code availability. MATLAB custom-written code is available upon request from the corresponding author.
Data availability. The datasets analyzed during the current study are available from the corresponding author on reasonable request.