letters
High-throughput fluorescence correlation spectroscopy enables analysis of proteome dynamics in living cells
© 2015 Nature America, Inc. All rights reserved.
Malte Wachsmuth1,5, Christian Conrad2,3,5, Jutta Bulkescher2,4, Birgit Koch1, Robert Mahen1, Mayumi Isokane1, Rainer Pepperkok1,2 & Jan Ellenberg1 To understand the function of cellular protein networks, spatial and temporal context is essential. Fluorescence correlation spectroscopy (FCS) is a single-molecule method to study the abundance, mobility and interactions of fluorescence-labeled biomolecules in living cells. However, manual acquisition and analysis procedures have restricted live-cell FCS to short-term experiments of a few proteins. Here, we present high-throughput (HT)-FCS, which automates screening and time-lapse acquisition of FCS data at specific subcellular locations and subsequent data analysis1,2. We demonstrate its utility by studying the dynamics of 53 nuclear proteins3,4. We made 60,000 measurements in 10,000 living human cells, to obtain biophysical parameters that allowed us to classify proteins according to their chromatin binding and complex formation. We also analyzed the cell-cycle-dependent dynamics of the mitotic kinase complex Aurora B/INCENP5 and showed how a rise in Aurora concentration triggers two-step complex formation. We expect that throughput and robustness will make HT-FCS a broadly applicable technology for characterizing protein network dynamics in cells. Characterizing proteins in their natural environment is essential for understanding their roles in cellular processes. Imaging of fluorescently labeled proteins in living cells is a powerful method for studying protein localization and interactions in their spatial and functional context, and genome editing methods now also allow physiological expression of such fusion proteins in human cell lines 6. Various techniques have been developed to extract from cells quantitative information about the dynamic behavior of labeled biomolecules7–9. Among these, FCS measures a particularly comprehensive set of parameters by analyzing local concentration fluctuations of single molecules to obtain information about free and bound fractions of labeled molecules, their diffusion properties and absolute concentrations8. The method can be extended to fluorescence cross-correlation spectroscopy (FCCS) experiments, where the fluctuation of differentially labeled molecules reveals quantitative bi- or multimolecular interaction properties2,10. Thus, FC(C)S measurements at specific sub cellular localizations identified by confocal fluorescence microscopy
is, in principle, ideally suited to explore the localization, diffusion and changing composition of protein complexes, allowing one, for example, to build three-dimensional cellular concentration maps11. Since its implementation on some commercial confocal microscopes, FC(C)S has successfully enabled studies, for example, on nuclear export–competent mRNA-protein particles12, transcription factors in developing embryos13, chromatin-associated and chromatinmodifying proteins14, and morphogen gradients15. However, confocal FC(C)S experiments in live cells are based upon a manual, labor-intensive workflow of image optimization and acquisition, decisions of where and when to acquire data, and the actual fluctuation-spectra recording. After acquisition, the processing and evaluation of the data is tedious manual work, because the method’s high sensitivity and time resolution reveal the strong variability of single-molecule properties in the cellular interior, requiring careful corrections to extract accurate information16. Also, biological variations in time and space have to be taken into account properly. In summary, because acquisition and analysis of FC(C)S data are carried out manually, require a great deal of time and are prone to bias, it has been difficult to exploit the potential of the technique for in vivo proteomics, that is, to study the dynamics of many different proteins or of the same protein at many different time points with sufficient sampling for robust statistical measurements. Recent developments in the automation of screening microscopy 17,18 have made it possible to overcome this challenge and fully automate live-cell FC(C)S with correlated confocal imaging and achieve high throughput and experimental and statistical robustness. Real-time image analysis by machine vision has already enabled unattended cellular classification19 and, combined with online feedback procedures using, for example, the ‘Micropilot’ software, has allowed fully automated high-resolution confocal imaging experiments 20,21. In addition, automatic FCS acquisition has allowed the analysis of a yeast library of more than 4,000 proteins 22, and more efficient processing of fluorescence fluctuation data became feasible for imaging FCS data23,24. Here, we have achieved full automation of the entire workflow of correlated confocal imaging and FC(C)S acquisition of live cells as well as data analysis to accomplish high-throughput characterization of a large
1Cell
Biology and Biophysics Unit, European Molecular Biology Laboratory, Heidelberg, Germany. 2Advanced Light Microscopy Facility, European Molecular Biology Laboratory, Heidelberg, Germany. 3Theoretical Bioinformatics, German Cancer Research Center/BioQuant, Heidelberg, Germany. 4Present address: Protein Imaging Center, Novo Nordisk Foundation Center for Protein Research and Danish Stem Cell Center, University of Copenhagen, Copenhagen, Denmark. 5These authors contributed equally to this work. Correspondence should be addressed to J.E. ([email protected]) or R.P. ([email protected]). Received 6 October 2014; accepted 24 December 2014; published online 16 March 2015; doi:10.1038/nbt.3146
nature biotechnology advance online publication
© 2015 Nature America, Inc. All rights reserved.
letters Figure 1 HT-FCS Workflow. (a) ‘Screening’ and ‘time-lapse’ are two instances of HT-FCS workflows that employ automated confocal imaging to control FC(C)S measurements of large sets of proteins or for detailed timeresolved analysis of smaller sets of proteins. In screening mode, a regular grid of positions in one or several wells of a plate is imaged in the confocal laser scanning microscope (CLSM). Position by position, an automatically focused high-resolution image is acquired, segmented and classified by Micropilot to identify cells suitable for FCS and automatically position FCS measurement points (marked with cross-hairs) in nucleus and cytoplasm. If no suitable cell is found, the position is skipped. In time-lapse mode, several positions with suitable cells are first selected; the system then tracks the cells over time at defined intervals. Also here, Micropilot automatically determines the subcellular coordinates of FCS measurement points. The system continues with the next position and, after all positions are measured, proceeds after the specified interval to the next time point. Scale bars, 5 µm. (b) Multistep automated processing of FCS data. The FluctuationAnalyzer module collects and processes the raw FCS single-molecule fluctuation data. For each measurement, correction factors for photobleaching and other slow fluctuations, background signal and spectral cross-talk are extracted from the raw data. Then the photobleaching-corrected correlation functions are computed and fitted with appropriate diffusion and interaction models. The resulting parameters are adjusted by the correction factors to yield physical parameters such as concentrations and diffusion coefficients. These are analyzed statistically for each protein in a screening application and for each time point in a time-lapse application, yielding tables, histograms and time-series of the parameters of interest.
set of proteins, as well as time-resolved characterization of protein complex dynamics in single cells over an entire cell cycle. To establish the HT-FCS workflow (Fig. 1a), we extended previously developed screening functionalities of a confocal microscope and related devices (Supplementary Fig. 1) to automatically acquire high-resolution images of cells expressing the protein of interest at regular grid positions in each of several wells of a multiwell plate. We extended the Micropilot software (Supplementary Table 1) to read each high-resolution image right after acquisition, to automatically identify a cell with fluorescent protein expression in the proper range to be suitable for FC(C)S, to determine nuclear and cytoplasmic protein positions for fluctuation measurements, and to execute FC(C)S acquisition. After that, the process is repeated at the next grid position, enabling us to sample tens to hundreds of cells per well and fluorescent fusion protein by HT-FCS in ‘screening’ mode. In addition, to follow the dynamic behavior of proteins in individual cells over time, we developed a ‘time-lapse’ HT-FCS workflow, by adding three-dimensional cell tracking after cell identification. We ran timelapse experiments to extract FC(C)S data from single cells through the cell cycle. In this way, up to several tens of cells could be followed individually with a time resolution as short as 10 min for 24 h. To automatically analyze large amounts of FC(C)S raw data, we developed the ‘FluctuationAnalyzer’ software package (Fig. 1b and Supplementary Software 1), which reads large numbers of raw data files, automatically corrects for slow processes such as photobleaching by adaptive local averaging (Supplementary Note and Supplementary Fig. 2) and extracts the molecular fluctuations of interest using temporal correlation. Rigorous mathematical analysis allowed us to correct
adaptively for photobleaching, background and cross-talk, after which the data were fitted with a two-component anomalous diffusion model initialized with unsupervised starting parameter estimations based on the raw fluctuation data (Online Methods and Supplementary Note). To validate the general applicability of the two-component model employed, we showed that the fit of a one-component system, EYFP expressed in HeLa cells (Supplementary Fig. 3a–d), converged to a single component25. All processing steps were implemented in one software package that enabled us to interactively explore, visualize and summarize relevant biochemical and biophysical properties for proteins of interest (Supplementary Software 2). Statistical analysis of the exported comprehensive results files showed that FC(C)S data from at least 100 cells were needed to robustly sample inherent cellto-cell variations (Supplementary Fig. 3e–h), defining the required scale of screening HT-FCS in human cells. Having developed the fully automatic data acquisition and analysis pipelines of HT-FCS technology, we next demonstrated the utility of the approach in two proof-of-concept biological applications. First, we set out to assess chromatin interactions of a set of 53 nuclear proteins in vivo. To this end we applied HT-FCS to characterize chromatin binding and complex formation properties of 53 EYFP-tagged cDNA constructs with nuclear localization4 in live HeLa cells stably co-expressing histone H2B-mCherry as a chromatin marker. On average, we acquired 1,133 FC(C)S measurements in 189 cells for each protein, totaling 60,072 measurements from 10,012 cells, collected in 69 unattended overnight sessions. Data analysis ran in parallel to acquisition with a one day delay. The resulting data set was used to extract 18 biochemical and biophysical parameters per protein advance online publication nature biotechnology
letters
1.0
0
0 0.01 1 100 Lag time (ms)
0 10 20 0 1 2 Mean diffusion Molecular coefficient brightness Dm (µm2s–1) CPM (#EYFP)
0 1 2 3 Concentration c (µM)
0
5 10 Crosscorrelation ratioG (%)
© 2015 Nature America, Inc. All rights reserved.
e
HP1BP3 H3F3B MECP2 DDX24 GATA3 HOXA9 SYF2 TCF12 OVOL2 RPL36 TARCBP PDZD4 CBX5 RNF113 RBM22 ZC3H13 RPL10 PNMAL1 SAMHD1 SF3B5 DDIT3 NEIL2 SENP3 DKC1 SAP130 AC004381.6 TFIP11 H2AFB3 H2A CGGBP1 USP37 APOLD1 LHX6 KIF22 ZSCAN22 USP1 CENPA PMF1 FAM50A SMARCE1 RFX7 EYFP UTP15 GADD45A UTP3 HORMAD1 XBP1 ZFP82 MED23 F8A2 ZBTB20 KIAA1967 DMAP1
SF3B5
d
Histograms
Cross-correlation ratioG (%)
0.001
MECP2
0
c
Correlation functions
SF3B5
1.0
MECP2
H2B-mCherry SF3B5-EYFP
b
Overlay
SF3B5
Cross Auto mCherry EYFP/mCherry
Overlay
MECP2
Auto EYFP
H2B-mCherry MECP2-EYFP
a
6
12
** 3
* *
0
Cluster I
*
*
*
Cluster II
*
*
Cluster III
Figure 2 Screening FCS. (a) Example of interphase HeLa cells Crosscorrelation transiently expressing 2 of the 53 nuclear proteins, MECP2-EYFP and SF3B5-EYFP, respectively, and the chromatin reference Concentration marker H2B-mCherry. At the positions marked with cross-hairs, Molecular corresponding two-color FCS measurements were acquired brightness and processed as described. Scale bar, 5 µm. (b) The resulting Diffusion autocorrelation functions (ACFs; Auto) of the EYFP constructs coefficient (blue, MECP2; orange, SF3B5) clearly show a difference in the Max. Free, fast Slow diffusion In complex, slower diffusion decay and thus in the mean diffusion coefficient as marked by the diffusion arrows in contrast to the very similar ACFs of H2B-mCherry. The Chromatin binding Not binding to chromatin Min. cross-correlation function (Cross) with H2B is higher for MECP2 than for SF3B5, indicating more interaction with chromatin (see arrows). (c) Histograms of the nuclear mobility parameters mean diffusion coefficient Dm, molecular brightness CPM (normalized to free EYFP), concentration c and cross-correlation ratioG resulting from 246 FCS measurements taken in 103 MECP2-EYFP-expressing cells and from 236 FCS measurements taken in 136 SF3B5-EYFP-expressing cells. (d) Cross-correlation ratioG for all proteins clustered in e (black line: mean value; gray bar: s.e.m.). Filled bars, ratioG values significantly higher than EYFP (significant chromatin binding); open bars, not significantly different from EYFP. HP1BP3 is shown out of scale as indicated by the percentage at the bar to improve the dynamic range of the graph. *, proteins that were also studied in cells with much higher expression levels. (e) Bispectral cluster analysis of the mobility parameters shown in c after bistochastic normalization, yielding three protein clusters with distinct functional properties in living cells.
(Supplementary List 1). The EYFP variant used to tag the cDNA library served as reference. EYFP alone showed a one-component diffusion and a monomodal brightness distribution (Supplementary Fig. 3i), indicating its predominantly monomeric state in cells under our expression conditions. For brevity, we concentrate here on nuclear HT-FCS measurements of two proteins, MECP2, a methyl-CpGbinding protein26, and SF3B5, a splicing factor27 (Fig. 2a) to illustrate how live-cell HT-FCS data can help to distinguish different functional properties. Compared to SF3B5, MECP2 exhibited slower diffusion (indicative of a larger complex), lower molecular brightness (indicative of a lower copy number in the complex), higher concentration (indicative of a higher expression level), and higher cross-correlation with the histone signal (indicative of stronger chromatin interaction) (Fig. 2b,c and Supplementary Figs. 4 and 5a,b). Different concentrations by overexpression were mostly uncorrelated with dynamic properties such as the mean diffusion coefficient (Supplementary Fig. 6). Compared to endogenous concentrations of ~100 nM for splicing factors like SF3B5 (ref. 12) and >10 µM for MECP2 (ref. 28), we observed about fivefold and
High-throughput fluorescence correlation spectroscopy enables analysis of proteome dynamics in living cells
© 2015 Nature America, Inc. All rights reserved.
Malte Wachsmuth1,5, Christian Conrad2,3,5, Jutta Bulkescher2,4, Birgit Koch1, Robert Mahen1, Mayumi Isokane1, Rainer Pepperkok1,2 & Jan Ellenberg1 To understand the function of cellular protein networks, spatial and temporal context is essential. Fluorescence correlation spectroscopy (FCS) is a single-molecule method to study the abundance, mobility and interactions of fluorescence-labeled biomolecules in living cells. However, manual acquisition and analysis procedures have restricted live-cell FCS to short-term experiments of a few proteins. Here, we present high-throughput (HT)-FCS, which automates screening and time-lapse acquisition of FCS data at specific subcellular locations and subsequent data analysis1,2. We demonstrate its utility by studying the dynamics of 53 nuclear proteins3,4. We made 60,000 measurements in 10,000 living human cells, to obtain biophysical parameters that allowed us to classify proteins according to their chromatin binding and complex formation. We also analyzed the cell-cycle-dependent dynamics of the mitotic kinase complex Aurora B/INCENP5 and showed how a rise in Aurora concentration triggers two-step complex formation. We expect that throughput and robustness will make HT-FCS a broadly applicable technology for characterizing protein network dynamics in cells. Characterizing proteins in their natural environment is essential for understanding their roles in cellular processes. Imaging of fluorescently labeled proteins in living cells is a powerful method for studying protein localization and interactions in their spatial and functional context, and genome editing methods now also allow physiological expression of such fusion proteins in human cell lines 6. Various techniques have been developed to extract from cells quantitative information about the dynamic behavior of labeled biomolecules7–9. Among these, FCS measures a particularly comprehensive set of parameters by analyzing local concentration fluctuations of single molecules to obtain information about free and bound fractions of labeled molecules, their diffusion properties and absolute concentrations8. The method can be extended to fluorescence cross-correlation spectroscopy (FCCS) experiments, where the fluctuation of differentially labeled molecules reveals quantitative bi- or multimolecular interaction properties2,10. Thus, FC(C)S measurements at specific sub cellular localizations identified by confocal fluorescence microscopy
is, in principle, ideally suited to explore the localization, diffusion and changing composition of protein complexes, allowing one, for example, to build three-dimensional cellular concentration maps11. Since its implementation on some commercial confocal microscopes, FC(C)S has successfully enabled studies, for example, on nuclear export–competent mRNA-protein particles12, transcription factors in developing embryos13, chromatin-associated and chromatinmodifying proteins14, and morphogen gradients15. However, confocal FC(C)S experiments in live cells are based upon a manual, labor-intensive workflow of image optimization and acquisition, decisions of where and when to acquire data, and the actual fluctuation-spectra recording. After acquisition, the processing and evaluation of the data is tedious manual work, because the method’s high sensitivity and time resolution reveal the strong variability of single-molecule properties in the cellular interior, requiring careful corrections to extract accurate information16. Also, biological variations in time and space have to be taken into account properly. In summary, because acquisition and analysis of FC(C)S data are carried out manually, require a great deal of time and are prone to bias, it has been difficult to exploit the potential of the technique for in vivo proteomics, that is, to study the dynamics of many different proteins or of the same protein at many different time points with sufficient sampling for robust statistical measurements. Recent developments in the automation of screening microscopy 17,18 have made it possible to overcome this challenge and fully automate live-cell FC(C)S with correlated confocal imaging and achieve high throughput and experimental and statistical robustness. Real-time image analysis by machine vision has already enabled unattended cellular classification19 and, combined with online feedback procedures using, for example, the ‘Micropilot’ software, has allowed fully automated high-resolution confocal imaging experiments 20,21. In addition, automatic FCS acquisition has allowed the analysis of a yeast library of more than 4,000 proteins 22, and more efficient processing of fluorescence fluctuation data became feasible for imaging FCS data23,24. Here, we have achieved full automation of the entire workflow of correlated confocal imaging and FC(C)S acquisition of live cells as well as data analysis to accomplish high-throughput characterization of a large
1Cell
Biology and Biophysics Unit, European Molecular Biology Laboratory, Heidelberg, Germany. 2Advanced Light Microscopy Facility, European Molecular Biology Laboratory, Heidelberg, Germany. 3Theoretical Bioinformatics, German Cancer Research Center/BioQuant, Heidelberg, Germany. 4Present address: Protein Imaging Center, Novo Nordisk Foundation Center for Protein Research and Danish Stem Cell Center, University of Copenhagen, Copenhagen, Denmark. 5These authors contributed equally to this work. Correspondence should be addressed to J.E. ([email protected]) or R.P. ([email protected]). Received 6 October 2014; accepted 24 December 2014; published online 16 March 2015; doi:10.1038/nbt.3146
nature biotechnology advance online publication
© 2015 Nature America, Inc. All rights reserved.
letters Figure 1 HT-FCS Workflow. (a) ‘Screening’ and ‘time-lapse’ are two instances of HT-FCS workflows that employ automated confocal imaging to control FC(C)S measurements of large sets of proteins or for detailed timeresolved analysis of smaller sets of proteins. In screening mode, a regular grid of positions in one or several wells of a plate is imaged in the confocal laser scanning microscope (CLSM). Position by position, an automatically focused high-resolution image is acquired, segmented and classified by Micropilot to identify cells suitable for FCS and automatically position FCS measurement points (marked with cross-hairs) in nucleus and cytoplasm. If no suitable cell is found, the position is skipped. In time-lapse mode, several positions with suitable cells are first selected; the system then tracks the cells over time at defined intervals. Also here, Micropilot automatically determines the subcellular coordinates of FCS measurement points. The system continues with the next position and, after all positions are measured, proceeds after the specified interval to the next time point. Scale bars, 5 µm. (b) Multistep automated processing of FCS data. The FluctuationAnalyzer module collects and processes the raw FCS single-molecule fluctuation data. For each measurement, correction factors for photobleaching and other slow fluctuations, background signal and spectral cross-talk are extracted from the raw data. Then the photobleaching-corrected correlation functions are computed and fitted with appropriate diffusion and interaction models. The resulting parameters are adjusted by the correction factors to yield physical parameters such as concentrations and diffusion coefficients. These are analyzed statistically for each protein in a screening application and for each time point in a time-lapse application, yielding tables, histograms and time-series of the parameters of interest.
set of proteins, as well as time-resolved characterization of protein complex dynamics in single cells over an entire cell cycle. To establish the HT-FCS workflow (Fig. 1a), we extended previously developed screening functionalities of a confocal microscope and related devices (Supplementary Fig. 1) to automatically acquire high-resolution images of cells expressing the protein of interest at regular grid positions in each of several wells of a multiwell plate. We extended the Micropilot software (Supplementary Table 1) to read each high-resolution image right after acquisition, to automatically identify a cell with fluorescent protein expression in the proper range to be suitable for FC(C)S, to determine nuclear and cytoplasmic protein positions for fluctuation measurements, and to execute FC(C)S acquisition. After that, the process is repeated at the next grid position, enabling us to sample tens to hundreds of cells per well and fluorescent fusion protein by HT-FCS in ‘screening’ mode. In addition, to follow the dynamic behavior of proteins in individual cells over time, we developed a ‘time-lapse’ HT-FCS workflow, by adding three-dimensional cell tracking after cell identification. We ran timelapse experiments to extract FC(C)S data from single cells through the cell cycle. In this way, up to several tens of cells could be followed individually with a time resolution as short as 10 min for 24 h. To automatically analyze large amounts of FC(C)S raw data, we developed the ‘FluctuationAnalyzer’ software package (Fig. 1b and Supplementary Software 1), which reads large numbers of raw data files, automatically corrects for slow processes such as photobleaching by adaptive local averaging (Supplementary Note and Supplementary Fig. 2) and extracts the molecular fluctuations of interest using temporal correlation. Rigorous mathematical analysis allowed us to correct
adaptively for photobleaching, background and cross-talk, after which the data were fitted with a two-component anomalous diffusion model initialized with unsupervised starting parameter estimations based on the raw fluctuation data (Online Methods and Supplementary Note). To validate the general applicability of the two-component model employed, we showed that the fit of a one-component system, EYFP expressed in HeLa cells (Supplementary Fig. 3a–d), converged to a single component25. All processing steps were implemented in one software package that enabled us to interactively explore, visualize and summarize relevant biochemical and biophysical properties for proteins of interest (Supplementary Software 2). Statistical analysis of the exported comprehensive results files showed that FC(C)S data from at least 100 cells were needed to robustly sample inherent cellto-cell variations (Supplementary Fig. 3e–h), defining the required scale of screening HT-FCS in human cells. Having developed the fully automatic data acquisition and analysis pipelines of HT-FCS technology, we next demonstrated the utility of the approach in two proof-of-concept biological applications. First, we set out to assess chromatin interactions of a set of 53 nuclear proteins in vivo. To this end we applied HT-FCS to characterize chromatin binding and complex formation properties of 53 EYFP-tagged cDNA constructs with nuclear localization4 in live HeLa cells stably co-expressing histone H2B-mCherry as a chromatin marker. On average, we acquired 1,133 FC(C)S measurements in 189 cells for each protein, totaling 60,072 measurements from 10,012 cells, collected in 69 unattended overnight sessions. Data analysis ran in parallel to acquisition with a one day delay. The resulting data set was used to extract 18 biochemical and biophysical parameters per protein advance online publication nature biotechnology
letters
1.0
0
0 0.01 1 100 Lag time (ms)
0 10 20 0 1 2 Mean diffusion Molecular coefficient brightness Dm (µm2s–1) CPM (#EYFP)
0 1 2 3 Concentration c (µM)
0
5 10 Crosscorrelation ratioG (%)
© 2015 Nature America, Inc. All rights reserved.
e
HP1BP3 H3F3B MECP2 DDX24 GATA3 HOXA9 SYF2 TCF12 OVOL2 RPL36 TARCBP PDZD4 CBX5 RNF113 RBM22 ZC3H13 RPL10 PNMAL1 SAMHD1 SF3B5 DDIT3 NEIL2 SENP3 DKC1 SAP130 AC004381.6 TFIP11 H2AFB3 H2A CGGBP1 USP37 APOLD1 LHX6 KIF22 ZSCAN22 USP1 CENPA PMF1 FAM50A SMARCE1 RFX7 EYFP UTP15 GADD45A UTP3 HORMAD1 XBP1 ZFP82 MED23 F8A2 ZBTB20 KIAA1967 DMAP1
SF3B5
d
Histograms
Cross-correlation ratioG (%)
0.001
MECP2
0
c
Correlation functions
SF3B5
1.0
MECP2
H2B-mCherry SF3B5-EYFP
b
Overlay
SF3B5
Cross Auto mCherry EYFP/mCherry
Overlay
MECP2
Auto EYFP
H2B-mCherry MECP2-EYFP
a
6
12
** 3
* *
0
Cluster I
*
*
*
Cluster II
*
*
Cluster III
Figure 2 Screening FCS. (a) Example of interphase HeLa cells Crosscorrelation transiently expressing 2 of the 53 nuclear proteins, MECP2-EYFP and SF3B5-EYFP, respectively, and the chromatin reference Concentration marker H2B-mCherry. At the positions marked with cross-hairs, Molecular corresponding two-color FCS measurements were acquired brightness and processed as described. Scale bar, 5 µm. (b) The resulting Diffusion autocorrelation functions (ACFs; Auto) of the EYFP constructs coefficient (blue, MECP2; orange, SF3B5) clearly show a difference in the Max. Free, fast Slow diffusion In complex, slower diffusion decay and thus in the mean diffusion coefficient as marked by the diffusion arrows in contrast to the very similar ACFs of H2B-mCherry. The Chromatin binding Not binding to chromatin Min. cross-correlation function (Cross) with H2B is higher for MECP2 than for SF3B5, indicating more interaction with chromatin (see arrows). (c) Histograms of the nuclear mobility parameters mean diffusion coefficient Dm, molecular brightness CPM (normalized to free EYFP), concentration c and cross-correlation ratioG resulting from 246 FCS measurements taken in 103 MECP2-EYFP-expressing cells and from 236 FCS measurements taken in 136 SF3B5-EYFP-expressing cells. (d) Cross-correlation ratioG for all proteins clustered in e (black line: mean value; gray bar: s.e.m.). Filled bars, ratioG values significantly higher than EYFP (significant chromatin binding); open bars, not significantly different from EYFP. HP1BP3 is shown out of scale as indicated by the percentage at the bar to improve the dynamic range of the graph. *, proteins that were also studied in cells with much higher expression levels. (e) Bispectral cluster analysis of the mobility parameters shown in c after bistochastic normalization, yielding three protein clusters with distinct functional properties in living cells.
(Supplementary List 1). The EYFP variant used to tag the cDNA library served as reference. EYFP alone showed a one-component diffusion and a monomodal brightness distribution (Supplementary Fig. 3i), indicating its predominantly monomeric state in cells under our expression conditions. For brevity, we concentrate here on nuclear HT-FCS measurements of two proteins, MECP2, a methyl-CpGbinding protein26, and SF3B5, a splicing factor27 (Fig. 2a) to illustrate how live-cell HT-FCS data can help to distinguish different functional properties. Compared to SF3B5, MECP2 exhibited slower diffusion (indicative of a larger complex), lower molecular brightness (indicative of a lower copy number in the complex), higher concentration (indicative of a higher expression level), and higher cross-correlation with the histone signal (indicative of stronger chromatin interaction) (Fig. 2b,c and Supplementary Figs. 4 and 5a,b). Different concentrations by overexpression were mostly uncorrelated with dynamic properties such as the mean diffusion coefficient (Supplementary Fig. 6). Compared to endogenous concentrations of ~100 nM for splicing factors like SF3B5 (ref. 12) and >10 µM for MECP2 (ref. 28), we observed about fivefold and
