Owing to the tremendous progress in microscopic imaging of fluorescently labeled proteins in living cells, the insight into the highly dynamic behavior of transcription factors has rapidly increased over the past decade. However, a consistent quantitative scheme of their action is still lacking. Using the androgen receptor (AR) as a model system, we combined three different fluorescence microscopy assays: single-molecule microscopy, photobleaching and correlation spectroscopy, to provide a quantitative model of the action of this transcription factor. This approach enabled us to distinguish two types of AR–DNA binding: very brief interactions, in the order of a few hundred milliseconds, and hormone-induced longer-lasting interactions, with a characteristic binding time of several seconds. In addition, freely mobile ARs were slowed down in the presence of hormone, suggesting the formation of large AR–co-regulator complexes in the nucleoplasm upon hormone activation. Our data suggest a model in which mobile hormone-induced complexes of transcription factors and co-regulators probe DNA by briefly binding at random sites, only forming relatively stable transcription initiation complexes when bound to specific recognition sequences.
The androgen receptor (AR) is a ligand-activated transcription factor that specifically regulates genes involved in the development and maintenance of the male phenotype; it also plays a role in the growth of prostate cancer. Like all steroid receptors (SRs), the AR has a modular structure composed of an N-terminal domain, a DNA-binding domain (DBD) and a C-terminal ligand-binding domain. Upon activation by agonistic ligand binding, SRs translocate from the cytoplasm to the nucleus where they bind hormone response elements in promoter and enhancer regions of target genes. When bound to the target sequences, SRs initiate the recruitment of specific transcriptional co-regulators, which alter local chromatin structure in order to enhance transcription initiation. Subsequently, the basal transcription machinery is recruited, inducing transcription of target genes (McKenna and O'Malley, 2002).
In the past decade, fluorescent labeling of proteins in living cells and advances in quantitative live-cell microscopy has greatly influenced our view of the organization of nuclear processes. Approaches like fluorescence recovery after photobleaching (FRAP) (Van Royen et al., 2009) and fluorescence correlation spectroscopy (FCS) (Weidtkamp-Peters et al., 2009) have provided novel insights into the mechanism of action of nuclear processes. Initial FRAP studies have revealed unexpectedly high mobilities and the occurrence of only brief immobilization events for proteins involved in many nuclear processes, including DNA replication (Leonhardt et al., 2000), DNA damage repair (Essers et al., 2002; Houtsmuller et al., 1999), gene transcription (Dundr et al., 2002; Kimura et al., 2002; McNally et al., 2000; Schaaf et al., 2006) and RNA processing (Kruhlak et al., 2000; Phair and Misteli, 2000). A multitude of FRAP studies have shown that SRs share this common behavior. Importantly, the observed transient immobilizations appear to be dependent on ligand activation and the DNA-binding ability of receptors, suggesting that activated SRs move freely through the nucleus and are bound to chromatin for only short time periods (Farla et al., 2004; Farla et al., 2005; Klokk et al., 2007; Marcelli et al., 2006; McNally et al., 2000; Meijsing et al., 2007; Mueller et al., 2008; Rayasam et al., 2005; Schaaf and Cidlowski, 2003; Schaaf et al., 2005; Stenoien et al., 2001; van Royen et al., 2007).
Using kinetic modeling, a quantitative analysis of FRAP has been performed in several studies (Farla et al., 2005; Hinow et al., 2006; Mueller et al., 2008; Phair et al., 2004; Sprague et al., 2004). However, because of the large number of variables (e.g. number of binding sites, on- and off-rates, and relative sizes of free and bound fractions), the variety of analytical approaches and the inaccuracy of FRAP at short time intervals, the results of these quantifications have not yet provided a consistent view on transcription factor mobility and the nature and timing of their interactions with DNA (van Royen et al., 2011).
To some extent, this problem can be addressed by a complementary approach. FCS has already been applied to SRs in several studies (Jankevics et al., 2005; Mikuni et al., 2007a; Mikuni et al., 2007b) and, recently, the group of McNally (Stasevich et al., 2010) cross-validated FRAP and FCS measurements on glucocorticoid receptor dynamics. Although the findings obtained with the two approaches were relatively consistent, uncertainties, because of a number of approximations in the FRAP and FCS analyses, call for additional approaches to obtain conclusive knowledge of the nature and dynamics of DNA interaction by nuclear proteins.
The most powerful approach to complement the limitations of both FRAP and FCS is to study protein behavior by single-molecule microscopy (SMM) in living cells. Using a laser-based fluorescence microscopy setup equipped with a high-sensitivity and high-speed charge-coupled device (CCD) camera (Schmidt et al., 1996), SMM has successfully been applied to proteins fused to autofluorescent proteins, like GFP, providing insight into the mobility patterns of several proteins at a time resolution of ∼5 ms and a positional accuracy of ∼40 nm (de Keijzer et al., 2008; Harms et al., 1999; Iino et al., 2001; Lommerse et al., 2005). Initially, these studies mostly focused on membrane proteins, but in recent years, data on the three-dimensional (3D) mobility of fluorescently labeled proteins in the nuclei of living cells have been extracted using this approach. The intra-nuclear mobility of fluorescently labeled inert proteins, such as streptavidin (Grünwald et al., 2008) and ovalbumin (Speil and Kubitscheck, 2010) was determined, showing that these proteins appear to be immobilized transiently inside the nucleoplasm for ∼10–20 ms. The first SMM study on a transcription factor in a living cell was performed on the lac repressor in Escherichia coli cells, in which brief immobilizations (<5 ms) were also observed (Elf et al., 2007). Recently, the transcription factor STAT1 has been studied using SMM, which revealed that activated STAT1 diffuses freely through the nucleus and is transiently immobilized, showing residence times of up to 5 s (Speil and Kubitscheck, 2010).
In the present paper, we have combined SMM with FRAP and FCS in order to study the intra-nuclear dynamics of the AR in detail. This combination of techniques provides consistent quantitative data on the mobility pattern of AR in the nucleus. Our results show the occurrence of a freely diffusing fraction and two different binding events, representing sequence-specific and nonspecific DNA binding. The combination of these three techniques enables the determination of the relative size of the different fractions, the diffusion coefficient of freely moving molecules and binding residence times.
Analysis of single YFP dynamics in 3D
In order to validate our methods for detecting molecules and analysis of the dynamic behavior in a 3D </emph>environment, we first studied the free diffusion of yellow fluorescent protein (YFP) in a 50% glycerol solution. Images of this solution, which were captured using an SMM setup (Schmidt et al., 1996), showed individual fluorescence intensity peaks (Fig. 1A) representing single molecules, identified because they fitted well to a Gaussian distribution with an intensity and width similar to single YFP fluorescence intensity peaks previously observed using an identical setup (Harms et al., 2001; Lommerse et al., 2004; Schaaf et al., 2009). The observed signal-to-noise ratio, defined as the fluorescence intensity of an individual fluorophore divided by the standard deviation of the background signal, was ∼17, resulting in a positional accuracy of the localization of these individual molecules of ∼33 nm (Schmidt et al., 1996).
Image sequences were acquired using time intervals of 6.25, 12.5 and 25 ms, and protein mobility was analyzed using the Particle Image Correlation Spectroscopy (PICS) analysis method described previously (Fig. 1B–D) (Semrau and Schmidt, 2007). The obtained cumulative distribution function of squared displacements Pcum(l) fitted well to a one-population model and, for each time lag used, the mean squared displacement (MSD) was calculated using this fit model (Fig. 1D). These values were plotted as a function of the time lag and the resulting curve showed a straight line, which reflected free diffusion of YFP molecules in the 50% glycerol solution (Fig. 1E) with a diffusion coefficient (D) of 7.35±0.99 µm2/s (all results are shown as ±s.e.m.). This D is in the range of the expected value (9.4 µm2/s) for YFP in 50% glycerol, which was determined based on the estimated hydrodynamic radius of YFP, using Eqn 4 (see Materials and Methods). Deviations in temperature, glycerol concentration or in the homogeneity of the solution could underlie the difference between the expected and determined value. Subsequently, we studied the dynamics of YFP-labeled histone protein H2B in the nuclei of living (Hep3B) cells. Histone proteins are known to be stably bound to DNA and are, therefore, predominantly immobile. Similarly to the data on free YFP, the data on H2B-YFP fitted to a one-population model. The MSD plot showed a straight line with a diffusion coefficient ∼200-fold lower than that of YFP in 50% glycerol (D = 0.040±0.0023 µm2/s</emph>), which probably reflects the slow movement of chromatin in these live nuclei (Fig. 1F).
In silico validation of analysis of 3D protein dynamics
Because we image 2D projections of molecules moving in three dimensions, and the thickness of the ‘optical slice’ from which this projection is made is limited, one can argue that molecules ‘escaping’ in the z-direction create a bias in our analysis. To determine the potential limitations of our analysis, we generated data in a series of Monte Carlo simulations and studied whether the optical slice thickness affected the analysis of molecular dynamics (see supplementary material Figs S1, S2). The results of these in silico experiments showed that our approach is well suited for analysis of 3D dynamics of a single fraction of freely diffusing molecules (D = 0–10 µm2/s) (supplementary material Fig. S1). In addition, when an immobile fraction was introduced, the simulations demonstrated that molecular dynamics can still be accurately determined using our approach when the immobile fraction was determined at the shortest time lag used (6.25 ms) (supplementary material Fig. S2).
Quantitative analysis of individual AR dynamics
To obtain a detailed description of the dynamic nuclear behavior of ARs in living cells, we applied SMM to Hep3B cells stably expressing ARs labeled with YFP that had been treated with the synthetic AR agonist metribolone (R1881) or the antagonist OH-flutamide (OHF) (Figs 2, 3). Unlike the data for YFP and H2B–YFP, the cumulative distribution function Pcum (l) of the ARs best fitted a two-population model (Fig. 3A). From this, the relative fraction size (α) and their mean squared displacements (MSD1 and MSD2) were determined and plotted as a function of the time lag.
For wild-type AR in the presence of R1881, the size of the fast fraction was found to be 46.0%±2.9 (Fig. 3B; Table 1). The MSDs of the fast and slow fraction plotted against the time lag fitted to a straight line, indicating free diffusion of both fractions through the nucleus at this timescale (Fig. 3C,D). The diffusion coefficient of the fast fraction (D1) was calculated to be 1.13±0.09 µm2/s (Fig. 3G; Table 1), whereas the diffusion coefficient for the slow fraction (D2) was 0.056±0.003 µm2/s (Fig. 3H; Table 1). The latter diffusion coefficient is in the same range as that found for H2B–YFP, strongly suggesting that this fraction of ARs is bound to chromatin (see also Fig. 1A,C).
Wt, wild-type; imm, immobilization; diff, diffusion. Results are means±s.e.m.
To study the role of AR activity on its mobility in more detail, we performed SMM on OHF-bound AR. The size of the fast fraction increased dramatically to 88.5%±2.8 of ARs (Fig. 3B; Table 1). The diffusion coefficient of this fast fraction was higher than in the presence of R1881 (2.31±0.10 µm2/s; Fig. 3C,G; Table 1), whereas the diffusion coefficient of the slow fraction was unchanged and in the same range as chromatin-bound H2B (0.063±0.008 µm2/s; Fig. 3D,H).
To verify that the slow fraction of ARs is a result of binding to DNA, an AR was used with a point mutation in the DBD (R585K). This mutant, in which the mutated amino acid is important for base-specific interaction with AR target sequences in DNA (Shaffer et al., 2004), has been found in a patient with complete androgen insensitivity syndrome (Sultan et al., 1993). In line with these findings, it was shown that this AR mutant is transcriptionally inactive (Lobaccaro et al., 1999) and lacks stable binding to DNA (van Royen et al., 2012). Taken together, these data suggest that the R585K mutation disrupts specific interaction of ARs with their target sites in the genome. Like wild-type AR, the AR R585K mutant showed two fractions of molecules. In the presence of the agonist R1881, the size of the fast fraction of mutant ARs (61.1±6.7%; Fig. 3B) was increased as compared with wild-type AR. In addition, the diffusion coefficients of both the fast (1.42±0.08 µm2/s; Fig. 3E,G; Table 1) and the slow fraction (0.077±0.005 µm2/s; Fig. 3F,H; Table 1) were only slightly increased. Thus, the results show that the AR R585K mutant dynamics are only slightly changed as compared with the wild-type receptor, indicating that the mutant is still able to bind to chromatin for periods within the timescale of the experiment (<50 ms).
The presence of the antagonist OHF slightly increased the size of the fast fraction of the mutant AR (79.7±4.0, Fig. 3B) in comparison with R1881, but the difference between R1881 and OHF is remarkably smaller than for the wild-type receptor (Fig. 3B; Table 1). The diffusion coefficient of the slow fraction (0.090±0.013 µm2/s; Fig. 3F,H) was unaltered, whereas the diffusion coefficient of the fast fraction (2.24±0.19 µm2/s; Fig. 3E,G; Table 1) was increased. Apparently, the difference between the sizes of the fast fractions of R1881-bound and OHF-bound wild-type ARs depends on the ability to bind DNA, whereas the difference in diffusion rate of the fast fraction is not dependent on the DNA-binding capacity of AR.
In summary, the results of our SMM experiments indicate the presence of two AR fractions, that both show free diffusion through the nucleus at the timescale of our experiments. The diffusion coefficient of the slow fraction is ∼20-fold lower than that of the fast fraction. Treatment with an antagonist dramatically decreased the size of the slow fraction and increased the diffusion rate of the fast fraction. Interestingly, a mutation in the DBD decreased the difference in the size of the fast fraction between agonist- and antagonist-bound ARs but left the difference in diffusion rate of the fast fraction intact.
Combining SMM analysis of ARs with FCS and FRAP
To verify the parameters obtained by SMM and expand the timescale of measurements on the dynamic behavior of the AR, the cell lines stably expressing AR–YFP and its R585K mutant, which had been used in the single-molecule analysis, were subjected to both FCS and FRAP (Figs 4 and 5, respectively). For accurate comparison, it must be noted that the FCS approach used in this study, in which intensity fluctuations are measured for 20 s, does not detect molecules that are immobile for periods in the range of seconds and longer because of photobleaching and the small number of long events in this time frame, and that inaccuracy in FRAP at short time intervals limits the ability to extract diffusion parameters, especially for highly mobile molecules.
Therefore in FCS, diffusion rates were only extracted from the retention times of the YFP tagged molecules in the confocal volume, using a two-population free-diffusion triplet-state model (Fig. 4A; supplementary material Fig. S3). The FCS data showed a lower diffusion coefficient (D) for wild-type AR in the presence of R1881 than in the presence of OHF (1.61±0.26 and 2.42±0.37 µm2/s, respectively; Fig. 4B). Although the absolute diffusion rates are slightly lower, they are in the same range as those found with SMM (1.13±0.09 µm2/s and 2.31±0.10 µm2/s; Fig. 3G; Table 1). This trend was also observed for the R585K mutant AR, for which the diffusion rates (D) determined using FCS are 1.78±0.19 µm2/s and 2.64±0.39 µm2/s in the presence of R1881 and OHF, respectively (Fig. 4B; supplementary material Fig. S3), and 1.42±0.08 µm2/s and 2.24±0.19 µm2/s in SMM (Fig. 3G; Table 1). Thus, the diffusion rates determined by FCS were consistent with the findings from our single-molecule experiments (Table 1).
Subsequently, FRAP experiments were performed and the resulting FRAP data were fitted to curves obtained using computer modeling described previously (e.g. Farla et al., 2005; Van Royen et al., 2009). The large immobile fractions for agonist-bound wild-type AR found in SMM could not be fully attributed to long immobilization events and required the inclusion of short immobilizations in the model. Note that, in previous reports (Farla et al., 2005), these short immobilizations have been explained by slower diffusion but, in combination with the SMM experiments presented here, this model is no longer sufficient.
The diffusion coefficients obtained using SMM and FCS were averaged and used as fixed parameters in the FRAP analysis, and the curves were fitted to a reaction diffusion model with two immobile fractions, one fraction was previously found to have a long interaction time (Farla et al., 2005) and one additional fraction of ARs that had short interactions with the DNA (Fig. 5; supplementary material Fig. S4). As SMM does not discriminate between these long and short immobilizations because of the temporal resolution of this technique (<50 ms), the sum of the two fractions in the FRAP data corroborates the results from the SMM experiments (Table 1).
FRAP of agonist-bound wild-type AR showed a fraction of ARs (28%±3) with a binding time of 8±2 s </emph>(Fig. 5B,C). FRAP data of antagonist-bound AR and the AR R585K mutant did not fit well to models that included these immobilizations. Thus, only the agonist-bound wild-type AR displays stable interactions with chromatin. In contrast, for all agonist- and antagonist-bound wild-type AR and AR R585K a substantial fraction (of ∼30%) with sub-second binding times (0.5–0.8 s) was found (Fig. 5C).
In summary, FCS confirms that antagonist-bound ARs show a faster diffusion rate in comparison with agonist-bound (wild-type and mutant) ARs. FRAP showed that only a substantially long immobilization was found for agonist bound wild-type AR, whereas an additional, briefly immobilized, fraction is found for all agonist- or antagonist-bound wild-type AR or AR R585K mutant (Table 1).
The view on how proteins find their way through the nucleus to identify and bind their target sites in the vast amount of DNA has been subject to intensive discussion (e.g. Erdel et al., 2011; Halford and Marko, 2004; Mueller et al., 2010; Mueller et al., 2008; Phair et al., 2004; Sprague and McNally, 2005; Sprague et al., 2004; van Royen et al., 2011). In a simple model, proteins diffuse freely though the nucleoplasm and find their targets by random collision, resulting in binding to specific and nonspecific binding sites (Gorski et al., 2006; Halford and Marko, 2004; Hoogstraten et al., 2008; Houtsmuller et al., 1999; McNally et al., 2000; Mueller et al., 2008). However, more sophisticated models based on in vitro experiments using isolated DNA suggest that proteins slide along the DNA strand (1D diffusion) or over the chromatin surface enabling proteins to bypass obstacles (2D diffusion) (Blainey et al., 2009; Gorman et al., 2007; Kampmann, 2004).
Although live-cell imaging methods, such as FRAP and FCS, have revealed high mobility of nuclear proteins and the very dynamic nature of their interactions with chromatin, a large variation in quantitative estimates of diffusion rates and DNA-binding kinetics still exists (van Royen et al., 2011; and references therein). This variation is mostly caused by differences in the choice of analytical methods and by different assessment of experimental parameters regarding microscopic properties, such as the laser intensity distribution or photophysical properties of fluorescent labels, such as blinking or photobleaching (e.g. Houtsmuller, 2005; Mueller et al., 2012). In addition, differences in the shape and size of the cell nucleus are often not taken into account, even though these might have considerable influence on FRAP recovery curves (as discussed in Houtsmuller, 2005; Mueller et al., 2010; van Royen et al., 2011). Here, we argue that errors caused by such methodological and analytical limitations can be largely eliminated by applying, and using, the strengths of several complementary approaches.
Therefore, we combined FRAP, FCS and SMM to provide a consistent quantitative model of the mobility and DNA interactions of the ligand-dependent transcription factor AR. We obtained mobility and interaction data at different timescales, from milliseconds in FCS, up to tens of milliseconds in single-molecule tracking assays and hundreds of milliseconds to seconds in FRAP. From these data, we determined diffusion rates using FCS and SMM, which gave consistent results (Table 1). In addition, we determined the fraction of immobile molecules using SMM and FRAP, which also yielded similar results. Further analysis of the FRAP data allowed us to dissect the fraction of immobile molecules into two fractions with distinct kinetics (Table 1).
The results are consistent with a model of activated ARs diffusing freely in the nucleoplasm with frequent, stochastically driven, short binding events, probably representing immobilizations by nonspecific DNA interactions in the sub-second range, as well as less frequent but more stable interactions, typically in the order of tens of seconds (Table 1; Fig. 6). The latter immobilization events most likely represent associations of transcriptionally active ARs with their cognate recognition sequence in promoter and/or enhancer regions of androgen-regulated genes, as these events are absent in an AR mutant (R585K) that is unable to identify its cognate recognition sequences in promoter and/or enhancer regions of androgen-regulated genes (Shaffer et al., 2004). The increased binding stability of wild-type AR might result from the association with stabilizing (co-regulating) factors during formation of transcription complexes or from changes in chromatin structure due to remodeling. This explains the absence of this fraction in the presence of the antagonist OHF, which does not result in binding of these factors (Fig. 6).
As well as the long-binding fraction (∼25%), we observed short immobilization events in not only agonist-bound wild-type AR but also in the R585K mutant and antagonist-bound ARs (Table 1). These short immobilizations might reflect a general nonspecific DNA-binding capacity, which is independent of sequence and agonist binding. This behavior could reflect a general mechanism by which nuclear proteins find their target sequences: free 3D diffusion through the nucleus combined with frequent random collisions with chromatin, leading to short interactions. It was previously hypothesized that nuclear proteins repeatedly bind in the same region, interspersed with only short 3D diffusion events to enhance their chances of finding their target sites (sometimes referred to as ‘hopping’ or ‘jumping’), (Gorski et al., 2006; Halford and Marko, 2004; Loverdo et al., 2009; and as previously discussed in van Royen et al., 2011). Although hopping and jumping, indeed, are often described as distinct models, molecules display essentially the same behavior in a model of 3D diffusion with random collisions, which is supported by our data. Importantly, because our data fitted well to a model in which the presented diffusion and binding parameters explain the data, it does not suggest the occurrence of 1D diffusion along the DNA helix (‘sliding’). However, the existence of very short 1D sliding behavior over a small distance cannot be ruled out as this might be undetectable by any of the technologies used. Furthermore, AR dimerization does, in theory, allow binding to a distant site that is brought into proximity by the looping of DNA before dissociating from the initial site (‘facilitated diffusion’), but these slowly moving molecules have not been detected by SMM in the present study. Moreover, the crowded nature of regulatory proteins that are bound to DNA will limit the ability of these proteins to scan the DNA for target sites, which by itself makes it an unlikely scenario.
Very recently, it has been reported that (by using a live cell study combining data from SMM, FCS and FRAP experiments) the binding of the transcription factor p53 to DNA showed a continuum in chromatin residence times, which included both sequence-specific and nonspecific binding (Mazza et al., 2012). However, in the present study, two distinct populations of residence times were found for AR, and only the short immobilization was found for the AR mutant R585K (Table 1). Although this seems to present a discrepancy, both studies suggest a very similar model of sequence-specific and nonspecific DNA binding. This is a similar model to that suggested for the lac repressor in E. coli (Elf et al., 2007). Numerical differences between p53 and AR could be explained by differences in the binding affinity or formation of complexes after binding to DNA.
Interestingly, SMM and FCS consistently indicated a substantial agonist-induced decrease of ∼2-fold in the diffusion rate of the freely mobile pool of wild-type and R585K mutant ARs (Table 1). This surprising decrease in diffusion rate could be due the formation of large hormone-induced AR complexes, which diffuse with lower diffusion coefficients (Fig. 6). Because the diffusion coefficient is linearly related to the molecule radius and, therefore, to the cube root of the molecular mass, the observed decrease of a factor of ∼1.6 would require a 4-fold increase in the molecular mass of such complexes. It has been shown that ARs dimerize upon agonist binding (van Royen et al., 2012). It is very conceivable that these activated AR dimers associate with a number of co-regulator proteins forming a complex with a molecular mass 4-fold higher than that of the AR monomer, thereby decreasing the diffusion coefficient of this complex 1.6-fold as compared with the AR monomer (Fig. 6). In addition, the data does not exclude the contribution of very brief (≤1 ms) binding events (on top of the previous described short interactions in the 0.5–0.8 ms range) to the diffusion rate decrease. These very short immobilizing interactions, representing an additional scanning behaviour, would be enhanced by agonist binding to result in a lower effective diffusion rate.
In conclusion, combining SMM, FCS and FRAP appears to be a powerful approach to obtain a detailed quantitative description of the dynamic behavior of nuclear proteins in living cells. The results presented here point to a model of free diffusion, where proteins randomly collide with DNA, and two classes of DNA-binding events: relatively long DNA binding, most likely in transcription complexes, and short interactions that might represent search mechanisms.
MATERIALS AND METHODS
Expression constructs and cell culture
The constructs expressing N-terminally YFP-tagged wild-type and mutant AR were generated as described previously (van Royen et al., 2012). In all constructs expressing the AR fusion proteins, the AR was separated from the fluorescent tag by a flexible (GlyAla)6 spacer (Farla et al., 2004). All new constructs were verified by sequencing and the size of expressed ARs was verified by using western blotting. The YFP–H2B expression plasmid was a generous gift from Hiroshi Kimura (Kyoto University, Kyoto, Japan).
Cell lines stably expressing YFP-labeled proteins at very low levels were generated as described previously (Van Royen et al., 2009). Stably expressing cell lines were maintained in α-MEM (Cambrex) supplemented with 5% fetal bovine serum (FBS) (HyClone), 2 mM L-glutamine, 100 µg/ml penicillin, 100 µg/ml streptomycin and 600 µg/ml G418 (active concentration).
Cultured Hep3B cells were studied by SMM at 37°C, using a previously described wide-field fluorescence microscopy setup (Harms et al., 2001; Lommerse et al., 2004; Schmidt et al., 1996). The microscope (Axiovert 100TV, Zeiss) was equipped with a 100× oil-immersion objective (NA = 1.4, Zeiss). A region of interest was set to 50×50 pixels at a pixel size of 220 nm. Excitation was performed using a 514-nm argon laser line (Spectra Physics, Mountain View, CA) combined with an acousto-optic tunable filter (AOTF), illuminating the region of interest for 3 ms using a power of ∼2 kW/cm2. The time lag between subsequent illuminations was either 6.25 or 25 ms and the camera frame rate was synchronized with the AOTF. Fluorescent light was filtered using a combination of filters [DCLP530, HQ570/80 (Chroma Technology, Brattleboro, VT) and OG530-3 (Schott, Mainz, Germany)] and detected by a liquid-nitrogen-cooled slow-scan CCD camera (Princeton Instruments, Trenton, NJ).
At least seven cells were studied by taking ten sequences of 120 images in each individual experiment. Positional data from three experiments were pooled for the PICS analysis of the mobility patterns. This way, positional information derived from ∼50,000–100,000 fluorescence intensity spots attributed to individual molecules was analyzed as a whole.
Three different time lags were used (6.25, 12.5, and 25 ms). For each time lag, data were obtained for different step sizes in the image sequence (e.g. using a time lag of 6.25 ms, data was obtained for 6.25, 12.5, 18.75, 25, 31.25, 37.5 and 43.75 ms). The generated series of data points ranged from 6.25 to 43.75 ms and for some time-points more than one data point was generated (e.g. the 12.5 ms data point was generated twice, using the 6.25- and 12.5-ms time lag).
Analysis of YFP–AR mobility patterns
Analysis of individual molecules was performed as described previously (Lommerse et al., 2004; Schütz et al., 1997). The signals from fluorescence intensity spots attributed to individual molecules were fitted to a 2D Gaussian surface. This permitted the localization of the molecule with a positional accuracy that is determined by the quotient of the full-width-at-half-maximum of the Gaussian fit and square root of the number of photons detected (Bobroff, 1986).
The 2D mobility patterns were analyzed using the PICS analysis method (Semrau and Schmidt, 2007). Briefly, the cross-correlation between single-molecule positions at two subsequent time-points was calculated (see Fig. 1B). To correct for the effect of random proximity, the contribution from uncorrelated molecules in close proximity (which is described by a linear function) was subtracted (Fig. 1C). This results in the cumulative distribution function Pcum(l, Δt) for length l of diffusion steps during time lag Δt (Fig. 1D).
This equation describes the second model, characterized by mean squared displacements MSD1 and MSD2, and relative fraction α and (1−α), respectively (Schütz et al., 1997). Subsequently, MSD1 and MSD2 are plotted against Δt. These plots (Fig. 3C–F) reveal the diffusional behavior of individual fractions.
Fluorescence recovery after photobleaching
The parameters of the top 7–10 best fitting (least square fitting) were averaged to represent the properties of the molecules in the experimental data (±2×s.e.m.) and to generate the fit curves in Fig. 5A and supplementary material Fig. S4 (for more details see Farla et al., 2004; Van Royen et al., 2009).
Fluorescence correlation spectroscopy
In order to determine the effect of the ligand and the mutation, the fraction sizes determined using SMM in three individual experiments (shown in Fig. 3B) were analyzed using two-way ANOVA. Average MSDs determined using SMM (shown in Fig. 3C–F) were analyzed using three-way ANOVA so the effect of the ligand, the mutation and the time-point was determined. Individual FCS and FRAP curves (of which averages are shown in Figs 4 and 5, respectively) were analyzed using three-way ANOVA in order to determine the effect of the ligand, the mutation and the time-point. In all analyses, interactions between variables were also determined. Statistical significance was accepted at P<0.05.
M.E.v.R., T.S., A.B.H. and M.J.M.S. designed the research. M.E.v.R. and M.J.M.S. performed the experiments. M.E.v.R., W.A.v.C., B.G., T.S., A.B.H. and M.J.M.S. contributed analysis tools and/or analyzed data. M.E.v.R. and M.J.M.S. wrote the paper.
This work was supported by grants from the Netherlands Organization for Scientific Research (NWO-STW) and the SmartMix Program of The Netherlands Ministry of Economic Affairs and the Ministry of Education, Culture and Science.
The authors declare no competing interests.