help button home button Endocrine Society Molecular Endocrinology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Molecular Endocrinology, doi:10.1210/me.2006-0091
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
20/11/2641    most recent
Author Manuscript (PDF)
Right arrow Purchase Article
Right arrow View Shopping Cart
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow NURSA Molecule Pages Link
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Voss, T. C.
Right arrow Articles by Hager, G. L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Voss, T. C.
Right arrow Articles by Hager, G. L.
Molecular Endocrinology 20 (11): 2641-2655
Copyright © 2006 by The Endocrine Society

Single-Cell Analysis of Glucocorticoid Receptor Action Reveals that Stochastic Post-Chromatin Association Mechanisms Regulate Ligand-Specific Transcription

Ty C. Voss, Sam John and Gordon L. Hager

Laboratory of Receptor Biology and Gene Expression, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892-5055

Address all correspondence and requests for reprints to: Gordon L. Hager, Laboratory of Receptor Biology and Gene Expression; Building 41, B602, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892-5055. E-mail: hagerg{at}exchange.nih.gov.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 REFERENCES
 
The glucocorticoid receptor (GR) dynamically interacts with response elements in the mouse mammary tumor virus (MMTV) promoter to regulate steroid-dependent transcription. In a clonal mammary carcinoma cell line containing a tandem array of MMTV promoter-reporter gene cassettes integrated at a single genomic locus, direct binding of a green fluorescent protein (GFP)-GR fusion protein to the MMTV regulatory elements can be observed in living cells. After ligand treatment, MMTV-dependent transcription in individual cells was detected by RNA fluorescence in situ hybridization (FISH). High-resolution fluorescence images were acquired from large numbers of randomly selected cells. Images were analyzed with a novel automated computer algorithm, measuring the RNA FISH signal and the relative GFP-GR fluorescence intensity at the MMTV array for each cell. Although dexamethasone increased the mean RNA FISH signal approximately 10-fold, RU486 produced only about a 2-fold induction, as expected for this mixed antagonist. For all treatment conditions, the relative GFP-GR fluorescence at the array for the averaged cells paralleled the RNA FISH measurements, suggesting that image analysis accurately detected an increase in steady-state GR association with the MMTV array that was responsible for the increase in transcriptional activity. The antagonist-dependent decreases in GR association with the MMTV promoter were confirmed by chromatin immunoprecipitation experiments, supporting the image analysis results. A pronounced cell-to-cell variability was observed in RNA FISH signal and GR-MMTV association within treatment groups. We observed a nonlinear relationship between GR-MMTV association and RNA FISH in individual cells, indicating that differences in GR-MMTV interaction account for some, but not all, of the transcriptional heterogeneity between individual cells. In selected cell subpopulations with equal levels of GR-MMTV association, there was a decrease in RNA FISH signal with RU486 treatment compared with dexamethasone treatment.

These results indicate that stochastic events occurring after GR-promoter association, such as the actions of chromatin remodeling complexes or other cofactors, change in a ligand-dependent manner and regulate heterogeneous transcription in individual cells.


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 REFERENCES
 
THE GLUCOCORTICOID RECEPTOR (GR), like other members of the steroid receptor superfamily, acts as a ligand-regulated transcription factor to modulate diverse genetic networks (1). In the absence of hormone, GR resides in the cytoplasm where it associates with chaperone complexes. Once bound to glucocorticoid, GR translocates into the nucleus and interacts with specific DNA response elements (glucocorticoid response elements, GREs) located in the promoters or enhancers of target genes. This interaction leads to the recruitment of coregulatory complexes that covalently modify chromatin and other components of the transcriptional apparatus (2). GR also recruits ATP-dependent chromatin remodeling complexes, altering locally repressive nucleosome-DNA conformations (3). These changes in local chromatin state and the GR-dependent recruitment of the transcriptional machinery play a central role in the regulation of gene expression by glucocorticoids.

Glucocorticoid agonists and antagonists, such as dexamethasone (Dex) and RU486 (RU), respectively, differentially regulate GR transcriptional activity by modulating the steady-state protein complexes associated with the target promoter. Coactivators preferentially interact with the Dex-bound GR, whereas the RU-bound GR associates more strongly with corepressors (4, 5). This ligand-dependent coregulator exchange is not absolute because corepressors can still influence the activity of the agonist-bound receptor (6, 7). In addition, recruitment of ATP-dependent chromatin remodeling complexes by receptors is also differentially affected by particular ligands (Klokk, T. I., P. Kurys, C. Elbi, A. K. Nagaich, A. Hendarwanto, T. Slagsvold, C.-Y. Chang, G. L. Hager, and F. Saatcioglu, manuscript submitted; and Ref. 9). Quantitative in vitro biochemistry demonstrates that GR binds to purified DNA GREs similarly in the presence of Dex or RU (10). However, in vivo footprinting experiments indicate that RU-bound GR interacts with chromatinized GREs much less efficiently compared with the Dex-bound GR (11, 12). Because GR binding to GREs is required for recruitment of coregulators and remodelers to target promoters, it remains an important question how these two events contribute to the regulation of ligand-specific in vivo transcription.

Essentially all experimental information regarding steroid-regulated transcriptional mechanisms has been derived from biochemical methods that average the behavior of large cell populations. Individual cells are often assumed to exhibit this averaged response, but early microscopy and cell-sorting experiments indicated that GR-dependent gene activation was highly heterogeneous, even among clonally derived cells (13). A rapidly emerging concept is that gene expression is often widely variable between individual isogenic cells due to the stochastic properties inherent in transcription (14). These stochastic mechanisms may be very significant in disease states, such as cancer, where heterogeneity could contribute to pathogenesis (15). Despite the potential importance of these mechanisms, it is unknown how cell-to-cell variability in specific mechanistic events, such as receptor-chromatin association and recruitment of coregulatory factors, impact steroid-dependent gene expression at the single-cell level.

We have recently characterized several clonal mouse mammary carcinoma cell lines, which contain at a single genomic locus an integrated 200-copy tandem array of the mouse mammary tumor virus (MMTV) long terminal repeat (LTR) driving expression of a reporter gene. The development of this cell system allows the interaction of GR with a target promoter to be observed in individual cells (9, 16, 17, 18). A functional green fluorescent protein (GFP)-GR fusion protein can associate with the six GREs in each LTR, making the MMTV array visible as a bright fluorescent domain in the nucleus when viewed by fluorescence microscopy. Live cell fluorescence photobleaching experiments reveal that GFP-GR exchanges very rapidly between the array and the surrounding nucleoplasm (17). Importantly, MMTV transcriptional regulation is very similar in low-copy cells and those containing the 200-copy array (19, 20, 21, 22). Subsequent biochemical experiments indicate that GR also interacts transiently with the LTR chromatin template in vitro (23, 24). Moreover, these transient GR interactions occur at genes other than MMTV (25). In light of these observations, we have proposed that the steady-state interactions measured biochemically for GR likely result from the balance of fast association and dissociation mechanisms that operate in individual living cells (23, 24, 26, 27). These properties make the MMTV chromatin array an excellent system for defining the connections between steroid receptor-chromatin interactions and transcriptional output at the single-cell level.

In the studies reported here, the steady-state interaction of GFP-GR with the MMTV chromatin array and transcriptional activity were characterized in individual cells using high-resolution fluorescence microscopy. The images of large numbers of cells were analyzed by a novel computer algorithm to develop quantitative models of the cell population response with single-cell resolution. The results of the averaged imaging data from agonist- and antagonist-treated cells agree with parallel biochemical assay data, validating the measurements of GR-chromatin association and transcriptional activity in the individual cells. Statistical analyses indicate that cell-to-cell variability in GR-chromatin association accounted for some but not all of the variability in transcriptional activity. These results suggest that one or more stochastic events, in addition to GR-chromatin association, are necessary for high-level transcriptional activation. Interestingly, agonists and antagonists differentially regulate both GR-chromatin association and the secondary stochastic event(s) through mechanisms that can be at least partially independent of one another. These findings indicate that stochastic molecular events subsequent to initial GR chromatin association are critical determinants in the extent of transcriptional activity induced by hormone within individual members of the cell population.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
DISCUSSION
 MATERIALS AND METHODS
 REFERENCES
 
Automated Measurement of the MMTV Array
The overlay images in Fig. 1Go show GFP-GR and RNA FISH fluorescence signals from 3617 cells treated for 30 min with vehicle alone, 100 nM Dex, or 100 nM RU (Fig. 1Go, top panels). The MMTV-LTR-regulated reporter gene transcripts temporarily accumulated at the array and appeared as a region of bright fluorescence in the nuclei. As previously reported, the hormone-bound GFP-GR fusion protein transiently binds to regulatory elements in the MMTV-LTR, and the steady-state concentration of GFP-GR increases at the MMTV array (17). This increased local concentration made GFP-GR visible on the array as a bright region of fluorescence. These representative images suggest that Dex caused more accumulation of GFP-GR on the MMTV array and increased transcriptional output compared with treatment with RU.


Figure 1
View larger version (43K):
[in this window]
[in a new window]
  		Fig. 1. Automated Identification of the MMTV Array

Top panels, Overlay images of MMTV RNA FISH and GFP-GR signals are shown for 3617 cells treated with (A) no hormone (or vehicle, Veh), (B) Dex, or (C) RU. Middle panels, the computer defined nROIs and aROIs (yellow outlines) are displayed in relation to each individual fluorophore image. The fluorescence intensities in subregions of the GFP-GR images (red squares, 5 µm per side) are detailed in the surface profile plots in the bottom panels. Differences in GFP-GR concentration at the array relative to the surrounding nucleoplasm are represented as changes in peak height and color of the surface profile plots.

 
To develop a quantitative model of GR-chromatin interactions and transcriptional output from individual cells in the ligand-treated populations, the images were analyzed with a custom-designed computer algorithm that automatically measured the RNA FISH signal and levels of GFP-GR concentrated on the array (see Materials and Methods). This algorithm identifies the region of interest for each nucleus (nROI) using the 4',6'-diamidino-2-phenylindole (DAPI) chromatin stain fluorescence signal (image not shown). The boundaries of the gene array ROI (aROI) are then automatically identified within each nucleus using the FISH image data. Taking advantage of robust signal-to-noise ratios in both the DAPI and the FISH images, the algorithm accurately identifies nROI and aROI in cells that were treated with vehicle, Dex, or RU (Fig. 1Go, middle panels). Importantly, the algorithm is sensitive enough to define the aROI in most vehicle-treated cells, using the FISH signal produced by low-level basal transcription (Fig. 1AGo). Although these images indicate an agonist-dependent increase in RNA FISH levels and concentration of GFP-GR on the array, additional analytical subroutines used the nROI and aROI to quantitatively confirm this preliminary observation (see Materials and Methods). The algorithm measures the relative amount of MMTV-driven transcription in each nucleus as the product of the aROI area and fluorescence intensity in the FISH image. As highlighted by the fluorescence intensity profile plots (Fig. 1Go, bottom panels), the relative concentration of GFP-GR increases at the array in the presence of agonist but to a lesser degree in the presence of antagonist. Because the subnuclear region containing the concentrated GFP-GR is spatially associated with the subnuclear position of the FISH signal (Fig. 1Go, top panels), the aROI can also be used to automatically measure the amount of GFP-GR loading on the array. To quantify the relative degree of array loading for each cell, the algorithm determines the ratio of the maximal peak height of GFP-GR fluorescence intensity in the aROI relative to the mean fluorescence intensity of GFP-GR in the nROI.

Ligand-Specific Behavior of the MMTV Array in Cell Populations
Before cell-population imaging studies, 3617 cells were treated for 30 min with vehicle, Dex, or RU and were then processed for RNA FISH. Multiple images were collected for each hormone treatment group and subjected to automated analysis as described in Materials and Methods. These regions were selected for imaging using only the DAPI chromatin fluorescence signal, minimizing user bias. The resulting morphometric data sets contained information for 230 vehicle-treated nuclei, 223 Dex-treated nuclei, and 200 RU-treated nuclei. The algorithm detected aROIs in 61, 99, and 65% of each respective hormone treatment group. The nuclei with no measurable aROIs could have contained FISH signals slightly below the detection limit of the image-analysis algorithm or, alternatively, could have no MMTV-driven transcriptional activity. Because it was difficult to ascertain which one of these conditions was true, only those cells with detectable aROIs were included in the subsequent statistical comparisons. This avoids corrupting the morphometric data sets with null values that may not have been accurate.

When all data were averaged for the sampled population, treatment with Dex increased the MMTV RNA FISH signal 10-fold, but RU was dramatically less efficient in activating transcription over the 30-min time frame (Fig. 2AGo). However, RU did stimulate RNA FISH signals 2.5-fold vs. the vehicle control, confirming the mixed antagonist property of this ligand. The analysis of GFP-GR loading ratio on the array parallels the pattern observed for RNA FISH. GFP-GR was concentrated 4.2-fold at the array relative to the surrounding nucleoplasm in the Dex-treated cell population but only 2.4-fold in RU-treated cell population (Fig. 2BGo).


Figure 2
View larger version (13K):
[in this window]
[in a new window]
  		Fig. 2. Ligand-Specific Regulation of MMTV Transcription and GFP-GR Array Loading Measured in Large Cell Populations

The 3617 cells were treated for 30 min with vehicle (Veh), Dex, or RU. After fixation and the RNA FISH procedure, z-stack images were collected, digitally deconvolved, maximally projected, and analyzed by the automated algorithm. Treatment-dependent changes are shown for (A) relative MMTV array RNA FISH signal and (B) GFP-GR array loading. The data shown for each treatment are the means of measurements from more than 125 individual cells, which contained arrays that were automatically identified using the FISH signal. Error bars denote SEM. C, Ligand-dependent changes in GR loading were biochemically confirmed by ChIP using anti-GR antibody and quantitative PCR detection of the MMTV LTR.

 
Chromatin immunoprecipitation (ChIP) results showed a similar pattern of ligand-specific biochemical interactions between GFP-GR and the MMTV array (Fig. 2CGo). ChIP with GR antibody captured the greatest amount of MMTV LTR chromatin after agonist treatment, whereas ChIP from antagonist-treated cells captured less MMTV LTR than from Dex-treated cells but captured more MMTV LTR compared with control-treated cells. In additional GR antibody ChIP experiments using primers to amplify regions flanking the MMTV LTR, very low amounts of flanking chromatin were captured regardless of hormone treatment (data not shown), indicating that the hormone-dependent GR-chromatin interactions are specifically mediated by the MMTV LTR. These data provide the first quantitative connection between relative steady-state concentrations of GFP-GR at GREs observed at both the cellular and biochemical levels.

Although the imaging data for GR-chromatin association agreed very well with the biochemical interaction data (Fig. 2Go, B and C), there was still the formal possibility that slight optical artifacts could adversely affect the imaging measurements. Variable spatial alignment of the GFP-GR and RNA FISH images due to chromatic aberration could potentially cause the aROI not to overlap with the region where the GFP-GR was concentrated on the array. Because smaller aROIs are more significantly shifted by small misalignment of the GFP-GR and FISH images, measurements of GFP-GR loading ratios that depend on smaller aROIs are more prone to chromatic aberration errors compared with loading ratio measurements that depend on larger aROIs. To allay this concern, we devised an experiment to determine whether chromatic aberration significantly affects the measurement of GFP-GR loading ratios. The image analysis algorithm was reprogrammed to place a square ROI of fixed size (9 µm2) over the GFP-GR array based on the location of RNA FISH signal (Fig. 3AGo). The size of this square ROI is approximately 2-fold larger than the average size of the irregularly shaped aROI from Dex-treated cells and 8.5-fold larger than the average aROI from the RU-treated cells. Automated analysis demonstrates that GFP-GR loading ratios are statistically identical for the irregularly shaped aROI and the larger square array ROI (Fig. 3BGo). In contrast, a significant decrease in the GFP-GR array loading ratios was observed for both Dex- and RU-treated cells when the large square ROI was placed at random positions within the nucleus (Fig. 3BGo). Taken together, these results indicate that chromatic aberration does not significantly affect the measurement GFP-GR array loading under these conditions.


Figure 3
View larger version (19K):
[in this window]
[in a new window]
  		Fig. 3. The Size of the RNA FISH Signal ROI Area Does Not Affect Automated Analysis of GFP-GR Array Loading Ratios

A, Representative GFP-GR image showing a fixed-size square ROI that was placed by the automated analysis algorithm over the array (yellow square) or at a random location in the nucleoplasm (red square). B, Comparison of GFP-GR array loading ratios for the same hormone-treated cell populations calculated using the irregularly shaped aROI, the square ROI placed over the array, or the square ROI placed randomly in the nucleoplasm. Error bars indicate SEM. Veh, Vehicle.

 
Heterogeneity in the Response of the Cell Population
Initial subjective comparison suggested that there was some variability of GR-chromatin association and transcriptional activity between individual cells of the population. To delineate ligand-specific responses at the cellular scale, automated image analysis was used to precisely quantify the heterogeneity in the populations after treatment with vehicle, Dex, or RU. The morphometric data from each cell was sorted into bins according to equal ranges of RNA FISH signal, or GFP-GR array loading (Fig. 4Go). In general, the percentage of cells in each bin of the resulting histograms showed that the distribution of cellular response to hormone was strikingly diverse.


Figure 4
View larger version (20K):
[in this window]
[in a new window]
  		Fig. 4. Ligands Cause Heterogeneous Responses in Individual Cells of the Population

Morphometric data from individual cells treated with vehicle (veh), Dex, or RU were sorted into bins according to value ranges of (A) relative RNA FISH signal, or (B) GFP-GR array loading. The percentages of the cell population that exhibit the indicated morphometric value ranges are shown in the histograms.

 
Although 95% of vehicle-treated cells sorted into the lowest RNA FISH signal bin, Dex-treated cells were shifted out of this low-level range into many different degrees of transcriptional activation with no more than 25% of cells sorted into any one RNA FISH range (Fig. 4AGo). The histogram analysis also indicated that the FISH signal varied by more than 8-fold within the agonist-treated population. After RU486 treatment, individual cells were also activated by varying amounts, which produced greater than a 5-fold range in FISH signals. In the RU-treated population, the number of cells in the basal level bin was much greater compared with the Dex-treated population. This distribution is consistent with the slight transcriptional activation observed in the averaged RU486 cell population (compare Figs. 4AGo and 2AGo).

The cells within each hormone-treated population also exhibited a broad distribution of GFP-GR loading on the array (Fig. 4BGo). For the vehicle-treated population, 75% of the cells had GR-loading values between 1 and 2. The Dex-treated population contained some cells with GR loading ratio values as high as 8-fold concentrated on the array relative to the surrounding nucleoplasm, but the bulk of cells were distributed over a wide range of loading values. Compared with the histogram analysis of Dex-treated cells, RU treatment caused a broad but less dramatic increase in loading of GFP-GR on the array. In sum, these results show that the consistent hormone-dependent behavior of the cell population is the average of many different graded responses produced by individual cells.

The cell-to-cell variation in the automated measurements warranted a careful comparison with the imaging data. The automated algorithm annotated the measurements for each cell with the source image file name, allowing a manual comparison. Overall, the values representing the RNA FISH signals and GR array loading ratios agreed very well with the heterogeneous morphology of individual cells. The images shown in Fig. 5Go exemplify the heterogeneity in the response of Dex-treated cells and the corresponding automated measurements. As predicted by the histogram analysis, the RNA FISH signal in the source images varied greatly between cells (Fig. 5Go, top panels). The aROIs and FISH signal values indicated that the algorithm correctly measured this heterogeneous behavior. The manual comparison of the images with the morphometric data suggested that the variation in the response of the cells was biological in origin and not merely the result of random error in the measurements.


Figure 5
View larger version (17K):
[in this window]
[in a new window]
  		Fig. 5. Cell-to-Cell Variability in the Transcriptional Response is Accurately Measured by the Automated Image Analysis Algorithm

Images of multiple Dex-treated cell nuclei (left, center, and right panels) exemplify the heterogeneity of MMTV array RNA FISH signal (top panels) and GFP-GR association with the array (bottom panels) in the cell population. The yellow outlines (top panels) indicate the boundaries of the aROIs defined by the automated image analysis algorithm. The arrows (bottom panels) highlight the corresponding positions of the MMTV array in the GFP-GR images. The RNA FISH and GR array loading ratio values measured by the automated image analysis algorithm are shown for each cell nucleus.

 
The GFP-GR loading on the arrays was consistently difficult to subjectively identify in cells where the algorithm measured low GR loading ratios (Fig. 5Go, left and right panels). To precisely determine the limit of detection for the GR array loading measurement, GR loading ratios were calculated for random regions of the nucleoplasm that did not overlap the RNA FISH signal. This mean random GR loading ratio was 2.5 for the Dex-treated cell population (Fig. 3BGo). Consistent with the example images (Fig. 5Go), this indicated that, when GFP-GR loading ratio measurements approach the 2.5 value, the GFP-GR array will become subjectively indistinguishable from the surrounding nucleoplasm. As expected, higher GR-array loading ratios were measured for arrays with a higher local concentration of GFP-GR, making the array clearly distinguishable from the surrounding nucleoplasm (Fig. 5Go, middle panels). Interestingly, some of the cells with relatively weak GFP-GR array loading had relatively high RNA FISH signals (Fig. 5Go, left panels). The inverse behavior was also observed; strong GR array loading can occur in cells with weak transcriptional output (Fig. 5Go, middle panels). These findings suggested that, in addition to the steady-state association of GR with the array, other events that are important in the final transcriptional output are highly heterogeneous between cells. Therefore, a more extensive statistical analysis was applied to the data sets.

Cellular Links between Transcriptional Activation and Chromatin Association
Because both FISH and GFP-GR array association were measured in many individual cells, statistical methods could be applied to detect significant relationships between these parameters at the cellular scale. Plotting the GR array loading ratio vs. the FISH signal for each measured cell reveals that, as Dex treatment increases the level of GR array association, there is a highly variable increase in the transcriptional output from individual cells (Fig. 6AGo). In these Dex-treated cells, low GR loading ratios (<3) clearly appear to correlate with lower FISH signals, but statistical analysis did not identify a significant linear correlation for the entire Dex-treated population. This suggests that there is a nonlinear connection between the array loading ratio and FISH signal in individual cells. Because non-linear systems can often be modeled by linear correlations that operate in specific subclasses of the response, subpopulations of the Dex-treated cells were systematically analyzed. When 50% of the population was considered, a significant positive linear correlation with FISH signal was found for cells having low to moderate GR array association (GR loading ratios ~1.5 to 4). The best-fit line for this correlated subregion is shown in Fig. 6AGo. The system is not positively correlated for Dex-treated cells containing higher GR loading ratios (>4), suggesting that other cofactors can become limiting under these conditions. These results indicate that some level of GR-array association is required for transcription, but GR association alone is not sufficient for full activity in all of the individual cells.


Figure 6
View larger version (39K):
[in this window]
[in a new window]
  		Fig. 6. Comparison of Ligand-Specific GR-Chromatin Association and Transcription in Individual Cells

The GFP-GR array loading ratio and RNA FISH signal were plotted for (A) Dex- and (B) RU-treated cells. The values of vehicle (Veh)-treated cells (square symbols) were included for reference. Each symbol represents the measurements from a single cell. The best-fit lines indicate the Dex- or RU-treated subpopulations that had a significant linear correlation between GR-loading ratio and FISH signal. C, The locally weighted (loess) average lines show the nonlinear relationship between the measurements in Dex- (top line) and RU-treated (bottom line) cells. The inset box (C) indicates the subpopulations of cells that were selected based on GFP-GR array loading ratio values. The mean values of (D) GFP-GR array loading ratio and (E) RNA FISH signal are shown for the selected subpopulations. Error bars denote SEM.

 
The individual antagonist-treated cells also showed a similar relationship between GR array loading ratio and FISH signal, although the increases in the ranges of these parameters were reduced compared with the Dex-treated cells (Fig. 6BGo). In RU-treated subpopulations with low to moderate GR array loading ratios, there was a significant positive linear correlation with RNA FISH signal in individual cells. Therefore, even in the presence of a mixed antagonist, increased GR response element association leads to higher levels of transcriptional activity. However, there was still a broad range of transcription levels associated with the higher levels of GR association, again suggesting that other heterogeneous cofactors must also regulate transcription in the individual cells.

The locally weighted scatter plot smoothing (loess) average line was calculated for the agonist- and antagonist-treated cells, allowing more detailed comparison of the nonlinear responses under different hormone conditions (Fig. 6CGo). The slopes of the loess average lines showed positive relationship between low to moderate GR array loading ratios and RNA FISH signals in both Dex- and RU-treated cells, consistent with the finding of significant positive linear correlations in these subpopulations. At higher GR loading ratios, the increase in RNA FISH signal plateaued, suggesting that the parameters were not correlated over these ranges. The difference in the two loess average lines suggested that, in cells with the same GR array loading ratio, the RNA FISH signal was reduced in the antagonist-treated cells compared with the agonist-treated cells. To test this observation statistically, subpopulations of cells with array loading ratios between 3 and 4 were selected for further analysis (Fig. 6CGo, inset rectangle shows selected cells). In this subpopulation, the mean GFP-GR array loading ratio was not significantly different between Dex- and RU-treated cells (Fig. 6DGo). In contrast, the mean RNA FISH signal in the RU-treated subpopulation was greater than 3-fold reduced compared with the Dex-treated subpopulation. This analysis clearly shows that cell subpopulations with identical levels of GR-MMTV array association exhibit ligand-specific differences in transcriptional output.

It was uncertain whether the maximal z-stack projection, the process of selecting the brightest pixels from individual optical planes of the three-dimensional imaging data for analysis, would accurately integrate the transcriptional signal at the single-cell level. To compare another method of integrating the FISH signal from the three-dimensional imaging data, a mean z-stack projection was performed that averaged the intensity values from all of the individual optical sections, and the automated analysis method was used to analyze the resulting two-dimensional mean projection images. For comparison to previous single-cell measurements, data from the mean projection images was subjected to scatter plot analysis (Fig. 7AGo). Although there is an overall reduction in the absolute FISH signal values from the mean projection images, we observe a pattern of cellular behavior that is essentially identical to that measured from the maximal projection images (compare Figs. 7AGo and 6CGo). Importantly, the loess average line shows the nonlinear relationship between GR array loading and RNA FISH signal in individual cells (Fig. 8AGo). Like the previous results determined with the maximal projection images, when Dex- and RU-treated cells with the same level of GFP-GR array loading are compared, there is an RU-dependent decrease in the RNA FISH signal (Fig. 7Go, B and C). Together, these data indicate that analysis of mean projection and maximal projection images support the same conclusions.


Figure 7
View larger version (20K):
[in this window]
[in a new window]
  		Fig. 7. Mean Projection of Three-Dimensional Images Yields Similar Measurements of Single-Cell Gene Expression compared with Data Derived from Previously Analyzed Maximal Projection Images

Mean projection of the previously analyzed RNA FISH three-dimensional image sets was performed by summing the fluorescence intensity in all optical sections and dividing by the constant number of total optical sections, and the mean projection images were analyzed by the automated algorithm. A, The resulting GFP-GR loading ratio and RNA FISH signal value are plotted for individual cells treated with Dex or RU. The inset box (A) indicates the subpopulations of cells that were selected based on GFP-GR array loading ratio values. The mean values of (B) GFP-GR array loading ratio and (C) RNA FISH signal are shown for the selected subpopulations. Error bars denote SEM.

 

Figure 8
View larger version (33K):
[in this window]
[in a new window]
  		Fig. 8. Comparison of Time-Dependent GR-Chromatin Association and Transcription

The RNA FISH signals and (A) GFP-GR array loading ratios (B) are shown averaged for cell populations treated with Dex for the indicated periods of time. Error bars denote SEM. C, The GFP-GR array loading ratios and RNA FISH signal values are plotted for the individual cells at two time points. The loess average lines show the nonlinear relationship between the measurements after 0.5 h (top line) and 8 h (bottom line) of Dex treatment. The inset box (C) indicates the subpopulations of cells that were selected based on GFP-GR array loading ratio values. The mean values of (D) GFP-GR array loading ratio and (E) RNA FISH signal are shown for the selected subpopulations. Error bars denote SEM.

 
The cellular behavior of GR association with the MMTV array and transcriptional activity was also measured after different lengths of incubation time with 100 nM Dex. After processing for RNA FISH, approximately 350 cells per time point were randomly selected and image analysis was performed as previously described. Consistent with the biphasic transcription profile reported for the Dex-induced MMTV LTR (3, 18), RNA FISH signals averaged for the cell populations peak at 30 min to 1 h of treatment and then decline significantly at the 4- and 8-h time points (Fig. 8AGo). The GFP-GR array loading ratios averaged for the populations also peak at the early time points and then exhibit a small but significant decrease after 4 and 8 h of treatment (Fig. 8BGo). ChIP assay results confirm similar time-dependent steady-state associations of GR with the MMTV LTR (data not shown). Plotting the GFP-GR array loading ratio vs. the RNA FISH signal for the individual cells reveals similar nonlinear relationships between these parameters for both the 30-min and 8-h time points (Fig. 8CGo). Even after 8 h, there is still a large degree of heterogeneity in FISH signal for any given level GFP-GR array association. The most pronounced difference between these two time points is a decrease in the RNA FISH signal plateau level at the 8-h time point. Supporting this finding, the intermediate time points show incremental decreases in the plateau level (data not shown). These results suggest that an activating event in addition to GR/MMTV interaction is less efficient at later time points. This was confirmed by selecting subpopulations of cells with a narrow range of GFP-GR array loading ratios for further analysis. Although the GR/MMTV association is statistically identical for these subpopulations at the 30-min and 8-h time points (Fig. 8DGo), there is significant decrease in the transcriptional output for the subpopulations at the later time (Fig. 8EGo). Consistent with these data, recent analysis of GR-associated protein complexes reveals that some components are covalently modified in a time-dependent manner, leading to down-regulation of promoter activity (28).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
MATERIALS AND METHODS
 REFERENCES
 
The critical microenvironments that regulate hormone-dependent gene expression are organized within individual cells. Although biochemical analyses of cell populations have provided many mechanistic insights into the control of transcription, these methods cannot resolve gene regulation at the level of individual cells. This is an important concern for understanding diseases such as cancer, which can arise from alterations in the behavior of single cells (15). To understand how cellular responses give rise to the hormone-dependent transcription observed in the population, automated analysis of high-resolution fluorescence microscopy images was used to measure the regulation of the MMTV array in many randomly sampled cells. Previous single-cell studies have used image analysis algorithms and statistical techniques to model transcription factor behavior or transcriptional output in separate experiments (29, 30, 31). However, the results reported here represent a major advance in that the algorithm simultaneously measured both steroid receptor behavior and the resulting transcriptional output at the single-cell level. Importantly, the single-cell data averaged for the sampled population closely agreed with parallel biochemical results, validating the analytical microscopy technique.

Consistent with the automated quantification of RNA FISH signals (Fig. 2AGo), previous nuclear run-on measurements of transcriptional initiation from the chromatinized MMTV- LTR showed a 7- to 10-fold increase after 30 min of Dex treatment (18, 19). Visualization of RNA polymerase II on the MMTV array also suggested a similar 10-fold increase in Dex-stimulated transcription (18). In contrast with Dex-treated cells, RU only slightly stimulated the RNA FISH signal measured for the cell population (Fig. 2AGo). This was in agreement with the finding of only minimal transcriptional activity of the chromatinized MMTV LTR in RU-treated Xenopus oocytes (32). In addition, reporter gene studies of transiently transfected cells have also demonstrated a pattern of high-level transcription after Dex treatment and low-level mixed antagonist stimulation after RU treatment (4, 33). Considering these observations, the averaged cell population RNA FISH signals confirmed previous biochemical measurements of transcriptional output.

Because hormone-bound GFP-GR transiently concentrates in a visible subnuclear region at the MMTV array in the 3617 cell line (17), this system also allowed fluorescence microscopy methods to measure the relative concentration of GFP-GR at sights of transcriptional regulation as a function of relative fluorescence intensity. In some cells, GFP-GR loaded on the array in a very robust way and was enriched more than 7-fold compared with the surrounding nucleoplasm (Fig. 4BGo). Because the local steady-state concentration of nuclear proteins may promote the formation of functional biochemical complexes, this parameter is likely to be a critical regulator of transcriptional output. Given the importance of this parameter, it was critical to establish how the GFP-GR imaging data related to underlying biochemical events. Both analytical microscopy and ChIP methods indicated that Dex-bound GR associated with the MMTV array to a greater degree than RU-bound GR (Fig. 2Go, B and C). The interpretation of these steady-state imaging and biochemical data must consider the dynamic nature of GR molecular interactions because both in vitro biochemical and live cell imaging methods have shown that the receptor is rapidly exchanged between chromatinized GREs and the surrounding environment (17, 23, 24). In light of this transient binding, the relative concentration of GR loaded on the array is determined at least in part by the balance of association and dissociation rates for the receptor. In support of this concept, RU-bound GR exchanges at the array more rapidly compared with the Dex-bound GR (34), which correlates with the decrease in steady-state array association (Fig. 2Go). Future photobleaching experiments on populations of individual living cells will be required to elucidate the kinetic details of this variable process.

In agreement with the results reported here, in vivo footprinting in Xenopus oocytes showed a similar pattern of MMTV association for agonist- and antagonist-bound GR (11, 12). According to these data, which average the response of all cells in the population, RU inefficiently activates transcription because steady-state GR association with the promoter is reduced. In contrast with the in vivo observations, both ligands caused GR to associate efficiently with the naked MMTV LTR or isolated GREs in vitro (10, 35). The differences reported for these divergent experimental systems are likely caused by the effects of chromatin organization that restricts DNA accessibility in vivo (36, 37, 38). In sum, the degree of GFP-GR loading on the MMTV array, which was measured in individual cells by the automated image analysis method, accurately quantified the steady-state physical interactions of ligand-bound GR with MMTV chromatin that regulate transcriptional output.

Transcriptional Variability within the Cell Population
Automated image analysis revealed that MMTV-driven transcription varied by more than 10-fold between individual Dex-stimulated cells (Fig. 6AGo). This heterogeneous behavior was expected based on previous reports of GRE-dependent transcriptional activity in individual cells (13, 39). Because the array consists of approximately 200 promoter-reporter gene units, the variable FISH signal measured for each cell represents the average output from interactions with many different promoters. There are many possible causes for this variability, including regulatory pathways connected to an unsynchronized cell cycle or genetic differences between nonclonal cells. The 3617 cell line used in these studies was clonally derived (16), suggesting that genetic differences between cells were not the cause of the variability. Supporting this conclusion, MMTV-dependent transcription was found to be equally heterogeneous even after multiple rounds of clonal selection (13).

It has become increasing clear that gene expression is an inherently probabilistic or stochastic process, which produces a wide range of expression profiles in the isogenic cell population (reviewed in Ref. 14). Studies of other systems indicate that this cellular behavior is typical of many expressed genes. In quantitative microscopy studies of synchronized normal fibroblast cells, acute serum stimulation caused a widely heterogeneous activation of multiple genes (29). This type of stochastic variability was not due to artifacts of cell culture or quantitative microscopy techniques because single-cell quantitative PCR-based transcription assays of apparently homogenous cells derived from normal tissues also yielded a high degree of heterogeneity (40, 41). These data suggest that some other combination of cellular conditions causes variable gene expression between individual cells. Based on the stochastic assembly observed for multiple components of nuclear protein complexes (42), it is likely that cell-to-cell differences in chromatin association or cofactor recruitment give rise to transcriptional heterogeneity.

Both Receptor Association with Chromatin and Additional Variable Events Contribute to Heterogeneous Transcriptional Activity
Even a simplified model of eukaryotic transcription involves many different events: the initiating DNA-binding factor associates with the promoter chromatin; protein-protein interactions with the DNA-binding factor allow the recruitment of many different cofactors; enzymatic action by cofactors to modify various proteins or remodel chromatin; and the resulting changes at the promoter that attract and activate the RNA polymerase complex. To reconcile the observations of rapid factor dynamics and slower changes in the steady-state association of factors with the promoter, the "return to template" hypothesis (23, 26, 43, 44) proposes that these events occur in rapid succession with each stage of the modified promoter complex serving as the template for the next productive interaction (Fig. 9Go; see also Fig. 4Go in Ref. 26). Although many of these associations are nonproductive, the high frequency of the interactions allows the formation of a sufficient steady-state concentration of active promoter complexes. The "return to template" hypothesis is also compatible with the stochastic view of gene expression, where any of the events involved in transcriptional activation could be a source of the gene intrinsic variability observed in individual cells (14). The link between dynamic protein interactions and stochastic mechanisms has also been recently demonstrated for nuclear factor-{kappa}B-dependent transcription (6), suggesting that these processes may contribute to the regulation of many genes. This is the first study to dissect which of the many possible regulating events gives rise to the heterogeneous transcription of steroid-dependent genes in individual cells.


Figure 9
View larger version (18K):
[in this window]
[in a new window]
  		Fig. 9. Stochastic Recruitment Model for Nuclear Receptor Function

A, Receptors exist in the nucleoplasmic space in a large variety of multiprotein complexes. These complexes are recruited randomly and stochastically to hormone response elements but remain template associated for brief periods of time. A fraction of these events, indicated by the probability term Ex for each stage, leads either to the correct template modification or secondary recruitment of the appropriate receptor-directed complex needed for the next step. Successful promoter activation requires the complete series of events, represented by the probability term P(T). B, The overall probability of a successful event series is diagramed schematically for a receptor agonist or antagonist. Antagonist-stimulated populations will have a few cells with transcriptional outputs almost as efficient as agonist-induced populations, but cells with high levels of expression will be much more frequent in agonist-activated populations. CARM1, Coactivator-associated arginine methyltransferase 1; GRIP1, GR-interacting protein 1; Swi/Snf complex, Swi/Snf nucleosome remodeling complex; Pol2, RNA polymerase 2.

 
The correlation between the GR array loading ratio and the RNA FISH signal in individual cells suggested that cell-to-cell variability in GR-promoter association was responsible for some of the transcriptional heterogeneity in the cell population. However, the wide cell-to-cell variability in FISH signals for any given level of steady-state GR-chromatin association indicates that other stochastic events in addition to GR-chromatin association are also required for full transcriptional activation. Because high levels of GR-chromatin association saturate the transcriptional response, these additional events appear to involve specific protein complexes that are present in limiting amounts. Cells with low GR array loading ratios exhibited an overall reduction in FISH signal, indicating that the additional activating events required an initial GR-chromatin association. These results are consistent with the additional event involving the stochastic recruitment of transcriptional cofactor complexes. Interestingly, a similar nonlinear relationship between GR/MMTV association and transcriptional output was observed after prolonged exposure to Dex. At these later time points, there remains a large degree of heterogeneity in transcriptional output for any given level of GR/MMTV association. Thus, the stochastic recruitment of additional factors is not merely the byproduct of the early stages of transcriptional stimulation when the steady-state activity of the promoters may not yet be achieved in all of the cells.

Agonists and antagonists cause changes in the affinity of GR for transcriptional cofactors, but these specific ligands also differentially regulate the affinity of the receptor for chromatinized GREs. Therefore, it has been difficult to determine whether the antagonist dependent reduction in the recruitment of cofactors to in vivo target promoters was due to changes in GR-chromatin- interactions, GR-cofactor interactions, or a combination of both. The novel single-cell microscopy approach presented here takes advantage of cellular heterogeneity to make the first comparison between agonist- and antagonist-treated cell subpopulations that have equal levels of GR-chromatin association. In these selected subpopulations, transcription is greatly reduced in the antagonist-treated cells vs. the agonist-treated cells. Similar analysis revealed that prolonged exposure to Dex causes a decrease in the transcriptional output in selected cell subpopulations exhibiting GR/MMTV association equal to that observed after acute Dex exposure. This indicates that time-dependent effects of specific ligands may differentially control transcription in individual cells by altering other regulatory events in addition to the ligand-specific changes in GR-chromatin interactions. Based on the current understanding of steroid-dependent gene expression, these additional events likely involve the recruitment and activation of transcriptional cofactors. This suggests the time-dependent mechanisms that are differentially influenced by specific ligands to regulate transcription are closely related to the mechanisms that give rise to heterogeneous gene expression within the cell population.

ATP-dependent chromatin remodeling complexes interact differentially with steroid receptors in the presence of specific ligands (Klokk, T. I., P. Kurys, C. Elbi, A. K. Nagaich, A. Hendarwanto, T. Slagsvold, C.-Y. Chang, G. L. Hager, and F. Saatcioglu, manuscript submitted; and Ref. 9). Interestingly, ATP-dependent chromatin remodeling factors are also involved in heterogeneous gene expression in individual yeast cells (8), suggesting that these complexes may also regulate stochastic gene expression in higher eucaryotes. Agonists and antagonists also differentially recruit coactivators, such as the p160 family of histone acetyltransferases (2). Thus, the variable actions of these protein complexes in individual cells may contribute to transcriptional heterogeneity.

We conclude that the transcriptional process induced by nuclear receptor activation involves a series of highly stochastic events, with considerable variation in efficiency possible at each stage (Fig. 9AGo). Thus, relatively inefficient steps, such as an initial antagonist-induced receptor binding event, can lead to downstream "jackpot" events that result in highly productive transcriptional output (Fig. 9BGo). Conversely, on an individual cell basis, highly efficient initial events do not necessarily lead to strong transcriptional activation (Fig. 9BGo), but these events occur more frequently in agonist-stimulated cell populations and thus result in more efficient output on a population basis. Receptor function can be accurately modeled only from this cell-by-cell perspective. Further analytical microscopy studies will be required to delineate which cofactors are involved in stochastic gene expression and how these factors are differentially regulated in individual cells.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
REFERENCES
 
Cell Culture and Hormone Treatment
Generation, characterization, and maintenance of the mouse mammary carcinoma 3617 cell line used throughout these studies have been previously described. For imaging experiments, cells were electroporated with expression vectors encoding the indicated cherry fusion proteins then plated on square 22-mm no. 1 German coverglass in six-well plates. After plating, cells were incubated for 24 h with 10% charcoal stripped serum in phenol red-free DMEM without tetracycline to induce expression of the GFP-GR fusion protein. Treatment medium containing 10% charcoal stripped serum supplemented with 100 nM Dex, 100 nM RU, or vehicle alone was added to the cells for 30 min. Cells were then fixed with formaldehyde and processed for RNA FISH or chromatin IP assay.

RNA Fluorescence in Situ Hybridization
Biotinylated DNA probe used in the RNA FISH experiments was prepared by nick translation and a plasmid encoding v-ras using standard methods. After 15 min fixation with 4% formaldehyde, cells on coverslips were permeabilized with 0.2% Triton X-100. For each coverslip, 100 ng biotinylated probe and 40 µg yeast tRNA were suspended in 100% formamide, denatured at 80 C, quick chilled, then added to hybridization buffer containing a final concentration of 25% formamide and 2x SSC. Hybridization buffer with probe was then added to the cells and incubated overnight at 37 C in a humidified chamber. After extensive washing, cells were incubated with Streptavidin-Alexa 680 (Molecular Probes, Eugene, OR) for 1 h to detect the MMTV driven v-ras transcripts. Coverslips were again washed extensively then incubated with DAPI for 5 min and mounted on slides with anti-fade reagent (Prolong Gold; Molecular Probes).

Automated Image Analysis
Images of fixed cells were captured using a TE 300 inverted epifluorescence microscope fitted with a 60x/1.4 numerical aperture oil objective (Nikon, Melville, NY), CCD camera (CoolSnap ES; Photometrics, Tucson, AZ), automated shutter, excitation and emission filter wheels (Sutter Instruments, Novato, CA), piezoelectric z-focus objective positioner (Physik Instrumente, Karlsruhe/Palmbach, Germany), and computer controlled x-y stage (Ludl, Hawthorne, NY). All electronic controls for image capture were integrated using Metamorph imaging software (Molecular Dynamics, Sunnyvale, CA). Fluorophores were distinguished using a four-color polychroic mirror and matched excitation and emission band pass filters (Chroma Technology Corp., Rockingham, VT). The wavelengths for each fluorophore that were selected by excitation and emission band pass filters were as follows: DAPI: excitation, 350/50, emission, 457/50; GFP: excitation, 490/20, emission, 528/38; Cherry: excitation, 555/28, emission, 617/73; Alexa 680: excitation, 635/20, emission 685/40. Using these optics, the pixel size in the resulting images was 132 nm.

For each experimental condition, the user sequentially selected nine different areas of the coverslip for later image capture. The relative x,y,z coordinates of the areas on the stage were stored using the Metamorph software. Imaged areas were at least 1000 µm apart to ensure that photobleaching did not adversely affect image quantification. During area selection, only the DAPI channel was viewed, eliminating user bias due to knowledge of the other fluorophore signals. Based on the stored coordinates for each selected area, the Metamorph software captured z-stack images spanning 7.8 µm of depth at a 0.3-µm step size for each fluorophore. The resulting images of each area contained full three-dimensional information (x,y,z) for five to 15 cells. All image files were batch converted from the proprietary Metamorph image file format to the proprietary Applied Precision image file format using a custom algorithm developed in the Matlab technical computing software (The Mathworks Inc., Natick, MA). Converted images were then batch processed with a commercially available constrained iterative digital deconvolution algorithm (Softworx Explorer Suite; Applied Precision Inc., Issaquah, WA) to quantitatively reassign out-of-focus light and restore signal contrast in the epifluorescence images. To reduce the amount of data processed in subsequent analysis steps, a custom Matlab algorithm batch converted the full three-dimensional data for each region to a maximal intensity z-stack projection image and saved the projection image in a 16-bit grayscale tiff file.

RNA FISH signal, relative levels of fluorescent proteins in the nucleus, and relative concentration of fluorescent protein at the MMTV array was measured for each cell using a completely automated image analysis algorithm that was developed with the Matlab Image Processing Toolbox. Before analysis, the user enters the computer file names for all maximal projection images into a spreadsheet file. The algorithm then loads the DAPI, GFP, Cherry, and RNA FISH images using the file names stored in the spreadsheet file. The background fluorescence intensity for each image is measured and subtracted. The ROI defining each nucleus (nROI) is selected from the DAPI image by an optimal thresholding subroutine. The relative levels of the fluorescent proteins in the nucleus are calculated as the mean intensity of the nROI area applied to the green and red images. The nROI positional information is also used to help define the RNA FISH signal. The FISH array ROI (aROI) is concisely identified by the algorithm as a contiguous region of FISH image pixels more than 16 pixels in size that is also greater than 2.5-fold brighter than the mean intensity of the FISH image nROI. The integrated FISH signal per nucleus is then calculated as the product of the aROI area and mean intensity of the aROI in the FISH image. The aROI positional information is also applied to the red and green images to quantify the degree of fluorescent protein loading on the array. For each fluorescent protein image, the algorithm calculates the relative concentration of fluorescent protein on the array as the maximal pixel intensity value in the aROI divided by the mean intensity of the fluorescent protein in the nucleus. All the morphometric data for each cell are automatically exported to spreadsheet files for further statistical analysis. The algorithm saves the ROI data in labeled images for each nucleus. These ROI images are used to manually inspect the regions that were measured by the algorithm. Less than 1% of cells are typically discarded from the analysis by the user due to errors in automated ROI selection. Chromatic aberration in the microscope was found to cause an approximately 0.6-µm misalignment of the GFP and FISH images in the optical axis. To ensure that this z-axis shift did not adversely affect the image analysis, a test set of three-dimensional imaging data was realigned to correct the differences in the optical axis and analyzed by the automated image analysis algorithm. When the values from the original and z-corrected image sets were compared, the average FISH signals and GFP-GR loading ratios for 88 cells varied by less than 2%. This result showed that chromatic aberration had negligible impact on the reported measurements. To confirm that the use of maximal intensity projection images did not cause errors in the results, the algorithm was also used to analyze mean projection images derived from the full z-stack imaging data. As described in the results, similar measurements were obtained from the mean projection images compared with the maximal projection images. However, the measurement derived from the maximal projection images were selected for presentation, except where specifically indicated, because the segmentation of the aROI was most accurate in this data set.

Chromatin Immunoprecipitation
Cells (3617) were treated with either vehicle or 100 nM Dex for 30 min. Cells were processed for ChIP as described elsewhere (Upstate Biotechnology, Lake Placid, NY). Briefly, soluble chromatin was immunoprecipitated with antibodies to the glucocorticoid receptor (Affinity BioReagents, Golden, CO; PA1-510). DNA isolated from immunoprecipitates was used as a template for real-time quantitative PCR amplification. Real-time assays were conducted on a Bio-Rad bicycler IQ system (Bio-Rad, Hercules, CA) using the intercalation dye SYBR Green as the fluorescence agent (Bio-Rad iQ SYBR Green Supermix) and the manufacturer-recommended conditions (10 nM FAM, 3 mM MgCl2, and 300 nM primers in a 25-µl reaction). PCR was performed by denaturing at 95 C for 15 sec and annealing/extending at 60 C for 60 sec. Standard curves were created for each run using a plasmid (pM18) that contained the MMTV LTR and primers that spanned the GR binding sites. Ten-fold serial dilutions of pM18 (over 4 logs) were used to generate the standard curve. All PCRs were subject to a melting curve to verify the integrity of the PCR product and to eliminate amplification of nonspecific products. Initial PCRs were also run on agarose gels to verify product size. The following primers were used for amplification: sense, 5'-ACAAGATATAAAAGAGTGC-3'; and antisense, 5'-ACACACAAGAGGTGAATGTT-3'.

Statistical Analysis
All data sets were derived from at least three cell cultures. Interactive data analysis, including means calculation and determination of SEM, and the calculation of the locally weighted scatter plot smoothing (loess) average line were performed using SPSS 13.0 software for Windows (SPSS, Chicago, IL). ANOVA and the Student-Newman-Keuls post hoc test (SPSS 13.0) were used to identify differences between cell subpopulations, with P < 0.05 considered significant.


   FOOTNOTES
 
DISCLOSURE STATEMENT: The authors report no conflicting financial interests.

First Published Online July 27, 2006

Abbreviations: aROI, Gene array ROI; ChIP, chromatin immunoprecipitation; DAPI, 4',6'-diamidino-2-phenylindole; Dex, dexamethasone; FISH, fluorescence in situ hybridization; GFP, green fluorescent protein; GR, glucocorticoid receptor; GRE, glucocorticoid response element; LTR, long terminal repeat; MMTV, mouse mammary tumor virus; nROI, region of interest for each nucleus; RU, RU486.

Received for publication February 21, 2006. Accepted for publication July 12, 2006.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 REFERENCES
 

  1. Mangelsdorf DJ, Thummel C, Beato M, Herrlich P, Schutz G, Umesono K, Blumberg B, Kastner P, Mark M, Chambon P, Evans RM 1995 The nuclear receptor superfamily: the second decade. Cell 83:835–839[CrossRef][Medline]
  2. McKenna NJ, O’Malley BW 2002 Combinatorial control of gene expression by nuclear receptors and coregulators. Cell 108:465–474[CrossRef][Medline]
  3. Kinyamu HK, Archer TK 2004 Modifying chromatin to permit steroid hormone receptor-dependent transcription. Biochim Biophys Acta 1677:30–45[Medline]
  4. Schulz M, Eggert M, Baniahmad A, Dostert A, Heinzel T, Renkawitz R 2002 RU486-induced glucocorticoid receptor agonism is controlled by the receptor N terminus and by corepressor binding. J Biol Chem 277:26238–26243[Abstract/Free Full Text]
  5. Stevens A, Garside H, Berry A, Waters C, White A, Ray D 2003 Dissociation of steroid receptor coactivator 1 and nuclear receptor corepressor recruitment to the human glucocorticoid receptor by modification of the ligand-receptor interface: the role of tyrosine 735. Mol Endocrinol 17:845–859[Abstract/Free Full Text]
  6. Bosisio D, Marazzi I, Agresti A, Shimizu N, Bianchi ME, Natoli G 2006 A hyper-dynamic equilibrium between promoter-bound and nucleoplasmic dimers controls NF-{kappa}B-dependent gene activity. EMBO J 25:798–810[CrossRef][Medline]
  7. Szapary D, Huang Y, Simons Jr SS 1999 Opposing effects of corepressor and coactivators in determining the dose-response curve of agonists, and residual agonist activity of antagonists, for glucocorticoid receptor-regulated gene expression. Mol Endocrinol 13:2108–2121[Abstract/Free Full Text]
  8. Raser JM, O’Shea EK 2004 Control of stochasticity in eukaryotic gene expression. Science 304:1811–1814[Abstract/Free Full Text]
  9. Rayasam GV, Elbi C, Walker DA, Wolford R, Fletcher TM, Edwards DP, Hager GL 2005 Ligand-specific dynamics of the progesterone receptor in living cells and during chromatin remodeling in vitro. Mol Cell Biol 25:2406–2418[Abstract/Free Full Text]
  10. Pandit S, Geissler W, Harris G, Sitlani A 2002 Allosteric effects of dexamethasone and RU486 on glucocorticoid receptor-DNA interactions. J Biol Chem 277:1538–1543[Abstract/Free Full Text]
  11. Belikov S, Astrand C, Holmqvist PH, Wrange O 2004 Chromatin-mediated restriction of nuclear factor 1/CTF binding in a repressed and hormone-activated promoter in vivo. Mol Cell Biol 24:3036–3047[Abstract/Free Full Text]
  12. Belikov S, Gelius B, Wrange O 2001 Hormone-induced nucleosome positioning in the MMTV promoter is reversible. EMBO J 20:2802–2811[CrossRef][Medline]
  13. Ko MS, Nakauchi H, Takahashi N 1990 The dose dependence of glucocorticoid-inducible gene expression results from changes in the number of transcriptionally active templates. EMBO J 9:2835–2842[Medline]
  14. Kaern M, Elston TC, Blake WJ, Collins JJ 2005 Stochasticity in gene expression: from theories to phenotypes. Nat Rev Genet 6:451–464[CrossRef][Medline]
  15. Capp JP 2005 Stochastic gene expression, disruption of tissue averaging effects and cancer as a disease of development. Bioessays 27:1277–1285[CrossRef][Medline]
  16. Walker D, Htun H, Hager GL 1999 Using inducible vectors to study intracellular trafficking of GFP-tagged steroid/nuclear receptors in living cells. Methods 19:386–393[CrossRef][Medline]
  17. McNally JG, Muller WG, Walker D, Wolford R, Hager GL 2000 The glucocorticoid receptor: rapid exchange with regulatory sites in living cells. Science 287:1262–1265[Abstract/Free Full Text]
  18. Becker M, Baumann C, John S, Walker DA, Vigneron M, McNally JG, Hager GL 2002 Dynamic behavior of transcription factors on a natural promoter in living cells. EMBO Rep 3:1188–1194[CrossRef][Medline]
  19. Archer TK, Lee HL, Cordingley MG, Mymryk JS, Fragoso G, Berard DS, Hager GL 1994 Differential steroid hormone induction of transcription from the mouse mammary tumor virus promoter. Mol Endocrinol 8:568–576[Abstract]
  20. Fragoso G, Pennie WD, John S, Hager GL 1998 The position and length of the steroid-dependent hypersensitive region in the mouse mammary tumor virus long terminal repeat are invariant despite multiple nucleosome B frames. Mol Cell Biol 18:3633–3644[Abstract/Free Full Text]
  21. Kramer PR, Fragoso G, Pennie W, Htun H, Hager GL, Sinden RR 1999 Transcriptional state of the mouse mammary tumor virus promoter can affect topological domain size in vivo. J Biol Chem 274:28590–28597[Abstract/Free Full Text]
  22. Smith CL, Htun H, Wolford RG, Hager GL 1997 Differential activity of progesterone and glucocorticoid receptors on mouse mammary tumor virus templates differing in chromatin structure. J Biol Chem 272:14227–14235[Abstract/Free Full Text]
  23. Nagaich AK, Walker DA, Wolford R, Hager GL 2004 Rapid periodic binding and displacement of the glucocorticoid receptor during chromatin remodeling. Mol Cell 14:163–174[CrossRef][Medline]
  24. Fletcher TM, Xiao N, Mautino G, Baumann CT, Wolford R, Warren BS, Hager GL 2002 ATP-dependent mobilization of the glucocorticoid receptor during chromatin remodeling. Mol Cell Biol 22:3255–3263[Abstract/Free Full Text]
  25. Rigaud G, Roux J, Pictet R, Grange T 1991 In vivo footprinting of rat TAT gene: dynamic interplay between the glucocorticoid receptor and a liver-specific factor. Cell 67:977–986[CrossRef][Medline]
  26. Hager GL, Nagaich AK, Johnson TA, Walker DA, John S 2004 Dynamics of nuclear receptor movement and transcription. Biochim Biophys Acta 1677:46–51[Medline]
  27. Hager GL, Elbi C, Johnson TA, Voss T, Nagaich AK, Schiltz RL, Qiu Y, John S 2006 Chromatin dynamics and the evolution of alternate promoter states. Chromosome Res 14:107–116[CrossRef][Medline]
  28. Qiu Y, Zhao Y, Becker M, John S, Parekh BS, Huang S, Hendarwanto A, Martinez ED, Chen Y, Lu H, Adkins NL, Stavreva DA, Wiench M, Georgel PT, Schiltz RL, Hager GL 2006 HDAC1 acetylation is linked to progressive modulation of steroid receptor-induced gene transcription. Mol Cell 22:669–679[CrossRef][Medline]
  29. Levsky JM, Shenoy SM, Pezo RC, Singer RH 2002 Single-cell gene expression profiling. Science 297:836–840[Abstract/Free Full Text]
  30. Voss TC, Demarco IA, Booker CF, Day RN 2005 Functional interactions with Pit-1 reorganize co-repressor complexes in the living cell nucleus. J Cell Sci 118:3277–3288[Abstract/Free Full Text]
  31. Voss TC, Demarco IA, Day RN 2005 Quantitative imaging of protein interactions in the cell nucleus. Biotechniques 38:413–424[Medline]
  32. Belikov S, Holmqvist PH, Astrand C, Wrange O 2004 Nuclear factor 1 and octamer transcription factor 1 binding preset the chromatin structure of the mouse mammary tumor virus promoter for hormone induction. J Biol Chem 279:49857–49867[Abstract/Free Full Text]
  33. Szapary D, Xu M, Simons Jr SS 1996 Induction properties of a transiently transfected glucocorticoid-responsive gene vary with glucocorticoid receptor concentration. J Biol Chem 271:30576–30582[Abstract/Free Full Text]
  34. Stavreva DA, Muller WG, Hager GL, Smith CL, McNally JG 2004 Rapid glucocorticoid receptor exchange at a promoter is coupled to transcription and regulated by chaperones and proteasomes. Mol Cell Biol 24:2682–2697[Abstract/Free Full Text]
  35. Cho S, Blackford Jr JA, Simons Jr SS 2005 Role of activation function domain-1, DNA binding, and coactivator GRIP1 in the expression of partial agonist activity of glucocorticoid receptor-antagonist complexes. Biochemistry 44:3547–3561[CrossRef][Medline]
  36. Fragoso G, John S, Roberts MS, Hager GL 1995 Nucleosome positioning on the MMTV LTR results from the frequency-biased occupancy of multiple frames. Genes Dev 9:1933–1947[Abstract/Free Full Text]
  37. Richard-Foy H, Hager GL 1987 Sequence-specific positioning of nucleosomes over the steroid-inducible MMTV promoter. EMBO J 6:2321–2328[Medline]
  38. Keeton EK, Fletcher TM, Baumann CT, Hager GL, Smith CL 2002 Glucocorticoid receptor domain requirements for chromatin remodeling and transcriptional activation of the mouse mammary tumor virus promoter in different nucleoprotein contexts. J Biol Chem 277:28247–28255[Abstract/Free Full Text]
  39. Necela BM, Cidlowski JA 2003 Development of a flow cytometric assay to study glucocorticoid receptor-mediated gene activation in living cells. Steroids 68:341–350[CrossRef][Medline]
  40. Kamme F, Salunga R