Abrupt and Dynamic Changes in Gene Expression Revealed by Live

Feb 8, 2012 - Live cell arrays enabled cell populations to be characterized temporally at ... and such variable activities within a population will pr...
0 downloads 0 Views 3MB Size
Article pubs.acs.org/ac

Abrupt and Dynamic Changes in Gene Expression Revealed by Live Cell Arrays Maureen A. Walling,† Hua Shi,‡ and Jason R. E. Shepard*,† †

Department of Chemistry and ‡Department of Biological Sciences, University at Albany, 1400 Washington Avenue, Albany, New York 12222, United States S Supporting Information *

ABSTRACT: A description of the noise associated with gene expression is presented, based on a simplified form of the combined multistep processes of transcription and translation. These processes are influenced by numerous factors, including the accessibility of promoter regions to the transcriptional machinery, the kinetics of assembly of the transcription complexes, and the synthesis and degradation of both mRNA and proteins, among others. Ultimately, stochasticity in cellular processes results in variation in protein levels. Here we constructed a rationally designed RNA-based transcriptional activator to reduce these variables and provide a cleaner, more detailed portrayal of cellular noise. Functioning at a level comparable to natural transcription activation, this activator is isolated to a lacZ reporter gene in yeast cells to quantitatively describe the efficiency of the combined processes of transcription and translation. By employing single-cell array techniques to monitor individual cells simultaneously and in real time, a statistical approach to investigate noise inherent in gene expression is possible. Live cell arrays enabled cell populations to be characterized temporally at the individual cell level. The array platform allowed for a relative measure of protein production in real time and could characterize protein bursts with variable size and random timing, such that bursts occurred in a temporally indiscriminate fashion. The inherent variability and randomness of these processes is characterized, with almost half (47%) of cells experiencing bursting behavior at least once over the course of the experiment. We demonstrate that cells identified on the upper periphery of activity exhibit behaviors that are substantially different from the majority of the population, and such variable activities within a population will provide a more accurate characterization of the population.

T

response based on cell population experiments at a single time point. Such analyses are problematic if the studied population exhibits a large cell-to-cell variability or variation is more pronounced over time. To understand the stochasticity associated within isogenic cells, it is thus essential to perform real-time observations of gene expression in individual living cells. Such heterogeneity studies in yeast (Saccharomyces cerevisiae) have proposed that transcription and translation proceed intermittently, resulting in cell-to-cell variation in the level of RNA and protein molecules.16−19 Quantitative realtime analysis on gene expression in a live cell has linked such stochasticity to “bursting” behavior.16,17,20,21 The phenomenon of bursting activity is defined as a cell exhibiting an increased level of expression separated by variable periods of low activity. The extent of such bursts in activity are described by the size, length, and frequency of the event.16 The experimental model system we present is based on the analysis of a clonal cell population of yeast as a basic framework for characterizing phenotypic heterogeneity. To capture the

he level of mRNA or protein within a cell is a balance between its rate of synthesis and its rate of degradation. Due to the stochastic nature of the myriad of biochemical reactions and coordinated recognition events involved in biosynthesis and metabolism, the probability that the two rates are synchronized is low from cell to cell, resulting in differential levels of mRNA and/or proteins.1−3 For the synthesis of mRNA or protein, transcription and translation are also stochastic processes themselves. Because mRNA and many proteins are found in a cell at low copy numbers, slight variations in mRNA and protein levels can dramatically influence the extent of gene expression. Ultimately, this differential extent of gene expression manifests as cellular heterogeneity, even in cell populations with identical genomes. This phenomenon carries important biological implications. As a wide range of behaviors is possible, individual cells found on the boundaries of their population activity may be phenotypically distinct.4−8 Specifically, certain outlier cells may be capable of playing a role qualitatively different from that of the bulk population in concerns like the onset of disease. Numerous pathways and processes have been studied to characterize the extent of the natural level of variation across an otherwise identical cell population. 9−15 However, most quantitative methods consider only a single “average cell” as a © 2012 American Chemical Society

Received: November 18, 2011 Accepted: February 8, 2012 Published: February 8, 2012 2737

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

loading on the array with a concanavalin A (ConA)− tetramethylrhodamine (TAMRA) conjugate or streptavidincoated quantum dots, or after addition to the array by the addition of ConA−TAMRA through the laminar flow strip. The cells were grown in liquid medium for 15−20 h, and the cell growth solution was rinsed with fresh medium prior to labeling. To label the cells with TAMRA, approximately 200 μL of ConA−TAMRA solution (250 μg/mL in medium; Invitrogen, Carlsbad, CA) was added to the cells, shaken for 45 min, and then washed three times with medium. To label cells with quantum dots, approximately 250 μL of ConA− biotin conjugate (500 μg/mL in medium; Invitrogen, Carlsbad, CA) was added to the cells and they were shaken for 45 min. After shaking, the cells were washed three times with medium, and the cells were resuspended in 100 μL of medium and 2.5 μL of 450NC streptavidin-coated quantum dots (eBioscience, Inc., San Diego, CA). The cells were shaken with the quantum dots for 45 min and then rinsed three times with fresh medium. Prior to loading on the array, the cell solutions were diluted to an appropriate density. To load the array with cells for analysis, approximately 10 μL of cell solution was aliquoted on the Primary Cell LiveCell Array microscope slides, which contain 15-μm-diameter wells (Molecular Cytomics Inc., Boston, MA). These slides hold an array of ∼16 000 wells in total and include a laminar flow strip that enables delivery or removal of medium or reagents to the cells localized on the arrays. As the slide platform is comparable to a microscope slide, the array portion of the device can be viewed directly with the microscope-based imaging system described below. To label cells on the array, 10 μL of ConA−TAMRA solution was added to the cells on the array and allowed to sit for 10 min. The array was washed with three 10-μL aliquots of medium before any images were acquired. All cell labeling was performed prior to the start of all gene expression assays. The array platform used for monitoring gene expression in individual live cells contained thousands of individual microwells to restrict cell positions,22,25,28 and fluorescence images were taken to register the cells’ positions on the array. The signals from 15−20 empty wells were also taken to serve as a background to correct for signal intensity. Cell signal data were determined by subtracting the average background intensity from individual cell signals for the length of the assay. Additional information regarding the array platform and image analysis are provided in the Supporting Information. Imaging System. Fluorescence images were acquired by use of a Hamamatsu ORCA-ER charge-coupled device (CCD) camera (Hamamatsu, Bridgewater, NJ) attached to an Olympus BX61 (Olympus America, Melville, NY) microscope. The microscope was also equipped with a mercury arc lamp and Prior filter wheels (Prior Scientific, Rockland, MA) containing multiple optical band-pass filters and dichroic mirrors (Chroma Technology, Rockingham, VT), which allow the use of various combinations of excitation and emission wavelengths. Each cell label and the fluorogenic substrate required a separate excitation and emission filter set for distinction (given as excitation filter, dichroic mirror, and emission filter, with the central wavelength in nanometers ± band-pass): TAMRA (540 ± 12.5, 565, 582 ± 25), 450NC (420 ± 20, 475, 500 ± 20), and C12FDG enzyme product (480 ± 15, 505, 535 ± 20). The instrument was controlled by IPLab software (Scanalytics, Fairfax, VA), which was also used in data extraction. Numerical data processing and analysis were performed with Microsoft Excel.

dynamical behavior of cell populations, live cell arrays were employed to study the expression of a lacZ reporter gene in individual cells over time. The lacZ gene expressed the enzyme β-galactosidase, whose activity was correlated as a level of expression within a cell. The substrate for the enzyme was a fluorescent precursor included in the medium, which was not chromogenic in its native unprocessed form but, upon enzymatic processing, provided a direct fluorescent response against the combined transcriptional and translational activity.22−25 To reduce the number of uncontrollable factors that amplify the noise, the reporter gene was embedded in the chromatin as a single copy in each cell but mechanistically isolated from the rest of the cellular transcription regulation system through a unique RNA-based transcriptional activator. Genes controlled under the same promoter elements are transcribed together into mRNA, which is then translated into protein. The resultant enzymatic activity of this protein, measured as fluorescence intensity increases, is relatable to the level of gene expression, in a similar fashion to routine selection techniques such as two-hybrid analysis where the successful binding of a transcriptional activator to a specific promoter sequence is determined. We demonstrate the randomness of protein production within a population of genetically uniform cells. A striking variability in bursting behavior was observed as discrete, sharp increases in fluorescence intensity that manifested with a high degree of randomness between cells. Bursting behavior occurred sporadically but throughout the experiment, both as single, discrete events as well as prolonged, steady levels of increased expression. Over the course of the experiment, almost half (47%) of the cells in the population exhibited bursting during at least one time point, and a substantial portion of the population (17%) exhibited multiple instances of bursting over an 8-h period. In total, the extent of bursting displayed by the cellular population demonstrates that the frequency and magnitude of gene expression bursts are not only prevalent but highly variable; this suggests the importance of stochastic processes to influence cellular heterogeneity, and ultimately phenotypic variation.



EXPERIMENTAL SECTION Strains, Media, and Plasmids. The Saccharomyces cerevisiae strain studied in this work was derived from the YBZ-1 strain [MATa ura3-52 leu2-3 112 his3-200 trp1-1 ade2 LYS2::(LexA op)-HIS3 ura3::(LexA op)-LacZ LexA−MS2-MS2 coat (N55K)].24,26 In this strain, the lacZ reporter gene was incorporated with single copy status as a lacZ coding sequence with a lexA operator inserted into the ura3 locus in the genome. The strain contained the plasmid pDB-m26−29, which encodes an RNA-based transcription activator for the reporter genes, including lacZ.24,27 The m26−29 is an RNA-based transcription activator,25 whose coding sequence is cloned into the NotI site of the plasmid pDB-sansA.22 pDB-sansA was derived from pIIIA/MS2-1, which encodes two copies of the MS2 coat protein ligand in a transcription unit driven by the yeast RPR1 promoter.24 Transformation was performed according to standard protocols by use of lithium acetate. The cell medium was a dropout medium consisting of yeast nitrogen base (U.S. Biological), 2% glucose, and synthetic dropout supplements lacking uracil (U.S. Biological). Cell Labeling and Array Fabrication. Fluorescent cell labels were utilized to verify individual cell positions on the array.22,24,25 The cells were labeled either in medium prior to 2738

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

Figure 1. Mechanism of activation for the reporter gene. (a) Secondary structure of the modified m26−29 moiety in the RNA-based transcriptional activator, as predicted by the mfold program.30 (b) Schematic diagram showing molecular events at the promoter leading to transcription of the lacZ gene. The reporter gene, under the control of LexA operators, was integrated into the chromosome. Factor X is the unknown target of m26−29 and is most likely a general transcription factor of RNA polymerase II.

Yeast β-Galactosidase Assay. All reagent and medium additions were performed in 10-μL aliquots. The fluorogenic substrate 5-dodecanoylaminofluorescein di-β-D-galactopyranoside (C12FDG, 1.2, 5.8, 14.5, and 30.0 μM in medium; Molecular Probes, Invitrogen, Carlsbad, CA) was added in 180min intervals, and fresh medium was added approximately 3 min before the addition of reagent to wash away old and unused reagent. Immediately before the start of the assay (T = 0), five sequential images with a 5-s exposure time were taken to serve as the background for the assay. A fluorescence image was acquired every 30 min for the length of the assay to monitor lacZ gene expression; the assay was such that an increase in fluorescence directly correlated to a level of gene expression for comparison across individual cells. Signal intensity data were extracted from the experimental images and corrected for background fluorescence by use of the average signal from neighboring empty wells. The population average and standard deviation were calculated from the individual cell signals for each time point, and the population average and individual cell signals were plotted versus time as net fluorescence increases. For the bursting experiments, the difference in fluorescence of individual cells between 30 min time points was determined and plotted to depict the extent of the population’s bursting behavior in relation to gene expression activity over time. Outlier cells exhibiting bursting behavior were determined on the basis of the standard deviation (σ) associated with each time point; for example, cells with a signal at least three standard deviations away from

the measurement average at a given time were considered 3σ outliers.



DISCUSSION We constructed a model system in the yeast S. cerevisiae from well-characterized parts, together serving as an artificial transcriptional activator. To reduce the number of uncontrollable factors and make the model more amenable to quantitative assessment, a single copy of the lacZ reporter gene was incorporated in the chromatin environment, whose constitutive expression was solely dependent on the presence of the RNA-based transcriptional activator, as described previously (Figure 1).24 In this case, the activation domain of the transcriptional activator was originally selected from a library of random sequences for its capability of sustaining a high level of transcription, presumably through the recruitment of a key factor in the assembly of the RNA polymerase II preinitiation complex.27 We took a high-performance variant of the original isolate, m26−29, and made it more compact by deleting a long stem and adding a GC clamp to form the secondary structure shown in Figure 1. With this configuration, the expression of the lacZ reporter gene was isolated from the rest of the regulatory network because it was controlled by the recognition site for the Escherichia coli LexA protein in the promoter, which recruits the RNA-based activator through a LexA−MS2 coat protein fusion construct. As a result, no alternative native mechanism existed that was capable of activating this gene, and there was no other gene in the cell that was regulated in the same way. In addition, the RNA-based factor and its protein 2739

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

adaptor were always present at a level that saturated the available LexA sites in the promoter,24 which should reduce the transient variability across the population. Yeast cells were grown overnight and labeled with an encoding dye that was used to identify their random positioning that results from the array-loading process.22,25 Once cells settled into the microscale wells of the cell chip (Figure 2), their position did not change and the cells were

Figure 2. Scanning electron microscopy of individual yeast cells confined to the microscale wells of the live cell array. Once localized into the array, the cell positions do not change, and they can be exposed to reagents and monitored over time. The diameter of the wells is 15 μm. Figure 3. (a) Average response and standard deviation of 879 individual yeast cells measured over 480 min, based on the enzyme βgalactosidase processing the substrate C12FDG, present at a concentration of 14.5 μM in cell medium. (b) Individual responses of 243 cells (randomly selected out of the same 879 yeast cells) to 14.5 μM C12FDG, highlighting the diversity in individual cell responses. The average of all 879 cells is included as a green line.

exposed to the reagent. The number of cells measured varied from hundreds to thousands depending on the field of view and resolution desired. Initially the yeast cells were monitored on the basis of exposure to different concentrations of the fluorescent precursor C12FDG in minimal medium. The reagent C12FDG is able to cross the yeast cell membrane where it is processed by β-galactosidase. Once the enzyme acts on its substrate, the product is fluorescent and unable to leave the cell. Accumulation of the fluorescent enzyme product inside the cell is used as a measure of cellular activity, with images of the array collected with a microscope system over time. The cells were monitored for 8 h, and the data produced allowed the statistical analysis that follows. At the 14.5 μM concentration of reagent, a total of 879 yeast cells were monitored. The cellular activity is shown in Figure 3, depicted as both the average population activity (Figure 3a) and the activity of individual cells (Figure 3b shows the response profile from 243 of 879 cells). The cells were monitored against their β-galactosidase activity over time, plotted as the net fluorescence increase over 8 h. As can be seen by the individual cell data, some cells responded rapidly, almost instantaneously, while others lingered around the baseline throughout the course of the experiment. However, these results confirm that protein levels are variable at the onset of the experiment, and changes in fluorescence can be observed within minutes via this method. By taking individual cell measurements, a true measure of the population standard deviation can be calculated and included with the average. As expected, the average fluorescence intensity increases over time and the spread of the population widens, along with an increase to the standard deviation. While the average and standard

deviation are useful in describing the general population response, the individual cell plot is much more informative as to the extent of noise in the cellular activity. Some cells exhibit little or no activity, with a fluorescent response hovering around the baseline, while other cells have activity greater than 6 times the average response. Figure 4 shows the average response of cell populations across three concentrations of C12FDG (1.2, 5.8, and 14.5 μM), which show predictable substrate concentration-dependent profiles that increase proportionally over time. Over the course of 8 h, there was a clear distinction between the population averages based on each assay’s C12FDG concentration, with higher substrate concentrations corresponding to greater average signal increases. However, this difference is less pronounced as individual cells are examined, where substantial overlap in cellular activity occurs at the margins (data not shown, but the cellular noise from Figure 3b exemplifies this concept). The relationship between substrate concentration and average fluorescence signal was highly predictable, as the average signal plotted versus substrate concentration across multiple time points exhibited a consistent linear fit based on regression analysis (Figure 4). At higher concentrations and longer assay times, linearity is maintained with R2 values remaining greater than 0.975 (data not shown). 2740

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

Figure 4. Predictive nature of the cellular response in characterizing global activities. In contrast to the individual cellular activities, the population activity shows averages that are concentration-dependent, with highly linear correlation, but mask the heterogeneity of the population.

Table 1. Bursting Activity from Each Time Point as the Number of Cells Exhibiting the Phenomenon and the Percentage of Cells from the Total (879 Cells)a time/min net avg std dev 3× std dev. no. of bursting cells % of cells bursting

60 1.2 3.3 10 12 1.4

90 0.0 2.7 8.0 7 0.8

120 1.6 2.5 7.6 14 1.6

150 4.7 3.3 10 63 7.2

180 0.0 3.2 9.5 8 0.9

210 18.1 12 36 60 6.8

240 7.1 6.3 19 35 4.0

270 9.1 7.3 22 46 5.2

300 6.8 5.3 16 44 5.0

330 4.4 6.8 21 16 1.8

360 5.7 5.2 15 33 3.8

390 22.8 13 40 89 10.1

420 20.1 8.2 25 102 11.6

450 5.4 7.6 23 15 1.7

480 19.0 14 43 9 1.0

a The net average and standard deviation is unique for each time measurement, so 3σ varies across the experiment. A list of bursting cells, denoted by arbitrary cell number, is included in the Supporting Information.

The problem with the average responses is that such data mask some of the more interesting details of cellular activity, including distribution profiles, dispersion, stochastic volatility, and the appearance of discrete instances of uncharacteristic individual behavior. To reveal the bursting behavior, data analysis was performed for individual time points throughout each experiment. Furthermore, to characterize the level and extent of bursting behavior, the individual net fluorescence increases and standard deviation at each time point were calculated. Because a Gaussian (normal) distribution is defined by the standard deviation, where 3σ defines ∼99.7% of the population, any cell (at any single time point) with net activity increases greater than described by a 3σ Gaussian distribution was considered as exhibiting bursting activity (Table 1). For

example, the data in Table 1 show that at 150 min into the assay, the average net increase over time 120 min was 4.7 ± 3.3 arbitrary fluorescence units (afu, as net corrected values). Sixtythree cells out of the 879 total, or 7.2% of the population, exhibited activity greater than 9.7 counts (3σ). When each time point was considered individually, the range of cells exhibiting bursting activity was from below 1% to 12%. However, the stochasticity of the population is apparent when it is considered that the bursting occurs randomly throughout the course of the experiment; if cellular activity is considered across all time points, bursting behavior is observed at least once in a total of 415 cells (47.2% of the population). A table that lists the cells exhibiting bursting at each time point is included in the Supporting Information. 2741

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

observed that spike above the general collective responses. Figure 5b shows smaller subsets of the population, including shaded areas denoting the 3σ thresholds. These data demonstrate the sheer heterogeneity found within the measured activity in the population of cells. However, reagent addition did influence cellular activity. While we observe bursting activity at each measurement, in the measurements following addition of fresh reagent, larger-than-normal increases in the average net signal were observed, along with an increased number of cells that showed bursting activity, even with their substantially larger 3σ thresholds. Nonetheless, taken together, the extreme variability in cellular response shows that both the frequency and magnitude of gene expression bursts can be quantitated and cellular activity can be categorized as highly variable and random. We define bursting quantitatively as an uncharacteristic increase in activity of an individual cell between two consecutive measurements. Individual cells displayed differential levels of stochastic volatility, with some displaying clearly evident bursts of protein production, while other cells would appear to be producing protein more consistently over time. What is uncertain is whether the consistent production of protein is a singular phenomenon or it is the manifestation of frequent, more regular random bursts of gene expression. It has been suggested that, in bacteria, bursting behavior with lacZ reporter is related to stochastic dissociation of the Lac repressor from the promoter, leading to variable transcription of mRNA and eventually protein levels.29 This natural feedback loop and the concomitant stochasticity associated with the lacZ response in bacteria are avoided by incorporating the gene via homologous displacement into yeast. As a single chromosomal copy of a synthetic gene, our model system is otherwise devoid of the complete Lac operon, and so does not fall under the activation/repression stochasticity previously described. The results herein characterize a population of genetically identical cells, providing a relative quantitation of an expressed gene product. Statistical parameters are provided that were used to identify and characterize peculiar cellular activity and recognize individual cells with outlier behavior. These results ultimately confirm that the aggregate stochasticity of the entire population may be more disproportionally influenced by cells exhibiting outlier behavior. It may be that as the number of cells with extreme behavior increases, the possibility of the population exhibiting deviant behavior, either positive or negative, is more probable. What is yet to be determined is the critical threshold that would serve as a “cut-off point” that differentially affects population activity, which might vary from experiment to experiment based on cell type and other variables such as substrate, concentration, temperature, etc. However, if a phenomenon exists that manifests as deviant or abnormal behavior, it is likely to have quantitative characteristics that can be identified as noticeable changes in individual activity. The extreme instance would be when aggregate individual activities lead to the existence of quantitatively different subpopulations, which lead to unique phenotypic behavior.

Overall, striking variations were observed related to increased activity or bursting behavior between cells in the population. Throughout the course of the experiment, the majority of cells (464/879 or 52.8%) did not show activity above this 3σ threshold. Some cells exhibited a discrete, single instance of bursting, surrounded by periods of low activity, while other cells showed random, multiple instances of bursting, flanked by variable, intermittent levels of lower, less pronounced activity. Finally, a small subset of cells showed prolonged, continuous increases in fluorescence over successive measurements. For a specific example, cell 357 increased its net fluorescence intensity a total of 48 afu between the measurements taken at 270 and 300 min, which represents the highest single intensity increase of any cell in the population between these two time points. However, cell 357 did not exhibit bursting behavior at any other time point in the experiment. Interestingly, a total of 268 cells showed bursting behavior only at a single time point, which relates to 65% of all cells within the population exhibiting bursting behavior (Table 2). While these data show the large Table 2. Summary of Bursting Behavior of 879 Cells Exposed to 14.5 μM C12FDGa

cells analyzed, 14.5 μM C12FDG cells without bursting behavior cells with bursting behavior cells with a single instance of bursting cells with multiple instances of bursting cells with 2 instances of bursting cells with 3+ instances of bursting cells with 4+ instances of bursting cells with 5+ instances of bursting cells with 9 instances of bursting cells with consecutive (sustained) bursts

total no.

% of population

% of bursting cells

879 464 415 268 of 415

100 52.8 47.2 30.5

100 64.6

147 of 415

16.7

35.4

89 of 415 58 of 415 20 of 415 8 of 415 2 of 415 46 of 147

10.1 6.6 2.3 0.9 0.2 5.2

21.4 14.0 4.8 1.9 0.5 11.1

a

Includes the number of cells exhibiting multiple instances of bursting and those with sustained bursting activity (showing bursting behavior over consecutive time periods).

majority of bursting activity occurred within the context of only a single instance, a more detailed examination of the remaining cells that exhibited bursting behavior is informative. A total of 147 cells (35% of the bursting population) exhibited bursting behavior at two or more instances. Of this 147 cells, 89 cells exhibited exactly two instances of bursting over the course of the experiment. However, of the cells exhibiting two instances of bursting, only 15 of these cells exhibited bursting activity in consecutive time points (sustained activity for over 1 h). The remaining 74 cells demonstrate two discrete instances of bursting activity; at two different, unconnected time points their activity increases >3σ in relation to the population. Conversely, there are 20 cells (2% of the population) that exhibit bursting activity four or more times out of the 15 measurements, and of these 20 cells, there are three cells that show sustained bursting activity over more than half of the measurements (8 and 9 measurements). The randomness of the bursting phenomenon is depicted in Figure 5, as plots of net fluorescence responses over time. Figure 5a shows the net response of 135 cells, and at each time point cell responses are



CONCLUSIONS Variations in gene expression products in single cells across a clonal cell population are influenced by the inherent stochasticity in these processes, and in the extreme may manifest as phenotypic heterogeneity among isogenic cells. This type of heterogeneity may influence cellular activity, response, and ultimately population evolution. As such, 2742

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

Figure 5. Individual cell responses showing some discrete levels of bursting activity. (a) Single incidences of bursting activity (135 random cells shown), characterized as the differential increase in net fluorescence intensity between each 30 min measurement. (b) Cells showing one instance of bursting (26 cells shown) and with 3σ shading, and cells showing four or more instances of bursting (10 cells shown) and with 3σ shading.

population methods are flawed for such analyses; averaged population measurements mask inherent cell−cell heterogeneity and limit the ability to differentiate between individual temporal responses. In contrast, live cell arrays have enabled the continuous analysis of hundreds of individual cells, providing real-time population “snapshots” of cellular activity, which allow quantitative characterization of the extent of bursting behavior in an otherwise homogeneous environment.

characterizing the relationship between noise in gene expression and the expression variation of a cell population will improve our understanding of fundamental cellular processes. Studying the subtleties of gene networks within individual cells is a critical approach to synthetic biology. Cellular microarray devices that are able to localize individual cells and/or small populations of cells for extended time measurements are ideal for such investigations. Traditional bulk 2743

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744

Analytical Chemistry

Article

(14) Swain, P. S.; Elowitz, M. B.; Siggia, E. D. Proc. Natl. Acad. Sci. U.S.A. 2002, 99, 12795−12800. (15) Blake, W. J.; Balázsi, G.; Kohanski, M. A.; Isaacs, F. J.; Murphy, K. F.; Kuang, Y.; Cantor, C. R.; Walt, D. R.; Collins, J. J. Mol. Cell 2006, 24, 853−865. (16) Cai, L.; Dalal, C. K.; Elowitz, M. B. Nature 2008, 455, 485−490. (17) Tan, R. Z.; van Oudenaarden, A. Mol. Syst. Biol. 2010, 6, 1−7. (18) Golding, I.; Paulsson, J.; Zawilski, S. M.; Cox, E. C. Cell 2005, 123, 1025−1036. (19) Chubb, J. R.; Trcek, T.; Shenoy, S. M.; Singer, R. H. Curr. Biol. 2006, 16, 1018−1025. (20) Franco, E.; Friedrichs, E.; Kim, J.; Jungmann, R.; Murray, R.; Winfree, E.; Simmel, F. C. Proc. Natl. Acad. Sci. U.S.A. 2011, 108, E784−E793. (21) Levsky, J. M.; Shenoy, S. M.; Pezo, R. C.; Singer, R. H. Science 2002, 297, 836−840. (22) Biran, I.; Walt, D. R. Anal. Chem. 2002, 74, 3046−3054. (23) Zhang, Y.; Naleway, J.; Larison, K.; Huang, Z.; Haugland, R. FASEB J. 1991, 5, 3108−3113. (24) Wang, S.; Shepard, J. R. E.; Shi, H. Nucleic Acids Res. 2010, 38, 2378−2386. (25) Walling, M. A.; Wang, S.; Shi, H.; Shepard, J. R. E. Anal. Bioanal. Chem. 2010, 398, 1263−1271. (26) Bernstein, D. S.; Buter, N.; Stumpf, C.; Wickens, M. Methods Mol. Biol. 2002, 26, 123−141. (27) Buskirk, A. R.; Kehayova, P. D.; Landrigan, A.; Liu, D. R. Chem. Biol. 2003, 10, 533−540. (28) Deutsch, M.; Deutsch, A.; Shirihai, O.; Hurevich, I.; Afrimzon, E.; Shafran, Y.; Zurgil, N. Lab Chip 2006, 6, 995−1000. (29) Cai, L.; Friedman, N.; Xie, X. S. Nature 2006, 440, 358−362. (30) Zuker, M. Nucleic Acids Res. 2003, 31, 3406−3415.

By monitoring the cells continuously over time, discrete cellular trends were identified that are particularly useful in detailing processes that ultimately relate to the underlying biological phenomena. Herein, a method and a statistical approach was presented for determining the stochasticity of cellular activity that provides a more characteristic description of the phenomenon of cellular bursting. Bursting behavior was evident throughout the temporal response of individual cells, manifesting as discrete, random increases of gene expression, ultimately leading to a wide range in protein levels. The number of cells manifesting a burst in activity, as well as the frequency and extent of such behavior, was described, with bursting behavior observed in close to half the population. This phenomenon occurs at a surprisingly substantial level, suggesting that a regular, random portion of the population exhibits bursting at any single time point.



ASSOCIATED CONTENT

S Supporting Information *

One table, listing cells exhibiting bursting across the 8-h experiment (PDF), and two figures, showing images of the live cell array and selected images from our experimental design (PDF). This material is available free of charge via the Internet at http://pubs.acs.org.

■ ■

AUTHOR INFORMATION

Notes

The authors declare no competing financial interest.

ACKNOWLEDGMENTS We thank the University at Albany, State University of New York, for the funds to support this project. H.S. also acknowledges a Concept Award from the U.S. Department of Defense (BC075466 to H.S.). The authors would like to thank Dr. Jeff Ault and the Electron Microscopy Core of The Wadsworth Center, New York State Department of Health, Albany, NY who performed the scanning electron microscopy for the image in Figure 2.



REFERENCES

(1) McAdams, H. H.; Arkin, A. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 814−819. (2) Xie, X. S.; Choi, P. J.; Li, G. W.; Lee, N. K.; Lia, G. Annu. Rev. Biophys. 2008, 37, 417−444. (3) Thattai, M.; van Oudenaarden, A. Proc. Natl. Acad. Sci. U.S.A. 2001, 98, 8614−8619. (4) Arkin, A.; Ross, J.; McAdams, H. H. Genetics 1998, 149, 1633− 1648. (5) Balaban, N. Q.; Merrin, J.; Chait, R.; Kowalik, L.; Leibler, S. Science 2004, 305, 1622−1625. (6) Bishop, A. L.; Rab, F. A.; Sumner, E. R.; Avery, S. V. Mol. Microbiol. 2007, 63, 507−520. (7) Burnett, J. C.; Miller-Jensen, K.; Shah, P. S.; Arkin, A. P.; Schaffer, D. V. PLoS Pathog. 2009, 5, No. e1000260. (8) Walling, M. A.; Shepard, J. R. E. Chem. Soc. Rev. 2011, 40, 4049− 4076. (9) Raser, J. M.; O’Shea, E. K. Science 2005, 309, 2010−2013. (10) Raj, A.; Peskin, C. S.; Tranchina, D.; Vargas, D. Y.; Tyagi, S. PLoS Biol. 2006, 4, No. e309. (11) Acar, M.; Mettetal, J. T.; van Oudenaarden, A. Nat. Genet. 2008, 40, 471−475. (12) Elowitz, M. B.; Levine, A. J.; Siggia, E. D.; Swain, P. S. Science 2002, 297, 1183−1186. (13) Weinberger, L. S.; Burnett, J. C.; Toettcher, J. E.; Arkin, A. P.; Schaffer, D. V. Cell 2005, 122, 169−182. 2744

dx.doi.org/10.1021/ac300344n | Anal. Chem. 2012, 84, 2737−2744