• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Apr 12, 2011; 108(15): 6329–6334.
Published online Mar 28, 2011. doi:  10.1073/pnas.1013148108
PMCID: PMC3076844
Systems Biology

Exploring transcription regulation through cell-to-cell variability

Abstract

The regulation of cellular protein levels is a complex process involving many regulatory mechanisms, each introducing stochastic events, leading to variability of protein levels between isogenic cells. Previous studies have shown that perturbing genes involved in transcription regulation affects the amount of cell-to-cell variability in protein levels, but to date there has been no systematic characterization of variability in expression as a phenotype. In this research, we use single-cell expression levels of two fluorescent reporters driven by two different promoters under a wide range of genetic perturbations in Saccharomyces cerevisiae, to identify proteins that affect variability in the expression of these reporters. We introduce computational methodology to determine the variability caused by each perturbation and distinguish between global variability, which affects both reporters in a coordinated manner (e.g., due to cell size variability), and local variability, which affects the individual reporters independently (e.g., due to stochastic events in transcription initiation). Classifying genes by their variability phenotype (the effect of their deletion on reporter variability) identifies functionally coherent groups, which broadly correlate with the different stages of transcriptional regulation. Specifically, we find that most processes whose perturbation affects global variability are related to protein synthesis, protein transport, and cell morphology, whereas most processes whose perturbations affect local variability are related to DNA maintenance, chromatin regulation, and RNA synthesis. Moreover, we demonstrate that the variability phenotypes of different protein complexes provide insights into their cellular functions. Our results establish the utility of variability phenotype for dissecting the regulatory mechanisms involved in gene expression.

Keywords: genetic screen, transcriptional noise, yeast

Understanding the cellular networks that regulate gene expression levels is an important goal in molecular biology. Traditionally, these networks were studied by measuring average gene expression levels in a population of cells. Recent technological advances have made it possible to measure protein levels reliably in single cells at a large scale. These assays allow identification of more subtle phenotypes, such as changes in cell-to-cell variability in gene expression that are not necessarily portrayed in the population mean. Recent experimental results show that the same mean protein levels can be achieved with very different variability (1, 2). For example, a gene whose promoter activation is slow and unstable, but produces many mRNA copies upon each activation event, might have the same mean protein level as a gene whose promoter has a slow transcription rate but is constantly active. The variability in expression levels of the first gene will be higher, as small differences in the number of activation events will propagate to larger fluctuations in mRNA copies and consequently to larger differences in protein levels (35).

Variability in protein levels (also referred to as noise) is driven by two fundamentally different sources, often referred to as intrinsic and extrinsic variability (6). The former, arising from the stochastic nature of the biochemical processes involved in transcription regulation, translation, and degradation, affects the expression of a protein encoded by a particular gene. The latter, affecting to a certain extent the expression of all proteins in the cell, arises from heterogeneities within a cell population that have broad effects on the protein levels. These heterogeneities include cell size, stage of cell cycle, and the concentration of the molecular machines involved in transcription and translation such as ribosomes, RNA polymerase, and RNA/protein degradation pathways.

Distinguishing between these very different sources of variability is essential for analyzing the observed reporter variability and understanding the mechanistic basis for such observations. However, disentangling the sources of variation is not trivial. One approach is to reduce extrinsic variability by constraining the analysis to a relatively homogenous subpopulation of cells (1, 2). An alternative approach is the dual-reporter system in which the expression of two reporter genes driven by identical promoters is measured simultaneously (4, 6, 7). This experimental approach decouples factors that affect both reporters in a correlated manner from those that affect them in an uncorrelated manner. Whereas the former is experimentally simpler, but forgoes extrinsic variations, the latter requires a specifically engineered experimental system.

Theoretical analysis of intrinsic variability shows that changing the rates of promoter activation, transcription, translation, and degradation results in a distinct influence on expression mean and variability (8, 9). These analytical predictions are consistent with observations of mean and variability under perturbation of key elements in these processes, including the chromatin remodeling complexes SAGA, INO80, and SWI/SNF (4); TATA-box sequence mutations (4, 10); MAP kinases (11); and transcriptional elongation factors (12). However, to date, no large-scale assay has been performed to systematically identify the genes and their products that affect variability. A notable exception is the use of reporter variability as a trait in a genome-wide association study (12). This approach, however, is limited to the range of genetic polymorphisms present in the parent strains.

Here, we use cell-to-cell variability of protein expression levels as a phenotype for a genetic screen. We use existing single-cell data of promoter activity reporter levels under different genetic perturbations (13) as a systematic screen for genes whose deletions affect expression variability of two reporters driven by two different promoters. We introduce a computational method to extract the expression variability phenotype of each perturbation and decompose its sources. We show how variability phenotypes allow us to identify cellular components that affect the expression of the two reporters and to gain insights into common and distinct mechanisms acting on the two promoters driving these reporters.

Results

Extracting Variability Phenotype from Single-Cell Measurements.

To systematically define the variability phenotype of strains carrying gene deletions, we used single-cell flow cytometry measurements of two fluorescent proteins regulated by two different Saccharomyces cerevisiae promoters: a high-expression, constitutive TEF2 promoter driving red fluorescent protein (RFP) and a synthetic enhancer that includes four unfolded protein response elements (UPRE) coupled to the CYC1 promoter driving GFP (13) (Fig. 1A). This synthetic promoter is designed to respond to unfolded protein response (UPR), a transcriptional response mediated by the Hac1 transcription factor to unfolded protein stress in the endoplasmic reticulum (ER). The original study (13) used these reporters to evaluate levels of unfolded protein stress in a genetic screen for functional ER components.

Fig. 1.
Measuring cell-to-cell variability. (A) A reporter system with red and green fluorescent proteins fused to a TEF2 promoter and an artificial promoter with UPR elements, respectively, was introduced into strains from the yeast deletion library (13). Flow ...

We next estimated reporter variability in a cell population. To understand the mechanistic causes for the observed variability, we must differentiate between sources of variability. Because the experimental system is based on two reporters, each driven by a very different promoter, we cannot assume that they have identical intrinsic variability and thus cannot use the dual-reporter assay analysis (6, 7). Moreover, we also had morphological measurements for each cell, which capture some of the heterogeneity in the population. We defined two sources of variability that we can expect to distinguish with these measurements. Global variability is defined as variability due to heterogeneity in general properties of the cell that affect both reporters (e.g., morphology and ribosome density). Local variability is defined as variability due to events that affect a specific reporter gene. These definitions are similar, but not identical, to intrinsic and extrinsic variability (6, 7) and are operational on the measurements available to us. Local variability aims to mostly capture intrinsic variability, although this is not guaranteed to be the case. For example, because the two promoters are different, specific regulatory events (e.g., ER stress in our case) can affect only one reporter and will be accounted for by local variability of that reporter.

There are two indicators of global variability: (i) coordinated changes of the two reporters in the cell population (also used in dual-reporter analysis) (6, 7) and (ii) coordinated changes of one reporter with cell morphology. For example, when reporter levels are correlated with cell size, we can attribute the variability in reporter levels to morphological heterogeneity rather than to reporter-specific events. In such a case, we would attribute this variability to global events.

We capture these intuitions with concise mathematical definitions. Assume that we have access to a probability distribution over cell morphology (M), green fluorescence intensity (G), and red fluorescence intensity (R) in cells with a specific genotype under defined experimental conditions. Using this distribution, we define the local variability of G as its conditional variance after we account for observation of R and M:

equation image

By conditioning on R and M we remove variability explained by the RFP reporter and by cell morphology, respectively. We define the global variability of G as the variability of the prediction of the expected value of G given values of R and M:

equation image

Intuitively, because the predictions are based on global factors, variance in the predictions is a result of variance in the global factors. Both these variances are taken over all of the cells in the population (SI Text). The variability of G is equal to the sum of local and global variability, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i1.jpg (SI Text). Moreover, we can identify what part of the global variability is accounted by the morphology alone,

equation image

The remaining variability An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i2.jpg captures the additional improvement in predicting G using R over the prediction from the morphology (SI Text). We define the variability terms of R in a symmetric manner.

To apply these definitions we need to approximate a multivariate distribution over M, G, and R from the observed data. As indicators of cell morphology we use forward-scatter (FSC) and side-scatter (SSC) measurements, which roughly correspond to cell size and granularity (14). We compared different approaches for representing the joint distribution and found that using a multivariate Gaussian over the logarithm of flow cytometery measurements leads to estimates similar to nonparametric regression (SI Text). We thus fit a multivariate Gaussian distribution to the logarithm of flow cytometry measurements of GFP, RFP, FSC, and SSC levels (Fig. 1B) for each cell population. We then evaluate the variability terms on the basis of this multivariate distribution (Fig. 1C, Materials and Methods, and SI Text) and report these values as coefficients of variation (CV, SD/mean). Comparison of our method with previous approaches shows good agreement in the relevant cases (SI Text, Fig. S1).

Systematic Genetic Screen for Variability Phenotype Across All Yeast Deletion Strains.

We applied our method to the data collected by Jonikas et al. (13), who measured protein levels expressed by the two reporters introduced into each strain within the entire yeast deletion library. We assigned to each strain local and global variability estimates for each reporter (Figs. 1D and and2).2). For both reporters we find that the global variability is dominated by morphological variability (average percentage of global variation accounted for by morphological variation 99% and 93% for RFP and GFP, respectively; Fig. S2 A and B).

Fig. 2.
Genetic screen for deletions that affect variability. (A–D) Scatter plots showing the reporter's coefficient of variation (y axis) against the reporter's mean expression level (x axis) for all strains (dots). Outlier strains are solid dots. A ...

The measurements for each strain have been repeated up to six times (Fig. S3A). We compared repeats of the same deletion strain and found good agreement for the CV estimates for the GFP reporter (Pearson correlation of 0.87 and 0.65 for local and global variability estimates, respectively; Fig S3 C and E). The RFP reporter CV estimates had a lower correlation (Pearson correlation of 0.14 and 0.64 for local and global variability estimates, respectively; Fig. S3 B and D). We believe that this lower agreement is a result of low levels of local variability in this reporter.

Theoretical and empirical results (1, 2, 4, 5) show that local variability (CV) levels depend on mean expression levels, as also seen in our results (Fig. 2 A and C). Thus, to eliminate such dependencies, we defined the CV residue of a strain as the difference between the observed and expected CVs for strains with similar mean expression levels (Fig. 2D, Inset, and Materials and Methods).

Variability Phenotype Is Distinct from Mean Expression Phenotype.

The dataset we used was designed as a screen to uncover genes that when deleted affect the mean expression of the GFP relative to a wild-type control (13). We asked if we could gain additional information from the variability of expression phenotype. Although variability tracks mean expression levels to a certain extent, there are large deviations from this trend. Namely, many deletion strains exhibit divergent variability (either local or global) despite having similar mean protein levels. For example, deletions of BEM2 and MSK1 result in a similar increase in mean GFP levels (10.4 and 10.33, respectively) but have markedly different effects on global variability for both reporters. Deletion of BEM2, a Rho GTPase activator involved in cytoskeleton organization, cellular morphogenesis, and budding (15), results in high global variability (GFP global CV = 0.08). On the other hand, deletion of MSK1, mitochondrial lysine-tRNA synthetase (16), leads to low global variability (GFP global CV = 0.04). This observation suggests that much of the variability in bem2Δ is due to high variance in cell state, whereas msk1Δ has a more homogenous population. Indeed, previous works show that the perturbation of BEM2 leads to increased morphological variability (17, 18) and that msk1Δ strains suffer from lack of aerobic respiration resulting in a relatively homogenous population of small cells (18). Similar examples are abundant also for the local variability phenotype.

To systematically examine the differences between these mean and variability phenotypes, and their overlap, we characterized the genes with an extreme phenotype in each category. We define outlier strains as ones where the CV residue is >2 SDs from the average (Fig. 2) in at least half of the repeats. This results in 539 genes that are outliers in at least one of four variability phenotypes (global and local variability for each of the reporters; Fig. 2, Materials and Methods, and Dataset S1). Similarly, we identified 367 genes with outlier mean protein levels (Materials and Methods and Dataset S1). Of these, 148 genes have both an extreme mean and variability phenotype. To understand whether the phenotypes highlight different cellular functions, we tested for GO categories and protein complexes (19, 20) that are enriched with genes in either phenotype. From 1,337 distinct gene sets we examined (Materials and Methods), 50 categories were significantly enriched in genes with a mean phenotype and 50 categories with variability phenotype (Dataset S1). Of these, 10 categories were enriched in both phenotypes. Thus, whereas some gene categories affect both phenotypes, each phenotype also captures a different aspect of cellular behavior.

Protein Expression Pathway and Cell Morphology Are the Most Dominant Processes Affecting Variability.

The GO categories that are enriched with variability phenotype genes can be characterized into cell morphology, DNA maintenance, and the different stages of gene expression: chromatin remodeling, transcription, protein synthesis, and protein maturation (Fig. 3). Annotations related to protein synthesis, transport, and maturation are enriched in genes whose deletions increase global variability. Notably, most ribosomal proteins were discarded from our analysis because their deletion decreases mean expression to the level of autofluorescence, and thus we could not assess the effect of their deletion on variability. The effects of deletions of genes associated with chromatin organization and transcription are more complex. For example, deletions of subunits of the SWR1 complex affected both local and global variability but in opposite directions. Whereas there are relatively few GO categories enriched in local RFP variability, there were many such annotations enriched in local GFP variability, mainly related to DNA, chromatin maintenance, and organization. Thus, we next asked if the screen provides clues about the different behavior of the promoters driving these two reporters.

Fig. 3.
Characterization of major functions affecting variability. Shown are the major processes that affect variability and a partial list of significantly enriched GO categories from these processes. Arrows indicate categories that are significantly enriched ...

Local Variability Phenotype Highlights Promoter-Specific Regulatory Mechanisms.

Comparing the variability phenotypes of genes shows that although global GFP and RFP variability phenotypes are mostly similar, there is much less agreement in local variability phenotypes (Fig. S4). Additionally, the reporters differ dramatically in the contribution of local variability to the total variability with average percent variance explained (PVE) of 83% for GFP and only 30% for RFP (Fig. S2 C and D). Importantly, it is possible for a perturbation to increase the variability of the two reporters without introducing correlation between them. Thus, this lack of agreement does not directly follow the definition of local variability and might be a result of the different properties of the two promoters: The TEF2 promoter is constitutively expressed, whereas the UPRE promoter is activated in response to signal transduction by the UPR. Accordingly, transcription of the two reporters involves many shared components, such as the basic transcription machinery and general transcription factors, but each promoter is regulated by distinct transcription factors and, potentially, different chromatin remodelers. Our premise is that complexes whose perturbation affects local variability in a coherent manner in both promoters are part of a common transcriptional apparatus in contrast to those that affect only one promoter or affect both promoters in an opposite manner.

To identify regulatory functions affecting variability in a robust manner, we screened for protein complexes (20) that have a coherent effect on local variability (Materials and Methods). We found that 22/198 protein complexes affect local variability significantly (Materials and Methods, Dataset S1, and Fig. 4). Comparing the effects of these complexes on the two reporters, we found that although some complexes affect local variability in the same direction in both reporters (e.g., mitochondrial ribosome), most have a significant effect on variability only for one reporter [e.g., the chromatin assembly factor 1 (CAF1) complex]. To understand how this comparison can teach us both about the role of these complexes in transcription regulation and about the difference in regulation programs of the two promoters, we examined the phenotypic effect of perturbing the CAF1 complex in more detail.

Fig. 4.
Complexes enriched for local variability phenotype. Each complex is shown as a gray dot with the mean local RFP CV residue (x axis) and mean local GFP CV residue (y axis). Complexes with a significant local variability phenotype (5% FDR) are indicated ...

Disparate Phenotypes of CAF1 Complex Are Indicative of Differences in Promoter Architectures.

The CAF1 complex has a distinctly different effect on local variability of the two reporters. CAF1 functions in association with Asf1 in deposition of histones H3/H4 during the assembly of chromatin following DNA replication (21). We find that deletions of each of the three subunits of CAF1, as well as of one of the two copies of genes encoding histones H3 and H4, result in low local GFP variability and in an increase in mean GFP expression (Fig. 5A). In addition, deletion of RTT106, a histone chaperone that functions in histone deposition and physically interacts with CAF1 (22), has a similar phenotype (Fig. 5A). On the other hand, these deletions cause only a slight increase in mean RFP expression and almost no change to local RFP variability (Fig. S5).

Fig. 5.
CAF1 perturbations reduce local variability in a UPRE-specific manner. (A) Scatter plot of GFP mean expression (x axis) and local variability (y axis) for all strains (gray and colored dots). Also shown is the LOWESS regression line (red). Deletion strains ...

We reasoned that there are two possible ways by which this UPRE-specific phenotype is achieved: indirectly by affecting the UPR or directly on the promoter (Fig. 5B). Loss of CAF1 activity results in underassembly of chromatin, defects in silencing, and genome instability (23). Thus, one hypothesis is that loss of CAF1 results in overproduction of proteins, increasing ER load. ER load activates Hac1-driven UPR transcriptional response (24), which would lead to the induction of the UPRE-driven GFP reporter. This hypothesis can explain the increase in GFP mean and the decrease in local variability (2, 4). Furthermore, because the TEF2 promoter is insensitive to ER stress, this hypothesis is also consistent with the lack of RFP phenotype. This hypothesis predicts that CAF1 deletions will behave similarly to other deletions that increase ER stress, such as the deletion of components of the pathway of ER-associated degradation of misfolded proteins (ERAD-M) (25) (Fig. 5B and Fig. S6A). Hac1 mediates the ER stress response; thus, we expect effects of ER stress to be buffered by deletion of HAC1. Indeed, deletions of ERAD-M components under hac1Δ background behave similarly to hac1Δ deletions, showing full epitasis (Fig. S6B). In contrast, we find that deletions of CAF1, RTT106, and histones H3/H4 under hac1Δ background still lead to reduced GFP local variability and increased mean expression levels (Fig. 5C). We thus conclude that the CAF1 effect is not UPR dependent and is likely to be a direct effect on the promoter.

The most likely explanation for the direct effect of CAF1 mutations on the UPRE promoter is that chromatin underassembly results in a more open promoter state and thus higher rate of HAC1-independent transcription. Whereas some promoter architectures require nucleosome remodeling to open them as a step in transcriptional induction, others are inherently open and thus less sensitive to chromatin underassembly (26, 27). In this case the lack of RFP phenotype can be explained if it is an inherently open promoter. Indeed, experimental evidence shows that the TEF2 promoter is nucleosome depleted (28). We thus reason that chromatin remodeling by CAF1 is not rate limiting in RFP expression, explaining why CAF1 deletion has little effect on RFP phenotypes. In contrast, the GFP phenotypes suggest that chromatin remodeling is rate limiting at the GFP promoter and thus that it is a covered promoter.

Variability Phenotype Implicates Elongator Complex Function in tRNA Modification.

Deletion of genes encoding components of the same protein complex or cellular function often (204/1,337) have a coherent effect on variability (Materials and Methods and Dataset S1). This result suggests that variability phenotype can elucidate protein function through “guilt by association.” As an illustration of this principle, we examine the function of the Elongator complex. There is an ongoing debate in the literature about the function(s) of the Elongator complex. Originally implicated in transcription elongation (29), it has since been suggested to play many additional roles, the most accepted being tRNA modification (30). In our analysis, this complex has distinctly high local and low local variability phenotypes (Fig. 6). We reasoned that examination of variability phenotypes of proteins with similar functions would provide clues about the role of the Elongator complex. Indeed, proteins involved in tRNA modification, such as KTI12 and NCS6, have variability phenotypes similar to Elongator proteins (Fig. 6). In contrast, most proteins involved in transcription elongation decrease global variability (Fig. 6). These results suggest that Elongator's function in tRNA modification is responsible for the distinctive variability we observe.

Fig. 6.
Variability phenotype of the Elongator complex is similar to genes involved in tRNA modification. Scatter plot is shown of GFP local CV residues (x axis) and global CV residues (y axis) for all strains (gray and colored dots). Deletion strains of Elongator ...

Discussion

Here we analyzed a large-scale dataset to identify genes that when deleted affect variability in gene expression. We developed methodology that uses single-cell measurements of reporter protein levels and morphology to estimate variability. Our method distinguishes between local variability, which cannot be predicted by other cell attributes (and thus is mostly due to stochastic events in the life cycle of proteins produced from a specific loci), and global variability, which is correlated to cell attributes (and mostly arises from heterogeneity in the cell population). We found the strongest local variability phenotype in genes involved in chromatin maintenance, transcriptional regulation, and transcription, whereas the strongest global variability phenotype appeared in a wide range of functions, including genes involved in protein synthesis, protein transport, and cell morphology.

Using the variability phenotype, we uncovered genes involved in regulating protein levels that do not have a mean expression phenotype. For example, the Elongator complex does not have any mean expression phenotype, but has a distinct variability phenotype (Fig. S7). Moreover, the mean expression level and variability are complementary phenotypes to our understanding of cellular processes. For example, the observation that a deletion results in both elevating the mean expression and decreasing local variability (e.g., deletion of CAF1 components) suggests an increase in the rate of transcription events (4).

Many complexes and functional groups have a coherent variability phenotype, suggesting that variability phenotypes can help elucidate the functional role of proteins. In some cases these phenotypes match a known function (e.g., CAF-1 complex), and for others the phenotype can elucidate the biological function (e.g., Elongator complex). Moreover, global variability in gene expression also leads to insights about protein function, and thus it is advantageous to distinguish between local and global variability types rather than simply eliminating global variability.

Measuring the effect of a gene deletion on the variability of two different reporters simultaneously in individual cells enables a comparative analysis that provides insights into the mechanisms of transcription regulation at the promoters driving these reporters. Here, we compared reporters driven by two very different promoters, one driving a constitutive gene and the other an artificial promoter induced by stress conditions. The difference between these promoters is evident from our results, as many of the perturbations have a distinctly different effect on the two reporters. Our approach is applicable to the study of any combination of promoters from two that are very different to two that are identical, laying the basis for a comparative examination of various transcription regulation programs. Taken together, our results establish variability as a highly informative phenotype for genetic dissection of gene expression.

Materials and Methods

Data.

We used flow cytometry data collected by Jonikas et al. (13). Strains were grown in YPD on 384-well plates. We discarded all strains and cells that did not pass quality control according to Jonikas et al. (13) and all cells that reached maximum value (i.e., 218) in one of the channels. Additionally, for each well we discarded cells with zero or negative fluorescent values after correction for autofluorescence (next section) and outlier cells of the top and bottom 1% of each channel. In our analysis we considered only the 6,554 wells that had >500 cells, representing 3,798 different deletion strains.

Correcting for Autofluorescence.

Yeast cells have relatively high GFP autofluorescence in the conditions in which this experiment was performed. Thus, the base level of GFP expression can affect the estimates of variability levels, especially in strains that have low GFP expression levels. Because the original dataset does not contain measurements of GFP-negative strains in the same experimental conditions, we could not directly measure the effect of autofluorescence. Therefore, we took a cautious approach and assumed that the fluorescent signal in strains with the lowest levels of GFP expression and variability is entirely due to autofluorescence. Measurements of such GFP-negative strains in our hands (SI Text and Fig. S8) showed that the autofluorescence levels depend linearly on the FSC and SSC values. Thus, we used a linear regression to learn the dependency of autofluorescence and forward and side scatter in these low-expression strains. Using these parameters we estimated the autofluorescence in each cell. For the rest of the analysis we defined the GFP values as the observed fluorescence minus the estimated autofluorescence. We discarded all cells in which all of the fluorescent signal is attributed to autofluorescence and disregard wells in which >20% of the cells were discarded.

Method for Estimating Local and Global Variability.

For each well, we estimated a multivariate Gaussian distribution of the logarithm of measurements of (autofluorescence corrected) GFP, RFP, and forward and side scatter (SI Text). We then applied the definition of local, global, and morphological variance to estimate the variance terms An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i3.jpg, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i4.jpg, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i5.jpg, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i6.jpg, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i7.jpg, and An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i8.jpg for GFP and RFP, respectively. We converted these variance terms to CV units. For example, An external file that holds a picture, illustration, etc.
Object name is pnas.1013148108i9.jpg. We defined other CV terms similarly.

Defining Outliers.

A locally weighted linear regression of the average CV as a function of the mean was calculated using the Matlab “lowess” function with span 0.1. CV residues were calculated as the difference of the measured CV from the LOWESS estimate at that mean. Deletion strains for which at least half of the repeats of the CV residues fell >2 SDs away from the mean were considered outlier strains. For the identification of protein-level outliers we used the raw data without autofluorescence correction.

Functional Enrichments.

The full GO term hierarchy was downloaded from the Saccharomyces Genome Database on July 17, 2008 (19). Complexes were downloaded from the CYC2008 complexes database (20). These categories were united, and redundant categories, which agree on all genes in our dataset, were merged, resulting in 1,337 categories. We checked for enrichments of all eight outlier groups (high/low, local/global, GFP/RFP) in these categories, using a hypergeometric P value and using a 5% false discovery rate (FDR) multiple-hypothesis correction.

When screening for complexes and cellular functions that affect variability, we used Student's t test to determine if the CV residues of a complex or GO category are significantly different from wild type. To be unbiased by the number of times each deletion strain was measured, we used for each strain the average CV residue (next section). For each of the 198 complexes from the CYC2008 database and 1,337 GO categories we performed a two-sample, two-tailed t test (using a 5% FDR) against 44 repeated measurements of the wild-type strain.

Averaging Repeats.

Many (2,272 of 3,798) deletion strains were measured more then once. In computing the average CV residues of these strains we discarded repeats that were far apart in terms of mean GFP expression, as these are suspected to be unreliable. To this end we calculated the average mean GFP SD of all genes measured more then once. Repeats that were >3 SDs apart from all other repeats of the strain were discarded. This process resulted in discarding 68 repeats and 24 strains (for which all repeats were discarded).

Supplementary Material

Supporting Information:

Acknowledgments

We thank N. Barkai, J. Bergman, M. Jonikas, T. Kaplan, R. Kupferman, A. Novogrodski, J. Paulsson, O. J. Rando, A. Regev, A. Rahat, M. Schuldiner, M. Yassour, and an anonymous reviewer for discussions and useful comments on the manuscript. We also thank M. Jonikas, J. S. Weissman, and M. Schuldiner for making their raw data available. This work was supported by the Rudin Foundation (R.R.) and a European Research Council grant (to N.F.).

Footnotes

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1013148108/-/DCSupplemental.

References

1. Newman JR, et al. Single-cell proteomic analysis of S. cerevisiae reveals the architecture of biological noise. Nature. 2006;441:840–846. [PubMed]
2. Bar-Even A, et al. Noise in protein expression scales with natural protein abundance. Nat Genet. 2006;38:636–643. [PubMed]
3. Thattai M, van Oudenaarden A. Intrinsic noise in gene regulatory networks. Proc Natl Acad Sci USA. 2001;98:8614–8619. [PMC free article] [PubMed]
4. Raser JM, O'Shea EK. Control of stochasticity in eukaryotic gene expression. Science. 2004;304:1811–1814. [PMC free article] [PubMed]
5. Pedraza JM, Paulsson J. Effects of molecular memory and bursting on fluctuations in gene expression. Science. 2008;319:339–343. [PubMed]
6. Elowitz MB, Levine AJ, Siggia ED, Swain PS. Stochastic gene expression in a single cell. Science. 2002;297:1183–1186. [PubMed]
7. Swain PS, Elowitz MB, Siggia ED. Intrinsic and extrinsic contributions to stochasticity in gene expression. Proc Natl Acad Sci USA. 2002;99:12795–12800. [PMC free article] [PubMed]
8. Paulsson J. Summing up the noise in gene networks. Nature. 2004;427:415–418. [PubMed]
9. Munsky B, Nemenman I, Bel G. Specificity and completion time distributions of biochemical processes. J Chem Phys. 2009;131:235103. [PubMed]
10. Blake WJ, et al. Phenotypic consequences of promoter-mediated transcriptional noise. Mol Cell. 2006;24:853–865. [PubMed]
11. Colman-Lerner A, et al. Regulated cell-to-cell variation in a cell-fate decision system. Nature. 2005;437:699–706. [PubMed]
12. Ansel J, et al. Cell-to-cell stochastic variation in gene expression is a complex genetic trait. PLoS Genet. 2008;4:e1000049. [PMC free article] [PubMed]
13. Jonikas MC, et al. Comprehensive characterization of genes required for protein folding in the endoplasmic reticulum. Science. 2009;323:1693–1697. [PMC free article] [PubMed]
14. Cookson S, Ostroff N, Pang WL, Volfson D, Hasty J. Monitoring dynamics of single-cell gene expression over multiple cell cycles. Mol Syst Biol. 2005;1 2005.0024. [PMC free article] [PubMed]
15. Kim YJ, Francisco L, Chen GC, Marcotte E, Chan CS. Control of cellular morphogenesis by the Ip12/Bem2 GTPase-activating protein: Possible role of protein phosphorylation. J Cell Biol. 1994;127:1381–1394. [PMC free article] [PubMed]
16. Tarassov I, Entelis N, Martin RP. Mitochondrial import of a cytoplasmic lysine-tRNA in yeast is mediated by cooperation of cytoplasmic and mitochondrial lysyl-tRNA synthetases. EMBO J. 1995;14:3461–3471. [PMC free article] [PubMed]
17. Levy SF, Siegal ML. Network hubs buffer environmental variation in Saccharomyces cerevisiae. PLoS Biol. 2008;6:e264. [PMC free article] [PubMed]
18. Saito TL, et al. SCMD: Saccharomyces cerevisiae Morphological Database. Nucleic Acids Res. 2004;32(Database issue):D319–D322. [PMC free article] [PubMed]
19. Hong EL, et al. Gene Ontology annotations at SGD: New data sources and annotation methods. Nucleic Acids Res. 2008;36(Database issue):D577–D581. [PMC free article] [PubMed]
20. Pu S, Wong J, Turner B, Cho E, Wodak SJ. Up-to-date catalogues of yeast protein complexes. Nucleic Acids Res. 2009;37:825–831. [PMC free article] [PubMed]
21. Ramirez-Parra E, Gutierrez C. The many faces of chromatin assembly factor 1. Trends Plant Sci. 2007;12:570–576. [PubMed]
22. Huang S, et al. Rtt106p is a histone chaperone involved in heterochromatin-mediated silencing. Proc Natl Acad Sci USA. 2005;102:13410–13415. [PMC free article] [PubMed]
23. Adkins MW, Tyler JK. The histone chaperone Asf1p mediates global chromatin disassembly in vivo. J Biol Chem. 2004;279:52069–52074. [PubMed]
24. Ron D, Walter P. Signal integration in the endoplasmic reticulum unfolded protein response. Nat Rev Mol Cell Biol. 2007;8:519–529. [PubMed]
25. Friedlander R, Jarosch E, Urban J, Volkwein C, Sommer T. A regulatory link between ER-associated protein degradation and the unfolded-protein response. Nat Cell Biol. 2000;2:379–384. [PubMed]
26. Cairns BR. The logic of chromatin architecture and remodelling at promoters. Nature. 2009;461:193–198. [PubMed]
27. Tirosh I, Barkai N. Two strategies for gene regulation by promoter nucleosomes. Genome Res. 2008;18:1084–1091. [PMC free article] [PubMed]
28. De Sanctis V, et al. In vivo topography of Rap1p-DNA complex at Saccharomyces cerevisiae TEF2 UAS(RPG) during transcriptional regulation. J Mol Biol. 2002;318:333–349. [PubMed]
29. Otero G, et al. Elongator, a multisubunit component of a novel RNA polymerase II holoenzyme for transcriptional elongation. Mol Cell. 1999;3:109–118. [PubMed]
30. Svejstrup JQ. Elongator complex: How many roles does it play? Curr Opin Cell Biol. 2007;19:331–336. [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • MedGen
    MedGen
    Related information in MedGen
  • PubMed
    PubMed
    PubMed citations for these articles
  • Substance
    Substance
    PubChem Substance links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...