Negative Feedback Confers Mutational Robustness in Yeast Transcription Factor Regulation

Negative feedback confers mutational robustness in yeast transcription factor regulation

Charles M. Denbya, Joo Hyun Ima, Richard C. Yub, C. Gustavo Pesceb, and Rachel B. Brema,1

aDepartment of Molecular and Cell Biology, University of California, Berkeley, CA 94720-3220; and bMolecular Sciences Institute, Berkeley, CA 94704

Edited* by Susan Lindquist, Whitehead Institute for Biomedical Research, Cambridge, MA, and approved January 26, 2012 (received for review October 5, 2011) Organismal fitness depends on the ability of gene networks to determinant of quantitative behaviors of inducible circuits (20– function robustly in the face of environmental and genetic 24), we also investigated the role of Rox1 feedback in expression perturbations. Understanding the mechanisms of this stability is regulation during oxygen response. one of the key aims of modern systems biology. Dissecting the basis of robustness to mutation has proven a particular challenge, Results with most experimental models relying on artificial DNA sequence We set out to establish a tractable model system for the study of variants engineered in the laboratory. In this work, we hypothe- feedback and robustness using yeast transcription factors. For sized that negative regulatory feedback could stabilize gene this purpose, we first screened transcription factor genes for expression against the disruptions that arise from natural genetic feedback on protein abundance. We used fluorescence micros- variation. We screened yeast transcription factors for feedback copy (25) to measure protein expression from a single genomic and used the results to establish ROX1 (Repressor of hypOXia) as copy of each factor tagged with GFP in a diploid strain (26) while a model system for the study of feedback in circuit behaviors and varying levels of an untagged copy of the factor (Fig. 1A). To its impact across genetically heterogeneous populations. Muta- maximize signal to noise, we analyzed the 23 most highly abun- genesis experiments revealed the mechanism of Rox1 as a direct dant factors in rich medium (Table S1) and found evidence for transcriptional repressor at its own gene, enabling a regulatory negative feedback in four cases (Fig. 1B). As expected (27, 28), program of rapid induction during environmental change that these screen hits included Rox1 and Swi4; we also found EVOLUTION reached a plateau of moderate steady-state expression. Addition- uncharacterized feedback loops by Mbp1 and Mot3. Independent ally, in a given environmental condition, Rox1 levels varied widely replicate experiments confirmed the ability of each factor to re- across genetically distinct strains; the ROX1 feedback loop regu- press its own abundance (Fig. 1C). lated this variation, in that the range of expression levels across To investigate the role of feedback in mutational robustness genetic backgrounds showed greater spread in ROX1 feedback and systems-level network behaviors, we focused on the tran- mutants than among strains with the ROX1 feedback loop intact. fi scription factor Rox1. This master regulator is repressed in Our ndings indicate that the ROX1 feedback circuit is tuned to hypoxic conditions and induced under normoxia to regulate respond to perturbations arising from natural genetic variation in biosynthetic pathways that use molecular oxygen as a substrate addition to its role in induction behavior. We suggest that regula- (29). Anticipating that Rox1 would act directly at the ROX1 tory feedback may be an important element of the network archi- promoter (27), we identified four candidate Rox1 binding sites in tectures that confer mutational robustness across biology. the 500 bp upstream of the ROX1 coding start site (Fig. S1). Mutagenesis confirmed the role of these sites in ROX1 feedback, obustness of organismal function in the face of perturbations with each site contributing incrementally to the strength of fi Ris critical for tness. Since the seminal work of Waddington autoregulation in an ROX1 transcriptional reporter (Fig. 2). (1), biologists have remarked on the stability of phenotypes Promoter response to changes in dose of ROX1 was markedly against environmental and genetic variation, and understanding reduced when all four sites were mutated in combination, in- how organisms achieve robustness remains one of the major dicating a near-complete abrogation of feedback (Fig. 2). We – challenges in systems biology (2 4). Much of the search for used these mutations to engineer a version of ROX1 in which the molecular mechanisms of robustness has focused on gene regu- feedback mutant promoter drove expression of Rox1 fused to lation. Characteristics of regulatory networks that confer ro- GFP; to avoid potentially confounding effects from elevated bustness include pathway redundancy and master regulatory Rox1-GFP steady-state levels in the presence of the feedback – organization (5), phenotypic capacitors (6 8), paired activating mutations, we manipulated the use of optimal codons (30) in the and inhibiting inputs (9), and cooperative and feed-forward ROX1-GFP coding region. The suboptimized sequence, in regulation (10). Additionally, negative regulatory feedback, in conjunction with mutated Rox1 binding sites in the ROX1 pro- which a biomolecule represses its own abundance, can buffer moter, gave rise to steady-state expression levels comparable variation in gene expression (11, 12), and negative feedback with those levels of the WT (Fig. S2). We refer to this version of loops have been shown to underlie robustness to variable envi- ROX1 as the suboptimized feedback mutant; a strain harboring – ronmental conditions and stochastic intracellular change (13 this gene grew indistinguishably from the WT across environ- 15). Negative feedback may also confer network stability against mental conditions (Fig. S3). the effects of mutations (3, 16), but evidence for negative feedback as a driver of mutational robustness in vivo has been at

a premium (17); the relevance of this principle to natural genetic Author contributions: C.M.D. and R.B.B. designed research; C.M.D. and J.H.I. performed variation remains largely unknown. research; R.C.Y. and C.G.P. contributed new reagents/analytic tools; C.M.D. and R.B.B. In this work, we focused on negative feedback in yeast hypoxia analyzed data; and C.M.D. and R.B.B. wrote the paper. regulation motivated by the extensive evidence for feedback in The authors declare no conﬂict of interest. oxygen response pathways across biology (18, 19). We charac- *This Direct Submission article had a prearranged editor. terized the feedback loop at the yeast hypoxia regulator ROX1 in Freely available online through the PNAS open access option. molecular detail, and we harnessed this system as a test bed to 1To whom correspondence should be addressed. E-mail: [email protected]. study how feedback confers stability against naturally occurring This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. mutations. Given the precedent for negative feedback as a 1073/pnas.1116360109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1116360109 PNAS Early Edition | 1of5 Downloaded by guest on September 30, 2021 A

B C

Fig. 1. Screening yeast transcription factors for regulatory feedback. (A) Each part of the schematic represents one yeast strain. Green rectangles indicate the coding sequence of GFP, and the oval on the right represents a plasmid carrying an overexpression construct controlled by the GAL1-10 promoter (44). △, whole-gene deletion. (B) Each data point represents the results of two analyses of nuclear abundance of a transcription factor fused to GFP (26) measured by quantitative microscopy in diploid strains. Each analysis, as indicated on the x or y axis, compared fluorescence from a given tagged factor in two strains encoding different doses of the untagged version of the factor with strains named according to the schematic in A.(C) Each set of bars reports nuclear fluorescence measurements of the indicated factor as a fusion with GFP measured by quantitative microscopy and normalized with respect to WT levels. Each bar reports measurements from one strain with names as in A; error bars represent the SD over biological replicates and microscope fields. *Comparisons relative to WT that are significant at Wilcoxon P < 0.001.

To analyze the role of feedback during ROX1 induction, we reporter construct (Fig. 3). We conclude that the ROX1 locus grew WT and feedback mutant strains under hypoxic conditions harbors regulatory information encoding strong activation during and measured Rox1-GFP expression during oxygen exposure. oxygen exposure. In WT cells, repression by Rox1 serves as For the feedback mutant promoter driving expression of the WT a brake on this induction signal, avoiding the deleterious effects ROX1-GFP reporter, levels during induction far overshot the of elevated expression. Thus, the feedback loop tunes the ki- steady state of the strain harboring a WT promoter (Fig. 3). Such netics of ROX1 induction, enabling a rapid approach to a mod- elevated expression was toxic to cells: this strain displayed erate level of steady-state expression during the transition from growth defects under a variety of environmental conditions (Fig. hypoxia to normoxia. S3). As expected, in the suboptimized feedback mutant, protein We next sought to evaluate Rox1 feedback as a mechanism for levels reached those of the WT at steady state but with slower robustness of gene expression to naturally occurring genetic ﬁ kinetics owing to the reduced translation ef ciency of the variation. For this purpose, we developed an assay to interrogate

Fig. 3. The ROX1 locus confers strong induction during oxygen exposure. Fig. 2. Transcription factor binding sites in the ROX1 promoter are required Each set of points reports fluorescence of a haploid yeast strain bearing an for transcriptional feedback. Each set of bars reports expression from one ROX1-GFP fusion reporter gene measured by quantitative microscopy after ROX1 transcriptional reporter in a diploid yeast strain measured by flow a transfer of a culture from hypoxia to normoxia. Each data point represents cytometry. Shading represents the presence or absence of a single untagged median nuclear fluorescence across cells of one culture at the indicated time copy of ROX1 in the strain background. Each set of bars labeled with Site after oxygenation. Each color represents one reporter. FB mutant indicates corresponds to a reporter with the indicated Rox1 binding sites (Fig. S1) a feedback mutant ROX1 promoter with all four Rox1 binding sites muta- mutagenized in the ROX1 promoter; 4 site indicates mutagenesis of all four genized, and subopt indicates that the ROX1 and GFP sequences were sites. Δ, whole-gene deletion of ROX1. encoded with suboptimized codons.

2of5 | www.pnas.org/cgi/doi/10.1073/pnas.1116360109 Denby et al. Downloaded by guest on September 30, 2021 the effects on ROX1 expression of the spectrum of variants yeast transcription factors. In particular, we reasoned that evo- present in a set of divergent yeast strains of environmental and lutionary pressures for buffering expression levels and thus for laboratory origin. For each such tester strain, we crossed it to feedback would be strongest among factors for which deviations a laboratory strain bearing the WT ROX1-GFP reporter, and we from homeostatic expression levels gives rise to fitness defects. In performed an analogous cross using the suboptimized feedback the case of Rox1, both increases and decreases in dosage were mutant. Haploid recombinant progeny, each a mosaic of in- toxic to yeast cells (Fig. S3). To test the relationship between heritance from the tester and laboratory parents, served as expression homeostasis and feedback more generally, we used a panel of genetically distinct strains among which Rox1 ex- growth rates of yeast strains manipulated to overexpress each pression could vary. For a given cross, we measured the median transcription factor in turn (31). We first integrated these data Rox1-GFP levels in a culture of each progeny strain. The results, with the results of our reporter-based screen for feedback among shown in Fig. 4, revealed deviation in Rox1-GFP abundance of transcription factors (Fig. 1). Conforming to our model, when up to eightfold across strain cultures, reflecting the impact of overexpressed, factors subject to feedback conferred a fitness naturally occurring genetic variation on Rox1 expression. Elim- defect ∼70% more severe than factors with no evidence for inating feedback compromised robustness to these variants, with feedback (Wilcoxon P = 0.02) (Fig. 5A). We next sought to a wider spread of median Rox1-GFP levels across genetic expand our analysis beyond the transcription factors detectable backgrounds in feedback mutant strains than in WT; the co- in our experimental screen. For this purpose, we used a bio- efficient of variation across strains was two- to fivefold higher in informatic approach (described in Materials and Methods)to the presence of the feedback mutation (Fig. 4). Control experi- predict instances of direct transcriptional feedback by tran- ments ruled out codon suboptimization as a predominant cause scription factors based on the presence of their binding sites in of this effect (Fig. S4). Interestingly, the extent of expression promoters of their own encoding genes. Even using this unvali- variation across recombinant progeny was a function of the tester dated set of feedback inferences, evidence for autoregulation strain parent, indicating that some testers harbored alleles with was again associated with overexpression toxicity, albeit to a less more dramatic consequences for Rox1-GFP expression than dramatic extent; toxicity effects were 13% more severe for fac- others (Fig. 4). We conclude that Rox1 autorepression buffers tors with inferred feedback relative to the remainder of the set the effects of natural genetic variation on ROX1 expression, (Wilcoxon P = 0.04) (Fig. 5B). To address the possibility that establishing regulatory feedback as a determinant of mutational computationally predicted binding sites were better specified for robustness in this system. factors with stronger overexpression toxicity, we tested the re- EVOLUTION We hypothesized that the ability to buffer gene expression lationship between overexpression growth rate for a factor and could be a driver of the prevalence of feedback circuits across the number of its predicted targets genome-wide, and we found no effect (regression P = 0.2). Taken together, our results highlight feedback by transcription factors as a correlate of the toxic effects of overexpression, lending credence to the notion that many such feedback loops function in vivo as a control against misregulation. Discussion Landmark studies have identified negative feedback loops that tune the quantitative properties of gene circuits (20–24, 32). Negative feedback has also been implicated in the robustness of gene networks to environmental and stochastic change (13–15), but the role of native feedback circuits as buffers against natural genetic variation has remained unknown. Addressing the ques- tion requires detailed molecular analysis of feedback regulation and its impact across genetically heterogeneous populations, for which we have established an experimental paradigm using yeast Rox1 as a model system. Rox1 regulates the expression of genes involved in oxygen- dependent sterol biogenesis and respiratory pathways. We have shown that cell growth is remarkably sensitive to changes in this activity even in normoxic conditions, with overexpression and deletion of Rox1 each conferring distinct defects. Consistent with a requirement for tight regulatory control of Rox1, our analysis has revealed two related ways in which Rox1 feedback Fig. 4. Rox1 feedback as a mechanism for mutational robustness. Each limits deviations from steady-state expression optima. In a con- panel shows ROX1 expression levels in haploid segregants from crosses between an Rox1-GFP strain and a tester strain: BY4741, a laboratory strain stant genetic background, negative feedback enables rapid derived from a fig tree isolate; YPS606, an isolate from an oak tree (49); SK1, ROX1 activation during oxygen exposure, minimizing the time a laboratory strain derived from a soil isolate (49); and 22:3:b, a strain of spent in intermediate expression states and avoiding toxic effects hybrid laboratory and vineyard origin (50). In a given panel, each distribu- of overexpression in a manner that dovetails with similar roles tion reports culture medians of fluorescence measured by flow cytometry. for autorepression in other networks (11, 33). Additionally, Each median was taken across cells of an isogenic culture of one segregant across genetic backgrounds, Rox1 feedback serves as a buffer and normalized with respect to the mean across all strains of the same cross. against the perturbations arising from naturally occurring se- Each color represents strains bearing an ROX1-GFP fusion with the indicated quence changes. A primary implication of our findings is thus fi modi cation. No feedback indicates the suboptimized feedback mutant that the Rox1 negative feedback circuit is tuned to respond (Figs. 2 and 3 and Fig. S2). p, Resampling P value evaluating the F statistic for differential variance between the feedback mutant and WT distributions. quantitatively to subtle up- and down-regulating effects of nat- Coefficients of variation of distributions of WT and feedback mutant strains ural genetic variation as well as dramatic shifts in regulatory trans are 0.01 and 0.07, respectively, for the BY4741 cross; 0.07 and 0.31, re- input when conditions change. In each case, -acting input spectively, for YPS606; 0.26 and 0.75, respectively, for SK1; and 0.10 and 0.18, that up-regulates ROX1 would be counterbalanced by increased respectively, for 22:3:b. Rox1 occupancy and repression at its own encoding locus, and

Denby et al. PNAS Early Edition | 3of5 Downloaded by guest on September 30, 2021 ABFig. 5. Transcription factors subject to feedback are toxic when overexpressed. Each panel represents analysis of growth rates across a panel of yeast strains, with each overexpressing a single transcription factor (31). Each distribution represents growth rates across the set of genes with or without evidence for feedback as indicated. The x axis reports growth rate in doublings per hour. (A) Blue represents factors emerging as screen hits in Fig. 1; red reports all other screened factors (n = 22). Distribution medians are 0.17 and 0.29, respectively. (B) Blue represents factors with signiﬁcant sequence matches to their own binding motif in their own promoters; red reports all other tested factors (n = 83). Distribution medians are 0.24 and 0.27, respectively. p, Wilcoxon P value evaluating the difference in growth rates between factors with and without evidence for feedback.

input that represses ROX1 would be counterbalanced by a re- We then permuted mutant and WT strain expression values 10,000 times, duction in Rox1 occupancy at its own gene. repeated the F calculation for each permuted dataset, and evaluated the Will regulatory feedback prove to underlie mutational ro- statistic from the real data against this null distribution to yield a one-sided bustness as a general mechanism across biology? Pathway-level empirical P value. feedback is common in yeast; in many cases, the network can fi Feedback Assays. For the feedback screen, one culture of each yeast strain detect the perturbation of an arti cially introduced genetic lesion was inoculated into complete supplemented media (CSM) minus uracil (MP and up-regulate functionally related genes to compensate (17). Biomedicals) with 2% raffinose and grown in 2-mL well 96-well plates at Whether gene-level feedback will be of similar importance for 30 8C with shaking to saturation. Strains were back-diluted into CSM minus robustness on a genomic scale depends, in part, on the prevalence uracil with 2% galactose to an OD of 0.1. Strains were imaged after ∼6hof of autoregulation in gene networks. Most highly expressed genes growth having reached OD of 0.5–1.0. On preparation of 96-well plates (see in yeast are not subject to complete dosage compensation when below), 100 μL cell suspension were added to each well and allowed to settle mutated (34), but feedback at the gene level may be particularly and bind for 30 min before imaging as described below. In the screen, only common among transcription factors (35, 36). In light of our transcription factors with putative feedback effects consistent between evidence that feedback may serve to constrain the deleterious overexpression and deletion assays were considered for additional study (Fig. S5). For confirmation of screen hits in Fig. 1C, three independent cul- effects of misregulation among transcription factors, a model tures of each strain were grown and prepared for imaging as above. invoking especially strong pressures for such feedback control would be consistent with the extreme overexpression toxicity Hypoxia Time Course. For induction experiments, a culture of each yeast strain observed across transcription factors in yeast (31). The emerging was grown in CSM with 2% galactose, back-diluted into deoxygenated picture is one in which both gene- and pathway-level feedback, at media at OD of 0.002, and grown for 40 h at 30 ºC in a 16 × 12-mm Hungate transcription factor genes and elsewhere in the yeast network, tube with septum stopper and screw cap (Bellco). From each culture, a 10-mL may be key elements of the architecture that mediates buffering aliquot was spun down, and cells were resuspended in 200 μL oxygenated of genetic change. Understanding which gene circuits are buff- media; 100 μL were added to concanavalin-treated, glass-bottomed plate ered and how will be of critical interest in the effort to interpret wells and allowed to settle for 1–5 min before imaging as described below at the extent and phenotypic penetrance of regulatory variation the time points indicated in Fig. 3. To deoxygenate media, 10 mL were (37–40) and the evolution of robustness mechanisms (16, 41). added to a stoppered tube and boiled for 5 min with a syringe through the stopper. After boiling, the syringe was removed, and each tube of media Addressing these questions and applying the emergent principles was cooled to <50 ºC and purged with N . to bioengineering (42) and human disease treatment (43–45) will 2 serve as continued motivation for the study of feedback and ro- Fluorescence Microscopy. Glass-bottomed 96-well plates (Falcon) were coated bustness in regulatory circuits. with 100 μL of a 100 μg/mL solution of Con A type V (Sigma-Aldrich), in- cubated for 1 h at room temperature, and then washed three times with Materials and Methods water. Once cells were added and allowed to adhere, wells were then Yeast Strain Construction and Analysis of Genetic Variation. For each of 67 washed and resuspended with media immediately before imaging. For im- yeast transcription factors (Table S1), strains bearing one copy of a GFP aging, we used a 60× PlanApo objective (N.A. = 1.4) under oil immersion in protein fusion of the respective factor gene (26) (Invitrogen) were mated to a Nikon TE2000 inverted microscope equipped with a mercury lamp, mo- strains deleted for the respective gene (46) (Invitrogen), and they were im- torized stage, and 512BFT MicroMax cooled CCD camera (Photometrix). We aged by microscopy as described below. For the 23 factors of highest abun- imaged three to six observation fields per strain, with each field containing dance (nuclear fluorescence > 30,000 arbitrary units) (Table S1), GFP-tagged ∼100 cells. For each field, we acquired a 0.5-s GFP exposure and a defocused haploid strains were then mated to strains bearing URAcil requiring (URA)3- bright-field image for cell identification and boundary determination using marked plasmids encoding each ORF under the GAL1,10 promoter (47) for Metamorph 7.0 software (Universal Imaging Corporation). Using Cell-ID 1.0 the complete screen as shown in Fig. 1. All plasmid transformations and yeast (25), for each cell, we estimated nuclear fluorescence by calculating average cell matings were performed by standard techniques (48). Construction of pixel intensity in the area of a 3-pixel radius around the brightest point. We reporters and mutagenesis protocols are described in detail in SI Materials estimated cytoplasmic fluorescence by calculating the average intensity of and Methods. The suboptimized ROX1-GFP sequence is provided in Fig. S6. pixels immediately inside the cell’s edge. We then calculated normalized Crosses for analysis of genetic variation were generated by mating ROX1- nuclear fluorescence by subtracting the estimated cytoplasmic fluorescence GFP strains with BY4741 (Open Biosystems), SK1 (49), YPS606 (49), and 22:3:b from the estimated nuclear fluorescence. (50). In the case of YPS606 and SK1, homothalic switching endonuclease was replaced with URA3 by cloning and transformation as above. For each cross, Flow Cytometry. Cultures of each strain were grown in CSM with 2% galactose hybrid diploids were sporulated on solid minimal sporulation medium (48), to saturation and back-diluted to OD of 0.1 6–8 h before measurement. haploids were isolated by spore enrichment (51), and 60–90 GFP+ recombi- Fluorescence measurements of ∼10,000 cells/sample were acquired with an nants were selected by growth on Yeast Peptone Dextrose (YPD) medium LSRFortesa (BD Biosciences) and analyzed in R with the flowCore package (48) supplemented with G418. To compare the variances of the feedback (www.bioconductor.org). A given strain was eliminated from analysis if the mutant and WT distributions for a given cross, we first calculated the F statistic majority of side scatter values across cells in its culture sample did not fall for differential variance using the var.test function in R (www.r-project.org). within a range typical of a nonflocculent laboratory strain growing at log

4of5 | www.pnas.org/cgi/doi/10.1073/pnas.1116360109 Denby et al. Downloaded by guest on September 30, 2021 phase. The median background fluorescence from cells of a strain bearing no Harbison/release_v24/final_set/Final_InTableS2_v24.motifs (53). We used GFP gene was subtracted from the median across cells of each culture MCAST (http://meme.sdsc.edu/meme) to predict binding to each upstream of interest. sequence for each factor. For analysis, we retained all matrix matches with −log2(p/0.0005) > 3.45, where the P value estimates the significance of Growth Condition Screen. Strains were grown overnight in CSM with 2% predicted binding, calculated with respect to a background model gener- glucose at 30 ºC to saturation and then reinoculated and grown for >2 ated from the complete set of nucleotide frequencies in all yeast upstream doublings before plating; 1:10 serial dilutions were performed in microtiter sequences (54). plates. Dilutions were transferred to plates using a multipronged inoculating device (frogger). Plate media were CSM with one of the following additions: ACKNOWLEDGMENTS. The authors thank Bhupinder Bhullar, Stephanie 2% glucose, 2% galactose, 3% glycerol, or 3.2% ethanol; 2% glucose with Mohr, and the Harvard Institute of Proteomics for providing overexpression 20 μg/mL fluoconazole (Sigma-Aldrich); or YPD (48). plasmids; Chris Phillips for technical assistance with media deoxygenation; Jeremy Roop for genome-scale sequence data; Roger Brent for generosity Bioinformatic Predictions of Feedback. with microscopy hardware and analysis tools; Meru Sadhu for comments on Sequences corresponding to the 1,000 the manuscript; and Mike Eisen, Jeremy Thorner, Adam Arkin, and members bp upstream of coding start for each of 5,922 yeast genes were downloaded of the R.B.B. laboratory for helpful discussions. This work was supported by a (52). Position weight matrices to score binding preferences for each Burroughs–Wellcome Career Award at the Scientific Interface and an Ellison yeast transcription factor were downloaded from http://fraenkel.mit.edu/ New Scholar Award in Aging (to R.B.B.).

1. Waddington CH (1942) Canalization of development and the inheritance of acquired 29. Kwast KE, et al. (2002) Genomic analyses of anaerobically induced genes in Saccha- characters. Nature 150:563–565. romyces cerevisiae: Functional roles of Rox1 and other factors in mediating the anoxic 2. Kirschner M, Gerhart J (1998) Evolvability. Proc Natl Acad Sci USA 95:8420–8427. response. J Bacteriol 184:250–265. 3. Masel J, Siegal ML (2009) Robustness: Mechanisms and consequences. Trends Genet 30. Jansen R, Bussemaker HJ, Gerstein M (2003) Revisiting the codon adaptation index 25:395–403. from a whole-genome perspective: Analyzing the relationship between gene ex- 4. Hartman JL, 4th, Garvik B, Hartwell L (2001) Principles for the buffering of genetic pression and codon occurrence in yeast using a variety of models. Nucleic Acids Res variation. Science 291:1001–1004. 31:2242–2251. 5. Macneil LT, Walhout AJ (2011) Gene regulatory networks and the role of robustness 31. Yoshikawa K, et al. (2011) Comprehensive phenotypic analysis of single-gene deletion – and stochasticity in the control of gene expression. Genome Res 21:645–657. and overexpression strains of Saccharomyces cerevisiae. Yeast 28:349 361. 6. Rutherford SL, Lindquist S (1998) Hsp90 as a capacitor for morphological evolution. 32. Mettetal JT, Muzzey D, Gómez-Uribe C, van Oudenaarden A (2008) The frequency – Nature 396:336–342. dependence of osmo-adaptation in Saccharomyces cerevisiae. Science 319:482 484. 7. Queitsch C, Sangster TA, Lindquist S (2002) Hsp90 as a capacitor of phenotypic vari- 33. Rosenfeld N, Elowitz MB, Alon U (2002) Negative autoregulation speeds the response times of transcription networks. J Mol Biol 323:785–793. ation. Nature 417:618–624.

34. Springer M, Weissman JS, Kirschner MW (2010) A general lack of compensation for EVOLUTION 8. Levy SF, Siegal ML (2008) Network hubs buffer environmental variation in Saccha- gene dosage in yeast. Mol Syst Biol 6:368. romyces cerevisiae. PLoS Biol 6:e264. 35. Thieffry D, Huerta AM, Pérez-Rueda E, Collado-Vides J (1998) From specific gene 9. Acar M, Pando BF, Arnold FH, Elowitz MB, van Oudenaarden A (2010) A general regulation to genomic networks: A global analysis of transcriptional regulation in mechanism for network-dosage compensation in gene circuits. Science 329: Escherichia coli. Bioessays 20:433–440. 1656–1660. 36. Wall ME, Hlavacek WS, Savageau MA (2004) Design of gene circuits: Lessons from 10. Alon U (2007) Network motifs: Theory and experimental approaches. Nat Rev Genet bacteria. Nat Rev Genet 5:34–42. – 8:450 461. 37. Raj A, Rifkin SA, Andersen E, van Oudenaarden A (2010) Variability in gene expression 11. Freeman M (2000) Feedback control of intercellular signalling in development. Na- underlies incomplete penetrance. Nature 463:913–918. – ture 408:313 319. 38. Rockman MV, Kruglyak L (2006) Genetics of global gene expression. Nat Rev Genet 7: 12. Brandman O, Meyer T (2008) Feedback loops shape cellular signals in space and time. 862–872. Science 322:390–395. 39. Burga A, Casanueva MO, Lehner B (2011) Predicting mutation outcome from early 13. Li X, Cassidy JJ, Reinke CA, Fischboeck S, Carthew RW (2009) A microRNA imparts stochastic variation in genetic interaction partners. Nature 480:250–253. robustness against environmental fluctuation during development. Cell 137:273–282. 40. Casanueva MO, Burga A, Lehner B (2012) Fitness trade-offs and environmentally in- 14. Paulsen M, Legewie S, Eils R, Karaulanov E, Niehrs C (2011) Negative feedback in the duced mutation buffering in isogenic C. elegans. Science 335:82–85. bone morphogenetic protein 4 (BMP4) synexpression group governs its dynamic 41. Lynch M (2007) The evolution of genetic networks by non-adaptive processes. Nat Rev signaling range and canalizes development. Proc Natl Acad Sci USA 108:10202–10207. Genet 8:803–813. 15. Ramsey SA, et al. (2006) Dual feedback loops in the GAL regulon suppress cellular 42. Andrianantoandro E, Basu S, Karig DK, Weiss R (2006) Synthetic biology: New engi- heterogeneity in yeast. Nat Genet 38:1082–1087. neering rules for an emerging discipline. Mol Syst Biol 2:2006.0028. 16. Meiklejohn CD, Hartl DL (2002) A single mode of canalization. Trends Ecol Evol 17: 43. Ideker T, Sharan R (2008) Protein networks in disease. Genome Res 18:644–652. 468–473. 44. Szalay MS, Kovács IA, Korcsmáros T, Böde C, Csermely P (2007) Stress-induced re- 17. Kafri R, Bar-Even A, Pilpel Y (2005) Transcription control reprogramming in genetic arrangements of cellular networks: Consequences for protection and drug design. backup circuits. Nat Genet 37:295–299. FEBS Lett 581:3675–3680. 18. Schwartsburd PM (2001) Self-cytoprotection against stress: Feedback regulation of 45. Amit I, Wides R, Yarden Y (2007) Evolvable signaling networks of receptor tyrosine heme-dependent metabolism. Cell Stress Chaperones 6:1–5. kinases: Relevance of robustness to malignancy and to cancer therapy. Mol Syst Biol 3: 19. Wu N, Yin L, Hanniman EA, Joshi S, Lazar MA (2009) Negative feedback maintenance 151. of heme homeostasis by its receptor, Rev-erbalpha. Genes Dev 23:2201–2209. 46. Tong AH, et al. (2001) Systematic genetic analysis with ordered arrays of yeast de- – 20. Yu RC, et al. (2008) Negative feedback that improves information transmission in letion mutants. Science 294:2364 2368. 47. Butcher RA, et al. (2006) Microarray-based method for monitoring yeast over- yeast signalling. Nature 456:755–761. expression strains reveals small-molecule targets in TOR pathway. Nat Chem Biol 2: 21. Acar M, Becskei A, van Oudenaarden A (2005) Enhancement of cellular memory by 103–109. reducing stochastic transitions. Nature 435:228–232. 48. Amberg DC, Burke D, Strathern JN (2005) Methods in Yeast Genetics: A Cold Spring 22. Kapuy O, He E, Uhlmann F, Novák B (2009) Mitotic exit in mammalian cells. Mol Syst Harbor Laboratory Course Manual (Cold Spring Harbor Laboratory Press, Plainview, Biol 5:324. NY). 23. Duong HA, Robles MS, Knutti D, Weitz CJ (2011) A molecular mechanism for circadian 49. Liti G, et al. (2009) Population genomics of domestic and wild yeasts. Nature 458: clock negative feedback. Science 332:1436–1439. 337–341. 24. Justman QA, Serber Z, Ferrell JE, Jr., El-Samad H, Shokat KM (2009) Tuning the acti- 50. Brem RB, Kruglyak L (2005) The landscape of genetic complexity across 5,700 gene – vation threshold of a kinase network by nested feedback loops. Science 324:509 512. expression traits in yeast. Proc Natl Acad Sci USA 102:1572–1577. fi 25. Gordon A, et al. (2007) Single-cell quanti cation of molecules and rates using open- 51. Rockmill B, Lambie EJ, Roeder GS (1991) Spore enrichment. Methods Enzymol 194: – source microscope-based cytometry. Nat Methods 4:175 181. 146–149. 26. Huh WK, et al. (2003) Global analysis of protein localization in budding yeast. Nature 52. Scannell DR, et al. (2011) The awesome power of yeast evolutionary genetics: New 425:686–691. genome sequences and strain resources for the Saccharomyces sensu stricto genus. G3 27. Deckert J, Perini R, Balasubramanian B, Zitomer RS (1995) Multiple elements and (Bethesda) 1:11–25. auto-repression regulate Rox1, a repressor of hypoxic genes in Saccharomyces cer- 53. MacIsaac KD, et al. (2006) An improved map of conserved regulatory sites for Sac- evisiae. Genetics 139:1149–1158. charomyces cerevisiae. BMC Bioinformatics 7:113. 28. Harbison CT, et al. (2004) Transcriptional regulatory code of a eukaryotic genome. 54. Bailey TL, Noble WS (2003) Searching for statistically significant regulatory modules. Nature 431:99–104. Bioinformatics 19(Suppl 2):ii16–ii25.

Denby et al. PNAS Early Edition | 5of5 Downloaded by guest on September 30, 2021