@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Pharmaceutical Sciences, Faculty of"@en, "Non UBC"@en ; edm:dataProvider "DSpace"@en ; ns0:identifierCitation "Genome Biology. 2014 Apr 10;15(4):R64"@en ; ns0:rightsCopyright "VanderSluis et al.; licensee BioMed Central Ltd."@en ; dcterms:creator "VanderSluis, Benjamin"@en, "Hess, David C"@en, "Pesyna, Colin"@en, "Krumholz, Elias W"@en, "Syed, Tahin"@en, "Szappanos, Balázs"@en, "Nislow, Corey"@en, "Papp, Balázs"@en, "Troyanskaya, Olga G"@en, "Myers, Chad L"@en, "Caudy, Amy A"@en ; dcterms:issued "2015-12-18T02:41:40"@en, "2014-04-10"@en ; dcterms:description """Background: Genome-wide sensitivity screens in yeast have been immensely popular following the construction of a collection of deletion mutants of non-essential genes. However, the auxotrophic markers in this collection preclude experiments on minimal growth medium, one of the most informative metabolic environments. Here we present quantitative growth analysis for mutants in all 4,772 non-essential genes from our prototrophic deletion collection across a large set of metabolic conditions. Results The complete collection was grown in environments consisting of one of four possible carbon sources paired with one of seven nitrogen sources, for a total of 28 different well-defined metabolic environments. The relative contributions to mutants' fitness of each carbon and nitrogen source were determined using multivariate statistical methods. The mutant profiling recovered known and novel genes specific to the processing of nutrients and accurately predicted functional relationships, especially for metabolic functions. A benchmark of genome-scale metabolic network modeling is also given to demonstrate the level of agreement between current in silico predictions and hitherto unavailable experimental data. Conclusions These data address a fundamental deficiency in our understanding of the model eukaryote Saccharomyces cerevisiae and its response to the most basic of environments. While choice of carbon source has the greatest impact on cell growth, specific effects due to nitrogen source and interactions between the nutrients are frequent. We demonstrate utility in characterizing genes of unknown function and illustrate how these data can be integrated with other whole-genome screens to interpret similarities between seemingly diverse perturbation types."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/55930?expand=metadata"@en ; skos:note "RESEARCHBroad metabolic sensitivitprototrophic yeast deletioliaLavwinsestruction techniques such as synthetic genetic array (SGA) eukaryote, a systematic exploration of mutant growthVanderSluis et al. Genome Biology 2014, 15:R64http://genomebiology.com/2014/15/4/R64provided additional nutrients. This requirement reflectsM5S 3E1, CanadaFull list of author information is available at the end of the articleanalysis [2] has further driven the creation of customizedyeast deletion arrays. While quantitative single mutantfitness assays have been performed [3], they are generallyacross basic nutrient environments is conspicuouslyabsent. These data would be valuable for metabolicresearchers and computational biologists that attemptto model the metabolic network of the cell using method-ologies such as flux balance analysis (FBA) [8] because thedefined growth conditions are amenable to modeling.Yeast strain collections used in previous high-throughputassays (that is, the deletion collection) are auxotrophic [1],and therefore unable to survive in minimal media unless* Correspondence: cmyers@cs.umn.edu; amy.caudy@utoronto.ca†Equal contributors1Department of Computer Science and Engineering, University of MinnesotaTwin Cities, 200 Union St SE, Minneapolis, MN 55455, USA8Donnelly Centre for Cellular and Biomolecular Research and Department ofMolecular Genetics, University of Toronto, 160 College Street, Toronto, ONrefinement of mating-based high-throughput strain con-across a large set of metabolic conditions.Results: The complete collection was grown in environments consisting of one of four possible carbon sourcespaired with one of seven nitrogen sources, for a total of 28 different well-defined metabolic environments. Therelative contributions to mutants' fitness of each carbon and nitrogen source were determined using multivariatestatistical methods. The mutant profiling recovered known and novel genes specific to the processing of nutrientsand accurately predicted functional relationships, especially for metabolic functions. A benchmark of genome-scalemetabolic network modeling is also given to demonstrate the level of agreement between current in silico predictionsand hitherto unavailable experimental data.Conclusions: These data address a fundamental deficiency in our understanding of the model eukaryoteSaccharomyces cerevisiae and its response to the most basic of environments. While choice of carbon sourcehas the greatest impact on cell growth, specific effects due to nitrogen source and interactions between thenutrients are frequent. We demonstrate utility in characterizing genes of unknown function and illustrate howthese data can be integrated with other whole-genome screens to interpret similarities between seeminglydiverse perturbation types.BackgroundLarge scale gene deletion screens have become commonin Saccharomyces cerevisiae due to efforts in the yeastcommunity to assemble a near complete collection ofnon-essential single-mutant strains [1]. The subsequentlimited to a single growth medium. A few notable ex-ceptions have begun to explore this space [4-7], butthe conditions of interest are often chosen with humantherapeutic ends in mind and are limited to knowndrugs or small molecules of unknown biological effect.A decade and a half after the sequencing of the best-studiedBenjamin VanderSluis1†, David C Hess2†, Colin Pesyna1, ECorey Nislow5, Balázs Papp4, Olga G Troyanskaya6,7, ChadAbstractBackground: Genome-wide sensitivity screens in yeast ha collection of deletion mutants of non-essential genes. Hoexperiments on minimal growth medium, one of the mostquantitative growth analysis for mutants in all 4,772 non-es© 2014 VanderSluis et al.; licensee BioMed CenCreative Commons Attribution License (http:/distribution, and reproduction in any mediumDomain Dedication waiver (http://creativecomarticle, unless otherwise stated.Open Accessy profiling of an collections W Krumholz3, Tahin Syed1, Balázs Szappanos4,Myers1* and Amy A Caudy8*e been immensely popular following the construction ofever, the auxotrophic markers in this collection precludeformative metabolic environments. Here we presentntial genes from our prototrophic deletion collectiontral Ltd. This is an Open Access article distributed under the terms of the/creativecommons.org/licenses/by/2.0), which permits unrestricted use,, provided the original work is properly credited. The Creative Commons Publicmons.org/publicdomain/zero/1.0/) applies to the data made available in thisthe historical use of auxotrophic markers for geneticselection. The resulting requirement for nutrient supple-mentation precludes systematic testing of the yeast deletioncollection on specific combinations of carbon and nitrogensources because the auxotrophic nutrient supplements canalso be used as carbon and nitrogen sources. Previous workhas shown not only that nutrient supplementation canAll 28 carbon:nitrogen combinations were included toproduce a broad set of well-defined metabolic conditions.The plates were imaged in time course in order toestimate growth rates from measurements of colonysize (Figure 1; see Materials and methods for details;Additional file 1).ditVanderSluis et al. Genome Biology 2014, 15:R64 Page 2 of 18http://genomebiology.com/2014/15/4/R64have different physiological consequences from geneticcomplementation [9] but also that auxotrophies can alterthe expression of many other genes [10].To address this deficiency in genome-scale data ongrowth in other, defined media, we constructed a proto-trophic version of the yeast deletion collection and thenscreened this collection of 4,772 mutants against 28defined minimal media conditions. These 28 conditionswere formed by using all pairwise combinations of fourcarbon sources and seven nitrogen sources (Table 1,Figure 1). These screens of the prototrophic collectionrevealed numerous interactions between carbon andnitrogen sources with respect to wild-type growth rate,underscoring the need to perform growth experiments ina combinatorial fashion. Mutant data revealed condition-specific sensitivities across all conditions, including manyeffects for uncharacterized genes and mutants that arehealthy under standard laboratory conditions. We showthat the data have power to predict functional relationshipsbetween genes and are otherwise validated via a separateliquid assay as well as through comparison with previousstudies involving galactose. We also present a methodfor distinguishing carbon and nitrogen effects from theircombined profiles and additionally provide a benchmarkof current constraint-based modeling techniques andtheir ability to predict our experimental data.Results and discussionPrototrophic deletion set construction and profilingBriefly, a MATα strain carrying the SGA marker [11,12]was crossed to the MATa yeast deletion set [1], selectedfor diploids, and sporulated. Prototrophic haploids wereselected using the SGA approach [11]. The final genotypeof these 4,772 strains is MATa yfgΔ0::KanMX can1Δ::STE2pr-SpHIS5 his3Δ1 lyp1Δ0. These strains were thenpinned out onto plates containing one of four differentcarbon sources along with one of seven nitrogen sources.Table 1 Summary of conditions and hits called in each conFast/slow Ammonium Proline GlutamateGlucose 41.5/41* 186/417 133/354Galactose 132/461 276/658 400/906Ribose 312/345 981/462 306/412Glycerol NA NA NA* Mean of six replicates. Note that this reflects the average number of false positiveNoise from extremely slow growth on glycerol precluded identification of significanYeast wild-type growth suggests carbon/nitrogeninteractionsThe mean growth rate of all wild-type replicates was calcu-lated in each condition, which revealed extensive variationacross the profiled conditions (Figure 2a; Additional file 2;Materials and methods). As expected, wild-type yeast growsubstantially faster on glucose or galactose than on glycerolor ribose. Similarly, urea is a consistently poor nitrogensource with glutamine and ammonium generally preferred.To systematically examine the interactions between carbonand nitrogen sources over our entire dataset, a linear modelwas fit to the logarithm of wild-type growth rates under theassumption that independent contributions to growth ratewould combine multiplicatively (a multiplicative model fitbetter than simple alternatives such as an additive formu-lation). Indeed, the model suggests that pairs of nitrogenand carbon sources commonly interact to produce a wild-type growth rate phenotype that is different from whatmight be predicted assuming independent contributions,evidenced by the fact that the majority of the interactionterms in the linear model were significant (Figure 2b). Forexample, consider the apparent increase in growth rateobserved under ribose:glutamate when compared toglucose:glutamate (Figure 2a), observable as a positiveinteraction between ribose and glutamate (Figure 2b).When paired with glucose, glutamate is the nitrogensource that yields the fourth fastest growth rate. However,when paired with a much poorer carbon source (for ex-ample, ribose or glycerol) glutamate becomes the nitrogensource that yields the fastest growth rate. This interactionis likely caused by the ability of the cell to utilize glutam-ate not only as a source of nitrogen, but as a secondarycarbon source in the presence of a poor primary carbonsource. When glutamate is deaminated for use as a nitrogensource, alpha-ketoglutarate is produced and can be subse-quently utilized for energy production via the tricarboxylicacid cycle. This dual role is not specific to glutamate. ForionGlutamine Arginine Urea Allantoin169/286 173/920 135/219 95/284270/877 452/530 154/216 124/545291/192 437/46 388/345 379/492NA NA NA NAs in the screen since glucose ammonium was the reference condition. NA,t individual mutant effects.timecolony sizeGlu:Amm Ratetimecolony sizeC:N RatezGlu:Amm RateC:N Ratelowesshaploidprototrophauxotrophicdeletioncollectionprototrophicdeletioncollection16 plate array4,772 mutants1 WT plate aggregatez scoresWT profile (701 reps)includes glycerol LEU2MET15URA3SGA sporulation/ selection4C x 7N = 28 conditions6 replicates Glu:Ammimage at:0, 5, 10, 24 hours4772 mutant strains21 conditionsMATa yfg 0::KanMX his3 1leu2 0 met15 0 ura3 0MATa yfg 0::KanMX can1 ::STE2pr SpHIS5his3 1 lyp1 0MAT can1 ::STE2pr SpHIS5his3 1 lyp1 0Figure 1 (See legend on next page.)VanderSluis et al. Genome Biology 2014, 15:R64 Page 3 of 18http://genomebiology.com/2014/15/4/R64(See figure on previous page.)Figure 1 Experimental overview. A custom prototrophic strain is mated to the entire deletion collection and haploids are selected via SGA. Theresulting prototrophic deletion collection is plated out onto 28 distinct metabolic media, and time course growth rate data are extracted fromplate images. Growth rates are normalized to a glucose:ammonia reference (constructed from six replicates) and z-scores are calculated for eachdeletion, in each condition (except glycerol). WT, wild-type.glycerolribosegalactoseglucose4 3 2 1 0 10.0 1.00.80.6 0.40.2source terminteraction termwild-type growth relative to glucose:ammoniummodel coefficients (log growth rate) relative to glucose:ammoniumglutamineammoniumallantoinglutamateprolinearginineureaglutamineammoniumallantoinglutamateprolinearginineureaglutamineammoniumallantoinglutamateprolinearginineureaglutamineammoniumallantoinglutamateprolinearginineureagalactosegalactose allantoingalactose argininegalactose glutamategalactose glutaminegalactose prolinegalactose ureariboseribose allantoinribose arginineribose glutamateribose glutamineribose prolineribose ureaglycerolglycerol allantoinglycerol arginineglycerol glutamateglycerol glutamineglycerol prolineglycerol ureaallantoinarginineglutamateglutamineprolineureaabFigure 2 Wild-type growth data in all conditions. (a) Average wild-type growth rates in all conditions. Conditions are grouped and colored bycarbon source. Nitrogen sources are ordered by growth rate when paired with glucose, and all values are relative to the glucose:ammonium rate.Error bars represent standard error from 701 wild-type replicates. (b) A linear model fit to log-transformed growth values. Terms forindividual carbon and nitrogen sources are colored, interaction terms are gray. All but the three terms marked with a black circle are significant(P < 0.01); error bars represent standard error.VanderSluis et al. Genome Biology 2014, 15:R64 Page 4 of 18http://genomebiology.com/2014/15/4/R64et al. [13], and Dudley et al. [7] each included a conditionin which galactose is the major source of carbon, and theoverlap between the deletions that we call as effects in ourgalactose conditions and sensitivities collected fromthese three experiments is highly significant (Figure 3a;Additional file 5). We define a galactose-sensitive genefor this purpose as having a significant fitness defect in atleast four of our seven galactose conditions and we obtaina list of 565 such genes (using FDR 20%; Additional files 3and 4). This list covers approximately 50% of the sensitivegenes identified in each of the three previous auxo-trophic screens (Giaever n = 23, P < 10−11; Kuepfer n = 120,P < 2 × 10−16; Dudley n = 16, P < 10−6; hypergeometric;Figure 3a; Additional file 5). Additionally, we discover385 mutants sensitive under galactose not revealed inany of these previous studies. For comparison, the over-lap between two of the previous genome-wide studiesa Galactose sensitivitescommon N = 4,456this studyKuepfer 2005Dudley 2005 Giaever 200251031002115620037143858669 genes show an effect in 1 or more studies3,787 genes show no effect in any studyVanderSluis et al. Genome Biology 2014, 15:R64 Page 5 of 18http://genomebiology.com/2014/15/4/R64example, glutamine is utilized in a similar manner, thoughthe ratio of 'free' carbon skeletons per nitrogen produced isless efficient (1:2 as opposed to 1:1). Despite the fact thatmany of the nitrogen sources share this property, we here-after continue to refer to them simply as 'nitrogen sources'for simplicity. Our results show that the wild-type growthrate can be predicted from independent contributions ofcarbon and nitrogen sources in only 3 of our 28 conditions(Figure 2b). Significant interaction terms in all butthree conditions signify the complex interdependenciesthroughout the metabolic network, thus underscoring theimportance of testing each pair of sources systematically.Fitness determination of deletion mutants over themedia conditionsIn an effort to identify mutant growth defects specificto particular conditions, we derived a model designedto score growth rate for each deletion strain in a givencondition relative to its growth under a reference con-dition (glucose:ammonium). First, the growth rate data(Additional file 1) were normalized for each experimentalcondition with respect to the glucose:ammonium refer-ence (see Materials and methods). This controlled for thegrowth rate differences observable in wild-type cells acrossthe different conditions (Figure 2; Additional file 2) andenabled us to focus on more subtle effects due only to thegenetic perturbation. A modified z-score was then calcu-lated for each mutant strain (see Materials and methods;Additional file 3). This measure is negative if the straingrew slower in the test condition than would be expecteddue to the nutrient environment alone, and positive if thestrain grew faster than expected. The distribution ofgrowth rates in the 701 wild-type replicates was used toassess the statistical significance of mutant effects in eachcondition and estimate a false discovery rate (FDR) forany gene-environment interactions (see Materials andmethods; Additional file 4). Table 1 shows the numberof deletions that grew slower or faster than expected atan FDR threshold of 20% (see Additional file 3 for acomplete list of z-scores). While the large number of wild-type replicates allowed for confidence in the small differ-ences in reference strain growth between various nitrogensources when paired with glycerol, the mutant data on gly-cerol proved to be too noisy due to extremely slow growthto call mutant effects. Therefore, no growth rate (z-score)data are presented for mutant strains on glycerol.Observations in galactose concur with previousauxotrophic studiesTo build additional confidence in our high-throughputdataset, we compared lists of mutants deficient for growthunder galactose to data from several previous studies thathad tested the auxotrophic deletion collection in a varietyof experimental conditions. Giaever et al. [1], Kuepfer100 104 102 103 104 105 1060.10.20.30.40.50.60.70.80.91RecallPrecisionAll Gene PairsMetabolic Pairs OnlybFigure 3 Overlap with known galactose sensitivities andprediction of known functional associations. (a) Overlap betweenmutants sensitive on galactose from several different studies. For thisstudy, galactose sensitivity is defined as a significant z-score in four ormore of our seven galactose conditions. N denotes the total number ofgenes the studies have in common. (b) Precision-recall analysis assessingthe ability of gene-gene similarity to predict co-annotation to specificterms in the Gene Ontology. Results for all gene pairs are shown in blue,and results for a subset of metabolism-related genes (included iniMM904 model) are shown in red.VanderSluis et al. Genome Biology 2014, 15:R64 Page 6 of 18http://genomebiology.com/2014/15/4/R64(Giaever et al. and Kuepfer et al.) was only 15 genes, 12of which are recovered in this study (Figure 3a). Wesuggest two primary reasons for the increased numberof galactose-sensitive mutants discovered in our study.The first is that 47% of these new galactose-sensitivegenes did not have a phenotype when only one nitrogensource (ammonium) was used. Thus, the testing ofa wide-range of nitrogen sources revealed additionalgalactose-sensitive mutants. The second reason is thatprevious studies used more stringent thresholds for galact-ose phenotypes. Smaller quantitative measurements offitness defects across multiple galactose:nitrogen sourcecombinations allow for increased sensitivity in detectinggalactose phenotypes compared with other studies.Another possible explanation for differences betweenour galactose results and those from the Dudley et al.study is the absence of antimycin A in our media. Anti-mycin A inhibits energy production from respiratorypathways and forces the strains to ferment galactose. Inour experiments, yeast had access to oxygen and couldperform both respiration and fermentation with galactoseas a carbon source, which is the natural metabolism ofgalactose by S. cerevisiae [14].Liquid culture validation of mutant fitness measurementsWe independently validated our single mutant fitnessmeasurements by measuring the growth rate of 40mutants in a liquid growth assay performed across 20 ofthe experimental conditions (excluding ribose:arginineand all glycerol pairings; see Materials and methods).The overall correlation between wild-type strain growthrates from these two different approaches was 0.65 (P <0.003; Pearson), suggesting general agreement betweengrowth rates determined on solid and liquid media. Wethen adjusted the liquid growth scores, controllingfor the wild-type rate in the given condition and therelevant mutant rate in glucose:ammonium so theywould reflect condition-specific effects, similar to ourmodified z-score derived from the agar experiment.The Spearman rank correlation between the adjustedliquid growth score and our agar z-score (for 40 mu-tant strains × 19 conditions) was 0.34 (P < 2.2 × 10−16).Further excluding glucose conditions (which are gen-erally sparser in the z-score data as a consequence ofour use of glucose:ammonium as a reference) increasesthis correlation to 0.38. Thus, we conclude that thereis reasonable agreement between the high-throughputmeasures and a lower-throughput liquid growth assay,including for condition-specific effects.Number of environmental sensitivities is correlated withsingle mutant fitness and genetic interaction degreeWe compared our growth measurements with otherquantitative phenotypes measured on the auxotrophicdeletion collection. For example, genetic interactionmapping efforts have measured the single mutant fitnessof all deletion strains from the auxotrophic backgroundon synthetic complete media [3,15] and found a correl-ation between the magnitude of the fitness defect and thenumber of genetic interactions for each single mutant(genetic interaction degree). The prevailing explanationfor this correlation is that genes that display a fitnessdefect represent the subset that are playing an activerole under the condition tested, are additionally notcompletely buffered by other genes, and/or contributeto a wider variety of cellular processes. We observe asimilar correlation between the single mutant fitnessdefect (as previously measured on synthetic completemedia [3]) and the number of significant condition-specific sensitivities in our study (r = 0.33, P < 5 × 10−100;Pearson). Additionally, there is a partial correlation be-tween the number of genetic interactions a gene has andthe number of environments with which it interacts, evenafter controlling for single mutant fitness defect (r = 0.18,P < 5 × 10−31; Pearson). This echoes a previously observedcorrelation between genetic interaction degree and sensi-tivities in more complex chemical environments (r = 0.4,P < 10−5) [6,15]. These results confirm that our study isuncovering more effects for genes known to be pleiotropicor central under a variety of environmental backgrounds[7]. These findings also suggest that hubs are conservedacross different network types, with many of the samegenes conferring robustness to genetic, chemical, andenvironmental perturbations.Mutant sensitivity profiles are predictive of gene functionPrevious genetic interaction studies have shown thathigh profile similarity for mutant sensitivity across variedenvironmental conditions or diverse genetic backgrounds(for example, genetic interaction profiles) is highly predict-ive of similar gene function [5,7,15]. We applied an analo-gous logic to our data to see if similar environmentalsensitivity profiles would also be predictive of similarfunction. Using co-annotation to an informative set ofGene Ontology (GO) terms [16,17] as our standard forfunctional similarity, we ranked all pairs of genes bytheir profile similarity (Pearson) and evaluated theserankings with respect to known functional relationships.We measured a precision of approximately 35% at a recallof 1,000 gene pairs (2-fold over a random baseline of 17%;Figure 3b). Additionally, when we restrict our predictionsto those genes with a known involvement in metabolism(663) we see a much higher precision (precision ~ 65% atrecall = 100), though a similar performance over the in-creased background rate (1.7-fold over 38%; see Materialsand methods). The higher performance for metabolism-related predictions is likely due to the direct relevance ofthe environmental conditions chosen to the study of basicVanderSluis et al. Genome Biology 2014, 15:R64 Page 7 of 18http://genomebiology.com/2014/15/4/R64metabolism. Thus, we have demonstrated an ability to pre-dict general gene function using the guilt-by-associationprinciple, and the diverse environments chosen for thisassay are well-suited to reveal sensitivities in the metabolicnetwork of this newly created prototrophic collection.Metabolic network models show modest ability to predictexperimental dataThe prototroph growth data on minimal media presentedhere are uniquely suited to bring experimental datato bear on theoretical predictions of constraint-basedanalysis of metabolic networks. Constraint-based modelingis a widely used approach to study the metabolic capacityof genome-scale biochemical networks in steady state with-out requiring detailed enzyme kinetic parameters [8]. FBAis the most popular constraint-based approach to computa-tionally predict the phenotypes under environmental andgenetic perturbations and has been shown to successfullypredict gene essentiality, and to a lesser extent, condition-specific essential status in yeast [13,18]. We used our sensi-tivity data to evaluate the ability of constraint-based modelsto predict subtler quantitative sensitivities in a condition-specific manner. We predicted biomass yield, a proxy forgrowth, in all conditions using two versions of the yeastmetabolic network reconstruction: the more recent Source-forge Yeast Consensus Reconstruction v5.35 (hereafterYeast5) [19], and iMM904 [20]. Additionally, we appliedtwo alternative algorithms to predict mutant phenotypes,namely standard FBA [21] and minimization of metabolicadjustment (MoMA) [22]. Predicted biomass productionfluxes were normalized with respect to every mutant's pre-dicted biomass production in glucose:ammonium and thewild-type prediction in each condition to make scoresanalogous to our experimental z-scores. The prediction ofz-scores as opposed to raw growth rates was chosen toassess the adaptability of each model's performance in theface of varied environments, an admittedly more difficultscenario than predicting global or condition-specific essen-tiality. Though the output of the models is quantitative,many conditions predict only a few discrete levels of result-ing biomass production and therefore yield identical predic-tions for the majority of mutants. The mode of the outputaccounted for between 39% and 95% of the predictions, sowe assessed model performance by comparing the pre-dicted set of slow mutants (below the mode biomassproduction) to our set of significant z-scores in eachcondition. Three metrics were collected to assess the per-formance of each model-method combination: averageprecision (across all 20 predicted conditions), average re-call, and the number of conditions in which precisionexceeded random expectation (at P < 0.05 hypergeometric;Figure 4; Additional file 6). Results for positive z-scoreprediction (above the mode biomass) are also available inAdditional file 6 (see Materials and methods).Prediction of condition-specific slow growth provedconsistently above random expectation (Figure 4), thoughvalues of precision are much lower than those previouslyreported in predicting qualitative essentiality (>90% [18]).One key difference between our study and Snitkin et al.[18] (as with Dudley et al. in the section on galactosesensitivity above) is the latter's inclusion of antimycin Ain the media, which inhibits energy production fromrespiration, whereas our strains could naturally respireand ferment. Our results show an advantage for themore recent Yeast5 model over the iMM904 model, aswell as a slight advantage for standard FBA over MoMA.The Yeast5 model was able to perform above random ex-pectation in 14 out of 20 conditions with a mean precisionof 25% and a mean recall of 18% (Figure 4; Additionalfile 6). Recall scores for MoMA were generally higherthan for FBA owing to a much smaller fraction of thepredictions equal to the mode, though this was generallyassociated with a loss of precision. Galactose conditionsappear to be well captured by the two models, and con-sistently perform above random. By contrast, all threeconditions for which no model-method achieved signifi-cance involved glucose (glucose:allantoin, glucose:glutam-ine, glucose:urea). Thus, while the overall performancedemonstrates an above-random ability of these modelsto predict quantitative and condition specific perturbationeffects, their modest precision and recall scores (<50%)suggest substantial room for improvement.An examination of false positives (predicted sensitiveby the model but not observed in the data) and falsenegatives (observed sensitive, not predicted) showedsome functional coherency. Specifically, Kyoto Encyclopediaof Genes and Genomes (KEGG) enrichment of false posi-tives in many conditions revealed connections to centralcarbon metabolism (for example, the tricarboxylic acidcycle), and half of the conditions showed enrichment forthe KEGG sulfur metabolism pathway in the model forfalse positives (Additional file 6). This suggests potentialpathways that may need attention for the development ofimproved models.We also attempted to leverage existing metabolic modelsto demonstrate the widespread metabolic consequencesof these common auxotrophies. To accomplish this, weran the models again using prototrophic and auxotrophicversions of the network on glucose:ammonium and char-acterized each metabolite as either: i) produced in theauxotroph and the prototroph; ii) produced in the proto-troph only; or iii) included in the model but not producedin an optimal solution (see Materials and methods). Thesimulations show that a significant proportion of produ-cible metabolites (18% in iMM904 and 7% in Yeast5;Figure 4c) are unavailable in the auxotrophic network.This means that consequences of using auxotrophicstrains, even under supplementation for their specific slonsenn)itesgluesd.VanderSluis et al. Genome Biology 2014, 15:R64 Page 8 of 18http://genomebiology.com/2014/15/4/R64deficiencies, may have a broader impact than expected.It is our hope that the collection and accompanyinggrowth data presented here will prove invaluable to themetabolic modeling community as it continues to refinethe structure of its models as well as their underlyingbiological assumptions.predicting negative z-scores(slower than reference)0 0.1 0.2mean precisionmean recallconditions exceedingrandom predictioniMM904MoMA FBA MoMAFBAYeast5Yeast5-FBAglucose allantoinglucose arginineglucose glutamateglucose glutamineglucose prolineglucose ureagalactose ammoniumgalactose allantoingalactose argininegalactose glutamategalactose glutaminegalactose prolinegalactose urearibose ammoniumribose allantoinribose arginineribose glutamateribose glutamineribose prolineribose urea00.10.20.30.40.50.60.70.80.91a bFigure 4 Agreement between constraint-based modeling predictiomodeling predictions for slow growth. Precision and recall (blue and greand glucose:ammonium excluded; Additional file 6) and means are showoverlap significantly with significant z-score effects is shown in purple. (bthe Yeast5 model and standard FBA. (c) Number of producible metabolnumber of producible metabolites was counted based on simulation inpeated for a model in which reactions involving auxotrophic marker genportion of metabolites that the auxotrophic model fails to produce in reBroad environmental surveys address incomplete geneannotationsA primary motivation for measuring fitness across diverseenvironments is the discovery of novel phenotypes formutants that have near wild-type fitness under previouslytested conditions. The existence of such mutants in aeukaryotic genome with approximately 6,000 genes isdriven by two main factors. The first is genetic redun-dancy, whereby genes are performing vital functionswithin the cell, but their importance is not captured bysingle mutant phenotypes because other genes are presentthat buffer the loss of function. This occurs at both thelevel of individual genes buffering one another (forexample, duplicate genes [23,24]) and the level of largernetwork structures (for example, parallel pathways). Thesebuffered functions are rapidly being mapped by geneticinteraction studies that delete multiple genes simultan-eously [2,4,11,15,25,26]. The remaining contributing factoris environmental robustness, whereby a gene presumablyhas an important function under some evolutionarilyrelevant circumstance that is not reflected in a laboratoryenvironment (for example, nutrients/media, temperature,stress). Thus, an important motivation for completepairwise coverage of basic metabolic conditions is thedetection of novel fitness defects for genes that becomenecessary only as the condition space is more broadlysurveyed. Interestingly, of the 729 remaining uncharacter-ized mutants in the auxotrophic collection for which wehave single mutant fitness measurements in syntheticcomplete media, a significant fraction of them (609) have0.3 0.4 0.5 0.6 0.7precisionrecallw prediction by conditioniMM904864 produciblemetabolites82%71218%15293%8457%62Yeast 5907 produciblemetabolitesproduced in auxotroph or prototrophproduced in prototroph model onlycand experimental observations. (a) Assessment of constraint based) were calculated for each model in each of 20 conditions (glycerol:*here. The fraction of conditions in which predicted model mutantsPrecision and recall scores (as in (a)) for each individual condition usingfor iMM905 and Yeast5 metabolic models. For each model the totalcose:ammonium (see Materials and methods). The procedure was re-(HIS3, URA3, LEU2, and MET15) were disabled. The chart shows the pro-a fitness greater than 99% of wild-type (hypergeometricP < 7 × 10−66) [27]. Despite the ever-increasing availabilityof high-throughput genomic data for these genes, the taskof eliminating this set has seen only marginal success since2007 [28]. It is possible that these genes (many of whichonly have orthologs in other yeasts) may be responsiblefor functions needed in the native environment of yeastbut unnecessary under standard laboratory conditions.Still others may be required in the lab, but only after vary-ing the nutrient conditions. The focus of recent chemicalgenomics work on subjecting yeast to an extremely broadrange of chemical environments is helping to addressthese genes [5,6], but auxotrophy in the deletion collectionhad precluded measurements of growth on simple butdirectly relevant metabolic conditions. Here we addressthe potential impact of these data on both uncharacterizedgenes and genes of little phenotypic consequence in stand-ard conditions.Novel effects for genes with high fitness in standardconditionsAs described earlier, we observed that the number ofsignificant effects in our data can be weakly predictedby single mutant fitness in synthetic complete media.However, nearly 40% of the S. cerevisiae genome showsVanderSluis et al. Genome Biology 2014, 15:R64 Page 9 of 18http://genomebiology.com/2014/15/4/R64little to no such effect. Of the genes in this study withsingle mutant fitness scores greater than 99% of wild-typeunder synthetic complete media, more than 50% of them(1,548/2,745; Figure 5a) show at least one significantslow-growth effect outside of glucose:ammonium. Mul-tiple random assignments of the number of expectedfalse positives (20% of effect counts listed in Table 1)demonstrate that only approximately 30% of genes shouldshow an effect. Additionally, 5% (142/2,745) show signifi-cant effects in five or more distinct non-glucose:ammoniumconditions compared to a random expectation of 2.6 × 10−5(<<1/2,745). For example, prs2Δ0 (the PRS2 gene encodesone of the four phosphoribosyl-pyrophosphate (PRPP)synthetases encoded in the genome; these synthetases arerequired for nucleotide, histidine, and tryptophan biosyn-thesis) has a single mutant fitness of 1.02 in syntheticcomplete media [3] but shows significant growth defectsin 14 different conditions These conditions are highly co-herent, including all seven galactose conditions, all riboseconditions (except ribose:arginine) and no conditions in-volving glucose except glucose:proline. PRS2 is highlyexpressed under fermentative conditions [29]. Anotherexample is ICL1, which facilitates a key reaction of theglyoxylate cycle, and shows slow growth effects in nine(non-glucose:ammonium) conditions despite a singlemutant fitness score slightly greater than that of wild-typeunder standard lab conditions (1.03) [3].Novel phenotypes for uncharacterized ORFsApproximately 13% of the S. cerevisiae deletion collectionis composed of uncharacterized ORFs [27], 692 of whichare included in this study. Nearly 25% of these unchar-acterized genes show a significant effect in two or morenon-glucose:ammonia conditions (172/692; Figure 5a)compared to the 4% expected given our FDR.One such example with a very specific nitrogen sensitiv-ity signature is FMP32. The fmp32Δ0 strain displays dra-matically decreased fitness under arginine and prolineconditions. While the protein product of FMP32 has beendetected in highly purified mitochondria [30], the geneis otherwise uncharacterized. The fmp32Δ0 strain wasincluded in our liquid confirmation assay and thesesensitivities were confirmed in this independent, small-scale assay (Figure 5b). This highly specific signatureappears to be completely unique to the fmp32Δ0 strain, asno other mutant in the collection shows a similar sensitiv-ity profile.The genes with the highest profile similarity to FMP32are PUT1, PUT3, and RRF1, which have been previouslyimplicated in proline utilization (PUT1, PUT3 [31]) andmitochondrial ribosome recycling/mitochondrial proteinsynthesis during respiration (RRF1 [32,33]). PUT3 inducesPUT1 transcription when proline is present as the bestavailable nitrogen source and the latter (along with PUT2)is responsible for the conversion of proline into glutamatefor further use as a nitrogen source. Our analysis suggeststhat FMP32 is similarly involved in the respiratoryresponse under proline, though the reason for its add-itional sensitivity under arginine remains unclear. Theseexamples show the utility of interactions between genesand simple environments in uncovering the function ofboth individual uncharacterized genes and genes withouta previously observed fitness defect in more completemedia.Clustering of metabolic conditions reveals carbon sourceas primary factor driving mutant profilesJust as gene-gene correlation predicts functional similar-ities, we expect a high correlation between condition pairsto reflect a substantial overlap in the cellular machinery re-quired to utilize the provided carbon and nitrogen sources.When our matrix of z-scores is hierarchically clustered inboth the gene and condition dimensions, a structure clearlydriven by carbon sources emerges (Figure 6; see Materialsand methods). All of the glucose conditions cluster to-gether, as do both the galactose and ribose conditions.The sole exception to this is glucose:proline, which fallsin the galactose cluster. We attribute this observation tothe fact that the utilization of proline as a nitrogensource requires some respiration. The glucose:prolinesignature reveals sensitivity in a number of respiratorydeficient mutants, which is atypical for glucose condi-tions in general since fermentation is generally preferredover respiration when cells are grown on glucose. Thisrespiration-dependent signature is strong enough toplace the glucose:proline profile in the galactose clusterwhere one would expect a modest profile contributionfrom both respiration- and fermentation-related processes(Figure 6), as is observed in growth on galactose [14].Matrix factorization distinguishes carbon fromnitrogen effectsFurther examination of gene and environmental profilesafter clustering revealed cases where a gene (for example,FMP32) exhibited an effect in multiple instances of a par-ticular nitrogen source (for example, proline or arginine),but without a specific pattern with regard to carbon source(or vice versa). This is expected behavior for genes requiredfor the utilization of a particular carbon/nitrogen sourceregardless of the context. In order to more formallyextract a list of sensitivities for each source of carbonor nitrogen regardless of its partner, we employed amethod known as non-negative matrix factorization (NMF)[34,35] to decompose our experimental data into a collec-tion of characteristic source signatures. When a matrix ofthese source signatures is multiplied by a matrix describingthe source composition in each of our conditions, the resultshould approximate our experimental observations. NMF0 1 2 3 4 5 6 0 1 2 3 4 5 6Uncharacterized Genes100200300400500600YPD-Healthy MutantsAt least x effects At least x effects2500200015001000500Observed EffectsExpected False PositivesObserved EffectsExpected False Positivesglucose prolinegalactose prolineribose prolineribose allantoinribose glutamateglucose argininegalactose arginineribose glutaminegalactose glutamineribose urearibose arginineglucose allantoingalactose ureaglucose ureagalactose ammoniumgalactose allantoinribose ammoniumglucose glutamineglucose glutamategalactose glutamatePUT1RRF1PUT3YER163CRXT2DEP1GDH2YBR027CFMP32YLR282C0 10 20 30 4000.20.40.60.81 galactose ammonium0 10 2000.20.40.60.81 glucose ammonium0 10 20 30 4000.20.40.60.81 galactose proline0 10 20 3000.20.40.60.81 glucose proline0 10 20 30 4000.20.40.60.81 galactose arginine0 5 10 15 2000.20.40.60.81 glucose arginineWild TypeFMP32 0 1Z-ScoreabcFigure 5 (See legend on next page.)VanderSluis et al. Genome Biology 2014, 15:R64 Page 10 of 18http://genomebiology.com/2014/15/4/R64allows us to run this multiplication in reverse and fit thesource signatures as an unknown factor. These sourcesignatures are available in Additional file 7, and severalof them demonstrate enrichment for related GO termsand KEGG pathways.One example of a decomposed signature involves genesthat are sensitive when glutamate is chosen as a nitrogensource. These genes are enriched for annotations relatingto endocytosis, endosome and vacuole related transport,and retrograde transport (Additional file 7). Extracellularinappropriate trafficking in these mutants causes highlevels of permease activity that inhibit cell growth.Many mutants (92) appear in both the galactose andribose signatures, and overlapping GO enrichments inthese conditions reveal many of these genes to haveknown involvement in various aspects of respiration. Forexample, enrichment for GO terms relating to mitochon-drial organization and translation, as well as 'aerobicrespiration', appear highly significant in both of these sig-natures (Additional file 7). Exceptions include GAL path-(See figure on previous page.)Figure 5 Measuring effects for poorly characterized genes. (a) Counting slow-growth effects for under-characterized genes. Histograms showthe total number of mutants with at least x significant slow-growth effects in our data from the set of uncharacterized genes (left, orange), andgenes with little to no fitness defect on synthetic complete media (right, blue; single mutant fitness > 98% of wild-type). As a control, the expectednumber of false positives (20% of significant effects in each condition) were randomly distributed among all genes, and the number of effectsfor each gene was counted again. Gray bars show the mean of 1,000 such randomizations. (b) Z-score data show specific growth defects forthe uncharacterized gene FMP32 when grown on proline or arginine. (c) Liquid growth confirmations for effects highlighted in X-axis, time in hours. Y-axis, optical density. (b). Two replicates of FMP32 mutants are shown (blue line) along with six replicates of a wild-type strain (black dashed line) in twoproline and two arginine conditions. The effects are pronounced when compared to observations in similar ammonium conditions.VanderSluis et al. Genome Biology 2014, 15:R64 Page 11 of 18http://genomebiology.com/2014/15/4/R64glutamate decreases cellular amino acid permease activityby redirecting intracellular trafficking of the permeaseGap1 from the plasma membrane to the vacuolar mem-brane [36]. Many of the mutations in our glutamatesignature increase Gap1 activity by misdirecting theprotein to the plasma membrane [37]. Although GAP1is transcribed at equal levels in cells grown on urea andglutamate, permease activity in urea grown cells is 100times higher than glutamate-grown cells [38]. Inappropri-ate Gap1 activity is toxic in the context of high concentra-tions of single amino acids [39], and we speculate that thegalactose glutaminegalactose prolinegalactose ammoniumgalactose allantoinglucose prolinegalactose glutamategalactose argininegalactose urearibose allantoinribose ammoniumribose prolineribose glutamateribose glutamineribose urearibose arginineglucose glutamateglucose arginineglucose ureaglucose allantoinglucose glutamineZ-ScoreFigure 6 A clustergram of Z-scores for the 500 mutants with the highdimensions. Conditions organize themselves primarily by carbon source, faway mutants that fall uniquely into the galactose carbonsignature ('galactose metabolic process' P < 1.3 × 10−4) andgenes involved in acetyl-CoA biosynthesis that appear tobe specifically sensitive under ribose (P < 1.4 × 10−6). Asmore complex environments are mapped, multivariatestatistical techniques will become increasingly important indetermining which environmental constituents are actuallyrelevant to which experimental observations, and careshould be taken when designing experiments to ensuretheir successful application (for example, complete com-binatorial coverage of relevant environmental factors).est variance. The data have been hierarchically clustered in bothlling into three distinct clusters.VanderSluis et al. Genome Biology 2014, 15:R64 Page 12 of 18http://genomebiology.com/2014/15/4/R64Environmental and genetic perturbations can provokesimilar cellular statesBeginning to test the immense space of possible environ-mental and chemical conditions combined with experi-ments that have queried the space of genetic perturbations[15] allows us to investigate how these spaces interrelate.For example, if mappings can be found between them, wecan apply knowledge from the already extensively mappedgenetic perturbation networks to the intractable space ofenvironmental variation. While the sensitivity profile for agiven condition most certainly includes genes directlyrequired for the processing of the provided raw materials(for example, the galactose metabolism pathway undergalactose conditions), it also contains information aboutgenes that, though not directly involved, are nonethelessindirectly required for optimal cell growth. These profilesthen reveal much more than the functions of genes forwhich we measure a fitness defect, and in fact give us ahigh dimensional fingerprint of the internal cellular state.We propose that genetic perturbations may put the cellinto a very similar state as would an alteration of theenvironment. For example, the deletion of a gene thatencodes a transporter may exhibit a profile that mimicsthe wild-type profile in an environment where the corre-sponding substrate is absent. Downstream consequencesof the environment or genetic perturbation may causesubtle and seemingly unexpected sensitivities. Thus, geneticperturbation experiments and environmental perturbationexperiments may both result in the same phenotypic pro-file. A similar principle has been demonstrated throughthe observation that deletion mutants with similar doublemutant sensitivity profiles tend to be functionally related[15]. Parsons et al. [5] first applied this principle to predictdrug targets, reasoning that a genetic sensitivity profile ona chemical that targets an individual gene would be similarto a sensitivity profile of a strain with the correspondinggene deleted. When we compared sensitivity profiles fromour condition experiments to that of query-deletionscrossed into the auxotrophic deletion collection viaSGA [15], we found several interesting cases wheregenetic perturbation profiles significantly overlapped withsensitivity profiles from our environmental perturbations(see Materials and methods). For example, the queries inthe top 10% in terms of similarity to galactose:urea areenriched for members of the threonine and methioninebiosynthesis pathway (hom2, hom3, hom6, thr4; Figure 7;GO:0006566 'threonine metabolic process' P < 4.5 × 10−2;KEGG 'glycine, serine and threonine metabolism' P <2.9 × 10−2). The strength and specificity of this similarity isnot driven by a handful of mutants in the collection, butinstead by trends across a much larger set of genes. Wespeculate that the profile similarity in this case may bedue to accumulation of aspartate, which is upstream ofhomoserine and threonine biosynthesis, and is excreted inpart through urea production. Growth on urea in thesetting of the respiratory growth of galactose may result inthe accumulation of aspartate.The idea of comparing environmental and genetic per-turbations can be generalized to other genome-wide per-turbation data as well. For example, we observe significantcorrelations between our glutamate signature and a rapa-mycin sensitivity profile as measured by two differentchemical genomic screens (Hillenmyer et al. P < 10−18 [6];Parsons et al. P < 10−9 [5]). The enrichment for transport-related terms observed in the glutamate signature (above),and its similarity to a rapamycin profile make sense giventhat rapamycin redirects trafficking of Gap1 from theplasma membrane to the vacuole [40]. Thus, the same setof mutations in vesicle trafficking that lead to inappropri-ate expression of Gap1 permease activity in cells grownon glutamate also cause inappropriate permease activityfollowing rapamycin treatment.ConclusionThe creation of the original yeast deletion collection hashad a profound impact on the way in which reverse gen-etic experiments are performed. Yet despite a staggeringnumber of successful studies, the inherent auxotrophiescreate a major blind-spot in a fundamental area of cellularfunction, and previous reviews of the topic have called forthe creation and use of standardized prototrophic strainsfor metabolic experiments [9]. Recently, Mülleder and col-leagues [41] have addressed the deletion collection auxot-rophies by introducing a plasmid containing sequences forHIS3, URA3, LEU2, andMET15. The resource used in thisstudy differs in that URA3, LEU2, and MET15 are in theirnative genomic locations, with the exception of HIS3,which is provided by Schizosaccharomyces pombe HIS5under the SGA reporter [11]. Without the necessity forplasmid selection, or possible effects on gene expressiondue to non-chromosomal location, we anticipate that ourdeletion collection will see frequent use by experimentalists.The use of a genome-wide prototrophic strain collectionenables truly informative sensitivity screening in metabol-ically controlled conditions. This represents a first step inprobing how nutrients in the environment jointly affectcellular response with or without additional genetic per-turbation. This study demonstrates that much work is yetto be done to understand growth in even simple environ-ments. A solid grasp of the surprisingly complex responsesto simple environments will add much needed context tostudies done in more complex environments.This study has demonstrated the potential of this col-lection, when screened against simple environments, touncover phenotypes for hundreds of mutants that arephenotypically normal in standard lab conditions. Webelieve that the stock of simple experiments that mightreveal a phenotype for these mutants has not yet beenged fThc guth oiatVanderSluis et al. Genome Biology 2014, 15:R64 Page 13 of 18http://genomebiology.com/2014/15/4/R64Figure 7 Comparison of sensitivity profiles from environmental tomutants in threonine biosynthetic pathway (circled in red) were obtaineprofile obtained in this study when strains are grown on galactose:urea.when grown in a specific environment, and when subjected to a specifimutants would all be expected to accumulate aspartate because these msimilarity in genetic interaction space between these mutants and growtthe internal accumulation of aspartate or some other metabolic intermedexhausted and expect that this whole-genome prototrophiccollection will be an invaluable resource to the community.The rising number of metabolomics studies, fueled inpart by the increasing accuracy of experimental mass-spectrometry, as well as the growing interest in metab-olism as central to many common ailments in humans,make it more important than ever to properly designmetabolically relevant experiments in the model eukaryoteS. cerevisiae. Central to that goal is a version of the deletioncollection that is unhindered by historical auxotrophicrequirements.For example, while central metabolism is unrivaledamong cellular processes with respect to our ability tomake in silico predictions from constraint-based meta-bolic models, it is far from a fully understood system.Our results show a generally weak ability to predictcondition-specific sensitivities, though performance isclearly above a random baseline. The prediction ofcondition-specific sensitivities is admittedly more diffi-cult than the prediction of sensitivities in general, butit was our estimation that FBA and MoMA would bewell suited to approximate our observations given oursimple experimental setup. Their only moderate successin doing so demonstrates the current limitations ofconstraint-based modeling and the difficulty of relatingmodels built from biomass predictions to quantitativegrowth rate data. There might be several possible reasonsnetic perturbations. High dimensional sensitivity information forrom SGA experiments [15]. These profiles correlate with the sensitivityis suggests a correspondence between the internal states of the cellsenetic perturbation. For example, hom2Δ, hom3Δ, hom6Δ, and thr4Δants shut down a major metabolic shunt for aspartate. The phenotypicn galactose:urea suggests that growth on galactose:urea may causee unique to the hom2Δ, hom3Δ, hom6Δ, and thr4Δ mutants.for the discrepancy between in silico and in vivo results.First, the success of predicting growth defects hinges onthe proper formulation of biomass composition. While asingle biomass composition is used for all our simulations,it likely changes across environmental conditions. Futurestudies could address this issue by measuring the com-position of yeast cells under different nutrient settings.A second limitation of purely flux-based models istheir inability to make predictions about componentsthat have an indirect effect on metabolism. Consider,for example, the enrichment for transport-related geneswhose deletion confers glutamate-specific sensitivities.Their putative role in nutrient sensing and signalingreflects the fact that, despite its constrained nature, themetabolic network operates as part of a much largerand more dynamic network. More generally, the basicconstraint-based modeling approaches ignore regulatorymechanisms. Several attempts have been made to bridgethis gap and they rely either on 'omic' data to constrainthe activity of specific reactions [42-44] or on integrating amathematical representation of gene regulation with themetabolic model [45-47]. We feel that the availability ofthis whole-genome collection and accompanying growthdata well suited to studies of metabolism will help thecommunity to develop and test novel models and methodsto better capture the operation of the greater cellularnetwork.VanderSluis et al. Genome Biology 2014, 15:R64 Page 14 of 18http://genomebiology.com/2014/15/4/R64Central to the understanding of the network as a wholeis the idea that a whole-genome screen reveals indirect aswell as direct consequences of the perturbation tested.Positive gene-environment interactions under ribose con-ditions may well illustrate this point. The median z-scorefor the 166 genes annotated to 'chromosome segregation'in GO is negative for all seven galactose conditions, yetpositive for all seven ribose conditions (binomial sign-testP < 6.2 × 10−5). We believe this shift may be explained byfundamental cellular rate limitations. Failure to segregatechromosomes in the midst of even moderate growth (forexample, galactose) can have very severe consequences,ultimately limiting growth rate, whereas comparativelyslow growth (for example, ribose) affords additional timefor slowly segregating mutants to complete segregation.These mutants grow faster than we expect despite no ap-parent link between carbon metabolism and chromosomesegregation. Thus, growth rates under one conditiondisclose information about the interplay between a widevariety of cellular subsystems, giving us a readout of theinternal cellular state. Similarly, a mutant profile acrossmany environments gives us information about how es-sential that gene may be in any of those various cellularstates, in addition to elucidating any direct role thatgene may have in direct utilization of the provided nu-trients. Analysis of our growth data recapitulated therole of vesicle trafficking in the regulation of the aminoacid permease Gap1, relating growth on glutamate tothe drug rapamycin. This broader view of whole-genomescreen information then allows for integration of profilesacross different perturbation types (chemical, genetic, envir-onmental), and should ultimately aid us in applying know-ledge gained in one arena to observations made in another.Materials and methodsConstruction of a prototrophic deletion collectionAs recently described [48], the strains in the standardMATa deletion collection (MATa yfgΔ0::KanMX his3Δ1leu2Δ0 met15Δ0 ura3Δ0) [1] were mated to a MATαcan1Δ::STE2pr-SpHIS5 his3Δ1 lyp1Δ0 strain, creatingdiploids (selection on minimal media + his + G418). Thesewere sporulated and successive pinnings on selectivemedia were used to select prototrophic MATa strainscarrying each deletion allele. These prototrophic strainswere organized into an array of 16 plates including oneentire plate of the wild-type strain (hoΔ::KanMX), withadditional wild-type replicates in each row and column ofevery plate (701 in all). The entire prototrophic collectionis available upon request, as is the individual SGA-readyprototroph strain for crossing into other collections.Media preparationMinimal growth media were prepared using yeast nitrogenbase (BD Difco, Sparks, Maryland,) with the specifiedcarbon and nitrogen sources. Carbon sources includedglucose, galactose, ribose, and glycerol. Nitrogen sourcesincluded ammonium, allantoin, arginine, glutamate, glu-tamine, proline, and urea. Carbon sources were providedat a concentration of 2%; nitrogen sources were 3.8 mMwith respect to nitrogen.Calculation of growth rateSixteen 16 × 24 well plates were grown in 28 chemicalconditions for 24 to 48 hours. Plates were scanned on aflatbed transparency scanner at 0, 5, 10, 24 and, in thecase of glycerol, 48 hours. Each condition is composedof one carbon source and one nitrogen source. In total,4,772 mutants were grown, and colony areas were ex-tracted from tiff images by CellProfiler [49] and precisetime points were taken from EXIF data in the digital im-ages. These values were used to compute an estimate ofthe growth rate of each colony equal to the slope of theleast-squares linear fit of area (pixels) to time (seconds).Colonies with insufficient data were given a growth rateof NaN, colonies with a negative calculated growth ratewere defined to have a growth rate of 0.Definition and construction of a reference conditionSix replicates of the glucose:ammonium combination weremerged to form a reference condition, establishing a base-line score for each deletion. The six replicates were firstnormalized to each other to control for differences in theoverall scale of growth rates, then averaged together ac-cording to the following procedure. For each array plate(p) the glucose:ammonium replicate with the fewest miss-ing data points was held out (PlateA) and the remainingfive replicates were LOWESS smoothed (window size = 50%of available data) and normalized by:GAplatep0 ¼ GAplatep  PlateAlowess GAplatep The result of this approach is quite robust to the choiceof PlateA, and so we used whichever replicate had the few-est number of missing values and would therefore providethe most complete LOWESS fit. After normalizing fivereplicates to the sixth, all six were averaged together tocreate one reference plate, and this procedure is repeated16 times to create a glucose:ammonium reference for eacharray plate.Normalization of experimental rates against referenceIn every experimental condition (Y), each plate wasLOWESS smoothed (window size = 50% of available data)against the constructed glucose:ammonium reference plate,then normalized:CondYplatep0 ¼ CondYplatep GAref plowess CondYplatep VanderSluis et al. Genome Biology 2014, 15:R64 Page 15 of 18http://genomebiology.com/2014/15/4/R64Recovery of missing dataIn certain cases, a growth rate of NaN was assigned toa colony due to insufficient data being collected byCellProfiler. In an effort to recover any good data,these cases were visually inspected by five researchersoperating independently and a vote was taken to deter-mine whether to leave it as missing data (NaN) or assignit a growth rate of 0, indicating that the colony appearedto be correctly plated but non-viable. In total, 1,362 of2,601 colonies were recovered this way.Transformation from normalized rates to z-scoresFor each array plate, at each position, a strain-wise stand-ard deviation is calculated across the residuals of the sixglucose:ammonium (GA) replicates.Similarly, a plate-wise standard deviation is calculatedthat accounts for the general growth variation on the plate,separately for each condition. These are then combined,and a z-score measure is calculated for each strain on eachexperimental plate:z ¼ CondYplatep0−GAref pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffistddev strainð Þ2 þ stddev plateð Þ2qThese z-scores are an expression of the difference inmagnitude and direction between the growth observedat each position of a plate under a given condition fromthe same position (and hence deletion) under the referenceGA model.Spatial smoothing procedureThe plate level spatial smoothing filter is similar to thatfound in [3]. First, temporarily replace any extremevalues (top and bottom 5%) along with NaNs with theplate mean. Second, replace previous NaN positions withvalues from a two-dimensional symmetric gaussian filter.Third, compute and subtract the residual between thetwo-dimensional smoothed plate and its mean.Choosing effect thresholdsEach condition had 701 wild-type replicates. The meanand standard deviation of the set of wild-type z-scoreswere used to define a normal distribution against whichP-values for the experimental z-scores could be calculated.This information allowed the use of Benjamini-Hochbergprocedure to establish condition-specific effect thresholdsas a function of a desired FDR (Additional file 4).Liquid growth confirmation assayThe growth rate of 40 mutants in a liquid growth assaywas measured across 20 of the experimental conditionsexcluding ribose:arginine and all glycerol pairings. Liquidculture assays were not performed for the ribose:arginineconditions because the combination of these carbon andnitrogen sources did not allow arginine to maintainadequate solubility over the duration of the experiment.The precipitation of arginine prevented accurate opticaldensity readings from being obtained and thus thesedata were excluded from our subsequent analyses. Sixreplicate wells contained the wild-type strain and eachmutant strain was represented twice. Cells were pre-grownon glucose:ammonia medium and diluted at a low densityinto the growth medium of interest. Growth rates weredetermined as the maximum optical density (saturation)divided by the time to saturation. A simple model wasfavored in order to robustly accommodate drastic differ-ences in curve characteristics between fast growth andslow growth conditions (for example, galactose versusribose).We adjusted the liquid growth scores by dividing themean of mutant growth slopes by the mean of wild-typegrowth slopes in the relevant condition. We further nor-malized these scores by dividing them by the correspondingadjusted mutant score in glucose:ammonium so they wouldreflect condition-specific effects, similar to our modifiedz-score derived from the agar experiment.Gene Ontology and KEGG enrichments and co-annotationstandardsGO and KEGG annotations were downloaded in January2011 [16,50].Genome-scale metabolic modeling (FBA and MoMA)Two S. cerevisiae metabolic models were used for mutantbiomass prediction. The Yeast Consensus Reconstructionversion 5.35 (Yeast5) [19] and iMM904 [20]. Yeast5 con-sisted of 898 ORFs, 2,031 reactions and 1,594 metabolitesand the iMM904 model contained 901 ORFs, 1,597 reac-tions and 1,234 metabolites. Default biomass descriptionswere used for both models.Wild-type biomass production flux for each conditionwas obtained using FBA [21] in MATLAB with theCOBRA Toolbox [51], which assumes optimal biomassproduction (that is, maximum biomass yield). Mutantbiomass flux was predicted using both FBA [21] inMATLAB with the COBRA Toolbox [51] and MoMA[22] in MATLAB with the ILOG CPLEX optimizationsuite. MoMA was formulated as a quadratic program-ming problem, whereby mutant fluxes were selectedthat minimized the Euclidean distance from an optimalwild-type flux distribution. The yeast wild-type fluxdistribution was calculated as a network flux solutionproducing maximal biomass flux, determined by FBA,with minimal total fluxes [52].FBA and MoMA biomass fluxes were correlated withboth raw and normalized (z-score) experimental growthrates using the Spearman rank correlation. Predictediterations, though in practice the results converged inVanderSluis et al. Genome Biology 2014, 15:R64 Page 16 of 18http://genomebiology.com/2014/15/4/R64biomass fluxes were also normalized for comparison toexperimental growth rate z-scores (separately in eachcondition):NormalizedFluxΔxCondY ¼ RawFluxΔxCondYRawFluxΔxCondGlu:Amm  RawFluxwild−typeCondYPrediction of positive z-scores was also carried out,though performance was generally below random ex-pectation (Additional file 6). This is likely due to thefact that many positive z-scores corresponded to rawgrowth rates for mutants that were faster than wild-typeunder the same condition, a consequence that FBA- andMoMA-based methods would find difficult or impossibleto predict.To calculate the effect of gene deletions on the meta-bolic network (Figure 4c), sets of producible metaboliteswere calculated for the complete model, and for a mutantwith all four auxotrophic marker genes deleted. Produ-cible metabolites were calculated for both iMM904 andYeast5 models in the glucose:ammonium media conditionby adding a special exchange reaction for each metaboliteand iteratively optimizing flux exported through that reac-tion. If the export flux for a given metabolite exceeded0.001 (with an upper and lower bound on internal reactionsset to ±1,000), it was classified as 'producible.' A non-zerothreshold is required to limit false positives as a resultof numerical errors. The threshold was determined tobe robust by scaling the upper and lower bounds, as wellas the threshold by a large constant and counting thenumber of producible metabolites. Obtaining consistentresults in these experiments led us to conclude that nu-merical errors are an order of magnitude smaller thancontributions from stoichiometry.Source signature decomposition via modified non-negativematrix factorizationGrowth data were decomposed using a variant of NMF[34]. Following transformation to z-scores, the data weremade binary using condition-specific FDR estimates asthresholds (20% FDR; Additional file 4). The resultingBoolean Data matrix was treated as numeric and servedas the target for decomposition. Genes without any signifi-cant z-scores in any condition (empty rows) were removed,as were the columns involving growth on glycerol. We thendefined a Coefficient matrix that related Condition rows inthe data to their component Sources. This matrix then hadC columns and S rows. For example, the glucose-urea col-umn has a 1 in the glucose row and a 1 in the urea row.Our task is then to find a Signatures matrix (Genes ×Sources) such that the difference between the Data matrixand the Signatures-Coefficients product is minimized:Data G;Cð Þ≈Signatures G; Sð Þ  Coefficients S;Cð Þfewer than 10, and repeated trials from different randominitializations of the Signature matrix showed the resultsto be quite stable. Genes were added to the signature listin Additional file 7 if their value exceeded 0.4.Comparison to SGA dataFor the comparison to auxotrophic SGA data representedin Figure 7, the SGA data were taken from [15]. The SGAdata and the z-score data (Additional file 3) were inde-pendently normalized so that row and column vectors hada euclidean length approximately equal to 1, and missingvalues were set to 0. Inner product was then used tomeasure the similarity between SGA 'queries' and environ-mental profiles. The top 10% of queries in each conditionwere checked for enrichment for GO terms and KEGGpathway annotations, and the resulting P-values wereBonferroni corrected to account for the number of terms/pathways tested against.Comparison with previous whole-genome screenson galactoseFigure 3a uses a publicly available image for the basis of theVenn diagram. This image is used with permission underthe terms of the Creative Commons Attribution-ShareAlike 3.0 Unported license [53].Additional filesAdditional file 1: Raw growth rates for all 28 conditions and all4,772 strains.Additional file 2: Raw growth rates for 701 wild-type replicates inall conditions.Additional file 3: z-score data for 21 conditions.Additional file 4: FDR 10% and 20% thresholds for z-score data.Additional file 5: A list of 565 galactose-sensitive genes as well asoverlap details between this set and three other full genomestudies using the auxotrophic collection on galactose.Additional file 6: Summary of fast/slow prediction accuracy of FBAmodels (sheets 1 and 2). Also contains KEGG pathway enrichments forsets of genes predicted to be sensitive but not observed and vice versaTo ensure linear independence among the columns ofthe Coefficient matrix, we removed all but one glucose:ammonium (glucose:ammonium01) column, removing thesame columns in the Data matrix. Traditional NMF woulduse a multiplicative update algorithm applied to both theSignature and the Coefficient matrix to find the best fit tothe data; however, we chose to fix the Coefficient matrix atthe initial defined values (0 or 1). This gives each Signaturecolumn equal weight and prevents over-fitting causedby the sparsity of the Data matrix and the dramaticallydifferent number of non-zero elements from one columnto the next. The multiplicative update was applied for 20(sheets 3 and 4).Additional file 7: List of genes in each signature and GO enrichments.VanderSluis et al. Genome Biology 2014, 15:R64 Page 17 of 18http://genomebiology.com/2014/15/4/R64AbbreviationsFBA: flux balance analysis; FDR: false discovery rate; GO: Gene Ontology;KEGG: Kyoto Encyclopedia of Genes and Genomes; MoMA: minimization ofmetabolic adjustment; NMF: non-negative matrix factorization; ORF: openreading frame; SGA: synthetic genetic array.Competing interestsThe authors declare that they have no competing interests.Authors’ contributionsDCH, OGT, CLM, and AAC conceived the study. DCH and AAC performed thelaboratory experiments. CN provided experimental resources for the liquidgrowth assay. BJV, CP, and TS analyzed the raw data. BJV, CP, and EKdesigned and performed further computational analysis and experiments.BJV, DCH, BS, BP, CLM, and AAC wrote the manuscript. All authors read andapproved the final manuscript.AcknowledgementsBJV and CLM are partially supported by grants from the National ScienceFoundation (DBI 0953881) and the National Institutes of Health (R01HG005084).BJV was also partially supported by the University of Minnesota DoctoralDissertation Fellowship. DCH is supported by the National Institutes of Health(R01 GM101091-01) and the National Science Foundation (11122240). BP wassupported by the Wellcome Trust and the ‘Lendület Program’ of the HungarianAcademy of Sciences. BS was supported by the European Union and the Stateof Hungary, co-financed by the European Social Fund in the framework ofTÁMOP 4.2.4. A/2-11-1-2012-0001 'National Excellence Program'. CN acknowledgesfunding from the CCSRI, grant number 20830. AAC is the Canada Research Chairin Metabolomics for Enzyme Discovery and is supported by the Ontario EarlyResearcher Award, by the Canadian Institutes for Health Research, and by theNatural Sciences and Engineering Research Council of Canada, and by theCanadian Foundation for Innovation and the Ontario Leader's OpportunityFund. DCH, CLM, OGT, and AAC were supported by the National Institute ofGeneral Medical Sciences (NIGMS) Center of Excellence P50 GM071508. Finally,we thank David Botstein for advice and insight throughout the project.Author details1Department of Computer Science and Engineering, University of MinnesotaTwin Cities, 200 Union St SE, Minneapolis, MN 55455, USA. 2Department ofBiology, Santa Clara University, 500 El Camino Real, Santa Clara, CA 95053,USA. 3Department of Plant Biology, University of Minnesota Twin Cities, 1445Gortner Avenue, Saint Paul, MN 55108, USA. 4Institute of Biochemistry,Biological Research Centre, Hungarian Academy of Sciences, H-6701, Szeged,Hungary. 5University of British Columbia, Pharmaceutical Sciences, 2405Wesbrook Mall, Vancouver, BC V6T 1Z3, Canada. 6Department of ComputerScience, Princeton University, Princeton, NJ 08540, USA. 7Lewis-Sigler Institutefor Integrative Genomics, Princeton University, Princeton, NJ 08544, USA.8Donnelly Centre for Cellular and Biomolecular Research and Department ofMolecular Genetics, University of Toronto, 160 College Street, Toronto, ONM5S 3E1, Canada.Received: 7 December 2013 Accepted: 10 April 2014Published: 10 April 2014References1. Giaever G, Chu AM, Ni L, Connelly C, Riles L, Véronneau S, Dow S, Lucau-DanilaA, Anderson K, André B, Arkin AP, Astromoff A, El-Bakkoury M, Bangham R,Benito R, Brachat S, Campanaro S, Curtiss M, Davis K, Deutschbauer A, EntianK-D, Flaherty P, Foury F, Garfinkel DJ, Gerstein M, Gotte D, Güldener U,Hegemann JH, Hempel S, Herman Z, et al: Functional profiling of theSaccharomyces cerevisiae genome. Nature 2002, 418:387–391.2. Tong AHY, Lesage G, Bader GD, Ding H, Xu H, Xin X, Young J, Berriz GF, Brost RL,Chang M, Chen Y, Cheng X, Chua G, Friesen H, Goldberg DS, Haynes J,Humphries C, He G, Hussein S, Ke L, Krogan N, Li Z, Levinson JN, Lu H, Ménard P,Munyana C, Parsons AB, Ryan O, Tonikian R, Roberts T, et al: Global mapping ofthe yeast genetic interaction network. Science 2004, 303:808–813.3. Baryshnikova A, Costanzo M, Kim Y, Ding H, Koh J, Toufighi K, Youn J-Y, OuJ, San Luis B-J, Bandyopadhyay S, Hibbs M, Hess D, Gingras A-C, Bader GD,Troyanskaya OG, Brown GW, Andrews B, Boone C, Myers CL: Quantitativeanalysis of fitness and genetic interactions in yeast on a genome scale.Nat Methods 2010, 7:1017–1024.4. Bandyopadhyay S, Mehta M, Kuo D, Sung M-K, Chuang R, Jaehnig EJ,Bodenmiller B, Licon K, Copeland W, Shales M, Fiedler D, Dutkowski J,Guénolé A, van Attikum H, Shokat KM, Kolodner RD, Huh W-K, Aebersold R,Keogh M-C, Krogan NJ, Ideker T: Rewiring of genetic networks in responseto DNA damage. Science 2010, 330:1385–1389.5. Parsons AB, Brost RL, Ding H, Li Z, Zhang C, Sheikh B, Brown GW, Kane PM,Hughes TR, Boone C: Integration of chemical-genetic and geneticinteraction data links bioactive compounds to cellular target pathways.Nat Biotechnol 2004, 22:62–69.6. Hillenmeyer ME, Fung E, Wildenhain J, Pierce SE, Hoon S, Lee W, Proctor M,St Onge RP, Tyers M, Koller D, Altman RB, Davis RW, Nislow C, Giaever G:The chemical genomic portrait of yeast: uncovering a phenotype for allgenes. Science 2008, 320:362–365.7. Dudley AM, Janse DM, Tanay A, Shamir R, Church GM: A global view ofpleiotropy and phenotypically derived gene function in yeast. Mol SystBiol 2005, 1:2005.0001.8. Price ND, Reed JL, Palsson BØ: Genome-scale models of microbial cells:evaluating the consequences of constraints. Nat Rev Microbiol 2004,2:886–897.9. Pronk JT: Auxotrophic yeast strains in fundamental and applied research.Appl Environ Microbiol 2002, 68:2095–2100.10. Brem RB, Yvert G, Clinton R, Kruglyak L: Genetic dissection oftranscriptional regulation in budding yeast. Science 2002, 296:752–755.11. Tong AHY, Evangelista M, Parsons AB, Xu H, Bader GD, Pagé N, Robinson M,Raghibizadeh S, Hogue CW, Bussey H, Andrews B, Tyers M, Boone C:Systematic genetic analysis with ordered arrays of yeast deletionmutants. Science 2001, 294:2364–2368.12. Tong AHY, Boone C: Synthetic genetic array analysis in Saccharomycescerevisiae. Methods Mol Biol 2006, 313:171–192.13. Kuepfer L, Sauer U, Blank LM: Metabolic functions of duplicate genes inSaccharomyces cerevisiae. Genome Res 2005, 15:1421–1430.14. Barford JP, Hall RJ: An examination of the crabtree effect inSaccharomyces cerevisiae: the role of respiratory adaptation. J GenMicrobiol 1979, 114:267–275.15. Costanzo M, Baryshnikova A, Bellay J, Kim Y, Spear ED, Sevier CS, Ding H,Koh JLY, Toufighi K, Mostafavi S, Prinz J, St Onge RP, VanderSluis B,Makhnevych T, Vizeacoumar FJ, Alizadeh S, Bahr S, Brost RL, Chen Y, CokolM, Deshpande R, Li Z, Lin Z-Y, Liang W, Marback M, Paw J, San Luis B-J,Shuteriqi E, Tong AHY, van Dyk N, et al: The genetic landscape of a cell.Science 2010, 327:425–431.16. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP,Dolinski K, Dwight SS, Eppig JT, Harris MA, Hill DP, Issel-Tarver L, Kasarskis A,Lewis S, Matese JC, Richardson JE, Ringwald M, Rubin GM, Sherlock G: Geneontology: tool for the unification of biology, The Gene OntologyConsortium. Nat Genet 2000, 25:25–29.17. Myers CL, Barrett DR, Hibbs MA, Huttenhower C, Troyanskaya OG: Findingfunction: evaluation methods for functional genomic data. BMC Genomics2006, 7:187.18. Snitkin ES, Dudley AM, Janse DM, Wong K, Church GM, Segrè D: Model-drivenanalysis of experimentally determined growth phenotypes for 465 yeastgene deletion mutants under 16 different conditions. Genome Biol 2008,9:R140.19. Heavner BD, Smallbone K, Barker B, Mendes P, Walker LP: Yeast 5 - anexpanded reconstruction of the Saccharomyces cerevisiae metabolicnetwork. BMC Syst Biol 2012, 6:55.20. Zomorrodi AAR, Maranas CCD: Improving the iMM904 S. cerevisiaemetabolic model using essentiality and synthetic lethality data. BMC SystBiol 2010, 4:178.21. Orth JD, Thiele I, Palsson BØ: What is flux balance analysis? Nat Biotechnol2010, 28:245–248.22. Segrè D, Vitkup D, Church GM: Analysis of optimality in natural andperturbed metabolic networks. Proc Natl Acad Sci U S A 2002,99:15112–15117.23. VanderSluis B, Bellay J, Musso G, Costanzo M, Papp B, Vizeacoumar FJ,Baryshnikova A, Andrews B, Boone C, Myers CL: Genetic interactionsreveal the evolutionary trajectories of duplicate genes. Mol Syst Biol2010, 6:429.24. Musso G, Costanzo M, Huangfu M, Smith AM, Paw J, San Luis B-J, Boone C,Giaever G, Nislow C, Emili A, Zhang Z: The extensive and condition-dependentnature of epistasis among whole-genome duplicates in yeast. Genome Res2008, 18:1092–1099.models: the COBRA Toolbox v2.0. Nat Protoc 2011, 6:1290–1307.52. Holzhütter H-G: The principle of flux minimization and its application toestimate stationary fluxes in metabolic networks. Eur J Biochem 2004,271:2905–2922.53. Venn’s four ellipse construction. http://en.wikipedia.org/wiki/File:Venn's_four_ellipse_construction.svg.doi:10.1186/gb-2014-15-4-r64Cite this article as: VanderSluis et al.: Broad metabolic sensitivityprofiling of a prototrophic yeast deletion collection. Genome Biology2014 15:R64.VanderSluis et al. Genome Biology 2014, 15:R64 Page 18 of 18http://genomebiology.com/2014/15/4/R6425. Ryan CJ, Roguev A, Patrick K, Xu J, Jahari H, Tong Z, Beltrao P, Shales M, QuH, Collins SR, Kliegman JI, Jiang L, Kuo D, Tosti E, Kim H-S, Edelmann W,Keogh M-C, Greene D, Tang C, Cunningham P, Shokat KM, Cagney G,Svensson JP, Guthrie C, Espenshade PJ, Ideker T, Krogan NJ: Hierarchicalmodularity and the evolution of genetic interactomes across species.Mol Cell 2012, 46:691–704.26. Schuldiner M, Collins SR, Thompson NJ, Denic V, Bhamidipati A, Punna T,Ihmels J, Andrews B, Boone C, Greenblatt JF, Weissman JS, Krogan NJ:Exploration of the function and organization of the yeast early secretorypathway through an epistatic miniarray profile. Cell 2005, 123:507–519.27. Cherry JM, Hong EL, Amundsen C, Balakrishnan R, Binkley G, Chan ET,Christie KR, Costanzo MC, Dwight SS, Engel SR, Fisk DG, Hirschman JE, HitzBC, Karra K, Krieger CJ, Miyasato SR, Nash RS, Park J, Skrzypek MS, Simison M,Weng S, Wong ED: Saccharomyces Genome Database: the genomicsresource of budding yeast. Nucleic Acids Res 2012, 40:D700–D705.28. Peña-Castillo L, Hughes TR: Why are there still over 1000 uncharacterizedyeast genes? Genetics 2007, 176:7–14.29. Landry CR, Oh J, Hartl DL, Cavalieri D: Genome-wide scan reveals thatgenetic variation for transcriptional plasticity in yeast is biased towardsmulti-copy and dispensable genes. Gene 2006, 366:343–351.30. Reinders J, Zahedi RP, Pfanner N, Meisinger C, Sickmann A: Toward thecomplete yeast mitochondrial proteome: multidimensional separationtechniques for mitochondrial proteomics. J Proteome Res 2006, 5:1543–1554.31. Huang HL, Brandriss MC: The regulator of the yeast proline utilizationpathway is differentially phosphorylated in response to the quality ofthe nitrogen source. Mol Cell Biol 2000, 20:892–899.32. Teyssier E, Hirokawa G, Tretiakova A, Jameson B, Kaji A, Kaji H:Temperature-sensitive mutation in yeast mitochondrial ribosomerecycling factor (RRF). Nucleic Acids Res 2003, 31:4218–4226.33. Kanai T, Takeshita S, Atomi H, Umemura K, Ueda M, Tanaka A: A regulatoryfactor, Fil1p, involved in derepression of the isocitrate lyase gene inSaccharomyces cerevisiae–a possible mitochondrial protein necessaryfor protein synthesis in mitochondria. Eur J Biochem 1998, 256:212–220.34. Lee DD, Seung HS: Learning the parts of objects by non-negative matrixfactorization. Nature 1999, 401:788–791.35. Lee DD, Seung HS: Algorithms for non-negative matrix factorization.Adv Neural Inf Process Syst 2001, 13:556–562.36. Roberg KJ: Physiological regulation of membrane protein sorting late inthe secretory pathway of Saccharomyces cerevisiae. J Cell Biol 1997,137:1469–1482.37. Chen EJ, Kaiser CA: Amino acids regulate the intracellular trafficking ofthe general amino acid permease of Saccharomyces cerevisiae. Proc NatlAcad Sci U S A 2002, 99:14837–14842.38. Stanbrough M, Magasanik B: Transcriptional and posttranslationalregulation of the general amino acid permease of Saccharomycescerevisiae. J Bacteriol 1995, 177:94–102.39. Risinger AL, Cain NE, Chen EJ, Kaiser CA: Activity-dependent reversibleinactivation of the general amino acid permease. Mol Biol Cell 2006,17:4411–4419.40. Chen EJ, Kaiser CA: LST8 negatively regulates amino acid biosynthesis asa component of the TOR pathway. J Cell Biol 2003, 161:333–347.41. Mülleder M, Capuano F, Pir P, Christen S, Sauer U, Oliver SG, Ralser M: Aprototrophic deletion mutant collection for yeast metabolomics andsystems biology. Nat Biotechnol 2012, 30:1176–1178.42. Becker SA, Palsson BO: Context-specific metabolic networks are consistentwith experiments. PLoS Comput Biol 2008, 4:e1000082.43. Shlomi T, Cabili MN, Herrgård MJ, Palsson BØ, Ruppin E: Network-basedprediction of human tissue-specific metabolism. Nat Biotechnol 2008,26:1003–1010.44. Jensen PA, Papin JA: Functional integration of a metabolic networkmodel and expression data without arbitrary thresholding. Bioinformatics2011, 27:541–547.45. Covert MW, Schilling CH, Palsson B: Regulation of gene expression in fluxbalance models of metabolism. J Theor Biol 2001, 213:73–88.46. Covert MW, Xiao N, Chen TJ, Karr JR: Integrating metabolic, transcriptionalregulatory and signal transduction models in Escherichia coli.Bioinformatics 2008, 24:2044–2050.47. Chandrasekaran S, Price ND: Probabilistic integrative modeling ofgenome-scale metabolic and regulatory networks in Escherichia coliand Mycobacterium tuberculosis. Proc Natl Acad Sci U S A 2010,107:17845–17850.Submit your next manuscript to BioMed Centraland take full advantage of: • Convenient online submission• Thorough peer review• No space constraints or color figure charges• Immediate publication on acceptance• Inclusion in PubMed, CAS, Scopus and Google Scholar• Research which is freely available for redistribution48. Gibney PA, Lu C, Caudy AA, Hess DC, Botstein D: Yeast metabolic andsignaling genes are required for heat shock survival and have littleoverlap with the heat-induced genes. Proc Natl Acad Sci U S A 2013,110:E4393–E4402.49. Jones TR, Kang IH, Wheeler DB, Lindquist RA, Papallo A, Sabatini DM,Golland P, Carpenter AE: Cell Profiler Analyst: data exploration andanalysis software for complex image-based screens. BMC Bioinformatics2008, 9:482.50. Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M: The KEGG resourcefor deciphering the genome. Nucleic Acids Res 2004, 32:D277–D280.51. Schellenberger J, Que R, Fleming RMT, Thiele I, Orth JD, Feist AM, ZielinskiDC, Bordbar A, Lewis NE, Rahmanian S, Kang J, Hyduke DR, Palsson BØ:Quantitative prediction of cellular metabolism with constraint-basedSubmit your manuscript at www.biomedcentral.com/submit"@en ; edm:hasType "Article"@en ; edm:isShownAt "10.14288/1.0221415"@en ; dcterms:language "eng"@en ; ns0:peerReviewStatus "Reviewed"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "BioMed Central"@en ; ns0:publisherDOI "10.1186/gb-2014-15-4-r64"@en ; dcterms:rights "Attribution 4.0 International (CC BY 4.0)"@en ; ns0:rightsURI "http://creativecommons.org/licenses/by/4.0/"@en ; ns0:scholarLevel "Faculty"@en ; dcterms:title "Broad metabolic sensitivity profiling of a prototrophic yeast deletion collection"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/55930"@en .