YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree...

11
RESEARCH ARTICLE SUMMARY YEAST GENOMICS Systematic analysis of complex genetic interactions Elena Kuzmin,* Benjamin VanderSluis,* Wen Wang, Guihong Tan, Raamesh Deshpande, Yiqun Chen, Matej Usaj, Attila Balint, Mojca Mattiazzi Usaj, Jolanda van Leeuwen, Elizabeth N. Koch, Carles Pons, Andrius J. Dagilis, Michael Pryszlak, Jason Zi Yang Wang, Julia Hanchard, Margot Riggi, Kaicong Xu, Hamed Heydari, Bryan-Joseph San Luis, Ermira Shuteriqi, Hongwei Zhu, Nydia Van Dyk, Sara Sharifpoor, Michael Costanzo, Robbie Loewith, Amy Caudy, Daniel Bolnick, Grant W. Brown, Brenda J. Andrews,Charles Boone,Chad L. MyersINTRODUCTION: Genetic interactions occur when mutations in different genes combine to result in a phenotype that is different from expectation based on those of the individual mutations. Negative genetic interactions oc- cur when a combination of mutations leads to a fitness defect that is more exacerbated than expected. For example, synthetic lethality oc- curs when two mutations, neither of which is lethal on its own, generate an inviable double mutant. Alterna- tively, positive genetic interactions occur when genetic perturbations combine to generate a double mutant with a greater fitness than expected. Global digenic interaction studies have been useful for understanding the functional wiring diagram of the cell and may also provide insight into the genotype-to-phenotype relationship, which is important for tracking the missing heritability of human health and disease. Here we describe a network of higher-order trigenic interactions and explore its implications. RATIONALE: Variation in phenotypic outcomes in different individuals is caused by genetic de- terminants that act as modifiers. Modifier loci are prevalent in human populations, but knowl- edge regarding how variants interact to modu- late phenotype in different individuals is lacking. Similarly, in yeast, traits including conditional essentialityin which certain genes are essen- tial in one genetic background but nonessential in anotheroften result from an interplay of multiple modifier loci. Because complex mod- ifiers may underlie the genetic basis of phys- iological states found in natural populations, it is critical to understand the landscape of higher-order genetic interactions. RESULTS: To survey trigenic interactions, we designed query strains that sampled key fea- tures of the global digenic interaction network: (i) digenic interaction strength, (ii) average number of digenic interactions, and (iii) di- genic interaction profile similarity. In total, we tested ~400,000 double and ~200,000 tri- ple mutants for fitness defects and identified ~9500 digenic and ~3200 trigenic negative interactions. Although tri- genic interactions tend to be weaker than digenic interactions, they were both enriched for functional relation- ships. About one-third of trigenic interactions identified novelcon- nections that were not observed in our di- genic control network, whereas the remaining approximately two-thirds of trigenic interac- tions modifieda digenic interaction, sug- gesting that the global digenic interaction network is important for understanding the trigenic interaction network. Despite their functional enrichment, trigenic interactions also bridged distant bioprocesses. We estimate that the global trigenic interaction network is ~100 times as large as the global digenic network, highlighting the potential for com- plex genetic interactions to affect the biol- ogy of inheritance. CONCLUSION: The extensive network of tri- genic interactions and their ability to gen- erate functionally diverse phenotypes suggest that higher-order genetic interactions may play a key role in the genotype-to-phenotype relationship, genome size, and speciation. RESEARCH Kuzmin et al., Science 360, 283 (2018) 20 April 2018 1 of 1 The list of author affiliations is available in the full article online. *These authors contributed equally to this work. Corresponding author. Email: [email protected] (C.L.M.); [email protected] (B.J.A.); charlie.boone@ utoronto.ca (C.B.) Cite this article as: E. Kuzmin et al., Science 360, eaao1729 (2018). DOI: 10.1126/science.aao1729 Trigenic interactions Digenic interactions Estimate of global negative genetic interaction networks ~ 5x10 5 ~ 1x10 8 Digenic interactions Trigenic interactions Trigenic interactions Digenic interactions Co-Localization Co-annotation Co-expression PPI Trigenic interactions are functionally informative Trigenic interactions often overlap digenic interactions Global digenic network properties of query genes enriched for trigenic interactions Digenic interaction profile similarity Neg. digenic interaction strength Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic interactions were scored in parallel for the corresponding single mutant and double mutant query strains (large nodes). Systematic analysis of trigenic interac- tions. We surveyed for trigenic interactions and found that they are ~100 times as prevalent as digenic interactions, often modify a digenic interaction, and connect functionally related genes as well as genes in more diverse bioprocesses (multicolored nodes). PPI, protein-protein interaction. ON OUR WEBSITE Read the full article athttp://dx.doi. org/10.1126/ science.aao1729 .................................................. on February 17, 2021 http://science.sciencemag.org/ Downloaded from

Transcript of YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree...

Page 1: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

RESEARCH ARTICLE SUMMARY

YEAST GENOMICS

Systematic analysis of complexgenetic interactionsElena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh DeshpandeYiqun Chen Matej Usaj Attila Balint Mojca Mattiazzi Usaj Jolanda van LeeuwenElizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Yang WangJulia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San LuisErmira Shuteriqi Hongwei Zhu Nydia Van Dyk Sara Sharifpoor Michael CostanzoRobbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Brenda J AndrewsdaggerCharles Boonedagger Chad L Myersdagger

INTRODUCTION Genetic interactions occurwhen mutations in different genes combineto result in a phenotype that is different fromexpectation based on those of the individualmutations Negative genetic interactions oc-cur when a combination of mutations leads toa fitness defect that is more exacerbated thanexpected For example synthetic lethality oc-curs when two mutations neither ofwhich is lethal on its own generatean inviable double mutant Alterna-tively positive genetic interactionsoccur when genetic perturbationscombine to generate a double mutantwith a greater fitness than expectedGlobal digenic interaction studies have beenuseful for understanding the functional wiringdiagramof the cell andmay also provide insightinto the genotype-to-phenotype relationshipwhich is important for tracking the missingheritability of human health and disease Herewe describe a network of higher-order trigenicinteractions and explore its implications

RATIONALEVariation in phenotypic outcomesin different individuals is caused by genetic de-terminants that act as modifiers Modifier lociare prevalent in human populations but knowl-edge regarding how variants interact to modu-late phenotype in different individuals is lackingSimilarly in yeast traits including conditionalessentialitymdashin which certain genes are essen-tial in one genetic background but nonessentialin anothermdashoften result from an interplay ofmultiple modifier loci Because complex mod-ifiers may underlie the genetic basis of phys-iological states found in natural populationsit is critical to understand the landscape ofhigher-order genetic interactions

RESULTS To survey trigenic interactions wedesigned query strains that sampled key fea-tures of the global digenic interaction network(i) digenic interaction strength (ii) averagenumber of digenic interactions and (iii) di-genic interaction profile similarity In totalwe tested ~400000 double and ~200000 tri-ple mutants for fitness defects and identified

~9500 digenic and ~3200 trigenicnegative interactions Although tri-genic interactions tend to be weakerthan digenic interactions they wereboth enriched for functional relation-ships About one-third of trigenicinteractions identified ldquonovelrdquo con-

nections that were not observed in our di-genic control network whereas the remainingapproximately two-thirds of trigenic interac-tions ldquomodifiedrdquo a digenic interaction sug-gesting that the global digenic interactionnetwork is important for understanding thetrigenic interaction network Despite theirfunctional enrichment trigenic interactionsalso bridged distant bioprocesses We estimatethat the global trigenic interaction networkis ~100 times as large as the global digenicnetwork highlighting the potential for com-plex genetic interactions to affect the biol-ogy of inheritance

CONCLUSION The extensive network of tri-genic interactions and their ability to gen-erate functionally diverse phenotypes suggestthat higher-order genetic interactions mayplay a key role in the genotype-to-phenotyperelationship genome size and speciation

RESEARCH

Kuzmin et al Science 360 283 (2018) 20 April 2018 1 of 1

The list of author affiliations is available in the full article onlineThese authors contributed equally to this workdaggerCorresponding author Email chadmumnedu (CLM)brendaandrewsutorontoca (BJA) charliebooneutorontoca (CB)Cite this article as E Kuzmin et al Science 360 eaao1729(2018) DOI 101126scienceaao1729 Trigenic interactionsDigenic interactions

Estimate of global negative genetic interaction networks

~ 5x105 ~ 1x108

Digenicinteractions

Trigenicinteractions

Trigenic interactionsDigenic interactions

Co-Localization

Co-annotation

Co-expression

PPI

Trigenic interactions are functionally informative

Trigenic interactions often overlap digenic interactions

Global digenic network properties of query genes enriched for trigenic interactions

Digenic interaction

profile similarity

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Mapping digenic and corresponding trigenic interaction networksDigenic and trigenic interactions were scored in parallel for the corresponding single mutant and double mutant query strains (large nodes)

Systematic analysis of trigenic interac-tionsWe surveyed for trigenic interactionsand found that they are ~100 times asprevalent as digenic interactions oftenmodify a digenic interaction and connectfunctionally related genes as well as genesin more diverse bioprocesses (multicolorednodes) PPI protein-protein interaction

ON OUR WEBSITE

Read the full articleat httpdxdoiorg101126scienceaao1729

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

RESEARCH ARTICLE

YEAST GENOMICS

Systematic analysis of complexgenetic interactionsElena Kuzmin12dagger Benjamin VanderSluis3 Wen Wang3 Guihong Tan1

Raamesh Deshpande3 Yiqun Chen1 Matej Usaj1 Attila Balint14DaggerMojca Mattiazzi Usaj1 Jolanda van Leeuwen1 Elizabeth N Koch3 Carles Pons3

Andrius J Dagilis5 Michael Pryszlak1 Jason Zi Yang Wang12 Julia Hanchard12

Margot Riggi6789 Kaicong Xu3 Hamed Heydari12 Bryan-Joseph San Luis1

Ermira Shuteriqi1 Hongwei Zhu1 Nydia Van Dyk1 Sara Sharifpoor1

Michael Costanzo1 Robbie Loewith689 Amy Caudy12 Daniel Bolnick5

Grant W Brown14 Brenda J Andrews12sect Charles Boone1210sect Chad L Myers3sect

To systematically explore complex genetic interactions we constructed ~200000 yeasttriple mutants and scored negative trigenic interactions We selected double-mutantquery genes across a broad spectrum of biological processes spanning a range ofquantitative features of the global digenic interaction network and tested for a geneticinteraction with a third mutation Trigenic interactions often occurred among functionallyrelated genes and essential genes were hubs on the trigenic network Despite theirfunctional enrichment trigenic interactions tended to link genes in distant bioprocessesand displayed a weaker magnitude than digenic interactions We estimate that theglobal trigenic interaction network is ~100 times as large as the global digenic networkhighlighting the potential for complex genetic interactions to affect the biology ofinheritance including the genotype-to-phenotype relationship

Genetic interactions occur when a combi-nation of mutations in different genesleads to an unexpected phenotype thatdeviates from a model incorporating thecombined effects of the corresponding

single-mutant phenotypes In humans each in-dividual carries thousands of different variantsthat may modulate gene function which meansthat there is incredible potential for combinato-rial genetic interactions to determine our person-al phenotype (1 2) Indeed genetic interactionsare thought to represent a substantial compo-nent of the missing heritability associated withcurrent genome-wide association studies (GWAS)(3) However the statistical limitations associatedwith GWAS data sets preclude the detection ofspecific genetic interactions and thus potentialgenetic networks underlying inherited traitsincluding diseases remain elusive (3ndash5) To ad-dress the role of genetic interactions in thegenotype-to-phenotype relationship we havebeen exploring their general principles throughsystematic analysis of genetic networks under-lying cellular fitness in a genetically tractablemodel system the budding yeast Saccharomyces

cerevisiae (6) Our previous studies focused pre-dominantly on genetic interactions involving twogenes (digenic interactions) (7) In this study weanalyzed a series of single- double- and triple-mutants by quantifying their colony size as aproxy for fitness to systematically explore com-plex genetic interactionsThere are two basic types of fitness-based ge-

netic interactions A negative genetic interac-tion refers to a combination of mutations thatresults in a fitness defect that is more severe thanexpected (8) Synthetic lethality is an extremeexample of a negative genetic interaction andoccurs when two mutations neither of which islethal on its own combine and lead to an inviabledouble-mutant phenotype (9 10) Converselya positive genetic interaction occurs when acombination of genetic perturbations gener-ates a double mutant with a greater fitness thanexpected one example is genetic suppression inwhich the fitness defect of a query mutant isalleviated by a mutation in a second gene (11)To map a global digenic interaction networkfor yeast we constructed millions of double mu-tants and identified hundreds of thousands of

negative and positive genetic interactions (7) Toput these results in perspective although only~1000 of the ~6000 total yeast genes are indi-vidually essential and cause lethality when deleted(12) and an equivalent number of nonessentialgenes cause a slow-growth defect under standardlaboratory conditions (13) ~550000 different yeastgene pairs display a combinatorial negative ge-netic interaction including a subset of ~10000extreme synthetic lethal interactions involvingnonessential gene pairs (7) Thus there are nu-merous potential ways to generate extreme lethalphenotypes through negative digenic interactionsof nonessential gene pairsThe set of digenic interactions for a query gene

its genetic interaction profile provides a quan-titative measure of function because genes withsimilar roles have overlapping profiles (14 15)Genes belonging to the same biological path-way or protein complex display highly similargenetic interaction profiles Moreover a globalnetwork based on digenic interaction profile sim-ilarity reveals a hierarchy of functional moduleswhich includes detailed pathways and complexesthat in turn cluster into larger modules corre-sponding to bioprocesses Those larger units sub-sequently assemble into modules correspondingto cellular compartments to outline the functionalarchitecture of a cell (7)A complete understanding of the role of ge-

netic interactions in the genotype-to-phenotyperelationship requires that we also investigate com-plex higher-order genetic interactions involvingmore than two genes Because there are ~2000times as many triple gene combinations as genepairs (~18 million) it is possible that there is asubstantially greater number of trigenic thandigenic interactions and that higher-order inter-actions may be important for driving inheritedtraits In this study we surveyed yeast trigenicinteractions sampling quantitative features ofthe digenic network and explored the impli-cations of the higher-order genetic interactionnetwork

Mapping trigenic interactionsquantitatively and surveying theglobal trigenic landscape

To explore the trigenic interaction landscape wedesigned query strains that sampled three keyquantitative features of our global digenic in-teraction network (7) We designed query strainscarrying mutations in two genes spanning arange of the following features (1) digenic in-teraction strength (2) number of digenic inter-actions (average digenic interaction degree) and(3) digenic interaction profile similarity (Fig 1Aand table S1) Gene pairs were selected to fill

RESEARCH

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 1 of 9

1The Donnelly Centre University of Toronto 160 College Street Toronto Ontario M5S 3E1 Canada 2Department of Molecular Genetics University of Toronto 160 College Street TorontoOntario M5S 3E1 Canada 3Department of Computer Science and Engineering University of MinnesotandashTwin Cities 200 Union Street Minneapolis MN 55455 USA 4Department ofBiochemistry University of Toronto 160 College Street Toronto Ontario M5S 3E1 Canada 5Department of Integrative Biology 1 University Station C0990 University of Texas at Austin AustinTX 78712 USA 6Department of Molecular Biology University of Geneva Geneva 1211 Switzerland 7Department of Biochemistry University of Geneva 1211 Geneva Switzerland 8iGE3 (Instituteof Genetics and Genomics of Geneva) 1211 Geneva Switzerland 9Swiss National Centre for Competence in Research Programme Chemical Biology 1211 Geneva Switzerland 10RIKEN Center forSustainable Resource Science Wako Saitama JapanThese authors contributed equally to this work daggerPresent address Rosalind and Morris Goodman Cancer Research Centre McGill University 1160 Avenue des Pins Ouest Montreal Quebec H3A 1A3 CanadaDaggerPresent address Center for Chromosome Stability Department of Cellular and Molecular Medicine University of Copenhagen Blegdamsvej 3B 2200 Copenhagen N DenmarksectCorresponding author Email chadmumnedu (CLM) brendaandrewsutorontoca (BJA) charliebooneutorontoca (CB)

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

bins of varying digenic interaction attributes andto cover all major biological processes in the cellthus producing a sample that would provide a di-verse survey of the trigenic interaction landscapeWe largely focused on unambiguous singletonsbecause duplicated genes represent a relativelysmall subset of genes and thus can only representa small fraction of the global trigenic interac-tion network For this survey we constructed151 double-mutant query strains and 302 single-mutant strains encompassing 47 temperature-sensitive alleles of different essential genes and255 deletion alleles of nonessential genes Thequery strains in this set were selected to spanthe different digenic attribute bins according topredefined thresholds (table S1) An additional31 double-mutant queries fell outside of the de-fined thresholds but were included for validationand comparison purposes (data S1 to S3) (16)The fitness of the resulting query strains was mea-sured using a quantitative growth assay and thebehavior of the single- and double-mutant querystrains showed strong agreement with our pre-viously published data set (figs S1 and S2 anddata S4) (7 15)

Trigenic interaction screening required devel-opment and implementation of three opera-tional components First synthetic genetic array(SGA) analysismdashan automated form of yeast ge-netics that is often used to cross a query genemutation into an array of single mutants togenerate a defined set of haploid double mutants(6)mdashwas adapted such that a double-mutantquery strain could be crossed into an array ofsingle mutants to generate triple mutants fortrigenic interaction analysis (Fig 1B) Becausethe identification of a trigenic interaction re-quires comparison with the corresponding dou-ble mutants we also conducted screens in whichthe individual mutants of the query gene pairwere scored for digenic interactions (Fig 1B)Second for experimental feasibility we assem-bled a diagnostic array of 1182 strains compris-ing 990 nonessential gene deletion mutants and192 essential gene mutants carrying temperature-sensitive alleles which combine to span ~20of the yeast genome (data S5) The diagnosticarray was designed to be highly representativeof the rest of the genome in terms of exhibitedgenetic interaction profiles (fig S3) Briefly ar-

ray strains were selected from a larger geneticinteraction data set for their ability to representdifferent regions of the global network in a min-imally redundant way This was accomplishedby iteratively selecting strains to maximize theperformance of profile similarities when predict-ing coannotations to a functional gold standard(17) Third we developed a scoring method thet-SGA score which combines double- and triple-mutant fitness estimates derived from colonysize measurements to identify trigenic interactionsquantitatively (Fig 1C) The t-SGA score differsfrom the MinDC score reported previously (18)because it accounts for all cases in which two ofthe genes are not independent resulting in an ex-pectation that contains digenic interaction effectsscaled by the fitness of the noninteracting genes(fig S4) (16) The final trigenic t-SGA interactionscore then accounts for digenic effects but alsoenables detection of trigenic interactions in whichdigenic effects of insufficient explanatory powercan be foundWe focused exclusively on the analysis of del-

eterious negative trigenic interactions for tworeasons First quantitative scoring of negative

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 2 of 9

BA

Diagnostic MATa Single Mutant Array

Mutant alleles (Δ or ts) Wild type alleles

MATα Mutant Query

Double mutantgene1Δ gene2Δ

Single mutantgene1Δ

Single mutantgene2Δ

SGA

CMATa Output Array

Triple mutant array

Double mutant array-1

Double mutant array-2

observed triple mutant fitness

expected triple mutant fitness

digenic interactions

2 Avg digenic interaction degree

1 Neg digenic interaction strength

low medium high

3 Interaction profile similarity

K

ML

OPQR

A B C D E F

N

Quantitative criteria for query gene selection

negative genetic interactionno genetic interactionpositive genetic interaction

Fig 1 Triple-mutant synthetic genetic array (SGA) analysis (A) Criteria for selecting query strains forsampling trigenic interaction landscape of singleton genes in yeast The gene pairs were grouped into threegeneral categories based on a range of features (1) Digenic interaction strength Gene pairs were directlyconnected by zero to very weak (digenic interaction score 0 to ndash008 n = 74 strains) weak (ndash008 to ndash01n = 32) or moderate (ltndash01 n = 45) negative digenic interactions (2) Number of digenic interactions Genepairs had a low (10 to 45 interactions n = 50) intermediate (46 to 70 n = 53) or high (gt71 n = 48) averagedigenic interaction degree (denoted by the number of black edges of each node) (3) Digenic interaction profilesimilarity Gene pairs had low (score ndash002 to 003 n = 46 represented by genes A and B which show arelatively low overlap of genetic interactions with genes K to R) intermediate (003 to 01 n = 59 representedby genes C and D which display an intermediate overlap of genetic interactions) or high (gt01 n = 46 rep-resented by genes E and F which display a relatively high level of overlap of genetic interactions) functionalsimilarity as measured by digenic interaction profile similarity and coannotation to the same GO term(s) Querymutant genes were either nonessential deletion mutant alleles (D) or conditional temperature-sensitive (ts)alleles of essential genes (B) Diagram illustrating the triple-mutant SGA experimental strategy To quantify atrigenic interaction three types of screens are conducted in parallel To estimate triple-mutant fitness a double-mutant query strain carrying two desired mutated genes of interest (red and blue filled circles) is crossedinto a diagnostic array of single mutants (black filled circle) Meiosis is induced in heterozygous triplemutants and haploid triple-mutant progeny is selected in sequential replica pinning steps In parallel single-mutant control query strains are used to generate double mutants for fitness analysis (C) Triple-mutant

SGA quantitative scoring strategy The top equation shows the quantification of a digenic interaction where eij is the digenic interaction score ƒij is theobserved double-mutant fitness and the expected double-mutant fitness is expressed as the product of single-mutant fitness estimates ƒiƒj In thebottom equation the trigenic interaction score (tijk) is derived from the digenic interaction score where ƒijk is the observed triple-mutant fitness and ƒiƒjƒkis the triple-mutant fitness expectation expressed as the product of three single-mutant fitness estimates The influence of digenic interactions issubtracted from the expectation and each digenic interaction is scaled by the fitness of the third mutation

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genetic interactions is often more accurate thanthat for positive interactions because there is agreater signal-to-noise ratio for negative geneticinteractions Hence negative genetic interac-tions are associated with lower false-positiveand false-negative rates than positive interactions(8) a feature that is important for the robuststatistical analysis necessary to differentiate truetrigenic interactions from the extensive back-ground digenic network Second negative digenicinteractions are generally more functionally in-formative than positive digenic interactions (8)and thus the large-scale mapping of a negativetrigenic interaction network is expected to pro-vide the most mechanistic insight into genefunction and pathway wiring

Trigenic interactions are enrichedfor functionally related genes

To obtain sufficient precision we carried outeach analysis which involved screening the in-dividual query genes for digenic interactionsand the double-mutant query for trigenic inter-actions in at least two replicate screens with fourcolonies per screen (fig S5) In total we tested410399 double and 195666 triple mutants forfitness defects meeting a previously establishedintermediate magnitude cutoff (15) (data S2) andidentified 9363 digenic and 3196 trigenic negativeinteractions From detailed validation of trigenicinteractions of our CLN1-CLN2 double-mutantquery which was screened previously (19) weestimated a false-negative rate of ~40 a false-positive rate of ~20 and a true-positive ratebetween ~60 and ~75 (table S2 and fig S6) (16)which is consistent with our previous global di-genic network analysis (8)The distribution of trigenic interaction degree

for array strains shows that the majority of low-degree genes (70) account for ~88 of all tri-genic interactions whereas highly connectedgenes contribute the remaining ~12 of interac-tions Thus the trigenic interactions are not as-sociated with a small set of highly connectedgenes rather the interactions are distributedacross many different genes (fig S7) On the otherhand with a smaller more biased set of double-mutant query genes the distribution of trigenicinteraction degree shows that ~22 of them ac-counted for 51 of trigenic interactions indicat-ing that a particular subset of the digenic querieswere enriched for trigenic interactions (fig S7)About one-third of the newly mapped trigenicinteractions identified connections that werenot observed in our digenic control network werefer to these as ldquonovelrdquo trigenic interactions Theremaining approximately two-thirds of the trigenicinteractions overlapped a digenic interaction whilestill exhibiting a stronger than expected fitnessdefect in the triple mutant these we refer to asldquomodifiedrdquo trigenic interactions (fig S8A) Thusalthough a substantial fraction of trigenic inter-actions elucidate totally new functional informationthe majority of the trigenic interactions we mappedexpand upon the digenic interaction networkWe first assessed the functional information

embedded in the trigenic network by comparing

the distributions of digenic and trigenic inter-actions across different biological processesAs observed previously (15) digenic interactionswere enriched among genes annotated to thesame biological process and although the mag-nitude of trigenic interaction enrichment wassomewhat lower they were comparably enrichedfor genes within the same bioprocess (Fig 2A)We also evaluated the enrichment of digenic andtrigenic interactions across common functionalstandards including annotation to the same GeneOntology (GO) biological process subcellular-localization pattern protein-protein interactionand gene coexpression (Fig 2B) Like digenicinteractions (7 15) genes involved in trigenicinteractions were significantly enriched for allof these standards with genes participating inthe ldquomodifiedrdquo class of trigenic interactions ex-hibiting stronger functional relationships (figS8B) as well as a stronger magnitude of inter-actions (fig S8C) Thus trigenic interactionsresemble digenic interactions in that they arerich in functional information which means thatgenes participating in many trigenic interactionscan be predicted from alternative data sets andgeneral knowledge of cellular function

Trigenic interactions expand functionalconnections mapped by the globaldigenic network

Our functional analysis revealed that trigenicinteractions have some properties distinct from

those of digenic interactions which suggests thattrigenic interactions may be useful for discover-ing previously unknown connections betweengenes and their corresponding pathways As anillustrative example we examined the MDY2-MTC1 double-mutant query which is a highlyconnected hub within the trigenic networkMDY2encodes a protein that interacts and functionswith components of the GET (guided entry oftail-anchor) pathway (20) which is importantfor Golgindashtondashendoplasmic reticulum (ER) traf-ficking and inserting tail-anchored proteins intoER membranes MTC1 encodes a protein of un-known function that localizes to the early Golgiapparatus The MTC1 digenic interaction profileis similar to that of USO1 which is involved invesicle-mediated ER-to-Golgi transport (21) andRUD3 which encodes a Golgi matrix proteinimportant for the structural organization of thecis-Golgi (22) suggesting that MTC1 also has arole in the early secretory pathwayAs expected from these previous findings the

MDY2 and MTC1 digenic interactions identifiedin our screen were enriched for genes involvedin cell polarity and the early secretory pathway(Fig 3 fig S9A and data S6) However theMDY2-MTC1 double-mutant query profile encompasseda much more functionally diverse set of genes(Fig 3) For example although we observed noveltrigenic interactions with ER-to-Golgi transportgenes we also observed trigenic interactionswith genes involved in other modes of vesicle

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 3 of 9

Fold change over background(all p lt 0008)

Co-localization

Co-annotation

Co-expression

PPI

0 2 4 6 8

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Withinprocess

|log2 fold change|

enrichment p lt 005p gt 005

05 2 35

BA

Trigenic interactionsDigenic interactions

Fig 2 Functional characterization of trigenic interactions (A) Frequency of negative geneticinteractions within biological processes For our analysis we used the fraction of screened query-array combinations exhibiting negative interactions belonging to functional gene sets annotated bySAFE (spatial analysis of functional enrichment) on the global genetic interaction network (55)The ldquowithin processrdquo category received a count for any combination in which both genes for digenicinteractions or all three genes for trigenic interactions were annotated to the same term The sizeof the circle assigned to each ldquowithin processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) Significance was assessedwith a hypergeometric test P lt 005 Blue circles represent significant enrichment gray circlesdenote no significant change (B) Enrichment of negative digenic and trigenic interactions acrossfour functional standards The dashed line indicates no enrichment The functional standards aremerged protein-protein interaction (PPI) (56ndash60) coannotation (based on SAFE terms) (7)coexpression (61) and colocalization (62) Significance was assessed with a hypergeometric test represents 10minus4 le P lt 001 represents P lt 10minus4

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

trafficking including endocytosis and peroxi-some biology Moreover MDY2-MTC1 trigenicinteractions identified connections to genes withmore diverse functions such as components ofthe elongator complex (Fig 3) which controlsthe modification of wobble nucleosides in tRNAsand several genes involved in DNA replicationand repair Notably the MDY2-MTC1 query alsoshowed a trigenic interaction with TOR1 (targetof rapamycin 1) which encodes the key kinasesubunit of the TORC1 complex that is requiredfor growth in response to nutrients by regulat-ing ribosome biogenesis nutrient transport andautophagy (23) Consistent with this observationthe MDY2-MTC1 trigenic interaction networkcaptures a set of genes that have a dual role inTORC1 signaling and sorting of the general ami-no acid permease including GTR1 MEH1 andLST4 (24ndash26)The spectrum of bioprocesses that are repre-

sented in genetic interaction profiles can be vis-ualized by mapping functional enrichment withinthe context of the global yeast digenic interac-tion profile similarity network which clustersgenes into 17 distinct bioprocesses (7 27) (Fig 4A)In comparison with the MDY2 and MTC1 digenicinteraction profiles (Fig 4 B and C) the MDY2-MTC1 trigenic interactions were enriched notonly for vesicle trafficking and cell polarity bio-process regions of the network but also in regionsencompassing genes annotated to the tRNA wob-ble modification bioprocess DNA replication andrepair as well as mitosis and chromosome segre-gation (Fig 4D) Thus the MDY2-MTC1 trigenicinteraction profile exhibited a more expandedand functionally diverse set of connections thaneither of the corresponding MDY2 or MTC1 di-genic interaction profilesWe used a variety of assays to test three func-

tional connections revealed by the MDY2-MTC1trigenic interaction profile First although the

MDY2-MTC1 double-mutant strain did not showan exaggerated cell biological phenotype asso-ciated with the early trafficking function (figS9B) it displayed a marked synthetic sick pheno-type when combined with deletion of SLA1 whichis involved in cortical actin assembly and endo-cytic vesicle formation which translates into anextended Sla1 patch lifetime reflecting a defectin endocytosis (Fig 5A) (28) Second given anegative trigenic interaction with OAF1 whichencodes an oleate-activated transcription factorinvolved in peroxisome organization and bio-genesis (29) we used fluorescence microscopyto explore peroxisome morphology The MDY2-MTC1 double mutant displayed an accumulationof relatively small peroxisomes (Fig 5B) whichmay be indicative of a defect in ER-derived per-oxisome membrane biogenesis (30) Third theMDY2-MTC1 double mutant showed pronouncedsensitivity to hydroxyurea (HU) but not methylmethanesulfonate (MMS) which is consistentwith a specific defect in DNA replication and re-flects the negative genetic interactions we ob-served with a number of DNA replication andrepair genes including NSE4 and NSE5 whichencode components of the Smc5-Smc6 complexthat mediates resolution of DNA structures span-ning sister chromatids (Fig 5C and fig S10 A toC) (31) We suspect that the MDY2-MTC1 doublemutant may be primarily defective in traffick-ing functions that can modulate signaling ormetabolic pathways and thereby influence DNAsynthesis and repair pathways indirectly (figS10 D to H)

Trigenic interaction profiles are morefunctionally diverse than theircorresponding digenic profiles

To test the generality of whether query genesconnect to more functionally divergent genesthrough trigenic interactions than through di-

genic interactions we compared digenic pro-file similarity of pairs of genes spanned by eitherdigenic or trigenic interactions Indeed genesinvolved in trigenic interactions tend to showprofiles that are less similar than those con-nected by digenic interactions which suggeststhat they are less functionally related than thoseconnected by digenic interactions (Fig 6A andfig S11A) We also found that trigenic interactionswere more enriched than digenic interactionsfor connections that bridge several differentbiological processes including mRNA and tRNAprocessing vesicle trafficking mitosis and chro-mosome segregation and glycosylation and pro-tein folding and targeting (Fig 6B) Moreoveras we showed for theMDY2-MTC1 double-mutantquery (Fig 4) trigenic interaction profiles weregenerally enriched for genes spanning more di-verse bioprocesses than the corresponding di-genic interaction profiles (Fig 6C and fig S11B to D) Genes involved in vesicle traffickingwere particularly enriched for trigenic interac-tions occurring between bioprocesses (Fig 6Band figs S7 and S11D) As we observed for MDY2-MTC1 other double-mutant queries carryingmutations in genes implicated in membrane traf-ficking were enriched for trigenic interactionswith genes involved in DNA replication and re-pair machinery which may indicate a generalconnection between these two bioprocesses Forexample the digenic query strain MVP1-MRL1which carries mutations in genes required forsorting proteins to the vacuole (32 33) and thestrain SEC27-GET4 which carries mutations ingenes involved in ER-to-Golgi transport (34) andthe insertion of tail-anchored proteins into ERmembrane (20) both exhibited an enrichmentof trigenic interactions with DNA replicationand repair machinery (fig S11D) In general ourfindings show that trigenic interaction profilesare composed of connections involving genes

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 4 of 9

EMC Sec62Sec63

EMC3 EMC4 SEC66

Trigenic interactions

GSE Lst4-Lst7Pan1Sla1End3

LST4SLA1 GTR1RVS161

OAFRvs161167

OAF1MEH1PAN1

Elongator

DNA replication factor C

CTF8 CTF18

ELP6ELP2 ELP3 ELP4ELP1

MRX

MRE11

TORC1

TOR1

MDY2 MTC1

GET

GET1 GET2

RIC1 YPT6

TRS23

COG8

BET3

COG2 COG7

TRS20 TRS85

COG6COG3 COG5

TRAPP

COG

Digenic interactions

Ric1Rgp1Ypt6

MDY2MTC1

GET3

RPN6 RPN7

1922S regulator

Smc5-Smc6

NSE4 NSE5

Fig 3 The MDY2-MTC1 double mutant a hub on the trigenicinteraction network Representative digenic interactions are high-lighted for MDY2 and MTC1 single-mutant query genes and rep-resentative trigenic interactions are shown for the MDY2-MTC1double-mutant query The network was visualized using Cytoscape(63) Genes were chosen from representative protein complexes (8) inwhich ge50 of members on the diagnostic array display geneticinteractions Negative genetic interactions (e or t lt ndash008 P lt 005)are depicted All of the digenic and trigenic interactions displayedhave been confirmed by tetrad analysis Nodes are color coded onthe basis of their biological roles and are labeled with gene namesGenes are grouped according to specific protein complexes

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 2: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

RESEARCH ARTICLE

YEAST GENOMICS

Systematic analysis of complexgenetic interactionsElena Kuzmin12dagger Benjamin VanderSluis3 Wen Wang3 Guihong Tan1

Raamesh Deshpande3 Yiqun Chen1 Matej Usaj1 Attila Balint14DaggerMojca Mattiazzi Usaj1 Jolanda van Leeuwen1 Elizabeth N Koch3 Carles Pons3

Andrius J Dagilis5 Michael Pryszlak1 Jason Zi Yang Wang12 Julia Hanchard12

Margot Riggi6789 Kaicong Xu3 Hamed Heydari12 Bryan-Joseph San Luis1

Ermira Shuteriqi1 Hongwei Zhu1 Nydia Van Dyk1 Sara Sharifpoor1

Michael Costanzo1 Robbie Loewith689 Amy Caudy12 Daniel Bolnick5

Grant W Brown14 Brenda J Andrews12sect Charles Boone1210sect Chad L Myers3sect

To systematically explore complex genetic interactions we constructed ~200000 yeasttriple mutants and scored negative trigenic interactions We selected double-mutantquery genes across a broad spectrum of biological processes spanning a range ofquantitative features of the global digenic interaction network and tested for a geneticinteraction with a third mutation Trigenic interactions often occurred among functionallyrelated genes and essential genes were hubs on the trigenic network Despite theirfunctional enrichment trigenic interactions tended to link genes in distant bioprocessesand displayed a weaker magnitude than digenic interactions We estimate that theglobal trigenic interaction network is ~100 times as large as the global digenic networkhighlighting the potential for complex genetic interactions to affect the biology ofinheritance including the genotype-to-phenotype relationship

Genetic interactions occur when a combi-nation of mutations in different genesleads to an unexpected phenotype thatdeviates from a model incorporating thecombined effects of the corresponding

single-mutant phenotypes In humans each in-dividual carries thousands of different variantsthat may modulate gene function which meansthat there is incredible potential for combinato-rial genetic interactions to determine our person-al phenotype (1 2) Indeed genetic interactionsare thought to represent a substantial compo-nent of the missing heritability associated withcurrent genome-wide association studies (GWAS)(3) However the statistical limitations associatedwith GWAS data sets preclude the detection ofspecific genetic interactions and thus potentialgenetic networks underlying inherited traitsincluding diseases remain elusive (3ndash5) To ad-dress the role of genetic interactions in thegenotype-to-phenotype relationship we havebeen exploring their general principles throughsystematic analysis of genetic networks under-lying cellular fitness in a genetically tractablemodel system the budding yeast Saccharomyces

cerevisiae (6) Our previous studies focused pre-dominantly on genetic interactions involving twogenes (digenic interactions) (7) In this study weanalyzed a series of single- double- and triple-mutants by quantifying their colony size as aproxy for fitness to systematically explore com-plex genetic interactionsThere are two basic types of fitness-based ge-

netic interactions A negative genetic interac-tion refers to a combination of mutations thatresults in a fitness defect that is more severe thanexpected (8) Synthetic lethality is an extremeexample of a negative genetic interaction andoccurs when two mutations neither of which islethal on its own combine and lead to an inviabledouble-mutant phenotype (9 10) Converselya positive genetic interaction occurs when acombination of genetic perturbations gener-ates a double mutant with a greater fitness thanexpected one example is genetic suppression inwhich the fitness defect of a query mutant isalleviated by a mutation in a second gene (11)To map a global digenic interaction networkfor yeast we constructed millions of double mu-tants and identified hundreds of thousands of

negative and positive genetic interactions (7) Toput these results in perspective although only~1000 of the ~6000 total yeast genes are indi-vidually essential and cause lethality when deleted(12) and an equivalent number of nonessentialgenes cause a slow-growth defect under standardlaboratory conditions (13) ~550000 different yeastgene pairs display a combinatorial negative ge-netic interaction including a subset of ~10000extreme synthetic lethal interactions involvingnonessential gene pairs (7) Thus there are nu-merous potential ways to generate extreme lethalphenotypes through negative digenic interactionsof nonessential gene pairsThe set of digenic interactions for a query gene

its genetic interaction profile provides a quan-titative measure of function because genes withsimilar roles have overlapping profiles (14 15)Genes belonging to the same biological path-way or protein complex display highly similargenetic interaction profiles Moreover a globalnetwork based on digenic interaction profile sim-ilarity reveals a hierarchy of functional moduleswhich includes detailed pathways and complexesthat in turn cluster into larger modules corre-sponding to bioprocesses Those larger units sub-sequently assemble into modules correspondingto cellular compartments to outline the functionalarchitecture of a cell (7)A complete understanding of the role of ge-

netic interactions in the genotype-to-phenotyperelationship requires that we also investigate com-plex higher-order genetic interactions involvingmore than two genes Because there are ~2000times as many triple gene combinations as genepairs (~18 million) it is possible that there is asubstantially greater number of trigenic thandigenic interactions and that higher-order inter-actions may be important for driving inheritedtraits In this study we surveyed yeast trigenicinteractions sampling quantitative features ofthe digenic network and explored the impli-cations of the higher-order genetic interactionnetwork

Mapping trigenic interactionsquantitatively and surveying theglobal trigenic landscape

To explore the trigenic interaction landscape wedesigned query strains that sampled three keyquantitative features of our global digenic in-teraction network (7) We designed query strainscarrying mutations in two genes spanning arange of the following features (1) digenic in-teraction strength (2) number of digenic inter-actions (average digenic interaction degree) and(3) digenic interaction profile similarity (Fig 1Aand table S1) Gene pairs were selected to fill

RESEARCH

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 1 of 9

1The Donnelly Centre University of Toronto 160 College Street Toronto Ontario M5S 3E1 Canada 2Department of Molecular Genetics University of Toronto 160 College Street TorontoOntario M5S 3E1 Canada 3Department of Computer Science and Engineering University of MinnesotandashTwin Cities 200 Union Street Minneapolis MN 55455 USA 4Department ofBiochemistry University of Toronto 160 College Street Toronto Ontario M5S 3E1 Canada 5Department of Integrative Biology 1 University Station C0990 University of Texas at Austin AustinTX 78712 USA 6Department of Molecular Biology University of Geneva Geneva 1211 Switzerland 7Department of Biochemistry University of Geneva 1211 Geneva Switzerland 8iGE3 (Instituteof Genetics and Genomics of Geneva) 1211 Geneva Switzerland 9Swiss National Centre for Competence in Research Programme Chemical Biology 1211 Geneva Switzerland 10RIKEN Center forSustainable Resource Science Wako Saitama JapanThese authors contributed equally to this work daggerPresent address Rosalind and Morris Goodman Cancer Research Centre McGill University 1160 Avenue des Pins Ouest Montreal Quebec H3A 1A3 CanadaDaggerPresent address Center for Chromosome Stability Department of Cellular and Molecular Medicine University of Copenhagen Blegdamsvej 3B 2200 Copenhagen N DenmarksectCorresponding author Email chadmumnedu (CLM) brendaandrewsutorontoca (BJA) charliebooneutorontoca (CB)

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

bins of varying digenic interaction attributes andto cover all major biological processes in the cellthus producing a sample that would provide a di-verse survey of the trigenic interaction landscapeWe largely focused on unambiguous singletonsbecause duplicated genes represent a relativelysmall subset of genes and thus can only representa small fraction of the global trigenic interac-tion network For this survey we constructed151 double-mutant query strains and 302 single-mutant strains encompassing 47 temperature-sensitive alleles of different essential genes and255 deletion alleles of nonessential genes Thequery strains in this set were selected to spanthe different digenic attribute bins according topredefined thresholds (table S1) An additional31 double-mutant queries fell outside of the de-fined thresholds but were included for validationand comparison purposes (data S1 to S3) (16)The fitness of the resulting query strains was mea-sured using a quantitative growth assay and thebehavior of the single- and double-mutant querystrains showed strong agreement with our pre-viously published data set (figs S1 and S2 anddata S4) (7 15)

Trigenic interaction screening required devel-opment and implementation of three opera-tional components First synthetic genetic array(SGA) analysismdashan automated form of yeast ge-netics that is often used to cross a query genemutation into an array of single mutants togenerate a defined set of haploid double mutants(6)mdashwas adapted such that a double-mutantquery strain could be crossed into an array ofsingle mutants to generate triple mutants fortrigenic interaction analysis (Fig 1B) Becausethe identification of a trigenic interaction re-quires comparison with the corresponding dou-ble mutants we also conducted screens in whichthe individual mutants of the query gene pairwere scored for digenic interactions (Fig 1B)Second for experimental feasibility we assem-bled a diagnostic array of 1182 strains compris-ing 990 nonessential gene deletion mutants and192 essential gene mutants carrying temperature-sensitive alleles which combine to span ~20of the yeast genome (data S5) The diagnosticarray was designed to be highly representativeof the rest of the genome in terms of exhibitedgenetic interaction profiles (fig S3) Briefly ar-

ray strains were selected from a larger geneticinteraction data set for their ability to representdifferent regions of the global network in a min-imally redundant way This was accomplishedby iteratively selecting strains to maximize theperformance of profile similarities when predict-ing coannotations to a functional gold standard(17) Third we developed a scoring method thet-SGA score which combines double- and triple-mutant fitness estimates derived from colonysize measurements to identify trigenic interactionsquantitatively (Fig 1C) The t-SGA score differsfrom the MinDC score reported previously (18)because it accounts for all cases in which two ofthe genes are not independent resulting in an ex-pectation that contains digenic interaction effectsscaled by the fitness of the noninteracting genes(fig S4) (16) The final trigenic t-SGA interactionscore then accounts for digenic effects but alsoenables detection of trigenic interactions in whichdigenic effects of insufficient explanatory powercan be foundWe focused exclusively on the analysis of del-

eterious negative trigenic interactions for tworeasons First quantitative scoring of negative

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 2 of 9

BA

Diagnostic MATa Single Mutant Array

Mutant alleles (Δ or ts) Wild type alleles

MATα Mutant Query

Double mutantgene1Δ gene2Δ

Single mutantgene1Δ

Single mutantgene2Δ

SGA

CMATa Output Array

Triple mutant array

Double mutant array-1

Double mutant array-2

observed triple mutant fitness

expected triple mutant fitness

digenic interactions

2 Avg digenic interaction degree

1 Neg digenic interaction strength

low medium high

3 Interaction profile similarity

K

ML

OPQR

A B C D E F

N

Quantitative criteria for query gene selection

negative genetic interactionno genetic interactionpositive genetic interaction

Fig 1 Triple-mutant synthetic genetic array (SGA) analysis (A) Criteria for selecting query strains forsampling trigenic interaction landscape of singleton genes in yeast The gene pairs were grouped into threegeneral categories based on a range of features (1) Digenic interaction strength Gene pairs were directlyconnected by zero to very weak (digenic interaction score 0 to ndash008 n = 74 strains) weak (ndash008 to ndash01n = 32) or moderate (ltndash01 n = 45) negative digenic interactions (2) Number of digenic interactions Genepairs had a low (10 to 45 interactions n = 50) intermediate (46 to 70 n = 53) or high (gt71 n = 48) averagedigenic interaction degree (denoted by the number of black edges of each node) (3) Digenic interaction profilesimilarity Gene pairs had low (score ndash002 to 003 n = 46 represented by genes A and B which show arelatively low overlap of genetic interactions with genes K to R) intermediate (003 to 01 n = 59 representedby genes C and D which display an intermediate overlap of genetic interactions) or high (gt01 n = 46 rep-resented by genes E and F which display a relatively high level of overlap of genetic interactions) functionalsimilarity as measured by digenic interaction profile similarity and coannotation to the same GO term(s) Querymutant genes were either nonessential deletion mutant alleles (D) or conditional temperature-sensitive (ts)alleles of essential genes (B) Diagram illustrating the triple-mutant SGA experimental strategy To quantify atrigenic interaction three types of screens are conducted in parallel To estimate triple-mutant fitness a double-mutant query strain carrying two desired mutated genes of interest (red and blue filled circles) is crossedinto a diagnostic array of single mutants (black filled circle) Meiosis is induced in heterozygous triplemutants and haploid triple-mutant progeny is selected in sequential replica pinning steps In parallel single-mutant control query strains are used to generate double mutants for fitness analysis (C) Triple-mutant

SGA quantitative scoring strategy The top equation shows the quantification of a digenic interaction where eij is the digenic interaction score ƒij is theobserved double-mutant fitness and the expected double-mutant fitness is expressed as the product of single-mutant fitness estimates ƒiƒj In thebottom equation the trigenic interaction score (tijk) is derived from the digenic interaction score where ƒijk is the observed triple-mutant fitness and ƒiƒjƒkis the triple-mutant fitness expectation expressed as the product of three single-mutant fitness estimates The influence of digenic interactions issubtracted from the expectation and each digenic interaction is scaled by the fitness of the third mutation

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genetic interactions is often more accurate thanthat for positive interactions because there is agreater signal-to-noise ratio for negative geneticinteractions Hence negative genetic interac-tions are associated with lower false-positiveand false-negative rates than positive interactions(8) a feature that is important for the robuststatistical analysis necessary to differentiate truetrigenic interactions from the extensive back-ground digenic network Second negative digenicinteractions are generally more functionally in-formative than positive digenic interactions (8)and thus the large-scale mapping of a negativetrigenic interaction network is expected to pro-vide the most mechanistic insight into genefunction and pathway wiring

Trigenic interactions are enrichedfor functionally related genes

To obtain sufficient precision we carried outeach analysis which involved screening the in-dividual query genes for digenic interactionsand the double-mutant query for trigenic inter-actions in at least two replicate screens with fourcolonies per screen (fig S5) In total we tested410399 double and 195666 triple mutants forfitness defects meeting a previously establishedintermediate magnitude cutoff (15) (data S2) andidentified 9363 digenic and 3196 trigenic negativeinteractions From detailed validation of trigenicinteractions of our CLN1-CLN2 double-mutantquery which was screened previously (19) weestimated a false-negative rate of ~40 a false-positive rate of ~20 and a true-positive ratebetween ~60 and ~75 (table S2 and fig S6) (16)which is consistent with our previous global di-genic network analysis (8)The distribution of trigenic interaction degree

for array strains shows that the majority of low-degree genes (70) account for ~88 of all tri-genic interactions whereas highly connectedgenes contribute the remaining ~12 of interac-tions Thus the trigenic interactions are not as-sociated with a small set of highly connectedgenes rather the interactions are distributedacross many different genes (fig S7) On the otherhand with a smaller more biased set of double-mutant query genes the distribution of trigenicinteraction degree shows that ~22 of them ac-counted for 51 of trigenic interactions indicat-ing that a particular subset of the digenic querieswere enriched for trigenic interactions (fig S7)About one-third of the newly mapped trigenicinteractions identified connections that werenot observed in our digenic control network werefer to these as ldquonovelrdquo trigenic interactions Theremaining approximately two-thirds of the trigenicinteractions overlapped a digenic interaction whilestill exhibiting a stronger than expected fitnessdefect in the triple mutant these we refer to asldquomodifiedrdquo trigenic interactions (fig S8A) Thusalthough a substantial fraction of trigenic inter-actions elucidate totally new functional informationthe majority of the trigenic interactions we mappedexpand upon the digenic interaction networkWe first assessed the functional information

embedded in the trigenic network by comparing

the distributions of digenic and trigenic inter-actions across different biological processesAs observed previously (15) digenic interactionswere enriched among genes annotated to thesame biological process and although the mag-nitude of trigenic interaction enrichment wassomewhat lower they were comparably enrichedfor genes within the same bioprocess (Fig 2A)We also evaluated the enrichment of digenic andtrigenic interactions across common functionalstandards including annotation to the same GeneOntology (GO) biological process subcellular-localization pattern protein-protein interactionand gene coexpression (Fig 2B) Like digenicinteractions (7 15) genes involved in trigenicinteractions were significantly enriched for allof these standards with genes participating inthe ldquomodifiedrdquo class of trigenic interactions ex-hibiting stronger functional relationships (figS8B) as well as a stronger magnitude of inter-actions (fig S8C) Thus trigenic interactionsresemble digenic interactions in that they arerich in functional information which means thatgenes participating in many trigenic interactionscan be predicted from alternative data sets andgeneral knowledge of cellular function

Trigenic interactions expand functionalconnections mapped by the globaldigenic network

Our functional analysis revealed that trigenicinteractions have some properties distinct from

those of digenic interactions which suggests thattrigenic interactions may be useful for discover-ing previously unknown connections betweengenes and their corresponding pathways As anillustrative example we examined the MDY2-MTC1 double-mutant query which is a highlyconnected hub within the trigenic networkMDY2encodes a protein that interacts and functionswith components of the GET (guided entry oftail-anchor) pathway (20) which is importantfor Golgindashtondashendoplasmic reticulum (ER) traf-ficking and inserting tail-anchored proteins intoER membranes MTC1 encodes a protein of un-known function that localizes to the early Golgiapparatus The MTC1 digenic interaction profileis similar to that of USO1 which is involved invesicle-mediated ER-to-Golgi transport (21) andRUD3 which encodes a Golgi matrix proteinimportant for the structural organization of thecis-Golgi (22) suggesting that MTC1 also has arole in the early secretory pathwayAs expected from these previous findings the

MDY2 and MTC1 digenic interactions identifiedin our screen were enriched for genes involvedin cell polarity and the early secretory pathway(Fig 3 fig S9A and data S6) However theMDY2-MTC1 double-mutant query profile encompasseda much more functionally diverse set of genes(Fig 3) For example although we observed noveltrigenic interactions with ER-to-Golgi transportgenes we also observed trigenic interactionswith genes involved in other modes of vesicle

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 3 of 9

Fold change over background(all p lt 0008)

Co-localization

Co-annotation

Co-expression

PPI

0 2 4 6 8

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Withinprocess

|log2 fold change|

enrichment p lt 005p gt 005

05 2 35

BA

Trigenic interactionsDigenic interactions

Fig 2 Functional characterization of trigenic interactions (A) Frequency of negative geneticinteractions within biological processes For our analysis we used the fraction of screened query-array combinations exhibiting negative interactions belonging to functional gene sets annotated bySAFE (spatial analysis of functional enrichment) on the global genetic interaction network (55)The ldquowithin processrdquo category received a count for any combination in which both genes for digenicinteractions or all three genes for trigenic interactions were annotated to the same term The sizeof the circle assigned to each ldquowithin processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) Significance was assessedwith a hypergeometric test P lt 005 Blue circles represent significant enrichment gray circlesdenote no significant change (B) Enrichment of negative digenic and trigenic interactions acrossfour functional standards The dashed line indicates no enrichment The functional standards aremerged protein-protein interaction (PPI) (56ndash60) coannotation (based on SAFE terms) (7)coexpression (61) and colocalization (62) Significance was assessed with a hypergeometric test represents 10minus4 le P lt 001 represents P lt 10minus4

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

trafficking including endocytosis and peroxi-some biology Moreover MDY2-MTC1 trigenicinteractions identified connections to genes withmore diverse functions such as components ofthe elongator complex (Fig 3) which controlsthe modification of wobble nucleosides in tRNAsand several genes involved in DNA replicationand repair Notably the MDY2-MTC1 query alsoshowed a trigenic interaction with TOR1 (targetof rapamycin 1) which encodes the key kinasesubunit of the TORC1 complex that is requiredfor growth in response to nutrients by regulat-ing ribosome biogenesis nutrient transport andautophagy (23) Consistent with this observationthe MDY2-MTC1 trigenic interaction networkcaptures a set of genes that have a dual role inTORC1 signaling and sorting of the general ami-no acid permease including GTR1 MEH1 andLST4 (24ndash26)The spectrum of bioprocesses that are repre-

sented in genetic interaction profiles can be vis-ualized by mapping functional enrichment withinthe context of the global yeast digenic interac-tion profile similarity network which clustersgenes into 17 distinct bioprocesses (7 27) (Fig 4A)In comparison with the MDY2 and MTC1 digenicinteraction profiles (Fig 4 B and C) the MDY2-MTC1 trigenic interactions were enriched notonly for vesicle trafficking and cell polarity bio-process regions of the network but also in regionsencompassing genes annotated to the tRNA wob-ble modification bioprocess DNA replication andrepair as well as mitosis and chromosome segre-gation (Fig 4D) Thus the MDY2-MTC1 trigenicinteraction profile exhibited a more expandedand functionally diverse set of connections thaneither of the corresponding MDY2 or MTC1 di-genic interaction profilesWe used a variety of assays to test three func-

tional connections revealed by the MDY2-MTC1trigenic interaction profile First although the

MDY2-MTC1 double-mutant strain did not showan exaggerated cell biological phenotype asso-ciated with the early trafficking function (figS9B) it displayed a marked synthetic sick pheno-type when combined with deletion of SLA1 whichis involved in cortical actin assembly and endo-cytic vesicle formation which translates into anextended Sla1 patch lifetime reflecting a defectin endocytosis (Fig 5A) (28) Second given anegative trigenic interaction with OAF1 whichencodes an oleate-activated transcription factorinvolved in peroxisome organization and bio-genesis (29) we used fluorescence microscopyto explore peroxisome morphology The MDY2-MTC1 double mutant displayed an accumulationof relatively small peroxisomes (Fig 5B) whichmay be indicative of a defect in ER-derived per-oxisome membrane biogenesis (30) Third theMDY2-MTC1 double mutant showed pronouncedsensitivity to hydroxyurea (HU) but not methylmethanesulfonate (MMS) which is consistentwith a specific defect in DNA replication and re-flects the negative genetic interactions we ob-served with a number of DNA replication andrepair genes including NSE4 and NSE5 whichencode components of the Smc5-Smc6 complexthat mediates resolution of DNA structures span-ning sister chromatids (Fig 5C and fig S10 A toC) (31) We suspect that the MDY2-MTC1 doublemutant may be primarily defective in traffick-ing functions that can modulate signaling ormetabolic pathways and thereby influence DNAsynthesis and repair pathways indirectly (figS10 D to H)

Trigenic interaction profiles are morefunctionally diverse than theircorresponding digenic profiles

To test the generality of whether query genesconnect to more functionally divergent genesthrough trigenic interactions than through di-

genic interactions we compared digenic pro-file similarity of pairs of genes spanned by eitherdigenic or trigenic interactions Indeed genesinvolved in trigenic interactions tend to showprofiles that are less similar than those con-nected by digenic interactions which suggeststhat they are less functionally related than thoseconnected by digenic interactions (Fig 6A andfig S11A) We also found that trigenic interactionswere more enriched than digenic interactionsfor connections that bridge several differentbiological processes including mRNA and tRNAprocessing vesicle trafficking mitosis and chro-mosome segregation and glycosylation and pro-tein folding and targeting (Fig 6B) Moreoveras we showed for theMDY2-MTC1 double-mutantquery (Fig 4) trigenic interaction profiles weregenerally enriched for genes spanning more di-verse bioprocesses than the corresponding di-genic interaction profiles (Fig 6C and fig S11B to D) Genes involved in vesicle traffickingwere particularly enriched for trigenic interac-tions occurring between bioprocesses (Fig 6Band figs S7 and S11D) As we observed for MDY2-MTC1 other double-mutant queries carryingmutations in genes implicated in membrane traf-ficking were enriched for trigenic interactionswith genes involved in DNA replication and re-pair machinery which may indicate a generalconnection between these two bioprocesses Forexample the digenic query strain MVP1-MRL1which carries mutations in genes required forsorting proteins to the vacuole (32 33) and thestrain SEC27-GET4 which carries mutations ingenes involved in ER-to-Golgi transport (34) andthe insertion of tail-anchored proteins into ERmembrane (20) both exhibited an enrichmentof trigenic interactions with DNA replicationand repair machinery (fig S11D) In general ourfindings show that trigenic interaction profilesare composed of connections involving genes

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 4 of 9

EMC Sec62Sec63

EMC3 EMC4 SEC66

Trigenic interactions

GSE Lst4-Lst7Pan1Sla1End3

LST4SLA1 GTR1RVS161

OAFRvs161167

OAF1MEH1PAN1

Elongator

DNA replication factor C

CTF8 CTF18

ELP6ELP2 ELP3 ELP4ELP1

MRX

MRE11

TORC1

TOR1

MDY2 MTC1

GET

GET1 GET2

RIC1 YPT6

TRS23

COG8

BET3

COG2 COG7

TRS20 TRS85

COG6COG3 COG5

TRAPP

COG

Digenic interactions

Ric1Rgp1Ypt6

MDY2MTC1

GET3

RPN6 RPN7

1922S regulator

Smc5-Smc6

NSE4 NSE5

Fig 3 The MDY2-MTC1 double mutant a hub on the trigenicinteraction network Representative digenic interactions are high-lighted for MDY2 and MTC1 single-mutant query genes and rep-resentative trigenic interactions are shown for the MDY2-MTC1double-mutant query The network was visualized using Cytoscape(63) Genes were chosen from representative protein complexes (8) inwhich ge50 of members on the diagnostic array display geneticinteractions Negative genetic interactions (e or t lt ndash008 P lt 005)are depicted All of the digenic and trigenic interactions displayedhave been confirmed by tetrad analysis Nodes are color coded onthe basis of their biological roles and are labeled with gene namesGenes are grouped according to specific protein complexes

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 3: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

bins of varying digenic interaction attributes andto cover all major biological processes in the cellthus producing a sample that would provide a di-verse survey of the trigenic interaction landscapeWe largely focused on unambiguous singletonsbecause duplicated genes represent a relativelysmall subset of genes and thus can only representa small fraction of the global trigenic interac-tion network For this survey we constructed151 double-mutant query strains and 302 single-mutant strains encompassing 47 temperature-sensitive alleles of different essential genes and255 deletion alleles of nonessential genes Thequery strains in this set were selected to spanthe different digenic attribute bins according topredefined thresholds (table S1) An additional31 double-mutant queries fell outside of the de-fined thresholds but were included for validationand comparison purposes (data S1 to S3) (16)The fitness of the resulting query strains was mea-sured using a quantitative growth assay and thebehavior of the single- and double-mutant querystrains showed strong agreement with our pre-viously published data set (figs S1 and S2 anddata S4) (7 15)

Trigenic interaction screening required devel-opment and implementation of three opera-tional components First synthetic genetic array(SGA) analysismdashan automated form of yeast ge-netics that is often used to cross a query genemutation into an array of single mutants togenerate a defined set of haploid double mutants(6)mdashwas adapted such that a double-mutantquery strain could be crossed into an array ofsingle mutants to generate triple mutants fortrigenic interaction analysis (Fig 1B) Becausethe identification of a trigenic interaction re-quires comparison with the corresponding dou-ble mutants we also conducted screens in whichthe individual mutants of the query gene pairwere scored for digenic interactions (Fig 1B)Second for experimental feasibility we assem-bled a diagnostic array of 1182 strains compris-ing 990 nonessential gene deletion mutants and192 essential gene mutants carrying temperature-sensitive alleles which combine to span ~20of the yeast genome (data S5) The diagnosticarray was designed to be highly representativeof the rest of the genome in terms of exhibitedgenetic interaction profiles (fig S3) Briefly ar-

ray strains were selected from a larger geneticinteraction data set for their ability to representdifferent regions of the global network in a min-imally redundant way This was accomplishedby iteratively selecting strains to maximize theperformance of profile similarities when predict-ing coannotations to a functional gold standard(17) Third we developed a scoring method thet-SGA score which combines double- and triple-mutant fitness estimates derived from colonysize measurements to identify trigenic interactionsquantitatively (Fig 1C) The t-SGA score differsfrom the MinDC score reported previously (18)because it accounts for all cases in which two ofthe genes are not independent resulting in an ex-pectation that contains digenic interaction effectsscaled by the fitness of the noninteracting genes(fig S4) (16) The final trigenic t-SGA interactionscore then accounts for digenic effects but alsoenables detection of trigenic interactions in whichdigenic effects of insufficient explanatory powercan be foundWe focused exclusively on the analysis of del-

eterious negative trigenic interactions for tworeasons First quantitative scoring of negative

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 2 of 9

BA

Diagnostic MATa Single Mutant Array

Mutant alleles (Δ or ts) Wild type alleles

MATα Mutant Query

Double mutantgene1Δ gene2Δ

Single mutantgene1Δ

Single mutantgene2Δ

SGA

CMATa Output Array

Triple mutant array

Double mutant array-1

Double mutant array-2

observed triple mutant fitness

expected triple mutant fitness

digenic interactions

2 Avg digenic interaction degree

1 Neg digenic interaction strength

low medium high

3 Interaction profile similarity

K

ML

OPQR

A B C D E F

N

Quantitative criteria for query gene selection

negative genetic interactionno genetic interactionpositive genetic interaction

Fig 1 Triple-mutant synthetic genetic array (SGA) analysis (A) Criteria for selecting query strains forsampling trigenic interaction landscape of singleton genes in yeast The gene pairs were grouped into threegeneral categories based on a range of features (1) Digenic interaction strength Gene pairs were directlyconnected by zero to very weak (digenic interaction score 0 to ndash008 n = 74 strains) weak (ndash008 to ndash01n = 32) or moderate (ltndash01 n = 45) negative digenic interactions (2) Number of digenic interactions Genepairs had a low (10 to 45 interactions n = 50) intermediate (46 to 70 n = 53) or high (gt71 n = 48) averagedigenic interaction degree (denoted by the number of black edges of each node) (3) Digenic interaction profilesimilarity Gene pairs had low (score ndash002 to 003 n = 46 represented by genes A and B which show arelatively low overlap of genetic interactions with genes K to R) intermediate (003 to 01 n = 59 representedby genes C and D which display an intermediate overlap of genetic interactions) or high (gt01 n = 46 rep-resented by genes E and F which display a relatively high level of overlap of genetic interactions) functionalsimilarity as measured by digenic interaction profile similarity and coannotation to the same GO term(s) Querymutant genes were either nonessential deletion mutant alleles (D) or conditional temperature-sensitive (ts)alleles of essential genes (B) Diagram illustrating the triple-mutant SGA experimental strategy To quantify atrigenic interaction three types of screens are conducted in parallel To estimate triple-mutant fitness a double-mutant query strain carrying two desired mutated genes of interest (red and blue filled circles) is crossedinto a diagnostic array of single mutants (black filled circle) Meiosis is induced in heterozygous triplemutants and haploid triple-mutant progeny is selected in sequential replica pinning steps In parallel single-mutant control query strains are used to generate double mutants for fitness analysis (C) Triple-mutant

SGA quantitative scoring strategy The top equation shows the quantification of a digenic interaction where eij is the digenic interaction score ƒij is theobserved double-mutant fitness and the expected double-mutant fitness is expressed as the product of single-mutant fitness estimates ƒiƒj In thebottom equation the trigenic interaction score (tijk) is derived from the digenic interaction score where ƒijk is the observed triple-mutant fitness and ƒiƒjƒkis the triple-mutant fitness expectation expressed as the product of three single-mutant fitness estimates The influence of digenic interactions issubtracted from the expectation and each digenic interaction is scaled by the fitness of the third mutation

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genetic interactions is often more accurate thanthat for positive interactions because there is agreater signal-to-noise ratio for negative geneticinteractions Hence negative genetic interac-tions are associated with lower false-positiveand false-negative rates than positive interactions(8) a feature that is important for the robuststatistical analysis necessary to differentiate truetrigenic interactions from the extensive back-ground digenic network Second negative digenicinteractions are generally more functionally in-formative than positive digenic interactions (8)and thus the large-scale mapping of a negativetrigenic interaction network is expected to pro-vide the most mechanistic insight into genefunction and pathway wiring

Trigenic interactions are enrichedfor functionally related genes

To obtain sufficient precision we carried outeach analysis which involved screening the in-dividual query genes for digenic interactionsand the double-mutant query for trigenic inter-actions in at least two replicate screens with fourcolonies per screen (fig S5) In total we tested410399 double and 195666 triple mutants forfitness defects meeting a previously establishedintermediate magnitude cutoff (15) (data S2) andidentified 9363 digenic and 3196 trigenic negativeinteractions From detailed validation of trigenicinteractions of our CLN1-CLN2 double-mutantquery which was screened previously (19) weestimated a false-negative rate of ~40 a false-positive rate of ~20 and a true-positive ratebetween ~60 and ~75 (table S2 and fig S6) (16)which is consistent with our previous global di-genic network analysis (8)The distribution of trigenic interaction degree

for array strains shows that the majority of low-degree genes (70) account for ~88 of all tri-genic interactions whereas highly connectedgenes contribute the remaining ~12 of interac-tions Thus the trigenic interactions are not as-sociated with a small set of highly connectedgenes rather the interactions are distributedacross many different genes (fig S7) On the otherhand with a smaller more biased set of double-mutant query genes the distribution of trigenicinteraction degree shows that ~22 of them ac-counted for 51 of trigenic interactions indicat-ing that a particular subset of the digenic querieswere enriched for trigenic interactions (fig S7)About one-third of the newly mapped trigenicinteractions identified connections that werenot observed in our digenic control network werefer to these as ldquonovelrdquo trigenic interactions Theremaining approximately two-thirds of the trigenicinteractions overlapped a digenic interaction whilestill exhibiting a stronger than expected fitnessdefect in the triple mutant these we refer to asldquomodifiedrdquo trigenic interactions (fig S8A) Thusalthough a substantial fraction of trigenic inter-actions elucidate totally new functional informationthe majority of the trigenic interactions we mappedexpand upon the digenic interaction networkWe first assessed the functional information

embedded in the trigenic network by comparing

the distributions of digenic and trigenic inter-actions across different biological processesAs observed previously (15) digenic interactionswere enriched among genes annotated to thesame biological process and although the mag-nitude of trigenic interaction enrichment wassomewhat lower they were comparably enrichedfor genes within the same bioprocess (Fig 2A)We also evaluated the enrichment of digenic andtrigenic interactions across common functionalstandards including annotation to the same GeneOntology (GO) biological process subcellular-localization pattern protein-protein interactionand gene coexpression (Fig 2B) Like digenicinteractions (7 15) genes involved in trigenicinteractions were significantly enriched for allof these standards with genes participating inthe ldquomodifiedrdquo class of trigenic interactions ex-hibiting stronger functional relationships (figS8B) as well as a stronger magnitude of inter-actions (fig S8C) Thus trigenic interactionsresemble digenic interactions in that they arerich in functional information which means thatgenes participating in many trigenic interactionscan be predicted from alternative data sets andgeneral knowledge of cellular function

Trigenic interactions expand functionalconnections mapped by the globaldigenic network

Our functional analysis revealed that trigenicinteractions have some properties distinct from

those of digenic interactions which suggests thattrigenic interactions may be useful for discover-ing previously unknown connections betweengenes and their corresponding pathways As anillustrative example we examined the MDY2-MTC1 double-mutant query which is a highlyconnected hub within the trigenic networkMDY2encodes a protein that interacts and functionswith components of the GET (guided entry oftail-anchor) pathway (20) which is importantfor Golgindashtondashendoplasmic reticulum (ER) traf-ficking and inserting tail-anchored proteins intoER membranes MTC1 encodes a protein of un-known function that localizes to the early Golgiapparatus The MTC1 digenic interaction profileis similar to that of USO1 which is involved invesicle-mediated ER-to-Golgi transport (21) andRUD3 which encodes a Golgi matrix proteinimportant for the structural organization of thecis-Golgi (22) suggesting that MTC1 also has arole in the early secretory pathwayAs expected from these previous findings the

MDY2 and MTC1 digenic interactions identifiedin our screen were enriched for genes involvedin cell polarity and the early secretory pathway(Fig 3 fig S9A and data S6) However theMDY2-MTC1 double-mutant query profile encompasseda much more functionally diverse set of genes(Fig 3) For example although we observed noveltrigenic interactions with ER-to-Golgi transportgenes we also observed trigenic interactionswith genes involved in other modes of vesicle

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 3 of 9

Fold change over background(all p lt 0008)

Co-localization

Co-annotation

Co-expression

PPI

0 2 4 6 8

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Withinprocess

|log2 fold change|

enrichment p lt 005p gt 005

05 2 35

BA

Trigenic interactionsDigenic interactions

Fig 2 Functional characterization of trigenic interactions (A) Frequency of negative geneticinteractions within biological processes For our analysis we used the fraction of screened query-array combinations exhibiting negative interactions belonging to functional gene sets annotated bySAFE (spatial analysis of functional enrichment) on the global genetic interaction network (55)The ldquowithin processrdquo category received a count for any combination in which both genes for digenicinteractions or all three genes for trigenic interactions were annotated to the same term The sizeof the circle assigned to each ldquowithin processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) Significance was assessedwith a hypergeometric test P lt 005 Blue circles represent significant enrichment gray circlesdenote no significant change (B) Enrichment of negative digenic and trigenic interactions acrossfour functional standards The dashed line indicates no enrichment The functional standards aremerged protein-protein interaction (PPI) (56ndash60) coannotation (based on SAFE terms) (7)coexpression (61) and colocalization (62) Significance was assessed with a hypergeometric test represents 10minus4 le P lt 001 represents P lt 10minus4

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

trafficking including endocytosis and peroxi-some biology Moreover MDY2-MTC1 trigenicinteractions identified connections to genes withmore diverse functions such as components ofthe elongator complex (Fig 3) which controlsthe modification of wobble nucleosides in tRNAsand several genes involved in DNA replicationand repair Notably the MDY2-MTC1 query alsoshowed a trigenic interaction with TOR1 (targetof rapamycin 1) which encodes the key kinasesubunit of the TORC1 complex that is requiredfor growth in response to nutrients by regulat-ing ribosome biogenesis nutrient transport andautophagy (23) Consistent with this observationthe MDY2-MTC1 trigenic interaction networkcaptures a set of genes that have a dual role inTORC1 signaling and sorting of the general ami-no acid permease including GTR1 MEH1 andLST4 (24ndash26)The spectrum of bioprocesses that are repre-

sented in genetic interaction profiles can be vis-ualized by mapping functional enrichment withinthe context of the global yeast digenic interac-tion profile similarity network which clustersgenes into 17 distinct bioprocesses (7 27) (Fig 4A)In comparison with the MDY2 and MTC1 digenicinteraction profiles (Fig 4 B and C) the MDY2-MTC1 trigenic interactions were enriched notonly for vesicle trafficking and cell polarity bio-process regions of the network but also in regionsencompassing genes annotated to the tRNA wob-ble modification bioprocess DNA replication andrepair as well as mitosis and chromosome segre-gation (Fig 4D) Thus the MDY2-MTC1 trigenicinteraction profile exhibited a more expandedand functionally diverse set of connections thaneither of the corresponding MDY2 or MTC1 di-genic interaction profilesWe used a variety of assays to test three func-

tional connections revealed by the MDY2-MTC1trigenic interaction profile First although the

MDY2-MTC1 double-mutant strain did not showan exaggerated cell biological phenotype asso-ciated with the early trafficking function (figS9B) it displayed a marked synthetic sick pheno-type when combined with deletion of SLA1 whichis involved in cortical actin assembly and endo-cytic vesicle formation which translates into anextended Sla1 patch lifetime reflecting a defectin endocytosis (Fig 5A) (28) Second given anegative trigenic interaction with OAF1 whichencodes an oleate-activated transcription factorinvolved in peroxisome organization and bio-genesis (29) we used fluorescence microscopyto explore peroxisome morphology The MDY2-MTC1 double mutant displayed an accumulationof relatively small peroxisomes (Fig 5B) whichmay be indicative of a defect in ER-derived per-oxisome membrane biogenesis (30) Third theMDY2-MTC1 double mutant showed pronouncedsensitivity to hydroxyurea (HU) but not methylmethanesulfonate (MMS) which is consistentwith a specific defect in DNA replication and re-flects the negative genetic interactions we ob-served with a number of DNA replication andrepair genes including NSE4 and NSE5 whichencode components of the Smc5-Smc6 complexthat mediates resolution of DNA structures span-ning sister chromatids (Fig 5C and fig S10 A toC) (31) We suspect that the MDY2-MTC1 doublemutant may be primarily defective in traffick-ing functions that can modulate signaling ormetabolic pathways and thereby influence DNAsynthesis and repair pathways indirectly (figS10 D to H)

Trigenic interaction profiles are morefunctionally diverse than theircorresponding digenic profiles

To test the generality of whether query genesconnect to more functionally divergent genesthrough trigenic interactions than through di-

genic interactions we compared digenic pro-file similarity of pairs of genes spanned by eitherdigenic or trigenic interactions Indeed genesinvolved in trigenic interactions tend to showprofiles that are less similar than those con-nected by digenic interactions which suggeststhat they are less functionally related than thoseconnected by digenic interactions (Fig 6A andfig S11A) We also found that trigenic interactionswere more enriched than digenic interactionsfor connections that bridge several differentbiological processes including mRNA and tRNAprocessing vesicle trafficking mitosis and chro-mosome segregation and glycosylation and pro-tein folding and targeting (Fig 6B) Moreoveras we showed for theMDY2-MTC1 double-mutantquery (Fig 4) trigenic interaction profiles weregenerally enriched for genes spanning more di-verse bioprocesses than the corresponding di-genic interaction profiles (Fig 6C and fig S11B to D) Genes involved in vesicle traffickingwere particularly enriched for trigenic interac-tions occurring between bioprocesses (Fig 6Band figs S7 and S11D) As we observed for MDY2-MTC1 other double-mutant queries carryingmutations in genes implicated in membrane traf-ficking were enriched for trigenic interactionswith genes involved in DNA replication and re-pair machinery which may indicate a generalconnection between these two bioprocesses Forexample the digenic query strain MVP1-MRL1which carries mutations in genes required forsorting proteins to the vacuole (32 33) and thestrain SEC27-GET4 which carries mutations ingenes involved in ER-to-Golgi transport (34) andthe insertion of tail-anchored proteins into ERmembrane (20) both exhibited an enrichmentof trigenic interactions with DNA replicationand repair machinery (fig S11D) In general ourfindings show that trigenic interaction profilesare composed of connections involving genes

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 4 of 9

EMC Sec62Sec63

EMC3 EMC4 SEC66

Trigenic interactions

GSE Lst4-Lst7Pan1Sla1End3

LST4SLA1 GTR1RVS161

OAFRvs161167

OAF1MEH1PAN1

Elongator

DNA replication factor C

CTF8 CTF18

ELP6ELP2 ELP3 ELP4ELP1

MRX

MRE11

TORC1

TOR1

MDY2 MTC1

GET

GET1 GET2

RIC1 YPT6

TRS23

COG8

BET3

COG2 COG7

TRS20 TRS85

COG6COG3 COG5

TRAPP

COG

Digenic interactions

Ric1Rgp1Ypt6

MDY2MTC1

GET3

RPN6 RPN7

1922S regulator

Smc5-Smc6

NSE4 NSE5

Fig 3 The MDY2-MTC1 double mutant a hub on the trigenicinteraction network Representative digenic interactions are high-lighted for MDY2 and MTC1 single-mutant query genes and rep-resentative trigenic interactions are shown for the MDY2-MTC1double-mutant query The network was visualized using Cytoscape(63) Genes were chosen from representative protein complexes (8) inwhich ge50 of members on the diagnostic array display geneticinteractions Negative genetic interactions (e or t lt ndash008 P lt 005)are depicted All of the digenic and trigenic interactions displayedhave been confirmed by tetrad analysis Nodes are color coded onthe basis of their biological roles and are labeled with gene namesGenes are grouped according to specific protein complexes

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 4: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

genetic interactions is often more accurate thanthat for positive interactions because there is agreater signal-to-noise ratio for negative geneticinteractions Hence negative genetic interac-tions are associated with lower false-positiveand false-negative rates than positive interactions(8) a feature that is important for the robuststatistical analysis necessary to differentiate truetrigenic interactions from the extensive back-ground digenic network Second negative digenicinteractions are generally more functionally in-formative than positive digenic interactions (8)and thus the large-scale mapping of a negativetrigenic interaction network is expected to pro-vide the most mechanistic insight into genefunction and pathway wiring

Trigenic interactions are enrichedfor functionally related genes

To obtain sufficient precision we carried outeach analysis which involved screening the in-dividual query genes for digenic interactionsand the double-mutant query for trigenic inter-actions in at least two replicate screens with fourcolonies per screen (fig S5) In total we tested410399 double and 195666 triple mutants forfitness defects meeting a previously establishedintermediate magnitude cutoff (15) (data S2) andidentified 9363 digenic and 3196 trigenic negativeinteractions From detailed validation of trigenicinteractions of our CLN1-CLN2 double-mutantquery which was screened previously (19) weestimated a false-negative rate of ~40 a false-positive rate of ~20 and a true-positive ratebetween ~60 and ~75 (table S2 and fig S6) (16)which is consistent with our previous global di-genic network analysis (8)The distribution of trigenic interaction degree

for array strains shows that the majority of low-degree genes (70) account for ~88 of all tri-genic interactions whereas highly connectedgenes contribute the remaining ~12 of interac-tions Thus the trigenic interactions are not as-sociated with a small set of highly connectedgenes rather the interactions are distributedacross many different genes (fig S7) On the otherhand with a smaller more biased set of double-mutant query genes the distribution of trigenicinteraction degree shows that ~22 of them ac-counted for 51 of trigenic interactions indicat-ing that a particular subset of the digenic querieswere enriched for trigenic interactions (fig S7)About one-third of the newly mapped trigenicinteractions identified connections that werenot observed in our digenic control network werefer to these as ldquonovelrdquo trigenic interactions Theremaining approximately two-thirds of the trigenicinteractions overlapped a digenic interaction whilestill exhibiting a stronger than expected fitnessdefect in the triple mutant these we refer to asldquomodifiedrdquo trigenic interactions (fig S8A) Thusalthough a substantial fraction of trigenic inter-actions elucidate totally new functional informationthe majority of the trigenic interactions we mappedexpand upon the digenic interaction networkWe first assessed the functional information

embedded in the trigenic network by comparing

the distributions of digenic and trigenic inter-actions across different biological processesAs observed previously (15) digenic interactionswere enriched among genes annotated to thesame biological process and although the mag-nitude of trigenic interaction enrichment wassomewhat lower they were comparably enrichedfor genes within the same bioprocess (Fig 2A)We also evaluated the enrichment of digenic andtrigenic interactions across common functionalstandards including annotation to the same GeneOntology (GO) biological process subcellular-localization pattern protein-protein interactionand gene coexpression (Fig 2B) Like digenicinteractions (7 15) genes involved in trigenicinteractions were significantly enriched for allof these standards with genes participating inthe ldquomodifiedrdquo class of trigenic interactions ex-hibiting stronger functional relationships (figS8B) as well as a stronger magnitude of inter-actions (fig S8C) Thus trigenic interactionsresemble digenic interactions in that they arerich in functional information which means thatgenes participating in many trigenic interactionscan be predicted from alternative data sets andgeneral knowledge of cellular function

Trigenic interactions expand functionalconnections mapped by the globaldigenic network

Our functional analysis revealed that trigenicinteractions have some properties distinct from

those of digenic interactions which suggests thattrigenic interactions may be useful for discover-ing previously unknown connections betweengenes and their corresponding pathways As anillustrative example we examined the MDY2-MTC1 double-mutant query which is a highlyconnected hub within the trigenic networkMDY2encodes a protein that interacts and functionswith components of the GET (guided entry oftail-anchor) pathway (20) which is importantfor Golgindashtondashendoplasmic reticulum (ER) traf-ficking and inserting tail-anchored proteins intoER membranes MTC1 encodes a protein of un-known function that localizes to the early Golgiapparatus The MTC1 digenic interaction profileis similar to that of USO1 which is involved invesicle-mediated ER-to-Golgi transport (21) andRUD3 which encodes a Golgi matrix proteinimportant for the structural organization of thecis-Golgi (22) suggesting that MTC1 also has arole in the early secretory pathwayAs expected from these previous findings the

MDY2 and MTC1 digenic interactions identifiedin our screen were enriched for genes involvedin cell polarity and the early secretory pathway(Fig 3 fig S9A and data S6) However theMDY2-MTC1 double-mutant query profile encompasseda much more functionally diverse set of genes(Fig 3) For example although we observed noveltrigenic interactions with ER-to-Golgi transportgenes we also observed trigenic interactionswith genes involved in other modes of vesicle

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 3 of 9

Fold change over background(all p lt 0008)

Co-localization

Co-annotation

Co-expression

PPI

0 2 4 6 8

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Withinprocess

|log2 fold change|

enrichment p lt 005p gt 005

05 2 35

BA

Trigenic interactionsDigenic interactions

Fig 2 Functional characterization of trigenic interactions (A) Frequency of negative geneticinteractions within biological processes For our analysis we used the fraction of screened query-array combinations exhibiting negative interactions belonging to functional gene sets annotated bySAFE (spatial analysis of functional enrichment) on the global genetic interaction network (55)The ldquowithin processrdquo category received a count for any combination in which both genes for digenicinteractions or all three genes for trigenic interactions were annotated to the same term The sizeof the circle assigned to each ldquowithin processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) Significance was assessedwith a hypergeometric test P lt 005 Blue circles represent significant enrichment gray circlesdenote no significant change (B) Enrichment of negative digenic and trigenic interactions acrossfour functional standards The dashed line indicates no enrichment The functional standards aremerged protein-protein interaction (PPI) (56ndash60) coannotation (based on SAFE terms) (7)coexpression (61) and colocalization (62) Significance was assessed with a hypergeometric test represents 10minus4 le P lt 001 represents P lt 10minus4

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

trafficking including endocytosis and peroxi-some biology Moreover MDY2-MTC1 trigenicinteractions identified connections to genes withmore diverse functions such as components ofthe elongator complex (Fig 3) which controlsthe modification of wobble nucleosides in tRNAsand several genes involved in DNA replicationand repair Notably the MDY2-MTC1 query alsoshowed a trigenic interaction with TOR1 (targetof rapamycin 1) which encodes the key kinasesubunit of the TORC1 complex that is requiredfor growth in response to nutrients by regulat-ing ribosome biogenesis nutrient transport andautophagy (23) Consistent with this observationthe MDY2-MTC1 trigenic interaction networkcaptures a set of genes that have a dual role inTORC1 signaling and sorting of the general ami-no acid permease including GTR1 MEH1 andLST4 (24ndash26)The spectrum of bioprocesses that are repre-

sented in genetic interaction profiles can be vis-ualized by mapping functional enrichment withinthe context of the global yeast digenic interac-tion profile similarity network which clustersgenes into 17 distinct bioprocesses (7 27) (Fig 4A)In comparison with the MDY2 and MTC1 digenicinteraction profiles (Fig 4 B and C) the MDY2-MTC1 trigenic interactions were enriched notonly for vesicle trafficking and cell polarity bio-process regions of the network but also in regionsencompassing genes annotated to the tRNA wob-ble modification bioprocess DNA replication andrepair as well as mitosis and chromosome segre-gation (Fig 4D) Thus the MDY2-MTC1 trigenicinteraction profile exhibited a more expandedand functionally diverse set of connections thaneither of the corresponding MDY2 or MTC1 di-genic interaction profilesWe used a variety of assays to test three func-

tional connections revealed by the MDY2-MTC1trigenic interaction profile First although the

MDY2-MTC1 double-mutant strain did not showan exaggerated cell biological phenotype asso-ciated with the early trafficking function (figS9B) it displayed a marked synthetic sick pheno-type when combined with deletion of SLA1 whichis involved in cortical actin assembly and endo-cytic vesicle formation which translates into anextended Sla1 patch lifetime reflecting a defectin endocytosis (Fig 5A) (28) Second given anegative trigenic interaction with OAF1 whichencodes an oleate-activated transcription factorinvolved in peroxisome organization and bio-genesis (29) we used fluorescence microscopyto explore peroxisome morphology The MDY2-MTC1 double mutant displayed an accumulationof relatively small peroxisomes (Fig 5B) whichmay be indicative of a defect in ER-derived per-oxisome membrane biogenesis (30) Third theMDY2-MTC1 double mutant showed pronouncedsensitivity to hydroxyurea (HU) but not methylmethanesulfonate (MMS) which is consistentwith a specific defect in DNA replication and re-flects the negative genetic interactions we ob-served with a number of DNA replication andrepair genes including NSE4 and NSE5 whichencode components of the Smc5-Smc6 complexthat mediates resolution of DNA structures span-ning sister chromatids (Fig 5C and fig S10 A toC) (31) We suspect that the MDY2-MTC1 doublemutant may be primarily defective in traffick-ing functions that can modulate signaling ormetabolic pathways and thereby influence DNAsynthesis and repair pathways indirectly (figS10 D to H)

Trigenic interaction profiles are morefunctionally diverse than theircorresponding digenic profiles

To test the generality of whether query genesconnect to more functionally divergent genesthrough trigenic interactions than through di-

genic interactions we compared digenic pro-file similarity of pairs of genes spanned by eitherdigenic or trigenic interactions Indeed genesinvolved in trigenic interactions tend to showprofiles that are less similar than those con-nected by digenic interactions which suggeststhat they are less functionally related than thoseconnected by digenic interactions (Fig 6A andfig S11A) We also found that trigenic interactionswere more enriched than digenic interactionsfor connections that bridge several differentbiological processes including mRNA and tRNAprocessing vesicle trafficking mitosis and chro-mosome segregation and glycosylation and pro-tein folding and targeting (Fig 6B) Moreoveras we showed for theMDY2-MTC1 double-mutantquery (Fig 4) trigenic interaction profiles weregenerally enriched for genes spanning more di-verse bioprocesses than the corresponding di-genic interaction profiles (Fig 6C and fig S11B to D) Genes involved in vesicle traffickingwere particularly enriched for trigenic interac-tions occurring between bioprocesses (Fig 6Band figs S7 and S11D) As we observed for MDY2-MTC1 other double-mutant queries carryingmutations in genes implicated in membrane traf-ficking were enriched for trigenic interactionswith genes involved in DNA replication and re-pair machinery which may indicate a generalconnection between these two bioprocesses Forexample the digenic query strain MVP1-MRL1which carries mutations in genes required forsorting proteins to the vacuole (32 33) and thestrain SEC27-GET4 which carries mutations ingenes involved in ER-to-Golgi transport (34) andthe insertion of tail-anchored proteins into ERmembrane (20) both exhibited an enrichmentof trigenic interactions with DNA replicationand repair machinery (fig S11D) In general ourfindings show that trigenic interaction profilesare composed of connections involving genes

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 4 of 9

EMC Sec62Sec63

EMC3 EMC4 SEC66

Trigenic interactions

GSE Lst4-Lst7Pan1Sla1End3

LST4SLA1 GTR1RVS161

OAFRvs161167

OAF1MEH1PAN1

Elongator

DNA replication factor C

CTF8 CTF18

ELP6ELP2 ELP3 ELP4ELP1

MRX

MRE11

TORC1

TOR1

MDY2 MTC1

GET

GET1 GET2

RIC1 YPT6

TRS23

COG8

BET3

COG2 COG7

TRS20 TRS85

COG6COG3 COG5

TRAPP

COG

Digenic interactions

Ric1Rgp1Ypt6

MDY2MTC1

GET3

RPN6 RPN7

1922S regulator

Smc5-Smc6

NSE4 NSE5

Fig 3 The MDY2-MTC1 double mutant a hub on the trigenicinteraction network Representative digenic interactions are high-lighted for MDY2 and MTC1 single-mutant query genes and rep-resentative trigenic interactions are shown for the MDY2-MTC1double-mutant query The network was visualized using Cytoscape(63) Genes were chosen from representative protein complexes (8) inwhich ge50 of members on the diagnostic array display geneticinteractions Negative genetic interactions (e or t lt ndash008 P lt 005)are depicted All of the digenic and trigenic interactions displayedhave been confirmed by tetrad analysis Nodes are color coded onthe basis of their biological roles and are labeled with gene namesGenes are grouped according to specific protein complexes

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 5: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

trafficking including endocytosis and peroxi-some biology Moreover MDY2-MTC1 trigenicinteractions identified connections to genes withmore diverse functions such as components ofthe elongator complex (Fig 3) which controlsthe modification of wobble nucleosides in tRNAsand several genes involved in DNA replicationand repair Notably the MDY2-MTC1 query alsoshowed a trigenic interaction with TOR1 (targetof rapamycin 1) which encodes the key kinasesubunit of the TORC1 complex that is requiredfor growth in response to nutrients by regulat-ing ribosome biogenesis nutrient transport andautophagy (23) Consistent with this observationthe MDY2-MTC1 trigenic interaction networkcaptures a set of genes that have a dual role inTORC1 signaling and sorting of the general ami-no acid permease including GTR1 MEH1 andLST4 (24ndash26)The spectrum of bioprocesses that are repre-

sented in genetic interaction profiles can be vis-ualized by mapping functional enrichment withinthe context of the global yeast digenic interac-tion profile similarity network which clustersgenes into 17 distinct bioprocesses (7 27) (Fig 4A)In comparison with the MDY2 and MTC1 digenicinteraction profiles (Fig 4 B and C) the MDY2-MTC1 trigenic interactions were enriched notonly for vesicle trafficking and cell polarity bio-process regions of the network but also in regionsencompassing genes annotated to the tRNA wob-ble modification bioprocess DNA replication andrepair as well as mitosis and chromosome segre-gation (Fig 4D) Thus the MDY2-MTC1 trigenicinteraction profile exhibited a more expandedand functionally diverse set of connections thaneither of the corresponding MDY2 or MTC1 di-genic interaction profilesWe used a variety of assays to test three func-

tional connections revealed by the MDY2-MTC1trigenic interaction profile First although the

MDY2-MTC1 double-mutant strain did not showan exaggerated cell biological phenotype asso-ciated with the early trafficking function (figS9B) it displayed a marked synthetic sick pheno-type when combined with deletion of SLA1 whichis involved in cortical actin assembly and endo-cytic vesicle formation which translates into anextended Sla1 patch lifetime reflecting a defectin endocytosis (Fig 5A) (28) Second given anegative trigenic interaction with OAF1 whichencodes an oleate-activated transcription factorinvolved in peroxisome organization and bio-genesis (29) we used fluorescence microscopyto explore peroxisome morphology The MDY2-MTC1 double mutant displayed an accumulationof relatively small peroxisomes (Fig 5B) whichmay be indicative of a defect in ER-derived per-oxisome membrane biogenesis (30) Third theMDY2-MTC1 double mutant showed pronouncedsensitivity to hydroxyurea (HU) but not methylmethanesulfonate (MMS) which is consistentwith a specific defect in DNA replication and re-flects the negative genetic interactions we ob-served with a number of DNA replication andrepair genes including NSE4 and NSE5 whichencode components of the Smc5-Smc6 complexthat mediates resolution of DNA structures span-ning sister chromatids (Fig 5C and fig S10 A toC) (31) We suspect that the MDY2-MTC1 doublemutant may be primarily defective in traffick-ing functions that can modulate signaling ormetabolic pathways and thereby influence DNAsynthesis and repair pathways indirectly (figS10 D to H)

Trigenic interaction profiles are morefunctionally diverse than theircorresponding digenic profiles

To test the generality of whether query genesconnect to more functionally divergent genesthrough trigenic interactions than through di-

genic interactions we compared digenic pro-file similarity of pairs of genes spanned by eitherdigenic or trigenic interactions Indeed genesinvolved in trigenic interactions tend to showprofiles that are less similar than those con-nected by digenic interactions which suggeststhat they are less functionally related than thoseconnected by digenic interactions (Fig 6A andfig S11A) We also found that trigenic interactionswere more enriched than digenic interactionsfor connections that bridge several differentbiological processes including mRNA and tRNAprocessing vesicle trafficking mitosis and chro-mosome segregation and glycosylation and pro-tein folding and targeting (Fig 6B) Moreoveras we showed for theMDY2-MTC1 double-mutantquery (Fig 4) trigenic interaction profiles weregenerally enriched for genes spanning more di-verse bioprocesses than the corresponding di-genic interaction profiles (Fig 6C and fig S11B to D) Genes involved in vesicle traffickingwere particularly enriched for trigenic interac-tions occurring between bioprocesses (Fig 6Band figs S7 and S11D) As we observed for MDY2-MTC1 other double-mutant queries carryingmutations in genes implicated in membrane traf-ficking were enriched for trigenic interactionswith genes involved in DNA replication and re-pair machinery which may indicate a generalconnection between these two bioprocesses Forexample the digenic query strain MVP1-MRL1which carries mutations in genes required forsorting proteins to the vacuole (32 33) and thestrain SEC27-GET4 which carries mutations ingenes involved in ER-to-Golgi transport (34) andthe insertion of tail-anchored proteins into ERmembrane (20) both exhibited an enrichmentof trigenic interactions with DNA replicationand repair machinery (fig S11D) In general ourfindings show that trigenic interaction profilesare composed of connections involving genes

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 4 of 9

EMC Sec62Sec63

EMC3 EMC4 SEC66

Trigenic interactions

GSE Lst4-Lst7Pan1Sla1End3

LST4SLA1 GTR1RVS161

OAFRvs161167

OAF1MEH1PAN1

Elongator

DNA replication factor C

CTF8 CTF18

ELP6ELP2 ELP3 ELP4ELP1

MRX

MRE11

TORC1

TOR1

MDY2 MTC1

GET

GET1 GET2

RIC1 YPT6

TRS23

COG8

BET3

COG2 COG7

TRS20 TRS85

COG6COG3 COG5

TRAPP

COG

Digenic interactions

Ric1Rgp1Ypt6

MDY2MTC1

GET3

RPN6 RPN7

1922S regulator

Smc5-Smc6

NSE4 NSE5

Fig 3 The MDY2-MTC1 double mutant a hub on the trigenicinteraction network Representative digenic interactions are high-lighted for MDY2 and MTC1 single-mutant query genes and rep-resentative trigenic interactions are shown for the MDY2-MTC1double-mutant query The network was visualized using Cytoscape(63) Genes were chosen from representative protein complexes (8) inwhich ge50 of members on the diagnostic array display geneticinteractions Negative genetic interactions (e or t lt ndash008 P lt 005)are depicted All of the digenic and trigenic interactions displayedhave been confirmed by tetrad analysis Nodes are color coded onthe basis of their biological roles and are labeled with gene namesGenes are grouped according to specific protein complexes

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 6: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

that are more functionally diverse than their cor-responding digenic interaction profiles (Fig 6C)However despite their higher tendency to con-nect diverse processes a significant fraction oftrigenic interactions occurs among genes withinthe same bioprocess (P lt 1 times 10minus16 hypergeo-metric test) (Fig 2A)

Gene features of trigenic interactionsand the expanse of the globaltrigenic landscape

Having selected the query gene pairs based onthe properties of the global digenic interactionnetwork we can assess how these properties relateto trigenic interaction frequency The strongestcorrelation with the number of observed trigenicinteractions for each gene pair was the digenicgenetic interaction profile similarity (correlationcoefficient r = 041 P = 12 times 10minus7) followed bythe average number of digenic interactions ofthe query genes (r = 025 P = 19 times 10minus3) and the

strength of a direct negative genetic interactionbetween the query gene pair (r = 023 P = 54 times10minus3) (fig S12 A to C) Thus numerous trigenicinteractions were observed for functionally re-lated query genes which display overlappingprofiles on the digenic similarity network andoften show a digenic interaction with each other(7) (Fig 7A) As observed for digenic interactions(7) the frequency of trigenic interactions washighly correlated with the fitness defect of thedouble-mutant query strain (fig S13) Consistentwith this observation essential genes exhibitedhigh connectivity on the trigenic interactionnetwork A double-mutant query that carries atleast one temperature-sensitive allele of an es-sential gene which is often associated with afitness defect at the semipermissive screeningtemperature exhibited more genetic interactionsthan a query deleted for a pair of nonessentialgenes (P = 0035) (Fig 7B) More generally querygenes that are highly connected on the digenic

network are also highly connected on the tri-genic network (fig S12D)Notably trigenic interactions tend to be ~25

weaker than digenic interactions (P lt 17 times 10minus98)(Fig 7C) which means the average digenic in-teraction often has a more profound phenotypethan the average trigenic interaction Howeverto fully understand the potential for trigenic in-teractions to drive fitness defects we also needto estimate the frequency at which they occurBecause we have mapped digenic interactionscomprehensively and we know the false-positiveand false-negative rates associated with thisanalysis we can estimate the number of digenicinteractions within the yeast genome reveal-ing a distribution that centers on ~6 times 105 totalnegative interactions (Fig 7D) (16) Further-more because digenic interaction propertiesare predictive of trigenic interaction degree wecan also extrapolate our findings to estimate thenumber of negative trigenic interactions across

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 5 of 9

Fig 4 Enrichment of genetic interactions within bioprocessesdefined by a global network of digenic interaction profilesimilarities (A) The global digenic interaction profile similarity network(7) was annotated using SAFE (55) identifying network regionsenriched for similar GO biological process terms as outlined by dashed

lines rRNA ribosomal RNA ncRNA noncoding RNA MVB multi-vesicular body (B) MDY2 digenic interactions showing bioprocessenrichments (C) MTC1 digenic interactions showing bioprocessenrichments (D) MDY2-MTC1 trigenic interactions showingbioprocess enrichments

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 7: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

the whole genome As noted earlier we selectedgene pairs for trigenic analysis to fill bins of vary-ing attributes including double-mutant querieswith either weak or strong interactions as wellas those with either sparse or rich genetic in-teraction profiles as depicted schematically inFigs 1A and 7A (table S1) For extrapolation tothe whole genome we used the mean trigenicinteraction degree of double mutants in a givenbin as the expected degree for any hypotheticalpair from the genome with similar character-istics (data S7 and fig S14) Integrating thisvalue across the total number of gene pairs withthe given characteristics which preserves thedouble-mutant distribution across different di-genic interaction features and summing overall bins yielded an estimate of the total numberof trigenic interactionsIn the yeast genome there are 2000 times as

many possible triple gene combinations (36 bil-lion) as possible gene pairs (18 million) but thedensity of interactions (as both observed andextrapolated) is similar reduced by only a factorof ~3 for trigenic interactions (table S3) Wepredict that ~108 trigenic combinations exhibita negative genetic interaction generating a con-servative estimate of on the order of 100 timesas many trigenic interactions as observed forthe global digenic network (Fig 7D and tableS3) To establish confidence intervals (CIs) forthe estimate we repeated the extrapolation pro-cess with 10000 bootstrapped samplings of the151 double-mutant query pairs keeping theirassociated trigenic interactions degrees and thecorresponding digenic interaction features con-stant The bin with the lowest digenic interac-tion degree encompasses a large fraction of thepotential double mutants in the genome and isassigned a low trigenic interaction degree whichmeans that the summarized estimate providedis likely a conservative underestimate More-over because our binning scheme restricts ourextrapolation to ~25 of the potential trigenicinteraction space (eg by omitting potentialdouble-mutant queries that show a positive di-genic interaction) we are underestimating itsextent and the true number of trigenic inter-actions is likely to be several times higher (figS15 and table S3) (16) The vast expanse of theglobal trigenic interaction network points tothe potential for higher-order interactions toaffect all aspects of the genetics of outbred pop-ulations including the genotype-to-phenotyperelationship

Discussion

Systematic mapping of trigenic interactions re-vealed that their properties resemble those ofdigenic interactions because they often connectfunctionally related genes which means that tri-genic interactions have the potential to con-tribute to our understanding of the functionalwiring diagram of the cell The global digenicnetwork is predictive of trigenic interactionsbecause query gene pairs showing propertiesassociated with shared functionality such asoverlapping digenic interaction profiles or a

negative digenic genetic interaction often mapnumerous trigenic interactions (Fig 7A) Thusif two query genes are in the same or similar bio-process cluster on the global digenic profile sim-ilarity network (Fig 4A) they will likely showa rich trigenic interaction profile as we observedfor the MDY2-MTC1 double mutant query (Figs 3and 4D) Gene essentiality and the average di-genic interaction degree of the query gene pairwere also correlated with trigenic connectivity(Fig 7B) indicating that highly connected hubsare consistent on both the digenic and trigenicinteraction networks (fig S12 C and D)Many of the trigenic interactions we observed

overlapped with at least one digenic interactionIn some cases we chose query gene pairs display-ing a negative genetic interaction and so all ofthe trigenic interactions in these profiles accen-tuated the query interaction (fig S8) Moreover

a substantial proportion of trigenic interac-tions measured for noninteracting query pairsexacerbated a digenic interaction that was pre-viously seen between one or both of the querygenes and the third gene (fig S8) Thus our find-ings show that negative trigenic interactions oftenhighlight the potential for a third mutation toamplify the phenotype of a digenic interactionAnalogously in human genetics the variation inan individualrsquos genetic background can have pro-found influence on the penetrance of the pheno-type associated with a disease gene (35)Although we found that trigenic interactions

tend to be slightly weaker than digenic interac-tions they are ~100 times more numerous andare more functionally diverse than their digeniccounterparts The expanse of possible three-genecombinations makes exhaustive trigenic inter-action mapping intractable with our current

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 6 of 9

wild type mdy2Δ

mtc1Δ mdy2Δ mtc1Δ

Sla1-GFP0 10 20 30 40 50

Sla1-GFP Patch Lifetime (s)

wild typemdy2Δmtc1Δ

mdy2Δ mtc1Δ

p lt 00001

0 05 1 15

wild type

mdy2Δ

mtc1Δ

mdy2Δ mtc1Δ

p = 001

0 05 1 15

Peroxisome area relative to wt

No peroxisomescellrelative to wt

scale bar = 10 s

Pex14-GFP

scale bar = 5 microm

A

mdy2Δ mtc1Δ

B

p = 0014

wild type

C

= Δ allele = wild type

= trigenic interaction

YPD 50 mM HU

mdy2ΔWT

mdy2Δ mtc1Δ mtc1Δ

0035 MMS

30degC

MDY2MTC1SLA1

+-+

---

++-

-++

+--

--+

-+-

+++

+-

Fig 5 Trigenic interactions reflect the physiology of the MDY2-MTC1 double-mutant querystrain (A) Endocytic membrane trafficking is impaired in the mdy2D-mtc1D double-mutant querystrain (Top) Example of tetrad analysis confirmations for the mdy2D-mtc1D-sla1D triple-mutantstrain (Bottom left) Endocytic uptake dynamics were examined with the Sla1ndashgreen fluorescent protein(GFP) reporter Representative kymographs are displayed for the wild type and the mdy2D-mtc1Ddouble mutant (Bottom right) Lifetime of Sla1-GFP endocytic vesicle formation was quantified across~100 different patches in two independent experiments Error bars represent SD (B) Peroxisomebiogenesis was monitored in the wild type (wt) and inmdy2D mtc1D and mdy2D-mtc1D mutants usingPex14p-GFP reporter (C) Growth response to HU and MMS for the wild type and mdy2D mtc1Dand mdy2D-mtc1D mutants YPD yeast extract peptone and dextrose

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 8: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

methodology However the substantial overlapof the digenic and trigenic networks indicatesthat the genetic landscape of the cell expandswith higher-order genetic interactions but doesnot change drastically in terms of its functionalmodularity Thus the global digenic network ishighly informative of potential trigenic inter-actions and can be used effectively to predictcandidate query gene pairs for efficient trigenicinteraction analysisTrigenic interaction data may inform a wide

variety of subjects within biology For exam-ple the number magnitude and properties ofdigenic and trigenic interactions clarifies as-pects of speciation theory in evolutionary biol-ogy (36 37) Hybrids between species exhibitreduced fitness which is usually attributed tonegative (epistatic) interactions among genesthat diverged in isolated populations Eachpopulation may evolve fixed variants that are

neutral or adaptive in their own genetic back-grounds When these variants are brought to-gether for the first time in hybrid genomes theymay cause deleterious genetic interactions alsotermed Dobzhansky-Muller incompatibilities Aspopulations diverge from one another the num-ber of potential digenic interactions increasesas the square of the number of substitutionsthe so-called ldquosnowball effectrdquo (36 38 39) Thatis each subsequent substitution in a distinctpopulation has the potential to interact withany substitution from the other population (andvice versa) and thus the probability of a specia-tion event grows with each step Most speci-ation genetics research has focused on thesedigenic interactions However the number oftrigenic combinations accumulates exponentiallyfaster than the number of digenic combinationsBoth digenic and trigenic interactions have beenimplicated in speciation (40 41) but the general

extent to which digenic or complex negativegenetic interactions drive speciation remainsunknown If digenic interactions do in factplay a major role in orchestrating speciationthen either the frequency andor the strengthof deleterious trigenic interactions must berelatively smaller than that of digenic inter-actions Our systematic analysis shows thattrigenic interactions are somewhat less likelyto occur (by a factor of 3 ~3 versus ~1 fordigenic versus trigenic respectively) and gener-ally weaker (~25 weaker) than digenic inter-actions Nevertheless modeling based on ourfindings suggests that trigenic interactions aresubstantially common and often strong enoughto play a key role in the evolution of hybrid in-viability (fig S16 and table S4) (16) Because theconnections associated with higher-order inter-actions may often overlap with those of simplerinteractions and because those simpler inter-actions require fewer substitutions and will oftenmanifest first our findings may also suggest thatthe evolution of even more highly complex in-teractions may be limited even though theirabsolute numbers increase exponentially apossibility that is consistent with evolutionarytheory (38)Our trigenic interaction study is also relevant

to synthetic biology efforts aimed at efficientsynthesis (42 43) and design of minimal ge-nomes (44 45) Digenic synthetic lethal interac-tions were recently noted as a major constraintin the design of the minimal genome for thebacteria Mycoplasma mycoides for which aviable genome could be constructed only afterresolving lethal interactions that arose betweennonessential genes (44) For species in whichsystematic gene perturbation studies have beenconducted the proportion of essential genes isrelatively small (eg ~20 in yeast ~10 inhuman cells which increases to 20 when onlyexpressed genes are considered) (13 46ndash48)However we expect that digenic and trigenicinteractions will dictate much larger minimalgenomes than the essential gene set even forgrowth under simple laboratory conditions Withthe complete digenic network (7) we estimatethat the minimal yeast genome would encom-pass more than ~70 of genes after accountingfor digenic interactions (table S5) (16) With theinclusion of constraints imposed by trigenic in-teractions we expect that a minimal genomewithout a substantial fitness defect may nearlyapproach the complete set of genes encoded inthe genome Thus genetic interactions may helpto explain the large gap between the number ofgenes with strong individual fitness defects andthe total genome size and the prevalence ofyeast negative trigenic interactions suggeststhat many genomes lack the potential for sub-stantial compression while maintaining normalfitnessIt is important to consider other types of ge-

netic interactions in addition to those associatedwith severe loss-of-function alleles due to entireopen reading frame deletions of nonessentialgenes or temperature-sensitive alleles of essential

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 7 of 9

Background

Digenic

Trigenic - modified

Trigenic - novel

Digenic interaction profile similarity

-015 0 025 050

20 40 60 80 100Frequency

p lt 10-30 p = 169 x 10-5

B

C

0123456

No

of b

iopr

oces

ses

Digenic interactions

Trigenic interactions

Digenic

inte

racti

ons

Trige

nic in

tera

ction

s

mRNA and tRNA processing

Nuclear-cytoplasmic transport

Vesicle trafficMitosis and chromosome segregation

DNA replication and repair

Transcription and chromatin organization

Glycosylation and protein foldingtargeting

Mitochondria and respiration

Between process

|log2 fold change|

enrichment p lt 005depletion p lt 005p gt 005

Trige

nic vs

Dige

nic

1 17Fold change

03 06 16

A

Fig 6 Trigenic interactions are more functionally distant than digenic interactions (A) Distribu-tion of genetic interaction profile similarities of genes showing digenic and trigenic interactions P valuesare based on a Wilcoxon rank sum test P lt 10minus30 (B) Frequency of negative genetic interactionsbetween biological processes using SAFE annotations for digenic and trigenic interactions (55) Thesize of the circle assigned to each ldquobetween processrdquo element reflects the fold increase over thebackground fraction of interactions (digenic = 0023 trigenic = 0016) P lt 005 based on ahypergeometric test The ldquobetween processrdquo category received a count for any combinations thatwere not counted in the ldquowithin processrdquo category shown in Fig 2A Filled blue circles representsignificant enrichment the open blue circle represents significant underenrichment and gray circlesdenote no significant change Trigenic versus digenic fold change (the ratio of trigenic interactionenrichment to digenic interaction enrichment) is represented by filled squares (black is maximal foldchange white is no fold change) In cases for which the ldquobetween processrdquo enrichment was observed butis not significant (P lt 005) the square is outlined with a dashed line (C) Number of SAFE bioprocessclusters enriched for digenic or trigenic interactions

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 9: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

genes Our analysis revealed that double mu-tants with strong fitness defects often show richtrigenic interaction profiles (fig S13C) Similarlysingle-mutant fitness defects also correlate withdigenic interaction degree (fig S13A) (7) Presum-ably the weaker fitness effects associated withthe variation found in natural populations mayrequire higher-order combinations involvingmore than two genes to influence trait herita-bility through genetic interactions (49) In theyeast model system genetic interactions werefound to play an important role in the herita-bility of a number of different quantitative traitspossibly with a greater contribution made bydigenic interactions versus higher-order inter-actions (4 5 50) The genetic mechanism un-derlying conditional essentiality in which a givenyeast gene is nonessential in one genetic back-ground but essential in another often appearsto be associated with a complex set of modifierloci (49) as do a number of other traits (51 52)

Thus both digenic and higher-order interactionsare established components of the genetic ar-chitecture of yeast complex traits and similarfindings have been made in a number of otherorganisms (53) In part because model organismpopulations have allele distributions that dif-fer from those in humans the degree to whichhigher-order genetic interactions will contrib-ute to the genetics of complex human diseaseremains to be seen (50 54) Nevertheless theextensive landscape of trigenic interactions re-vealed here for yeast as well as their capacityfor generating functionally diverse phenotypesand driving speciation suggests that higher-order genetic interactions may play a key rolein the genotype-to-phenotype relationship

Materials and methods summary

The supplementary materials contain a detaileddescription of materials and methods for theconstruction of yeast single- double- and triple-

mutant strains as well as quantification of ge-netic interactions and any associated analysesGeneral methodological information and ref-erences to specific sections appear throughoutthe text

REFERENCES AND NOTES

1 J L Hartman IV B Garvik L Hartwell Principles for thebuffering of genetic variation Science 291 1001ndash1004 (2001)doi 101126science29155061001 pmid 11232561

2 1000 Genomes Project Consortium A global reference forhuman genetic variation Nature 526 68ndash74 (2015)doi 101038nature15393 pmid 26432245

3 O Zuk E Hechter S R Sunyaev E S Lander The mystery ofmissing heritability Genetic interactions create phantomheritability Proc Natl Acad Sci USA 109 1193ndash1198 (2012)doi 101073pnas1119675109 pmid 22223662

4 J S Bloom I M Ehrenreich W T Loo T L Lite L KruglyakFinding the sources of missing heritability in a yeast cross Nature494 234ndash237 (2013) doi 101038nature11867 pmid 23376951

5 J S Bloom et al Genetic interactions contribute less thanadditive effects to quantitative trait variation in yeastNat Commun 6 8712 (2015) doi 101038ncomms9712pmid 26537231

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 8 of 9

Trigenic interactionsDigenic interactions

Estimate of global negative interaction network105 106 107 109B

oots

trap

ped

freq

uenc

y

0

500

1000

1500

Trigenic interactionsDigenic interactions

p = 49 x 10-106

Freq

uenc

y

0

01

02

04

Genetic interaction scoreminus1 minus08 minus06 minus04 minus02 0minus12

108

A C

Digenic interaction

profile similarity

Ess

entia

lN

ones

sent

ial

Ess

entia

lN

ones

sent

ial

Single mutantquery

Double mutantquery

Avg

gen

etic

inte

ract

ion

degr

ee

B

0

25

50

p = 398x10-7

p = 0035

D

Neg digenic

interaction strength

Avg

dig

enic

inte

ract

ion

degr

ee

Fig 7 Relation of digenic and trigenic interaction networks(A) Trigenic interaction degree distribution correlated with three quanti-tative features of genes on the digenic interaction network (i) Interactionprofile similarity of the two genes in the double-mutant query gene pair(bin thresholds ndash002 003 01 +1) which generates three bins foraverage digenic interaction profile similarity (r) ndash002 lt r lt 003003 le r lt 01 01 le r (ii) Negative digenic interaction strength associatedwith the double-mutant query gene pair (bin thresholds 0 ndash008ndash01 ndash1) which generates three bins for digenic interaction score (e)e lt ndash01 ndash01 le e lt ndash008 ndash008 le e lt 0 (iii) Average digenic interactiondegree which represents the average number of negative geneticinteractions associated with each of the genes of the double-mutantquery gene pair (bin thresholds 10 45 70 +1) which generates threebins for average digenic interaction degree 10 le degree lt 45 45 le degreelt70 70 le degree The bin with the highest average negative trigenicinteraction degree at the intermediate interaction score cutoff

(t lt ndash008) of 635 is shown in dark blue (B) Essentiality determinestrigenic interaction degree Number of single mutants 254 nonessentialgenes 47 essential genes Number of double mutants 111 nonessentialgene pairs 40 essential or mixed essentiality gene pairs Mean geneticinteraction is represented error bars indicate SEM P values are based ona t test Negative genetic interactions (e or t lt ndash008 P lt 005) are depicted(C) Cumulative distribution of negative digenic and trigenic interactionscore magnitudes Pairwise significance was assessed with a Wilcoxon ranksum test (D) Estimates of the number of digenic and trigenic interactionsat the intermediate score cutoff (e or t lt ndash008 P lt 005) Bootstrappingwas used to generate the estimate by sampling 10000 times withreplacement Dashed lines indicate the 95 CIs solid lines denotethe estimated extent of the trigenic interaction landscape Thisconservative estimate of the total number of trigenic interactionsin the yeast genome covers ~26 of the interaction space For thetotal genome-wide estimate see fig S15B and table S3

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 10: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

6 A H Tong et al Systematic genetic analysis with orderedarrays of yeast deletion mutants Science 294 2364ndash2368(2001) doi 101126science1065810 pmid 11743205

7 M Costanzo et al A global genetic interaction network maps awiring diagram of cellular function Science 353 aaf1420(2016) doi 101126scienceaaf1420 pmid 27708008

8 A Baryshnikova et al Quantitative analysis of fitness and geneticinteractions in yeast on a genome scale Nat Methods 71017ndash1024 (2010) doi 101038nmeth1534 pmid 21076421

9 P Novick D Botstein Phenotypic analysis of temperature-sensitive yeast actin mutants Cell 40 405ndash416 (1985)doi 1010160092-8674(85)90154-0 pmid 3967297

10 A Bender J R Pringle Use of a screen for synthetic lethal andmulticopy suppressee mutants to identify two new genesinvolved in morphogenesis in Saccharomyces cerevisiaeMol Cell Biol 11 1295ndash1305 (1991) doi 101128MCB1131295 pmid 1996092

11 J van Leeuwen et al Exploring genetic suppressioninteractions on a global scale Science 354 aag0839 (2016)doi 101126scienceaag0839 pmid 27811238

12 E A Winzeler et al Functional characterization of theS cerevisiae genome by gene deletion and parallel analysisScience 285 901ndash906 (1999) doi 101126science2855429901 pmid 10436161

13 G Giaever et al Functional profiling of the Saccharomycescerevisiae genome Nature 418 387ndash391 (2002)doi 101038nature00935 pmid 12140549

14 A H Tong et al Global mapping of the yeast geneticinteraction network Science 303 808ndash813 (2004)doi 101126science1091317 pmid 14764870

15 M Costanzo et al The genetic landscape of a cell Science 327425ndash431 (2010) doi 101126science1180823 pmid 20093466

16 Materials and methods are available as supplementarymaterials

17 R Deshpande J Nelson S W Simpkins M CostanzoJ S Piotrowski S C Li C Boone C L Myers Efficient strategiesfor screening large-scale genetic interaction networks bioRxiv159632 [Preprint] 5 July 2017 doi 101101159632

18 J E Haber et al Systematic triple-mutant analysis uncoversfunctional connectivity between pathways involved inchromosome regulation Cell Rep 3 2168ndash2178 (2013)doi 101016jcelrep201305007 pmid 23746449

19 J Zou et al Regulation of cell polarity throughphosphorylation of Bni4 by Pho85 G1 cyclin-dependent kinasesin Saccharomyces cerevisiae Mol Biol Cell 20 3239ndash3250(2009) doi 101091mbcE08-12-1255 pmid 19458192

20 M C Jonikas et al Comprehensive characterization of genesrequired for protein folding in the endoplasmic reticulumScience 323 1693ndash1697 (2009) doi 101126science1167983pmid 19325107

21 Y Noda T Yamagishi K Yoda Specific membrane recruitmentof Uso1 protein the essential endoplasmic reticulum-to-Golgitethering factor in yeast vesicular transport J Cell Biochem 101686ndash694 (2007) doi 101002jcb21225 pmid 17192843

22 A K Gillingham A H Tong C Boone S Munro The GTPaseArf1p and the ER to Golgi cargo receptor Erv14p cooperate torecruit the golgin Rud3p to the cis-Golgi J Cell Biol 167281ndash292 (2004) doi 101083jcb200407088 pmid 15504911

23 A Gonzaacutelez M N Hall Nutrient sensing and TOR signalingin yeast and mammals EMBO J 36 397ndash408 (2017)doi 1015252embj201696010 pmid 28096180

24 A van der Zand J Gent I Braakman H F Tabak Biochemicallydistinct vesicles from the endoplasmic reticulum fuse to formperoxisomes Cell 149 397ndash409 (2012) doi 101016jcell201201054 pmid 22500805

25 S Kira et al Dynamic relocation of the TORC1-Gtr12-Ego123complex is regulated by Gtr1 and Gtr2 Mol Biol Cell 27 382ndash396(2016) doi 101091mbcE15-07-0470 pmid 26609069

26 M P Peacuteli-Gulli A Sardu N Panchaud S Raucci C De VirgilioAmino acids stimulate TORC1 through Lst4-Lst7 a GTPase-activating protein complex for the Rag family GTPase Gtr2Cell Rep 13 1ndash7 (2015) doi 101016jcelrep201508059pmid 26387955

27 M Usaj et al TheCellMaporg A web-accessible database forvisualizing and mining the global yeast genetic interactionnetwork G3 7 1539ndash1549 (2017) doi 101534g3117040220pmid 28325812

28 Y Sun N T Leong T Wong D G Drubin A Pan1End3Sla1complex links Arp23-mediated actin assembly to sites ofclathrin-mediated endocytosis Mol Biol Cell 26 3841ndash3856(2015) doi 101091mbcE15-04-0252 pmid 26337384

29 I V Karpichev G M Small Global regulatory functions ofOaf1p and Pip2p (Oaf2p) transcription factors that regulate

genes encoding peroxisomal proteins in Saccharomycescerevisiae Mol Cell Biol 18 6560ndash6570 (1998) doi 101128MCB18116560 pmid 9774671

30 E H Hettema R Erdmann I van der Klei M Veenhuis Evolvingmodels for peroxisome biogenesis Curr Opin Cell Biol 29 25ndash30(2014) doi 101016jceb201402002 pmid 24681485

31 D E Bustard et al During replication stress non-SMC element5 (NSE5) is required for Smc56 protein complex functionalityat stalled forks J Biol Chem 287 11374ndash11383 (2012)doi 101074jbcM111336263 pmid 22303010

32 R J Chi et al Fission of SNX-BAR-coated endosomal retrogradetransport carriers is promoted by the dynamin-related proteinVps1 J Cell Biol 204 793ndash806 (2014) doi 101083jcb201309084 pmid 24567361

33 J R Whyte S Munro A yeast homolog of the mammalianmannose 6-phosphate receptors contributes to the sorting ofvacuolar hydrolases Curr Biol 11 1074ndash1078 (2001)doi 101016S0960-9822(01)00273-1 pmid 11470415

34 A Eugster G Frigerio M Dale R Duden The a- and bprime-COPWD40 domains mediate cargo-selective interactions withdistinct di-lysine motifs Mol Biol Cell 15 1011ndash1023 (2004)doi 101091mbcE03-10-0724 pmid 14699056

35 C A Argmann S M Houten J Zhu E E Schadt A nextgeneration multiscale view of inborn errors of metabolismCell Metab 23 13ndash26 (2016) doi 101016jcmet201511012pmid 26712461

36 S Gavrilets in Fitness Landscapes and the Origin of SpeciesS A Levin H S Horn Eds (Monographs in PopulationBiology Princeton Univ Press 2004) pp 149ndash194

37 B R Foley C G Rose D E Rundle W Leong S EdmandsPostzygotic isolation involves strong mitochondrial andsex-specific effects in Tigriopus californicus a species lackingheteromorphic sex chromosomes Heredity 111 391ndash401(2013) doi 101038hdy201361 pmid 23860232

38 H A Orr The population genetics of speciation The evolutionof hybrid incompatibilities Genetics 139 1805ndash1813 (1995)pmid 7789779

39 J J Welch Accumulating Dobzhansky-Muller incompatibilitiesReconciling theory and data Evolution 58 1145ndash1156 (2004)doi 101111j0014-38202004tb01695x pmid 15266965

40 L C Moyle T Nakazato Hybrid incompatibility ldquosnowballsrdquobetween Solanum species Science 329 1521ndash1523 (2010)doi 101126science1193063 pmid 20847271

41 S Tang D C Presgraves Evolution of the Drosophila nuclearpore complex results in multiple hybrid incompatibilitiesScience 323 779ndash782 (2009) doi 101126science1169123pmid 19197064

42 S M Richardson et al Design of a synthetic yeast genomeScience 355 1040ndash1044 (2017) doi 101126scienceaaf4557pmid 28280199

43 D G Gibson et al One-step assembly in yeast of 25 overlappingDNA fragments to form a complete synthetic Mycoplasmagenitalium genome Proc Natl Acad Sci USA 105 20404ndash20409(2008) doi 101073pnas0811011106 pmid 19073939

44 C A Hutchison 3rd et al Design and synthesis of a minimalbacterial genome Science 351 aad6253 (2016) doi 101126scienceaad6253 pmid 27013737

45 D G Gibson et al Creation of a bacterial cell controlled bya chemically synthesized genome Science 329 52ndash56 (2010)doi 101126science1190719 pmid 20488990

46 T Wang et al Identification and characterization of essentialgenes in the human genome Science 350 1096ndash1101(2015) doi 101126scienceaac7041 pmid 26472758

47 T Hart et al High-resolution CRISPR screens reveal fitness genesand genotype-specific cancer liabilities Cell 163 1515ndash1526(2015) doi 101016jcell201511015 pmid 26627737

48 V A Blomen et al Gene essentiality and synthetic lethalityin haploid human cells Science 350 1092ndash1096 (2015)doi 101126scienceaac7557 pmid 26472760

49 R D Dowell et al Genotype to phenotype A complex problemScience 328 469 (2010) doi 101126science1189015pmid 20413493

50 S K Forsberg J S Bloom M J Sadhu L Kruglyak Ouml CarlborgAccounting for genetic interactions improves modeling ofindividual quantitative trait phenotypes in yeast Nat Genet 49497ndash503 (2017) doi 101038ng3800 pmid 28250458

51 M B Taylor I M Ehrenreich Genetic interactions involvingfive or more genes contribute to a complex trait in yeastPLOS Genet 10 e1004324 (2014) doi 101371journalpgen1004324 pmid 24784154

52 M B Taylor I M Ehrenreich Higher-order genetic interactionsand their contribution to complex traits Trends Genet 3134ndash40 (2015) doi 101016jtig201409001 pmid 25284288

53 P C Phillips EpistasismdashThe essential role of gene interactionsin the structure and evolution of genetic systems Nat Rev Genet9 855ndash867 (2008) doi 101038nrg2452 pmid 18852697

54 T B Sackton D L Hartl Genotypic context and epistasis inindividuals and populations Cell 166 279ndash287 (2016)doi 101016jcell201606047 pmid 27419868

55 A Baryshnikova Systematic functional annotation andvisualization of biological networks Cell Syst 2 412ndash421(2016) doi 101016jcels201604014 pmid 27237738

56 A C Gavin et al Proteome survey reveals modularity ofthe yeast cell machinery Nature 440 631ndash636 (2006)doi 101038nature04532 pmid 16429126

57 N J Krogan et al Global landscape of protein complexes inthe yeast Saccharomyces cerevisiae Nature 440 637ndash643(2006) doi 101038nature04670 pmid 16554755

58 K Tarassov et al An in vivo map of the yeast proteininteractome Science 320 1465ndash1470 (2008) doi 101126science1153878 pmid 18467557

59 H Yu et al High-quality binary protein interaction map ofthe yeast interactome network Science 322 104ndash110(2008) doi 101126science1158684 pmid 18719252

60 M Babu et al Interaction landscape of membrane-proteincomplexes in Saccharomyces cerevisiae Nature 489 585ndash589(2012) doi 101038nature11354 pmid 22940862

61 C Huttenhower M Hibbs C Myers O G TroyanskayaA scalable method for integration and functional analysis ofmultiple microarray datasets Bioinformatics 22 2890ndash2897(2006) doi 101093bioinformaticsbtl492 pmid 17005538

62 Y T Chong et al Yeast proteome dynamics from single cellimaging and automated analysis Cell 161 1413ndash1424(2015) doi 101016jcell201504051 pmid 26046442

63 P Shannon et al Cytoscape A software environment forintegrated models of biomolecular interaction networksGenome Res 13 2498ndash2504 (2003) doi 101101gr1239303pmid 14597658

ACKNOWLEDGMENTS

We thank H Friesen J Hou and J Moffat for discussions andM P Masinas and J Nelson for logistical help Funding This workwas primarily supported by the NIH (grant R01HG005853 to CBBJA and CLM and grants R01HG005084 and R01GM104975to CLM) the Canadian Institutes of Health Research (CIHR) (grantsFDN-143264 and FDN-143265 to CB and BJA) and the NSF (grantDBI0953881 to CLM) Computing resources and data storageservices were partially provided by the Minnesota SupercomputingInstitute and the University of Minnesota Office of InformationTechnology respectively Additional support was provided by theCIHR (grant MOP-79368 to GWB) the NSF (DEB-1456462 to DB)the Swiss National Science Foundation (RL) the Canton of Genevaand the European Research Council Consolidator Grant program(RL) Natural Science and Engineering Research Council of CanadaPostgraduate Scholarship-Doctoral PGS D2 (EK) a University ofToronto Open Fellowship (EK) and a University of MinnesotaDoctoral Dissertation Fellowship (BV) CLM BJA and CB arefellows of the Canadian Institute for Advanced Research Authorcontributions Conceptualization EK BV BJA CB and CLMMethodology and investigation EK BV WW RD YC ABMMU JvL ENK CP AJD MP JZYW JH MR KX HHB-JSL ES and HZ Formal analysis EK BV WW MMUENK CP AJD JH KX HH MC RL AC DB and GWBResources GT Data curation MU Writing ndash original draft EKBV BJA CB and CLM Writing ndash review and editing EK BVWW RD AB MMU JvL ENK CP MC DB GWB BJACB and CLM Supervision BJA CB and CLM Projectadministration NVD and SS Funding acquisition BJA CB andCLM Competing interests The authors declare no competinginterests Data and materials availability All data files (data S1 toS7) associated with this study are described in detail and available in thesupplementary materials and can be downloaded from httpboonelabccbrutorontocasupplementkuzmin2018supplementhtml Data files S1 to S7 have also been deposited in the DRYADDigital Repository (doi 105061dryadtt367)

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3606386eaao1729supplDC1Materials and MethodsFigs S1 to S16Tables S1 to S5References (64ndash82)Data S1 to S7

25 June 2017 accepted 23 February 2018101126scienceaao1729

Kuzmin et al Science 360 eaao1729 (2018) 20 April 2018 9 of 9

RESEARCH | RESEARCH ARTICLEon F

ebruary 17 2021

httpsciencesciencemagorg

Dow

nloaded from

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from

Page 11: YEAST GENOMICS trigenic interaction networks Systematic ...enic Avg. digenic interaction degree Mapping digenic and corresponding trigenic interaction networks Digenic and trigenic

Systematic analysis of complex genetic interactions

Brenda J Andrews Charles Boone and Chad L MyersZhu Nydia Van Dyk Sara Sharifpoor Michael Costanzo Robbie Loewith Amy Caudy Daniel Bolnick Grant W Brown Yang Wang Julia Hanchard Margot Riggi Kaicong Xu Hamed Heydari Bryan-Joseph San Luis Ermira Shuteriqi HongweiMojca Mattiazzi Usaj Jolanda van Leeuwen Elizabeth N Koch Carles Pons Andrius J Dagilis Michael Pryszlak Jason Zi Elena Kuzmin Benjamin VanderSluis Wen Wang Guihong Tan Raamesh Deshpande Yiqun Chen Matej Usaj Attila Balint

DOI 101126scienceaao1729 (6386) eaao1729360Science

this issue p eaao1729 see also p 269Scienceinteractions were more likely to bridge biological processes in the celldigenic interaction landscape Although the overall effects were weaker for trigenic than for digenic interactions trigenictrigenic associations functioned within the same biological processes These converged on networks identified in the

withsynthetic genetic array (see the Perspective by Walhout) Triple-mutant analyses indicated that the majority of genes created a trigenic landscape of yeast by using aet alBuilding on the digenic protein-protein interaction network Kuzmin

To dissect the genotype-phenotype landscape of a cell it is necessary to understand interactions between genesTrigenic interactions in yeast link bioprocesses

ARTICLE TOOLS httpsciencesciencemagorgcontent3606386eaao1729

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl201804183606386eaao1729DC1

CONTENTRELATED httpsciencesciencemagorgcontentsci3606386269full

REFERENCES

httpsciencesciencemagorgcontent3606386eaao1729BIBLThis article cites 79 articles 40 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2018 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on February 17 2021

httpsciencesciencem

agorgD

ownloaded from