INVESTIGATION

Genetic Architecture of Ethanol-Responsive Transcriptome Variation in Saccharomyces cerevisiae Strains

Jeffrey A. Lewis,*,†,‡ Aimee T. Broman,§ Jessica Will,† and Audrey P. Gasch†,‡,1 *Department of Biological Sciences, University of Arkansas, Fayetteville, Arkansas 72701, and †Laboratory of Genetics, ‡Great Lakes Bioenergy Research Center, and §Department of Biostatistics and Medical Informatics, University of Wisconsin, Madison, Wisconsin 53704

ABSTRACT Natural variation in gene expression is pervasive within and between species, and it likely explains a significant fraction of phenotypic variation between individuals. Phenotypic variation in acute systemic responses can also be leveraged to reveal physiological differences in how individuals perceive and respond to environmental perturbations. We previously found extensive variation in the transcriptomic response to acute ethanol exposure in two wild isolates and a common laboratory strain of Saccharo- myces cerevisiae. Many expression differences persisted across several modules of coregulated genes, implicating trans-acting systemic differences in ethanol sensing and/or response. Here, we conducted expression QTL mapping of the ethanol response in two strain crosses to identify the genetic basis for these differences. To understand systemic differences, we focused on “hotspot” loci that affect many transcripts in trans. Candidate causal regulators contained within hotspots implicate upstream regulators as well as downstream effectors of the ethanol response. Overlap in hotspot targets revealed additive genetic effects of trans-acting loci as well as “epi- hotspots,” in which epistatic interactions between two loci affected the same suites of downstream targets. One epi-hotspot impli- cated interactions between Mkt1p and proteins linked to translational regulation, prompting us to show that Mkt1p localizes to P bodies upon ethanol stress in a strain-specific manner. Our results provide a glimpse into the genetic architecture underlying natural variation in a stress response and present new details on how yeast respond to ethanol stress.

ATURAL variation in gene expression is hypothesized to Expression quantitative trait loci (eQTL) mapping (reviewed Nbe a major source of phenotypic variation between indi- in Gilad et al. 2008) is a powerful approach to dissect the viduals (King and Wilson 1975; Oleksiak et al. 2002). Gene genetic basis of expression differences. Transcript abundance expression variation underlies differences in susceptibility to is treated as a quantitative trait whose genetic determinants infectious disease (Li et al. 2010; Barreiro et al. 2012), drug can be implicated by well-established linkage mapping tech- sensitivity (Fay et al. 2004; Kvitek et al. 2008; Maranville niques. The first eQTL studies in yeast illuminated the genetic et al. 2011; Hodgins-Davis et al. 2012; Chang et al. 2013), landscape of gene expression-variation determinants under inflammation (Gargalovic et al. 2006; Orozco et al. 2012), standard growth conditions (Brem et al. 2002; Yvert et al. cardiovascular disease (Romanoski et al. 2010), metabolism 2003). Subsequent eQTL studies surveyed the genetic control (Fraser et al. 2010; Rossouw et al. 2012; Skelly et al. 2013), of transcript abundance in Arabidopsis (Keurentjes et al. morphology (Yvert et al. 2003; Chin et al. 2012; Skelly et al. 2007; West et al. 2007), maize (Schadt et al. 2003), mice 2013), and even behavior (Ziebarth et al. 2012). Still, iden- (Schadt et al. 2003), worms (Li et al. 2006), and humans tifying the genetic and molecular mechanisms that underlie (Cowles et al. 2002; Schadt et al. 2003). Collectively, these the expression variation is a major challenge. studies have catalogued the number of variable traits caused by cis-polymorphisms that occur locally on the affected DNA Copyright © 2014 by the Genetics Society of America molecule (e.g., promoter sequence variation) vs. those affected doi: 10.1534/genetics.114.167429 Manuscript received March 31, 2014; accepted for publication June 23, 2014; by distant polymorphisms that act in trans to the affected gene published Early Online June 26, 2014. (e.g., upstream physiological or regulatory differences that can Supporting information is available online at http://www.genetics.org/lookup/suppl/ doi:10.1534/genetics.114.167429/-/DC1. affect many target genes). Many studies have detected a larger 1Corresponding author: 425 Henry Mall, Madison, WI 53706. E-mail: [email protected] fraction of locally acting polymorphisms than trans effects

Genetics, Vol. 198, 369–382 September 2014 369 Figure 1 Extensive variation in the eth- anol response between S288c, M22, and YPS163. A total of 3515 genes with differential ethanol responses in any strain relative to the average (FDR = 0.01) were organized by model-based clustering into 46 distinct clusters (see Materials and Methods). The left portion of the heat map displays expression changes in response to ethanol across six biological replicates for S288c-derived strain DBY8268 (S), M22 (M), and YPS163 (Y). Differences in etha- nol response in each wild strain vs. the lab strain are shown in the right portion of the figure. Each row represents a gene, and each column represents a single microarray experiment or strain comparison. Red indicates induced and green indicates repressed expression in response to eth- anol, while yellow indicates higher expres- sion and blue indicates lower expression for M22 or YPS163 vs. S288c, according to the key. Enriched functional groups (Bonferroni corrected P , 0.01) are an- notated to the right. ESR, environmental stress response; RiBi, ribosome biogenesis genes.

(Gilad et al. 2008) (arguably due to the higher statistical power et al. (2013) examined the response of mouse dendritic cells for their detection; Rockman and Kruglyak 2006; Kliebenstein to multiple distinct stimuli, leading to the discovery of eQTL 2009), and a notable number of genes are affected by both that were either common to many stimuli or specific to par- local and distant effectors (Brem et al. 2002; Smith and ticular conditions. Leveraging “common” and “specific” Kruglyak 2008). eQTL allowed the authors to both refine known signaling The effects of genetic variation often differ depending on pathways and pinpoint the true causative loci for eQTL. environmental conditions, and several studies have charac- Here we sought to identify the genetic determinants of terized the effects of gene–environment interaction (GEI) on variation in the ethanol-stress response in several strains of expression variation when cells are grown continuously in Saccharomyces cerevisiae. A prior study performed eQTL map- different environments (Smith and Kruglyak 2008; Gagneur ping of steady-state growth on ethanol as a carbon source, but et al. 2013). However, many environment-responsive tran- did not interrogate the dynamic response to ethanol stress at script changes occur transiently when cells are first shifted high concentrations (Smith and Kruglyak 2008). We previ- between conditions (Gasch et al. 2000; Gunasekera et al. ously showed extensive natural variation in the response to 2008; Marks et al. 2008; Eng et al. 2010; Stanley et al. acute ethanol stress in three yeast strains: a lab strain derived 2010). As such, the genetic determinants of environmental from the commonly used S288c, vineyard isolate M22, and responses are less well characterized by genetic mapping. To oak-soil strain YPS163 (Lewis et al. 2010). Across these strains, date, a small number of eQTL studies have examined dy- we identified thousands of gene expression differences in re- namic responses to changing conditions (Gargalovic et al. sponse to ethanol. Many of these differences persisted across 2006; Li et al. 2006; Smirnov et al. 2009; Yang et al. modules of coregulated genes, suggesting trans-acting varia- 2009; Barreiro et al. 2012; Gat-Viks et al. 2013). Gat-Viks tion in regulators, signaling proteins, and/or environmental

370 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch Table 1 eQTL mapping summary

Total transcripts Number (%) mapping Number (%) mapping Number (%) mapping mappeda to 1 locus to 2 loci to 3 or more loci Basal abundance mapping 2675 2149 (80) 440 (16) 85 (4) Ethanol abundance mapping 2499 2091(83) 350 (14) 58 (3) Ethanol response mapping 1847 1612 (87) 213 (12) 21 (1) One or more mapping regime 4055 a The mapping was performed using the combined crosses. sensing and response. For example, targets of the “general with the dTomato coding sequence from pFA6a-TEF2Pr- stress” transcription factor Msn2p, the oxidative stress factor dTomato-ADH1-NATMX4 (Breslow et al. 2008) (kindly pro- Yap1p, and the proteasome regulator Rpn4p were all affected vided by J. Weissman). Diploid hybrid strains for reciprocal coordinately across the strains. At least in the case of Rpn4p hemizygosity were generated by mating YPS163-1 hoD::HygMX targets, the genetic determinant lies outside of the RPN4 gene, mkt1D::KanMX to BY4741 MKT1-GFP, or YPS163-1 hoD:: since swapping RPN4 alleles across strains did not affect ex- HygMX MKT1-GFP to BY4741 mkt1D::KanMX. Images of cells pression variation in downstream targets (J. Lewis and A. expressing Mkt1p-GFP and/or Edc3p-dTomato were taken with Gasch, unpublished data). a Carl Zeiss AxioVision microscope. To elucidate the genetic basis for these expression For all growth experiments, cells were grown in batch differences, we here conducted an eQTL analysis of 100 culture in YPD (1% yeast extract, 2% peptone, 2% glucose) F2 progeny of S288c crossed to M22 or to YPS163. Although at 30°. For ethanol shock experiments, cells were grown fi “ local linkages were prevalent, we identi ed scores of hot- four generations to an optical density (OD600) of 0.3. A spot” loci that affected tens to thousands of transcripts in sample of unstressed cells was collected, and then ethanol trans. Several of these hotspots included transcription fac- was added to a final concentration of 5% (v/v) for 30 min, tors and upstream regulators, pointing to their role in the which comprises the peak expression response for each ethanol response. Overlap in hotspot transcripts pointed to strain (Lewis et al. 2010). additive interactions between implicated regulators as well as “epi-hotspots” of interacting determinants. Together, our Microarray hybridization and analysis results shed new light on the ethanol signaling system and Six replicate cultures of each parental strain and duplicate increase our understanding of the genetic complexity of the cultures for each QTL mapping strain were used to collect yeast ethanol response. unstressed and ethanol-stressed samples. Cell collection, RNA isolation, and cDNA labeling were performed as described in Materials and Methods Lewis and Gasch (2012). cDNA was labeled with amino-allyl dUTP (Ambion) and cyanine dyes (Flownamics). Inverse dye Strains and growth conditions labeling was used in replicates to control for dye-specific Strains are listed in Supporting Information, File S1.Thepa- effects. The labeled cDNA sample from each ethanol-treated rental strains for QTL mapping were YPS163, M22, and culture was mixed with the corresponding sample from the S288c-derived DBY8268 (hereafter referred to as S288c). unstressed culture and hybridized to custom Nimblegen tiled The QTL mapping strains from the S288c 3 M22 and arrays designed against the S288c genome (previously vali- S288c 3 YPS163 crosses are described in Kim and Fay dated for gene expression analysis in Lee et al. (2011) and (2007) (kindly provided by J. Fay). Deletions in the BY4741 Huebert et al. (2012), spanning both strands of the yeast ’ background were obtained from Open Biosystems and veri- genome, according to the manufacturer s instructions (Roche- fied by diagnostic PCR. MKT1 was deleted from a haploid Nimblegen). To eliminate hybridization differences across derivative of YPS163 (YPS163-1 hoD::HygMX; Lewis et al. strains, we removed from analysis all probes with sequence 2010) by homologous recombination with the mkt1::KanMX variation between S288c, M22, and YPS163 (Doniger et al. cassette amplified from the yeast knockout strain (Winzeler 2008). Arrays were scanned and analyzed with a Gene- et al. 1999) and subsequently verified by diagnostic PCR. Pix4000 scanner (Molecular Devices, Sunnyvale, CA), and C-terminal Mkt1p-GFP and Edc3p-dTomato fusions were the background-subtracted signal from both channels was genomically integrated into the MKT1 or EDC3 locus of both extracted with the program NimbleScan. Data normalization the BY4741 and YPS163-1 hoD::HygMX backgrounds (verified was performed using quantile normalization of the pooled via diagnostic PCR and sequencing). For GFP fusions, pFA6a- array data as in Wohlbach et al. (2011). Expression differ- GFP(S65T)-NatMX6 was used as a PCR template (Van Driessche ences in response to ethanol were taken as the log2 ratio of et al. 2005). To generate dTomato fusions, we used restriction- the red/green signal from the arrays. To compare absolute free (RF) cloning (van den Ent and Lowe 2006) to construct transcript abundance for unstressed or ethanol-treated cells, a new template vector (pFA6a-dTomato-KanMX6) by replacing the signal from the individual channels was extracted, com- the 13myc tag in pFA6a-13myc-KanMX6 (Longtine et al. 1998) pared, and quantile normalized as described above. File S2,

eQTL Mapping of the Yeast Ethanol Response 371 File S3, and File S4 contain the normalized expression data. All microarray data are available through the National Insti- tutes of Health Gene Expression Omnibus (GEO) database under accession no. GSE54196. Genes with significant differences in ethanol responsive expression in the parental strains (compared to the mean expression across all strains) were identified using the BioConductor package Limma v 2.10.2 (Smyth 2004) and q-value correction (Storey and Tibshirani 2003) taking P , 0.01 as significant (see File S5 for Limma output). Gene expression differences were organized by model-based mul- tinomial clustering, using the VII model in mclust and the Bayesian information criterion (BIC) (Fraley and Raftery 2002). The cluster centroids were organized by hierarchical clustering, and highly similar clusters were manually col- lapsed into 46 distinct clusters. Enrichment of gene ontology (GO) functional categories was performed using GO terms (5% false discovery rate, FDR) (Boyle et al. 2004) or com- piled lists of transcription factor and RNA binding proteins (1% FDR) taken from Harbison et al. (2004); MacIsaac et al. (2006); Teixeira et al. (2006); Monteiro et al. (2008); and Abdulrehman et al. (2011). Figure 2 Local and distant mapping of transcripts. The genetic position of ethanol-responsive eQTL (x-axis) is plotted against the genetic position eQTL mapping of the transcript trait (y-axis). The greyscale heat map denotes the LOD – scores of local eQTL (diagonal), and the colored heat map indicates LOD Single mapping scans were performed using Haley Knott re- scores of distant eQTL (see Materials and Methods), where the color is gression (Haley and Knott 1992) implemented through the proportionate to LOD score as shown in the key. R/QTL software package (Broman et al. 2003). Single-cross scans were performed for the S288c-M22 and S288c-YPS163 crosses separately in three different “regimes”:byscoring S288c-YPS163 cross. All significant linkages are reported in absolute transcript abundance in unstressed (“basal”)or File S6. Linkages were considered local (cis) if they mapped ethanol-stressed cells, and by mapping the log2 (fold-change) within 620 cM of the ORF of the trait. This threshold was in abundance upon ethanol treatment (“ethanol response”). chosen to increase our ability to detect local linkages that fall Because both crosses involve the S288c strain, we also per- directly in-between markers, which were spaced 30 cM formed a combined cross across both mapping sets, for each apart. mapping regime. For markers where S288c differed from We identified hotspots as loci that affect many different both M22 and YPS163, we set the M22 and YPS163 markers traits. To infer the hotspot architecture, we counted the to be equivalent, and we included “cross” as an interactive number of traits mapping significantly to each locus along covariate. In addition, “ploidy” was included as an additive the chromosome using a sliding 5-cM window. We defined covariate, since the segregants are either haploid or diploid the hotspot “peak” as the position with the maximum num- strains depending on whether they inherited the wild or lab ber of traits mapped to it. Traits were assigned to a particular allele of the HO endonuclease (conferring mating type hotspot if their 1.5 LOD support interval fell within 10 cM switching or not, respectively). Significant LOD scores (5% of that hotspot’speak.Weidentified significant eQTL hot- FDR) were determined by estimating the distribution of spots affecting at least 20 traits in one condition using the genome-wide maximum LOD scores under the global null quantile-based permutation (NL)methodofNetoet al. (2012), hypothesis of no segregating QTL. This was performed via implemented through R/qtlhot. One thousand permutations 100,000 simulations as follows. First, phenotypes for the seg- were performed for each mapping regime, and hotspots regants were randomly simulated using the rnorm function in passing a 5% FDR cutoff were called significant. To over- the R stats package v. 2.15.2 (vector of means set to 0, vector come differences in statistical power for each mapping re- of standard deviations set to 1). Single mapping scans were gime, transcripts that mapped to the same hotspot (within then performed on each set of simulated phenotypes using 20 cM) in one or more mapping regime were pooled to the same parameters as above, including appropriate covari- identify 37 unique hotspots. Cross-specific hotspots were ates. This generated a simulated distribution of 100,000 max- those identified in only one of the two single-cross map- imum LOD scores. The maximum LOD scores were sorted, pings. Enrichment of gene functional groups or known regu- and the 95th percentile was used to set the global FDR at 5%. lators was performed using GO terms (5% FDR) (Boyle et al. This corresponded to a LOD = 3.96 for the combined cross, 2004) or compiled lists of transcription factor and RNA bind- LOD = 3.29 for the S288c-M22 cross, and LOD = 3.27 for the ing proteins (1% FDR) taken from Harbison et al. (2004);

372 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch Table 2 Local vs. distant linkages

% mapping to both % local linkagesa % distant linkages local and distant loci Total linkages Basal abundance mapping 29.7 70.2 287 (8.7%) 3296 Ethanol abundance mapping 28.4 71.6 220 (7.4%) 2975 Ethanol response mapping 24.5 75.5 113 (5.3%) 2108 a The mapping was performed using the combined crosses.

MacIsaac et al. (2006); Teixeira et al. (2006); Monteiro et al. related to glycogen (P =23 1025) and trehalose metabolism (2008); and Abdulrehman et al. (2011). (P =53 1027), asparagine metabolism (P =13 1025), and Interactions between hotspots and other loci were identi- the responses to heat (P =83 1027) and oxidative stress fied using the method described in Broman and Sen (2009). (P = 0.001). Several other clusters represented strain-specific For a given hotspot, mapping of the combined crosses was responses between the oak and vineyard strains, suggesting reperformed by holding the marker nearest to that hotspot natural variation in the ethanol response across wild isolates. as an interactive covariate (LODf) or as an additive cova- M22-specific clusters were enriched for genes involved in telo- riate (LODa). The interaction term (LODi)wascalculated mere maintenance (0.001), cytoplasmic translation (2 3 2 fi 2190 210 as LODf LODa. Traits were considered to have a signi - 10 ), and translation initiation (3 3 10 ). YPS163-specific 213 cant hotspot interaction if both LODf and LODi were signif- clusters were enriched for proteolysis (1 3 10 ), regulation icant at a 5% FDR (calculated via 100,000 simulations as of translation (2 3 1028), cellular localization (1 3 1024), and described above). ribosome biogenesis (1 3 10234). Enrichment analyses were done using the hypergeomet- eQTL mapping of the ethanol response ric distribution, calculating the probability of observing the number of observed occurrences or more from the appro- The striking differences in the ethanol-dependent expression priate background set of genes. Transcripts mapping to the modules across strains suggested key differences in ethanol MKT1 locus were clustered using Mclust as described above sensing and/or signaling. To explore the genetic basis for to identify 26 discrete clusters. Enrichment of regulator tar- these differences, we measured changes in expression at gets (including kinases, phosphatases, transcription factors, 30 min after 5% ethanol treatment in 45 F2 strains from two and RNA binding proteins) was performed using targets de- crosses: S288c 3 M22 and S288c 3 YPS163, in biological fined by D. Berry, Y. Ho, M. MacGilvray, A. Gasch, unpub- duplicate. We performed eQTL mapping in three separate lished data. regimes for each cross: by comparing differences in absolute transcript abundance before (basal) and after ethanol treat- Results ment, and by scoring the fold change in transcript abundance upon ethanol treatment, which we refer to as the ethanol Extensive natural variation in ethanol-responsive response (see Materials and Methods). Mapping the ethanol transcriptional modules response inherently captures GEI interactions, because it To confirm the extent of natural variation in the parental compares expression across strains and across environments. strains responding to ethanol stress, we first analyzed Because of the prominent differences in the common S288c changes to the transcriptome in the three diploid strains strain, we also performed a “combined-cross” analysis by (S288c-derived DBY8268, M22, and YPS163) responding to pooling cross data together and using the cross identifier 5% ethanol for 30 min, in six biological replicates. Over half as a covariate in the mapping (see Materials and Methods). of the genome (3287/6532 genes) exhibited differential This combined strategy allowed us to identify cross-specific ethanol-responsive expression (FDR , 0.01) in at least one eQTL, implicating effects that were specific to M22, YPS163, strain (File S5). This fraction is significantly higher than that or S288c. of our previous study (Lewis et al. 2010), likely a reflection Our initial mapping revealed that 1200 transcripts of increased statistical power from additional replicates and mapped to the HO locus, strongly suggesting that differen- improved array technology (Brem et al. 2002; Smith and ces in ploidy have a significant effect on the ethanol re- Kruglyak 2008). sponse (Figure S1). The F2 strains resulting from the cross We used model-based clustering (see Materials and Meth- are either diploid or haploid, depending on whether they ods) to categorize the gene expression differences into 46 inherited a functional HO allele from the wild parent that distinct clusters (Figure 1). Most of these clusters were allows mating-type switching and subsequent selfing to form enriched for functionally related genes, as well as known tar- diploids (Taxis et al. 2005). The ploidy-affected transcripts gets of various transcription factors or RNA-binding proteins were strongly enriched for a variety of functions, including (File S7). This strongly suggests the prevalence of trans-acting cytoplasmic translation (P =43 10288), ribosome biogen- effectsongeneexpressionvariation.Muchoftheexpression esis (P =53 10224), nitrogen compound metabolism (1 3 variation was unique to the lab strain, including gene clusters 10222), glucose metabolism (1 3 1029), and response to

eQTL Mapping of the Yeast Ethanol Response 373 Table 3 eQTL hotspots

Peak Mappinga Crossb Total linkages Enrichment summary Chr. Candidate gene 1 B, E Both 53 Fatty acid metabolism I OAF1 (Litvin et al. 2009) 2 B, E, R Both 98 Cell division II AMN1 (Yvert et al. 2003) 3 B, E YS 85 4 B, E, R Both 37 Mating, HMR III MAT-ALPHA1 5 B MS 28 Electron transport, iron-sulfur cluster, metal binding IV RGT2 6 B, E MS 32 Y9 elements IV 7 B Both 81 Fatty acid oxidation, cell wall V URA3 8 B, E, R YS 285 iESR V TSC11/DOT6 9 B, E MS 124 VII VPS73 10 B, R MS 74 Methionine/sulfate metabolism VII CYS4 (Kim and Fay 2007) 11 B, E, R YS 126 Mitochondrial RPs VII CIR1 12 B, E, R Both 152 VIII 13 B, E, R Both 376 Sterol biosynthesis, respiration XII HAP1 (Brem et al. 2002) 14 B, E YS 47 Dubious ORFs/Y9 elements XII 15 B, E, R Both 152 Vacuole, arginine metabolism, phosphate metabolism XIII PHO84 (Perlstein et al. 2007) 16 B MS 24 XIII PSE1 17 B, E, R Both 1264 Mitochondrion, mitochondrial RPs XIV MKT1 (Lee et al. 2009) 18 B, R Both 47 XV 19 B Both 35 XV 20 B MS 28 XV 21 B YS 33 Phosphorous metabolism XVI FUM1 22 E YS 26 IV NDE2 23 E YS 35 IV 24 E Both 33 mRNA processing VII RNA15 25 E YS 30 X 26 E, R MS 60 iESR, trehalose XII PHD1 27 E MS 35 vacuole XIII 28 E MS 25 XIV 29 E MS 36 XV 30 E Both 25 XV PRE6 31 E, R Both 79 V FAA2 32 R Both 48 Ribi cluster, nucleolus VIII SBP1, RPL8A 33 R YS 20 X 34 R MS 22 IV 35 R MS 39 Ribosome biogenesis VII 36 E Both 27 III 37 E YS 28 II iESR, induced ESR genes; rESR, repressed ESR genes; RiBi: ribosome biogenesis genes in the ESR. References are provided for candidate genes observed in other QTL studies. a Mapping regime: basal abundance (B), ethanol abundance (E), and ethanol response (R). b Cross where hotspot was observed: S288c 3 M22 (MS), S288c 3 YPS163 (YS), or both. pheromone (3 3 1027). To better detect eQTL that repre- regulated transcript’s open reading frame) vs. other distant sent natural variation across strain lineages, we therefore loci (Figure 2). Roughly 25–30% of mapped transcripts focused our attention on mapping regimes that included (depending on mapping regime) were linked locally to their ploidy as a covariate to mitigate its effect. own locus, suggesting possible cis-variation at those loci (Ta- Our analysis identified 4055 transcripts that could be ble 2). Depending on the mapping regime, 5–10% of tran- mapped significantly to at least one locus, in one or more scripts mapped both locally and distantly, and these mapping regimes (Table 1, Figure S1, and Figure S2). Eighty transcripts made up half of all of those that mapped to more to 87% of transcripts (depending on the mapping regime) than one locus. Cis-acting eQTL are reported to have larger mapped to a single locus, while 12–16% and 1–3% of eQTL effect sizes, making them easier to detect (Petretto et al. mapped to two or more than two loci, respectively. Mapping 2006). Consistently, we observed that cis-andtrans-acting the fold change in transcript abundance (ethanol response) eQTL had mean LOD scores of 8.7 and 5.6, respectively, a dif- identified significantly fewer eQTL linkages overall and ference that was highly significant (P =23 102118, t-test). pointed to a greater number of transcripts mapping to only eQTL hotspots point to known and novel regulators a single locus, compared to other mapping regimes. This may reflect reduced statistical power when comparing fold The image in Figure 2 revealed that a significant number of changes in abundance across strains. transcripts map distantly to the same eQTL loci, implicating We also looked at the proportion of mapped transcripts the existence of hotspot loci affecting many traits. We there- that were linked to their own locus (within 620 cM from the fore scanned the genome for hotspots that were linked to at

374 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch S288c cross (Figure 3). Overall, 24 of the 37 hotspots were significantly enriched for functionally related genes and/or known targets of transcription factors or RNA-binding pro- teins (Table 3 and File S8). Candidate causal genes were identified by manually interrogating genes encompassed by each hotspot; several candidates represented known polymorphisms while others implicated novel regulators. For example, the large hotspot on chromosome XII captures a known S288c-specific poly- morphism in the transcription factor HAP1 (Gaisne et al. 1999; Brem et al. 2002); accordingly, transcripts associated with this hotspot include Hap1 targets and capture Hap1p- regulated processes, including ergosterol metabolism (3 3 10221) and cellular respiration (1 3 10210). The S288c 3 M22-specific hotspot on chromosome VII points to a known polymorphism in the cystathionine beta-synthase CYS4 (Kim and Fay 2007; Kim et al. 2009), with targets enriched for genes involved in methionine (6 3 10218) and cysteine (6 3 10210) biosynthesis. The largest hotspot from all three map- ping regimes was found on chromosome XIV, spanning a known causal polymorphism in the MKT1 gene (Deutschbauer and Davis 2005; Smith and Kruglyak 2008; Zhu et al. 2008). Natural variation in MKT1 has been implicated in a large num- ber of phenotypic differences, ranging from mitochondrial stability (Nickoloff et al. 1986), drug resistance (Kim and Fay 2009), thermotolerance (Steinmetz et al. 2002; Sinha et al. 2006), sporulation (Deutschbauer and Davis 2005), and DNA-damage sensitivity (Ehrenreich et al. 2010). Impor- tantly, natural variation in MKT1 has been linked to high ethanol tolerance in a Brazilian bioethanol production strain compared to a lab strain (Swinnen et al. 2012). Our observation that MKT1 affects a large number of transcripts responding Figure 3 Cross-specific and condition-specific hotspots. The average to ethanol suggests a potential link between Mkt1p-affected number of transcripts (using a sliding 5-cM window) mapping to each transcriptional responses and ethanol defense (see below locus across the 16 yeast chromosomes (x-axis) is plotted for each of the and Discussion). three mapping regimes, for the S288c 3 YPS163 cross (red) and the Several other hotspots point to previously unknown loci 3 S288c M22 cross (blue). that contain known regulators. For example, the enrichment for genes involved in respiration and in iron–sulfur metab- least 20 transcripts in any of the mapping regimes. Because olism (which occurs in mitochondria and is required for spurious hotspots may arise due to nongenetic correlations respiration) among transcripts that link to hotspot peak 5 between expression traits (Neto et al. 2012), we used per- hints at a role for RGT2, which encodes a key membrane- mutation tests to assess hotspot significance (see Materials bound glucose sensor in yeast (Ozcan et al. 1996, 1998). and Methods). To overcome differences in statistical power, The enrichment for amino acidmetabolismtranscripts transcripts that mapped to the same hotspot region in one or among those that link to hotspot peak 8 points to TSC11,a more mapping regimes from the same cross were pooled. member of the amino-acid-responsive TOR complex (Loewith We identified 37 unique hotspots (5% FDR) that together et al. 2002); it is notable that transcripts encoding translation encompassed 2797 different transcripts (Table 3, Figure 3, machinery, which are repressed in the yeast environmental Figure S2). Only 1752 (63%) of these represented genes stress response (ESR), are also linked to this hotspot but may that were scored as differentially expressed in the parental instead be due to variation at DOT6, a gene near TSC11 that strains, suggesting that .1000 transcripts mapping to hot- is the known repressor of ESR genes (Lippman and Broach spots may be affected by transgressive segregation that 2009; Huber et al. 2011). Other hotspots enriched for tran- emerges in the F2’s. The number of transcripts mapping to scripts related to fatty acid metabolism (peak 1), mating each hotspot ranged from 20 to .1200, with a median of 37 (peak 4), and stress and trehalose metabolism (peak 26) transcripts per hotspot. Of the 37 hotspots, 15 were ob- suggest the effects of transcription factors Oaf1p,Mat- served in both crosses, whereas 12 were unique to the alpha, and Phd1p that are encoded within those hotspot loci, M22 3 S288c cross and 10 were unique to the YPS163 3 respectively.

eQTL Mapping of the Yeast Ethanol Response 375 Figure 4 Significant overlap in hotspot targets. Each node represents a hotspot, where the size is propor- tionate to the number of transcripts mapping to it. Edges between nodes represent statistically significant overlap (FDR , 0.05), where the thickness of the edge is proportionate to the fraction of the smaller node’s targets in the overlap and where the color of the edge is proportionate to the 2log(P-value) of the overlap (with black = 5). Hotspots are colored according to the cross in which they were identified, according to the key.

Interactions between hotspots indicate additive and epi-hotspot peak (File S9). None of the pairs of hotspots epistatic relationships with overlapping target sets shown in Figure 4 were identi- fi Several of the hotspots were enriched for similar functional ed as epi-hotspots, indicating that most of the overlap we categories, raising the possibility that the same genes were initially observed represents additive hotspot effects. How- linked to multiple hotspots. Indeed, we found significant ever, we uncovered several other epi-hotspots involving two overlap in the targets of a number of hotspots (Figure 4). or more regions implicated from the 1D mapping regimes: Notably the MKT1 and HAP1 hotspots were hubs of overlap, for example the interaction between peak 13 (Hap1p) and — since both their target sets overlapped with targets of five peak 1 (Oaf1p) represents an epi-hotspot the candidate fl other hotspots. In several of these cases, we observed func- regulators encoded at these loci both affect membrane uid- tional similarities between the candidate causal genes within ity through ergosterol (Kennedy et al. 1999; Jensen-Pergakes those hotspots. For example, the overlap in targets linked et al. 2001; Tamura et al. 2004) and fatty acid composition to Hap1p, Rgt2p, Fum1p,andNde2p points to respiratory (Karpichev et al. 1997; Karpichev et al. 2008), respectively. effects: Hap1p and Rgt2p both regulate genes related to Several loci were associated with multiple epi-hotspots, in- respiration and glucose metabolism, while fumarase Fum1p cludingpeak13(HAP1) linked to four epi-hotspots, and peak and NADH dehydrogenase Nde2p are critical for respira- 4(MAT-ALPHA1), peak 7 (URA3), peak 12, peak 17 (MKT1), tion. Transcripts linked to Mkt1p, which binds to the HO and peak 31 (FAA2)thatwereeachlinkedtothreeepi-hotspots. transcript to mediate translational regulation (Tadauchi Several other epi-hotspots implicated new peaks that were not et al. 2004), overlap with transcripts linked to peak 24, foundinthe1Dmappingscans. which encodes Rna15p, a protein that regulates mRNA Epistatic interactions have been extensively interrogated polyadenylation and stability (Minvielle-Sebastia et al. in gene-deletion mutants in laboratory strains (Decourty 1991; Minvielle-Sebastia et al. 1994; Gross and Moore et al. 2008; Costanzo et al. 2010). We reasoned that lab- 2001). strain deletion data could inform on the interacting alleles The overlap in hotspot targets could represent additive within epi-hotspots and found several epi-hotpots that encom- interactions if the causal loci independently modulate passed genes with genetic interactions in the lab strain. Three of transcript abundance; alternatively, the overlap may arise the 24 epi-hotspots included genes with genetic interactions due to epistatic interactions between causal genes within measured in the lab strain. Although this could arise by chance, those hotspots. To investigate this, we explicitly tested for interactions between genes in the MKT1 epi-hotpots pointed to epistatic interactions by reperforming the mapping for each functionality: MKT1 deletion in the lab strain produces epistatic regime, with each hotspot fixed as an interactive covariate, effectswhencombinedwithdeletionofSKI7 (Ridley et al. to identify eQTL with a significant interactive term (see 1984) and DIA2 (Costanzo et al. 2010), encompassed in epi- Materials and Methods). We then identified hotspots from hotspot peak 8. In contrast, Sbp1p, encoded in epi-hotspot 32, the interactive scans as epistatic hotspots, or epi-hotspots. is reported to have a physical interaction with Mkt1p in the lab To minimize spurious interactions, we focused only on epi- strain (Mitchell et al. 2013). The probability of observing this hotspots that displayed significant functional enrichment of number of genetic or physical interactions by chance, given the the associated transcripts. number of genes under the interacting hotspots, is relatively We identified 24 epi-hotspots whose targets are affected low (P = 0.047, random sampling). Ski7p plays a role in reg- by epistatic interactions between the two eQTL effectors ulating RNA turnover (Anderson and Parker 1998), while (Table 4 and Figure 5). In all, 803 unique transcripts map- Sbp1p is a component of P bodies that has a known function ped to an epi-hotspot, with 10 to 118 (25) transcripts per in translational regulation (Segal et al. 2006; Mitchell et al.

376 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch Table 4 Genetic interactions between hotspots

Peak ID Interacting hotspot Mapping regimea Total linkages Enrichment summary Chr. Candidate gene 1 Peak12 B, E 52 Ribi cluster III 2 Peak12 R 12 RPs, translation, ribosome X 3 Peak13 (HAP1) B, E 62 iESR, response to heat I OAF1 4 Peak13 (HAP1) E, R 66 RPs, translation, ribosome V TSC11/DOT6 5 Peak15 (PHO84) B, E, R 118 iESR V URA3 6 Peak15 (PHO84) R 21 RPs V 7 Peak17 (MKT1) R 22 Intracellular mRNA localization XIII 8 Peak17 (MKT1) B, E 41 Mitochondrial RPs, mitochondrial translation XV 9 Peak18 E, R 61 RPs, translation, ribosome V TSC11/DOT6 10 Peak19 B, E, R 50 mitochondrial RPs X 1D Peak25 11 Peak2 (AMN1) E 17 RPs, translation, ribosome VII CYS4 12 Peak24 (RNA15) B 24 Electron transport chain, aerobic respiration IV NDE2 13 Peak30 (PRE6) E 10 Asparagine metabolism, response to nitrogen XIV starvation 14 Peak31 (FAA2) B 22 RiBi cluster VII 15 Peak31 (FAA2) R 21 RiBi cluster VII CIR1 16 Peak31 (FAA2) E 12 iESR XI PHD1 17 Peak32 B, E, R 93 Mitochondrial RPs XIV MKT1 18 Peak36 B, E 25 Mating IV 19 Peak4 (MAT-ALPHA1) B, E, R 24 Mating IV 20 Peak4 (MAT-ALPHA1) E 34 RPs, translation, ribosome VIII 21 Peak4 (MAT-ALPHA1) B, E 48 iESR XII 22 Peak7 (URA3) B, R 55 RPs, translation, ribosome XIII 23 Peak12 B 16 Cell cycle XII HAP1 24 Peak13 (HAP1) B, E 17 Cell cycle IX iESR, induced ESR genes; rESR, repressed ESR genes; RiBi: ribosome biogenesis genes in the ESR. a Mapping regime: basal abundance (B), ethanol abundance (E), and ethanol response (R); the mapping was performed using the combined crosses (see Materials and Methods).

2013). This information, together with the previous connection abundance, likely contributing to the pleiotropic effects of Mkt1 to P-body-related genes (Lee et al. 2009) and trans- reported in other mapping studies (see Discussion). lational regulation (Tadauchi et al. 2004), suggests that Mkt1p Mkt1p localizes to P bodies upon ethanol stress in the plays a role in these processes to affect downstream transcript S288c strain only abundance (see below). Given the links between Mkt1p, translation, and P-body for- Stress-regulated transcripts map to MKT1 mation, we wondered if Mkt1p might in fact be a P-body The MKT1 hotspot represented the largest in our study, component. Indeed, we found that ethanol stress induced affecting .1000 transcripts across mapping regimes. Inter- Mkt1p-GFP colocalization with P-body marker Edc3p- estingly, transcripts mapping to MKT1 are enriched for dTomato (Decker et al. 2007)—but only in the S288c strain messages encoding protein kinases (P , 1 3 10214)and and not in YPS163 (Figure 6). Reciprocal hemizygosity includes many signaling proteins and transcriptional regu- analysis (in which hemizygous hybrid strains lack either the lators, particularly those linked to stress signaling. This S288c or YPS163 allele of MKT1) indicated that the differ- includes regulators in the HOG (SSK2 and SKO1), PKC ential localization is due to the MKT1 alleles, since only the (RHO2 and BCK1), TOR (TOR1 and SCH9), RAS/PKA hemizygote carrying the BY4741_Mkt1p-GFP allele formed (RAS2 and GPB1), MEC (MEC1 and DUN1), and cell cycle foci during ethanol stress (Figure S3). The YPS163_Mkt1p- networks (including cyclins), as well as other stress-activated GFP allele did localize to P bodies under hypoosmotic stress, transcriptional regulators (YAP1, YAP5, SKN7, PDR1, DOT6, indicating that YPS163_Mkt1p can form foci under certain and others) (File S10). Transcripts of several candidate conditions (Figure S4). We hypothesize that Mtk1p may regulators within hotspots (Table 3 and Table 4), includ- differentially affect translation of specific mRNAs in S288c ing HAP1, OAF1,andPHD1,alsomappedtotheMKT1 and the wild strains responding to ethanol stress (see locus. The effect of Mkt1 on regulator abundance could Discussion). produce secondary effects on the regulators’ targets. In- deed, we find the known targets of many of these regu- Discussion lators or pathways to be enriched among the transcripts mapping to Mkt1 (File S10). Thus, polymorphisms at the Natural variation in a gene’s expression represents a complex MKT1 gene lead to widespread effects on stress and envi- trait subject to myriad levels of genetic control, including ronmental signaling at the level of differential transcript local (cis) and distant (trans)influences that may interact

eQTL Mapping of the Yeast Ethanol Response 377 hurdle by focusing on interactions with hotspot loci from the 1D mapping scan—this revealed hotspots of additive ge- netic interaction as well as a surprising number of epi-hotspots. We conservatively identified 803 expression traits mapped to an epi-hotspot (using our stringent criteria, see Materials and Methods), indicating a lower bound of 12% of all mapped traits affected by epistatic regulation. While this fraction is similar to previous reports citing that 14–16% of traits are affected by epistasis (Brem and Kruglyak 2005; Storey et al. 2005), these values are likely a significant underestimate. Nonetheless, they demonstrate the prevalence of epistatic interactions on expres- sion control. This study also provides new insight into the genetics underlying natural variation in the yeast ethanol-stress response. A significant fraction of hotspot loci, including additive and epi-hotspots, include candidate causal regulators and transcription factors; others implicate genes that in- Figure 5 Epi-hotspots affect transcript variation. Each circle represents an fluence physiology via enzymatic or transporter functions. x y epi-hotspot, where the coordinates on the - and -axes represent the Given that we previously showed that many of the transcript interacting loci from the 1D scan and the 2D scan, respectively. The size of the circle represents the number of transcripts mapping with statistical differences affect ethanol tolerance (Lewis et al. 2010), these significance to each epi-hotspot, according to the key (see Materials and results suggest insights into ethanol-stress defense. A recur- Methods). Colors correspond to the same 1D hotspot. ring theme among affected genes links to respiration and carbon metabolism (Table 3 and Figure 4); this is consistent with a prior screen of the yeast deletion library that found additively or epistatically and in environment-specific ways that strains lacking respiratory components showed increased (Brem et al. 2002; Brem and Kruglyak 2005; Gibson and ethanol sensitivity (Yoshikawa et al. 2009). The S288c back- Weir 2005; Gilad et al. 2008; Smith and Kruglyak 2008; ground has several mutations in regulators of respiration and Zhu et al. 2008, 2012; Gagneur et al. 2013; Gat-Viks et al. carbon metabolism, including Hap1p and Mig3p (Gaisne 2013). The consequential effects on organismal phenotypes et al. 1999; Lewis and Gasch 2012); however several of the emerge from the suite of differences across all genes in the hotspots we observe are cross specific, indicating that natural genome, highlighting the importance of considering net- variation in carbon responses in wild strains also contributes work-level interactions that give rise to these differences. to differences in ethanol tolerance (Lewis et al. 2010). Hap1p Our results together illuminate the architecture of natural also controls genes involved in ergosterol biosynthesis, which variation in gene expression while implicating systemic dif- affects membrane fluidity and thus ethanol tolerance (Kennedy ferences in how yeast isolates perceive and/or respond to et al. 1999; Jensen-Pergakes et al. 2001; Tamura et al. 2004). ethanol stress. Likewise, variation in fatty acid consumption, controlled in part Similar to recent eQTL studies in yeast (Brem and Kruglyak by the Oaf1 transcription factor, can alter the membrane 2005; Smith and Kruglyak 2008), we identified thousands content of cells to directly affect ethanol resistance (Chi and of significant linkages. While our study had a smaller sample Arneborg 1999; You et al. 2003; Lockshon et al. 2007). The size than that of Smith and Kruglyak (2008) (89 segregants OAF1 and HAP1 loci represent an epi-hotspot and could to- in our combined cross vs. 109 segregants in their single gether impart synergistic effects on ethanol tolerance. Such cross), and thus captured fewer eQTL for absolute transcript genetic interactions in natural isolates could provide a novel abundance, we captured more GEI eQTL in our study (2018 route to engineering increased ethanol tolerance in industrial vs. 1555). One possibility is that the larger expression changes ethanologenic strains. triggered by acute ethanol stress, compared to continuous By far the hotspot with the greatest influence on the yeast growth after cells have acclimated, boost power to find transcriptome is the MKT1 locus. Natural variation in MKT1 strain-specific differences in expression. In addition, transient has been implicated in a large number of phenotypic differ- GEI exposed during acute stress responses may also uncover ences between yeast strains (Nickoloff et al. 1986; Steinmetz more GEI affects compared to continuous-growth experiments et al. 2002; Deutschbauer and Davis 2005; Sinha et al. 2006; (Eng et al. 2010). Zhu et al. 2008; Kim and Fay 2009; Ehrenreich et al. 2010), Of the thousands of transcripts mapped in this study, including tolerance to ethanol (Swinnen et al. 2012) and three-quarters were linked to one or more hotspot loci that other stresses (Steinmetz et al. 2002; Demogines et al. each explained tens to .1000 traits. A lingering question of 2008), though the mechanism of its effect remains obscure. debate is the role of epistatic interactions between QTL, since Our results strongly suggest that the pleiotropic and condi- detecting epistatic interactions remains a statistical challenge tion-specific effects of Mkt1p are linked to widespread dif- (Kapur et al. 2011; Huang et al. 2013). We overcame this ferences in transcriptional regulation and stress-activated

378 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch Figure 6 Mkt1p-GFP from S288c, but not YPS163, localizes to P bodies in response to ethanol stress. Lo- calization of Mkt1p-GFP and Edc3p-dTomato was ob- served via fluorescence microscopy before and after 30-min treatment with 5% ethanol.

signaling. The MKT1 linkage of transcripts encoding regula- Biosciences Institute (Arkansas Settlement Proceeds Act tors, and in turn the linkage of many of their targets, sug- of 2000); and by the Clinical and Translational Science gests that Mkt1p affects the abundance, and thus activity, of Award program, through the National Institutes of Health signaling proteins. It is notable that MKT1-linked traits in National Center for Advancing Translational Sciences, our study are enriched for transcripts required for respira- grant UL1TR000427. tion (P =93 1010) and heat survival (P =23 1026), perhaps explaining prior mapping studies linking MKT1 to mitochondrial genome stability and thermotolerance (Steinmetz Literature Cited et al. 2002; Dimitrov et al. 2009). We find that the S288c allele of Mkt1p colocalizes to P bodies upon ethanol stress. Abdulrehman, D., P. T. Monteiro, M. C. Teixeira, N. P. Mira, A. B. Given its role in translational regulation of the HO transcript Lourenco et al., 2011 YEASTRACT: providing a programmatic access to curated transcriptional regulatory associations in Sac- (Tadauchi et al. 2004), our results present a model in which charomyces cerevisiae through a web services interface. Nucleic S288c_Mkt1p translationally silences specificmRNAsvia Acids Res. 39: D136–D140. P-body association during ethanol stress. The encoded regu- Anderson, J. S., and R. P. Parker, 1998 The 39 to 59 degradation of lators may affect the abundance of their own transcripts yeast mRNAs is a general mechanism for mRNA turnover that 9 9 through feedback regulation, which is common in stress- requires the SKI2 DEVH box protein and 3 to 5 exonucleases of the exosome complex. EMBO J. 17: 1497–1506. regulated signaling. Barreiro, L. B., L. Tailleux, A. A. Pai, B. Gicquel, J. C. Marioni et al., Gene expression variation is hypothesized to play a large 2012 Deciphering the genetic architecture of variation in the role in phenotypic variation at the organismal level. How- immune response to Mycobacterium tuberculosis infection. Proc. ever, identifying the variable transcripts that cause differ- Natl. Acad. Sci. USA 109: 1204–1209. ences in organismal fitness is challenging, given the hundreds Boyle, E. I., S. Weng, J. Gollub, H. Jin, D. Botstein et al., 2004 GO: TermFinder–open source software for accessing Gene Ontology to thousands of gene expression differences across individu- information and finding significantly enriched Gene Ontology als. Future studies to couple variation in F2 ethanol tolerance terms associated with a list of genes. Bioinformatics 20: 3710– andinF2 transcript regulation will further implicate transcript 3715. differences that underlie stress tolerance. Brem, R. B., and L. Kruglyak, 2005 The landscape of genetic com- plexity across 5,700 gene expression traits in yeast. Proc. Natl. Acad. Sci. USA 102: 1572–1577. Acknowledgments Brem, R. B., G. Yvert, R. Clinton, and L. Kruglyak, 2002 Genetic dissection of transcriptional regulation in budding yeast. Sci- We thank Justin Fay for providing strain panels and Karl ence 296: 752–755. Broman, Brian Yandell, Shuyun Ye, and Christina Kendziorski Breslow, D. K., D. M. Cameron, S. R. Collins, M. Schuldiner, J. for helpful conversations. This work was supported in part by Stewart-Ornstein et al., 2008 A comprehensive strategy en- abling high-resolution functional analysis of the yeast genome. the Department of Energy Great Lakes Bioenergy Research Nat. Methods 5: 711–718. Center (Office of Science DE-FC02-07ER64494); startup Broman, K. W., and S. Sen, 2009 A Guide to QTL Mapping with R/qtl, funds provided by the University of Arkansas; the Arkansas Springer-Verlag, New York.

eQTL Mapping of the Yeast Ethanol Response 379 Broman, K. W., H. Wu, S. Sen, and G. A. Churchill, 2003 R/qtl: on variations of human endothelial cell responses to oxidized QTL mapping in experimental crosses. Bioinformatics 19: 889– lipids. Proc. Natl. Acad. Sci. USA 103: 12741–12746. 890. Gasch, A. P., P. T. Spellman, C. M. Kao, O. Carmel-Harel, M. B. Chang, J., Y. Zhou, X. Hu, L. Lam, C. Henry et al., 2013 The Eisen et al., 2000 Genomic expression programs in the re- molecular mechanism of a cis-regulatory adaptation in yeast. sponse of yeast cells to environmental changes. Mol. Biol. Cell PLoS Genet. 9: e1003813. 11: 4241–4257. Chi, Z., and N. Arneborg, 1999 Relationship between lipid com- Gat-Viks,I.,N.Chevrier,R.Wilentzik,T.Eisenhaure,R.Raychowdhury position, frequency of ethanol-induced respiratory deficient mu- et al., 2013 Deciphering molecular circuits from genetic var- tants, and ethanol tolerance in Saccharomyces cerevisiae. J. Appl. iation underlying transcriptional responsiveness to stimuli. Nat. Microbiol. 86: 1047–1052. Biotechnol. 31: 342–349. Chin, B. L., O. Ryan, F. Lewitter, C. Boone, and G. R. Fink, Gibson, G., and B. Weir, 2005 The quantitative genetics of tran- 2012 Genetic variation in Saccharomyces cerevisiae: circuit di- scription. Trends Genet. 21: 616–623. versification in a signal transduction network. Genetics 192: Gilad, Y., S. A. Rifkin, and J. K. Pritchard, 2008 Revealing the 1523–1532. architecture of gene regulation: the promise of eQTL studies. Costanzo, M., A. Baryshnikova, J. Bellay, Y. Kim, E. D. Spear et al., Trends Genet. 24: 408–415. 2010 The genetic landscape of a cell. Science 327: 425–431. Gross, S., and C. L. Moore, 2001 Rna15 interaction with the A- Cowles, C. R., J. N. Hirschhorn, D. Altshuler, and E. S. Lander, rich yeast polyadenylation signal is an essential step in mRNA 2002 Detection of regulatory variation in mouse genes. Nat. 39-end formation. Mol. Cell. Biol. 21: 8045–8055. Genet. 32: 432–437. Gunasekera, T. S., L. N. Csonka, and O. Paliy, 2008 Genome-wide Decker, C. J., D. Teixeira, and R. Parker, 2007 Edc3p and a gluta- transcriptional responses of Escherichia coli K-12 to continuous mine/asparagine-rich domain of Lsm4p function in processing osmotic and heat stresses. J. Bacteriol. 190: 3712–3720. body assembly in Saccharomyces cerevisiae. J. Cell Biol. 179: Haley, C. S., and S. A. Knott, 1992 A simple regression method for 437–449. mapping quantitative trait loci in line crosses using flanking Decourty, L., C. Saveanu, K. Zemam, F. Hantraye, E. Frachon et al., markers. Heredity (Edinb) 69: 315–324. 2008 Linking functionally related genes by sensitive and quan- Harbison, C. T., D. B. Gordon, T. I. Lee, N. J. Rinaldi, K. D. Macisaac titative characterization of genetic interaction profiles. Proc. et al., 2004 Transcriptional regulatory code of a eukaryotic Natl. Acad. Sci. USA 105: 5821–5826. genome. Nature 431: 99–104. Demogines,A.,E.Smith,L.Kruglyak,andE.Alani,2008 Identification Hodgins-Davis, A., A. B. Adomas, J. Warringer, and J. P. Townsend, and dissection of a complex DNA repair sensitivity phenotype in 2012 Abundant gene-by-environment interactions in gene ex- Baker’s yeast. PLoS Genet. 4: e1000123. pression reaction norms to copper within Saccharomyces cerevi- Deutschbauer, A. M., and R. W. Davis, 2005 Quantitative trait loci siae. Genome Biol. Evol. 4: 1061–1079. mapped to single-nucleotide resolution in yeast. Nat. Genet. 37: Huang, Y., S. Wuchty, and T. M. Przytycka, 2013 eQTL epistasis: 1333–1340. challenges and computational approaches. Front. Genet. 4: 51. Dimitrov, L. N., R. B. Brem, L. Kruglyak, and D. E. Gottschling, Huber, A., S. L. French, H. Tekotte, S. Yerlikaya, M. Stahl et al., 2009 Polymorphisms in multiple genes contribute to the spon- 2011 Sch9 regulates ribosome biogenesis via Stb3, Dot6 and taneous mitochondrial genome instability of Saccharomyces cer- Tod6 and the histone deacetylase complex RPD3L. EMBO J. 30: evisiae S288C strains. Genetics 183: 365–383. 3052–3064. Doniger, S. W., H. S. Kim, D. Swain, D. Corcuera, M. Williams et al., Huebert, D. J., P. F. Kuan, S. Keles, and A. P. Gasch, 2012 Dynamic 2008 A catalog of neutral and deleterious polymorphism in changesinnucleosomeoccupancyarenotpredictiveofgeneex- yeast. PLoS Genet. 4: e1000183. pression dynamics but are linked to transcription and chromatin Ehrenreich, I. M., N. Torabi, Y. Jia, J. Kent, S. Martis et al., regulators. Mol. Cell. Biol. 32: 1645–1653. 2010 Dissection of genetically complex traits with extremely Jensen-Pergakes, K., Z. Guo, M. Giattina, S. L. Sturley, and M. Bard, large pools of yeast segregants. Nature 464: 1039–1042. 2001 Transcriptional regulation of the two sterol esterification Eng, K. H., D. J. Kvitek, S. Keles, and A. P. Gasch, 2010 Transient genes in the yeast Saccharomyces cerevisiae. J. Bacteriol. 183: genotype-by-environment interactions following environmental 4950–4957. shock provide a source of expression variation for essential Kapur, K., T. Schupbach, I. Xenarios, Z. Kutalik, and S. Bergmann, genes. Genetics 184: 587–593. 2011 Comparison of strategies to detect epistasis from eQTL Fay, J. C., H. L. McCullough, P. D. Sniegowski, and M. B. Eisen, data. PLoS ONE 6: e28415. 2004 Population genetic variation in gene expression is asso- Karpichev, I. V., Y. Luo, R. C. Marians, and G. M. Small, 1997 A ciated with phenotypic variation in Saccharomyces cerevisiae. complex containing two transcription factors regulates per- Genome Biol. 5: R26. oxisome proliferation and the coordinate induction of beta- Fraley, C., and A. E. Raftery, 2002 Model-based clustering, dis- oxidation enzymes in Saccharomyces cerevisiae.Mol.Cell.Biol. criminant analysis, and density estimation. J. Am. Stat. Assoc. 17: 69–80. 97: 611–631. Karpichev, I. V., J. M. Durand-Heredia, Y. Luo, and G. M. Small, Fraser, H. B., A. M. Moses, and E. E. Schadt, 2010 Evidence for 2008 Binding characteristics and regulatory mechanisms of widespread adaptive evolution of gene expression in budding the transcription factors controlling oleate-responsive genes in yeast. Proc. Natl. Acad. Sci. USA 107: 2977–2982. Saccharomyces cerevisiae. J. Biol. Chem. 283: 10264–10275. Gagneur, J., O. Stegle, C. Zhu, P. Jakob, M. M. Tekkedil et al., Kennedy, M. A., R. Barbuch, and M. Bard, 1999 Transcriptional 2013 Genotype-environment interactions reveal causal path- regulation of the squalene synthase gene (ERG9) in the yeast ways that mediate genetic effects on phenotype. PLoS Genet. Saccharomyces cerevisiae. Biochim. Biophys. Acta 1445: 110–122. 9: e1003803. Keurentjes, J. J., J. Fu, I. R. Terpstra, J. M. Garcia, G. van den Gaisne, M., A. M. Becam, J. Verdiere, and C. J. Herbert, 1999 A Ackerveken et al., 2007 Regulatory network construction in ’natural’ mutation in Saccharomyces cerevisiae strains derived Arabidopsis by using genome-wide gene expression quantitative from S288c affects the complex regulatory gene HAP1 (CYP1). trait loci. Proc. Natl. Acad. Sci. USA 104: 1708–1713. Curr. Genet. 36: 195–200. Kim, H. S., and J. C. Fay, 2007 Genetic variation in the cysteine Gargalovic, P. S., M. Imura, B. Zhang, N. M. Gharavi, M. J. Clark biosynthesis pathway causes sensitivity to pharmacological com- et al., 2006 Identification of inflammatory gene modules based pounds. Proc. Natl. Acad. Sci. USA 104: 19387–19391.

380 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch Kim, H. S., and J. C. Fay, 2009 A combined-cross analysis reveals in an abnormal mRNA decay rate; sequence analysis reveals an genes with drug-specific and background-dependent effects on RNA-binding domain in the RNA15 protein. Mol. Cell. Biol. 11: drug sensitivity in Saccharomyces cerevisiae. Genetics 183: 3075–3087. 1141–1151. Minvielle-Sebastia, L., P. J. Preker, and W. Keller, 1994 RNA14 Kim, H. S., J. Huh, and J. C. Fay, 2009 Dissecting the pleiotropic and RNA15 proteins as components of a yeast pre-mRNA 39-end consequences of a quantitative trait nucleotide. FEMS Yeast Res. processing factor. Science 266: 1702–1705. 9: 713–722. Mitchell, S. F., S. Jain, M. She, and R. Parker, 2013 Global anal- King, M. C., and A. C. Wilson, 1975 Evolution at two levels in ysis of yeast mRNPs. Nat. Struct. Mol. Biol. 20: 127–133. humans and chimpanzees. Science 188: 107–116. Monteiro, P. T., N. D. Mendes, M. C. Teixeira, S. d’Orey, S. Tenreiro Kliebenstein, D., 2009 Quantitative genomics: analyzing intraspe- et al., 2008 YEASTRACT-DISCOVERER: new tools to improve cific variation using global gene expression polymorphisms or the analysis of transcriptional regulatory associations in Saccha- eQTLs. Annu. Rev. Plant Biol. 60: 93–114. romyces cerevisiae. Nucleic Acids Res. 36: D132–D136. Kvitek, D. J., J. L. Will, and A. P. Gasch, 2008 Variations in stress Neto, E. C., M. P. Keller, A. F. Broman, A. D. Attie, R. C. Jansen sensitivity and genomic expression in diverse S. cerevisiae iso- et al., 2012 Quantile-based permutation thresholds for quan- lates. PLoS Genet. 4: e1000223. titative trait loci hotspots. Genetics 191: 1355–1365. Lee, M. V., S. E. Topper, S. L. Hubler, J. Hose, C. D. Wenger et al., Nickoloff, J. A., E. Y. Chen, and F. Heffron, 1986 A 24-base-pair 2011 A dynamic model of proteome changes reveals new roles DNA sequence from the MAT locus stimulates intergenic recom- for transcript alteration in yeast. Mol. Syst. Biol. 7: 514. bination in yeast. Proc. Natl. Acad. Sci. USA 83: 7831–7835. Lee, S. I., A. M. Dudley, D. Drubin, P. A. Silver, N. J. Krogan et al., Oleksiak, M. F., G. A. Churchill, and D. L. Crawford, 2002 Variation 2009 Learning a prior on regulatory potential from eQTL data. in gene expression within and among natural populations. Nat. PLoS Genet. 5: e1000358. Genet. 32: 261–266. Lewis, J. A., and A. P. Gasch, 2012 Natural variation in the yeast Orozco, L. D., B. J. Bennett, C. R. Farber, A. Ghazalpour, C. Pan glucose-signaling network reveals a new role for the Mig3p et al., 2012 Unraveling inflammatory responses using systems transcription factor. G3 (Bethesda) 2: 1607–1612. genetics and gene-environment interactions in macrophages. Lewis, J. A., I. M. Elkon, M. A. McGee, A. J. Higbee, and A. P. Gasch, Cell 151: 658–670. 2010 Exploiting natural variation in Saccharomyces cerevisiae Ozcan, S., J. Dover, A. G. Rosenwald, S. Wolfl, and M. Johnston, to identify genes for increased ethanol resistance. Genetics 186: 1996 Two glucose transporters in Saccharomyces cerevisiae are 1197–1205. glucose sensors that generate a signal for induction of gene Li, J., Y. Liu, T. Kim, R. Min, and Z. Zhang, 2010 Gene expression expression. Proc. Natl. Acad. Sci. USA 93: 12428–12432. variability within and between human populations and implica- Ozcan, S., J. Dover, and M. Johnston, 1998 Glucose sensing and tions toward disease susceptibility. PLOS Comput. Biol. 6(8): pii: signaling by two glucose receptors in the yeast Saccharomyces e1000910. cerevisiae. EMBO J. 17: 2566–2573. Li, Y., O. A. Alvarez, E. W. Gutteling, M. Tijsterman, J. Fu et al., Perlstein, E. O., D. M. Ruderfer, D. C. Roberts, S. L. Schreiber, and L. 2006 Mapping determinants of gene expression plasticity by Kruglyak, 2007 Genetic basis of individual differences in the genetical genomics in C. elegans. PLoS Genet. 2: e222. response to small-molecule drugs in yeast. Nat. Genet. 39: 496– Lippman, S. I., and J. R. Broach, 2009 Protein kinase A and 502. TORC1 activate genes for ribosomal biogenesis by inactivating Petretto, E., J. Mangion, N. J. Dickens, S. A. Cook, M. K. Kumaran repressors encoded by Dot6 and its homolog Tod6. Proc. Natl. et al., 2006 Heritability and tissue specificity of expression Acad. Sci. USA 106: 19928–19933. quantitative trait loci. PLoS Genet. 2: e172. Litvin, O., H. C. Causton, B. J. Chen, and D. Pe’er, 2009 Modularity Ridley, S. P., S. S. Sommer, and R. B. Wickner, 1984 Superkiller and interactions in the genetics of gene expression. Proc. Natl. mutations in Saccharomyces cerevisiae suppress exclusion of M2 Acad. Sci. USA 106: 6441–6446. double-stranded RNA by L-A-HN and confer cold sensitivity in Lockshon, D., L. E. Surface, E. O. Kerr, M. Kaeberlein, and B. K. the presence of M and L-A-HN. Mol. Cell. Biol. 4: 761–770. Kennedy, 2007 The sensitivity of yeast mutants to oleic acid Rockman, M. V., and L. Kruglyak, 2006 Genetics of global gene implicates the peroxisome and other processes in membrane expression. Nat. Rev. Genet. 7: 862–872. function. Genetics 175: 77–91. Romanoski, C. E., S. Lee, M. J. Kim, L. Ingram-Drake, C. L. Plaisier Loewith, R., E. Jacinto, S. Wullschleger, A. Lorberg, J. L. Crespo et al., 2010 Systems genetics analysis of gene-by-environment et al., 2002 Two TOR complexes, only one of which is rapa- interactions in human cells. Am. J. Hum. Genet. 86: 399–410. mycin sensitive, have distinct roles in cell growth control. Mol. Rossouw, D., D. Jacobson, and F. F. Bauer, 2012 Transcriptional Cell 10: 457–468. regulation and the diversification of metabolism in wine yeast Longtine, M. S., A. McKenzie, 3rd, D. J. Demarini, N. G. Shah, A. strains. Genetics 190: 251–261. Wach et al., 1998 Additional modules for versatile and eco- Schadt, E. E., S. A. Monks, T. A. Drake, A. J. Lusis, N. Che et al., nomical PCR-based gene deletion and modification in Saccharo- 2003 Genetics of gene expression surveyed in maize, mouse myces cerevisiae. Yeast 14: 953–961. and man. Nature 422: 297–302. MacIsaac, K. D., T. Wang, D. B. Gordon, D. K. Gifford, G. D. Stormo Segal, S. P., T. Dunckley, and R. Parker, 2006 Sbp1p affects trans- et al., 2006 An improved map of conserved regulatory sites for lational repression and decapping in Saccharomyces cerevisiae. Saccharomyces cerevisiae. BMC Bioinformatics 7: 113. Mol. Cell. Biol. 26: 5120–5130. Maranville, J. C., F. Luca, A. L. Richards, X. Wen, D. B. Witonsky Sinha, H., B. P. Nicholson, L. M. Steinmetz, and J. H. McCusker, et al., 2011 Interactions between glucocorticoid treatment and 2006 Complex genetic interactions in a quantitative trait lo- cis-regulatory polymorphisms contribute to cellular response cus. PLoS Genet. 2: e13. phenotypes. PLoS Genet. 7: e1002162. Skelly, D. A., G. E. Merrihew, M. Riffle, C. F. Connelly, E. O. Kerr Marks, V. D., S. J. Ho Sui, D. Erasmus, G. K. van der Merwe, J. et al., 2013 Integrative phenomics reveals insight into the Brumm et al., 2008 Dynamics of the yeast transcriptome dur- structure of phenotypic diversity in budding yeast. Genome ing wine fermentation reveals a novel fermentation stress re- Res. 23: 1496–1504. sponse. FEMS Yeast Res. 8: 35–52. Smirnov, D. A., M. Morley, E. Shin, R. S. Spielman, and V. G. Minvielle-Sebastia, L., B. Winsor, N. Bonneaud, and F. Lacroute, Cheung, 2009 Genetic analysis of radiation-induced changes 1991 Mutations in the yeast RNA14 and RNA15 genes result in human gene expression. Nature 459: 587–591.

eQTL Mapping of the Yeast Ethanol Response 381 Smith, E. N., and L. Kruglyak, 2008 Gene-environment interac- tagging in Saccharomyces cerevisiae and Schizosaccharomyces tion in yeast gene expression. PLoS Biol. 6: e83. pombe using nourseothricin resistance. Yeast 22: 1061–1068. Smyth, G. K., 2004 Linear models and empirical bayes methods West, M. A., K. Kim, D. J. Kliebenstein, H. van Leeuwen, R. W. for assessing differential expression in microarray experiments. Michelmore et al., 2007 Global eQTL mapping reveals the Stat. Appl. Genet. Mol. Biol. 3: Article3. complex genetic architecture of transcript-level variation in Ara- Stanley, D., P. J. Chambers, G. A. Stanley, A. Borneman, and S. bidopsis. Genetics 175: 1441–1450. Fraser, 2010 Transcriptional changes associated with ethanol Winzeler, E. A., D. D. Shoemaker, A. Astromoff, H. Liang, K. Ander- tolerance in Saccharomyces cerevisiae. Appl. Microbiol. Biotech- son et al., 1999 Functional characterization of the S. cerevisiae nol. 88: 231–239. genome by gene deletion and parallel analysis. Science 285: Steinmetz, L. M., H. Sinha, D. R. Richards, J. I. Spiegelman, P. J. 901–906. Oefner et al., 2002 Dissecting the architecture of a quantitative Wohlbach, D. J., A. Kuo, T. K. Sato, K. M. Potts, A. A. Salamov et al., – trait locus in yeast. Nature 416: 326 330. 2011 Comparative genomics of xylose-fermenting fungi for fi Storey, J. D., and R. Tibshirani, 2003 Statistical signi cance for enhanced biofuel production. Proc. Natl. Acad. Sci. USA 108: – genomewide studies. Proc. Natl. Acad. Sci. USA 100: 9440 9445. 13212–13217. Storey, J. D., J. M. Akey, and L. Kruglyak, 2005 Multiple locus Yang, I. V., C. M. Wade, H. M. Kang, S. Alper, H. Rutledge et al., linkage analysis of genomewide expression in yeast. PLoS Biol. 2009 Identification of novel genes that mediate innate immu- 3: e267. nity using inbred mice. Genetics 183: 1535–1544. Swinnen, S., K. Schaerlaekens, T. Pais, J. Claesen, G. Hubmann Yoshikawa, K., T. Tanaka, C. Furusawa, K. Nagahisa, T. Hirasawa et al., 2012 Identification of novel causative genes determin- et al., 2009 Comprehensive phenotypic analysis for identifica- ing the complex trait of high ethanol tolerance in yeast using tion of genes affecting growth under ethanol stress in Saccharo- pooled-segregant whole-genome sequence analysis. Genome myces cerevisiae. FEMS Yeast Res. 9: 32–44. Res. 22: 975–984. You, K. M., C. L. Rosenfield, and D. C. Knipple, 2003 Ethanol Tadauchi, T., T. Inada, K. Matsumoto, and K. Irie, 2004 Post- tolerance in the yeast Saccharomyces cerevisiae is dependent transcriptional regulation of HO expression by the Mkt1-Pbp1 – on cellular oleic acid content. Appl. Environ. Microbiol. 69: complex. Mol. Cell. Biol. 24: 3670 3681. – Tamura, K., Y. Gu, Q. Wang, T. Yamada, K. Ito et al., 2004 A hap1 1499 1503. mutation in a laboratory strain of Saccharomyces cerevisiae re- Yvert,G.,R.B.Brem,J.Whittle,J.M.Akey,E.Fosset al., sults in decreased expression of ergosterol-related genes and 2003 Trans-acting regulatory variation in Saccharomyces cer- cellular ergosterol content compared to sake yeast. J. Biosci. evisiae and the role of transcription factors. Nat. Genet. 35: – Bioeng. 98: 159–166. 57 64. Taxis, C., P. Keller, Z. Kavagiou, L. J. Jensen, J. Colombelli et al., Zhu,J.,B.Zhang,E.N.Smith,B.Drees,R.B.Bremet al., 2005 Spore number control and breeding in Saccharomyces 2008 Integrating large-scale functional genomic data to dis- cerevisiae: a key role for a self-organizing system. J. Cell Biol. sect the complexity of yeast regulatory networks. Nat. Genet. 171: 627–640. 40: 854–861. Teixeira, M. C., P. Monteiro, P. Jain, S. Tenreiro, A. R. Fernandes Zhu, J., P. Sova, Q. Xu, K. M. Dombek, E. Y. Xu et al., et al., 2006 The YEASTRACT database: a tool for the analysis 2012 Stitching together multiple data dimensions reveals in- of transcription regulatory associations in Saccharomyces cere- teracting metabolomic and transcriptomic networks that modu- visiae. Nucleic Acids Res. 34: D446–D451. late cell regulation. PLoS Biol. 10: e1001301. van den Ent, F., and J. Lowe, 2006 RF cloning: a restriction-free Ziebarth, J. D., M. N. Cook, X. Wang, R. W. Williams, L. Lu et al., method for inserting target genes into plasmids. J. Biochem. 2012 Treatment- and population-dependent activity patterns Biophys. Methods 67: 67–74. of behavioral and expression QTLs. PLoS ONE 7: e31805. Van Driessche, B., L. Tafforeau, P. Hentges, A. M. Carr, and J. Vandenhaute, 2005 Additional vectors for PCR-based gene Communicating editor: J. Akey

382 J. A. Lewis, A. T. Broman, J. Will, and A. P. Gasch GENETICS

Supporting Information http://www.genetics.org/lookup/suppl/doi:10.1534/genetics.114.167429/-/DC1

Genetic Architecture of Ethanol-Responsive Transcriptome Variation in Saccharomyces cerevisiae Strains

Jeffrey A. Lewis, Aimee T. Broman, Jessica Will, and Audrey P. Gasch

Copyright © 2014 by the Genetics Society of America DOI: 10.1534/genetics.114.167429

Figure S1 Ploidy affects hundreds of ethanol‐responsive transcripts. The number of linkages falling within a sliding 5 cM window is plotted across the genome. The large hotspot on chromosome II coincides with the HO locus.

2 SI J. A. Lewis et al.

Figure S2 Hotspot linkage association for the combined crosses across mapping regimes. The average number of transcripts (using a sliding 5cM window) mapping to each locus across the sixteen yeast chromosomes (x‐axis) is plotted for each of the three mapping regimes.

J. A. Lewis et al. 3 SI

Figure S3 Reciprocal Hemizygosity analysis of Mkt1p‐GFP localization. Localization of Mkt1p‐GFP was observed via fluorescent microscopy before and after 30 min treatment with 5% ethanol.

4 SI J. A. Lewis et al.

Figure S4 YPS_Mkt1p‐GFP co‐localizes to P‐bodies during hypo‐osmotic stress. Localization of Mkt1p‐GFP and Edc3p‐ dTomato was observed via fluorescence microscopy following resuspension of cells in water for 30 min.

J. A. Lewis et al. 5 SI

Files S1‐S10

Available for download as Excel files at http://www.genetics.org/lookup/suppl/doi:10.1534/genetics.114.167429/‐/DC1

File S1 Strains Used in this Study File S2 Normalized Microarray Data for Basal Transcript Abundance File S3 Normalized Microarray Data for Ethanol Transcript Abundance File S4 Normalized Microarray Data for Ethanol Response File S5 Limma Output for Analysis of Parental Microarray Data File S6 List of All Significant Linkages File S7 Functional Enrichments for Parental Gene Expression Clusters File S8 eQTL Hotspot Coordinates and Enrichments File S9 Epi‐Hotspot Coordinates and Enrichments File S10 Regulator Enrichments for MKT1 Hotspot Traits

6 SI J. A. Lewis et al.

Table S1 eQTL mapping summary for S288c x M22 cross

Total Number (and %) Number (and %) Number (and %) Transcripts mapping to 1 mapping to 2 mapping to 3 or Mapped locus loci more loci

Basal Abundance 1480 1370 (93%) 100 (7%) 10 (1%) Mapping

Ethanol Abundance 1349 1230 (91%) 111 (8%) 8 (1%) Mapping Ethanol Response 1127 1048 (93%) 76 (7%) 3 (0%) Mapping One or more mapping regime 2664

J. A. Lewis et al. 7 SI

Table S2 eQTL mapping summary for S288c x YPS163 cross

Total Number (and %) Number (and %) Number (and %) Transcripts mapping to 1 mapping to 2 mapping to 3 or Mapped locus loci more loci

Basal Abundance 2314 1928 (83%) 313 (14%) 73 (3%) Mapping

Ethanol Abundance 2260 1875 (83%) 321 (14%) 64 (3%) Mapping Ethanol Response 1631 1432 (88%) 188 (12%) 11 (1%) Mapping One or more mapping regime 3764

8 SI J. A. Lewis et al.