Genetic basis of individual differences in the response to small-molecule drugs in yeast Ethan O Perlstein1,2, Douglas M Ruderfer3, David C Roberts4, Stuart L Schreiber1,2,5 & Leonid Kruglyak3 Individual response to small-molecule drugs is variable a drug that provides a cure for some may confer no therapeutic benefit or trigger an adverse reaction in others. To begin to understand such differences systematically, we treated 104 genotyped segregants from a cross between two yeast strains with a collection of 100 diverse small molecules. We used linkage analysis to identify 124 distinct linkages between genetic markers and response to 83 compounds. The linked markers clustered at eight genomic locations, or quantitative-trait locus ���hotspots���, that contain one or more polymorphisms that affect response to multiple small molecules. We also experimentally verified that a deficiency in leucine biosynthesis caused by a deletion of LEU2 underlies sensitivity to niguldipine, which is structurally related to therapeutic calcium channel blockers, and that a natural coding-region polymorphism in the inorganic phosphate transporter PHO84 underlies sensitivity to two polychlorinated phenols that uncouple oxidative phosphorylation. Our results provide a step toward a systematic understanding of small-molecule drug action in genetically distinct individuals. Most therapeutic drugs are small molecules. The genetic basis of small-molecule activity in living cells can be complex and is often evolutionarily conserved across different taxa. In humans, many pharmacogenomic studies have been performed to assess the role of natural genetic variation in the cellular response to small-molecule drugs1���8. However, these studies are limited by small sample sizes and the inability to rapidly screen large numbers of drugs and phenotypes. Saccharomyces cerevisiae is a well studied unicellular yeast with extensive genetic and physiological homology to multicellular eukar- yotes9. For this reason, budding yeast has been an attractive model for the study of many human diseases10, including metabolic disorders11 and neurodegeneration12. Yeast has also been effectively exploited in large-scale studies by consortia (for example, artificial gene deletion collections) to systematically assess individual gene function in response to small-molecule perturbation13. However, those studies have focused primarily on identifying the molecular targets of small molecules rather than on identifying the genes up- and downstream of these targets that modulate the physiological response (that is, resistance or sensitivity) to small molecules. A growing, complemen- tary trend in yeast biology that is generalizable to other laboratory model organisms involves examining the effects of natural genetic variation, as opposed to engineered mutations, on the cellular phenotypes associated with complex traits such as high-temperature growth14,15, sporulation16,17 and genome-wide mRNA expression18,19. Expression traits have also been studied in mammalian cells20,21. Here, we greatly expand upon a previous study22 of compound response traits in yeast by analyzing 104 genotyped segregants of Saccharomyces cerevisiae treated with 100 diverse compounds that we term small-molecule perturbagens (SMPs). Eighteen SMPs are Food and Drug Administration (FDA)-approved drugs that modulate evolutionarily conserved targets or cellular processes. The 104 segre- gants were derived from a cross between a laboratory strain (BY4716, hereafter ���BY���) and a vineyard isolate (RM11-1a, hereafter ���RM���) and were the focus of an extensive linkage study of steady-state mRNA expression traits23. The parent strains differ in DNA sequence at 0.6% of nucleotides24. RESULTS Related compounds cluster based on cellular response We first performed dose-response experiments on the two parent strains, BYand RM, in order to determine the half-maximal inhibitory concentration (IC50) for each SMP (data not shown). We then measured in parallel the growth of all 104 segregants and both parent strains in the presence of one or more concentrations near the IC50 of each SMP at multiple time points post-inoculation (for complete list of SMPs, see Supplementary Table 1 online). In this overlapping fashion, we better resolved the variance in segregant final yield. The median growth-inhibitory SMP concentration tested was 21.6 mM. We phenotyped seven SMPs���cetylpyridinium chloride (766 nM), cyclo- heximide (356 nM), clotrimazole (483 nM), flutrimazole (723 nM), Received 6 November 2006 accepted 2 January 2007 published online 4 March 2007 doi:10.1038/ng1991 1Howard Hughes Medical Institute, Broad Institute of Harvard and MIT, 7 Cambridge Center, Cambridge, Massachusetts 02142, USA. 2Department of Molecular and Cellular Biology, Harvard University, 7 Divinity Avenue, Cambridge, Massachusetts 02138, USA. 3Lewis-Sigler Institute for Integrative Genomics and Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544, USA. 4Theoretical Division and Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, New Mexico 87544, USA. 5Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford Street, Cambridge, Massachusetts 02138, USA. Correspondence should be addressed to S.L.S. (stuart_schreiber@harvard.edu) or L.K. (leonid@genomics.princeton.edu). 496 VOLUME 39 [ NUMBER 4 [ APRIL 2007 NATURE GENETICS A RT I C L E S �� 200 7 Nature Publishing Group http://www.nature.com/naturegenetics

ketoconazole (940 nM), rapamycin (50 nM) and tunicamycin (592 nM)���at submicromolar concentrations. In order to investigate the underlying genetic basis of compound response in the 104 segregants, we first performed two-dimensional hierarchical clustering on all 324 phenotypes (100 compounds at multiple time points and concentrations) after regressing out the effect of growth in SMP-free medium (for complete clustergram and raw data, see Supplementary Fig. 1 and Supplementary Table 2 online, respectively). As expected, SMPs having common physiological effects on cells yet lacking structural similarity (so-called functional analogs) clustered together. For example, one cluster contained cycloheximide, anisomy- cin and rapamycin. These three SMPs are structurally unrelated, and they target different proteins, but their common physiological effect on cells is inhibition of protein translation. Another cluster contained FCCP, a proton ionophore, and nocodazole, a microtubule depoly- merizer (Fig. 1a���c). A previously published report states that chronic exposure of mammalian cells to FCCP depolymerizes microtubules in a mitochondrial-dependent fashion through its destabilizing effects on the mitochondrial proton gradient25. Notably, another cluster con- tained FDA-approved therapeutic drugs of three distinct pharmaco- logical classes, as well as the structurally unrelated natural product E6 berbamine, which is found in Chinese herbal remedies (Fig. 1d���g). The three pharmacological classes are the phenothiazines (for example, chlorpromazine), which are clinical antipsychotics, the selective serotonin reuptake inhibitors (for example, sertraline) and the tricyc- lics (for example, nortriptyline) the latter two classes are clinical antidepressants. Although the precise mechanism of action of anti- depressant cytotoxicity in yeast is unknown, the clustering results suggest that there are evolutionarily conserved pathways associated with single-cell physiology that are relevant to multicellular human psychiatric disease. Linkage analysis identifies loci that affect response to SMPs We carried out linkage analysis between segregant final yield at each SMP concentration and time point (324 total phenotypes) and 2,956 genetic markers previously genotyped in the segregants23. We identi- fied 219 QTLs with a logarithm of the odds (lod) score Z4 (Supplementary Table 3 online). Only 3.8 loci are expected by chance at this significance threshold based on empirical permutation tests thus, the detected loci had a false discovery rate of o2%. Because all time points and concentrations of each compound were used as independent phenotypes, some of the 219 QTLs were duplicate linkages of the same SMP to the same locus. We collapsed all such duplicate loci for a given SMP with overlapping confidence intervals (1 lod drop) to obtain a set of 124 unique SMP-locus pairs (Supple- mentary Tables 4 and 5 online). Eighty-three of the tested SMPs linked to at least one locus, 25 SMPs linked to two loci each and eight SMPs linked to three loci each. At lower lod score thresholds, we observed many more additional loci at rates much higher than chance, providing further evidence that response to most SMPs is a genetically complex trait. For 17 of the 33 compounds with multiple detected QTLs, we observed both BY- and RM-derived genotypes that FCCP 64 h FCCP 74 h Nocodazole 48 h Nocodazole 60 h CF3O H N N CN CN S O N NHCO2Me N H O O O O H N O O O OH OH N HN Me MeHN CI CI E6-berbamine 64 h E6-berbamine 74 h Nortriptyline 118 h Sertraline 68 h Sertraline 78 h Chlorpromazine 70 h Trifluoperazine 90 h Trimeprazine 80 h Wiskostatin 64 h Chlorpromazine 48 h Trifluoperazine 66 h Sertraline 30 h Sertraline 44 h Sertraline 52 h H N a b e d f g c Sertraline 90 h Figure 1 Hierarchical clustering of final yield measurements of 104 BY and RM segregants treated with selected SMPs shows clustering of functional analogs, including a cluster of psychiatric disease drugs. Red indicates resistance green indicates sensitivity. Columns represent segregants rows represent responses to SMPs. SMP names and time points are listed on the right-hand side of the clustergram. (a) Clustergram containing nocodazole and FCCP. (b) Structure of nocodazole. (c) Structure of FCCP. (d) Clustergram containing psychoactive therapeutic drugs. (e) Structure of the natural product E6-berbamine. (f) Structure of the tricyclic antidepressant nortriptyline. (g) Structure of the selective serotonin reuptake inhibitor sertraline. NATURE GENETICS VOLUME 39 [ NUMBER 4 [ APRIL 2007 497 A RT I C L E S �� 200 7 Nature Publishing Group http://www.nature.com/naturegenetics