Yeast genome-wide drug-induced haploinsufficiency screen to determine drug mode of action

Abstract

Methods to systematically test drugs against all possible proteins in a cell are needed to identify the targets underlying their therapeutic action and unwanted effects. Here, we show that a genome-wide drug-induced haploinsufficiency screen by using yeast can reveal drug mode of action in yeast and can be used to predict drug mode of action in human cells. We demonstrate that dihydromotuporamine C, a compound in preclinical development that inhibits angiogenesis and metastasis by an unknown mechanism, targets sphingolipid metabolism. The systematic, unbiased and genome-wide nature of this technique makes it attractive as a general approach to identify cellular pathways affected by drugs.

Cell-based screening is increasingly used to identify chemicals that can suppress or enhance particular cellular phenotypes. Chemicals identified by phenotypic screening are valuable chemical genetic tools to study complex cellular processes and are often attractive candidates for drug development (1–3). Identification of the mechanism of action of these chemicals is an integral part of the chemical genetic approach and is also critical for evaluating and optimizing therapeutic agents, yet no systematic target identification approach exists (3). The classical method is to purify the target from cellular extracts by using immobilized or radiolabeled versions of the chemical or to guess candidates and test them one by one in binding or functional assays (3). However, without prior knowledge of the pathways affected by the drug, these approaches have met with little success and systematic genomic and proteomic methods to test a wide range of drugs against all possible targets in a cell are clearly needed.

The budding yeast Saccharomyces cerevisiae has been proposed as a tool for identifying human drug targets because >40% of yeast proteins share some conserved sequence with at least one known or predicted human protein, including several hundred genes implicated in human disease (4, 5). Recent technological advances in yeast genomics, including gene expression profiling, chemical-genetic synthetic profiling, and drug-induced haploinsufficiency, show promise as tools to study proteins and pathways affected by drugs at a genome-wide level (4, 5). Drug-induced haploinsufficiency occurs when lowering the dosage of a single gene from two copies to one copy in diploid cells results in a heterozygote that displays increased drug sensitivity compared with wild-type strains (6). Giaever et al. (6) pooled 233 “bar-coded” heterozygous yeast strains, grew them competitively in the presence of a drug whose target in yeast cells was known, and successfully used DNA microarrays to identify the heterozygous drug target strain whose growth was compromised in the presence of the drug. Here, we asked whether drug-induced haploinsufficiency could be applied genome-wide to identify the mode of action of a drug with an unknown target. We used dihydromotuporamine C (dhMotC) (Fig. 1A), a member of the motuporamine family of agents discovered in a phenotypic screen for small molecules that inhibit invasion of human carcinoma cell lines (2, 7). The motuporamines inhibit the formation of invadopodia, decrease tumor cell motility, and alter the organization of filamentous actin at the cell margin (2). Identifying the mechanism of action of dhMotC is critical for evaluating its potential for cancer therapy but it is a difficult task, because dhMotC has no other known biological activities, it does not resemble other bioactive molecules, and its chemical structure revealed no clues to guide target identification.

Fig. 1.

Sensitivity of yeast to dhMotC. (A) Structural formula of dhMotC. (B) Growth of wild-type (BY4743) strain (OD₆₀₀) as a function of time in hours. BY4743 was grown in YPD liquid culture with or without dhMotC treatment as indicated. Growth curves were performed in triplicate and represent the average of three experiments.

By using a genome-wide haploinsufficiency screen, we show that dhMotC targets sphingolipid metabolism, resulting in decreased ceramide levels in yeast. Furthermore, exogenous ceramide partially rescues the effect of dhMotC on human cell survival, demonstrating that systematic genome-wide analysis of drug effects in yeast can be used to elucidate the mechanism of action of drugs in yeast and in human cells.

Materials and Methods

Reagents. dhMotC was a generous gift of D. Williams (University of British Columbia). It was synthesized as described (7) and was solubilized in DMSO. Dihydrosphingosine (DHS) and C6-ceramide were purchased from Sigma.

Yeast Strains, Media, and Genetic Manipulations. The diploid heterozygous deletion mutants generated by the International Deletion Consortium (8) were obtained from Research Genetics/Invitrogen. BY4743 was used as the diploid wild-type control. TDY2037 (wild-type) and TDY2038 (csg2Δ) MATα haploid strains were described (9). Yeast was grown on standard rich medium [yeast extract/peptone/dextrose (YPD)] as described (10).

Individual Growth Rate Analysis. Growth rates of strains were measured under various drug conditions essentially as described (6). Cells were diluted from an overnight culture to an OD₆₀₀ of ≈0.01 and were allowed to grow until the OD₆₀₀ reached ≈0.05 (t = 0) to ensure that cells were in logarithmic phase. Drug was added and growth rates were measured by determination of OD₆₀₀ as a function of time by using a Pharmacia Biotech Ultrospec 3000 or a Tecan Sunrise plate reader. Growth curves were carried out in triplicate and curves shown are averages of three experiments with error bars representing 1 SD.

Drug-Induced Haploinsufficiency Screen. The deletion mutant arrays (DMA) were propagated on rich liquid medium containing G418 (200 mg/liter, GIBCO/BRL). The DMA were transferred by hand with a model VP408FH 96-floating pin replicator (V & P Scientific, San Diego) and a VP380 colony copier (V & P Scientific) into 96-well plates containing 100 μl of liquid YPD and were grown overnight at 25°C to saturation. Replicator sterilization procedures were as described (11). Saturated cultures compensated for differences in growth rate between strains, to ensure that roughly equal amounts of cells were deposited on the agar plates. After 24 h, the DMA were pinned onto plates (OmniTray, Nunc) containing YPD-agar and DMSO (6 μl/ml) or 60 μM dhMotC and DMSO (6 μl/ml). DMA were arrayed at high density by using a BioRobotics TAS1 robot with a 0.4-mm diameter 96-pin tool that spots ≈20 nl or ≈250 cells. Strains were pinned in duplicate with a density of 1,536 per plate. Pins were sterilized by agitating for 30 s in 10% bleach followed by 95% ethanol for 15 s and drying with warm air for 15 s. Plates were grown at 25°C and strains were scored for growth, slow growth, or no growth, compared with no drug over a period of 4 days.

Integrative Growth Curve Difference (IGCD) Analysis. We analyzed growth curves for 29 strains, each grown without drug and in the presence of 20 μM dhMotC. OD₆₀₀ measurements of cell density were taken hourly for 10 time points. The typical level of replication was three growth curves per strain per condition, for a total of six curves per strain. The cultures underwent exponential growth during the time points and conditions studied. Therefore, we used a log-transformed linear model to relate growth to time by using the model: log OD_{600 g}(t) = β_0g + β_1g M + β_2g t + β_3g tM + ε_g where g indicates the strain, M indicates the presence of dhMotC, and t indicates time. The intercept of this model, equal to β_0g + β_1g M, captures early growth and β_1g is the adjustment due to dhMotC. The slope, equal to β_2g + β_3g M, reflects the steady-state doubling time and β_3g is the adjustment due to dhMotC. Both the intercept and the slope were modified by the presence of dhMotC, indicating early and persistent effects on growth, respectively. The dhMotC effect on growth was highly statistically significant (P < 10^–11) for all 29 strains evaluated, when quantified by using an F test for a restricted model that does not include the dhMotC terms. Strains were ranked according to the cumulative dhMotC effect or IGCDs, denoted D_g, as measured by the area between the expected growth curves with and without drug. Because these curves were linear on the transformed scale, the integral has the following simple form D_g = (–Tβ_1,g – T ²/2) β_3,g. Therefore, the final quantitative measure of cumulative dhMotC effect on each strain is the estimated value of D_g, obtained by evaluating the above expression at the estimated dhMotC regression coefficients.

Immunolocalization Studies. Cells were grown to early log phase in YPD at 25°C and treated with either DMSO, 60 μM dhMotC, or 60 μM dhMotC plus 50 μM DHS for 1.5 h. Cells were then fixed in formaldehyde and were stained with rhodamine-phalloidin (R-415, Molecular Probes) to visualize F-actin and 4′,6-diamidino-2-phenylindole (Sigma) to visualize DNA as described (12).

Whole-Cell Ceramide Isolation. Wild-type (TDY2037) and csg2Δ (TDY2038) cells were grown overnight in YPD to an OD₆₀₀ near 1.0. Cells were diluted to 0.2–0.4 OD₆₀₀ in YPD containing 10 μCi/ml (1 Ci = 37 GBq) ³H-serine and were treated with either 5 μM dhMotC or the equivalent amount of DMSO. After 8 h, 12 OD₆₀₀ units of cells were harvested by centrifugation, were washed with water, were dried thoroughly, and 20 volumes of hexane:ethanol (95:5) were added. The cells were extracted by bead beating and bath sonication. The resulting suspension was centrifuged to remove cell debris. The hexane:ethanol fraction was dried under nitrogen and was subjected to mild alkaline hydrolysis in ethanol:water:ether:pyridine (15:15:5:1) with 0.1 M KOH added. After 3 h at 37°C, the solution was neutralized by the addition of 0.1 M acetic acid and was dried under nitrogen. The dried extract was desalted by resuspension in water-saturated butanol. An equal volume of water was added and the butanol layer was extracted after centrifugation and was dried under nitrogen. A total of 10 OD₆₀₀ of each sample were run on TLC plates by using chloroform:methanol:acetic acid (95:4.5:0.5), the TLC plates were dried, and were rerun on the same system. After the second run, the plates were sprayed with 10% CuSO₄ in 8% phosphoric acid, were dried completely, and were heated at 80° until the ceramide was visible. The ceramide bands were scraped off the TLC plate and the amount of serine incorporated into ceramide was determined by liquid scintillation counting.

Mammalian Cell Survival Assay. MDA-231 cells were plated at 10,000 cells per well in 96-well culture dishes and were allowed to grow overnight. They were subsequently exposed to different concentrations of dhMotC and/or C6-ceramide (Sigma) for 24 h. Cell survival was measured by using the (3-(4,5-dimethylthiazolyl-2)-2,5-diphenyltetrazolium bromide) assay (Sigma).

Database. For further information, refer to the Saccharomyces Genome Database, which can be accessed at www.yeastgenome.org.

Results

We developed a pathway/target identification procedure by using drug-induced haploinsufficiency in yeast that consists of five stages: (i) selection of a drug-induced phenotype; (ii) systematic high-throughput phenotypic screen of a yeast heterozygous deletion diploid set for enhanced or suppressed drug-induced phenotype; (iii) selection of potential pathways/targets based on quantitative ranking of drug sensitivity; (iv) confirmation of drug mode of action in yeast; and (v) assessment of cognate mode of action in the mammalian system.

Sensitivity of Yeast to dhMotC. As a prerequisite, treatment of yeast with a drug of interest must cause a phenotype that is easily scored, such as growth inhibition. dhMotC inhibited the growth of wild-type diploid yeast (BY4743) in liquid cultures in a concentration-dependent manner (Fig. 1B). Concentrations >60 μM severely inhibited growth and resulted in a terminal phenotype that was not cell cycle-specific (data not shown). This result suggests that dhMotC targets at least one gene product involved in an essential function in S. cerevisiae.

Genome-Wide Drug-Induced Haploinsufficiency Screen. For high-throughput phenotypic screening, a set of >5,000 heterozygous diploid deletion strains covering most yeast genes was robotically arrayed onto agar plates containing either DMSO (no drug control) or a sublethal dhMotC concentration of 60 μM. The plates were incubated at 25°C and strain growth on DMSO- and dhMotC-treated plates was compared over a period of 4 days. An extended monitoring period was necessary for scoring of strains that grew slowly because of their intrinsic haploinsufficiency. The screen was carried out twice and strains displaying increased sensitivity to dhMotC in both screens are shown in Table 1. The list includes six essential genes and 15 nonessential genes implicated in a variety of functions.

Table 1. Heterozygous deletion strains sensitive to dhMotC.

ORF	Name	Biological process†
YCL034W	LSB5	Actin filament organization
YNL314W	DAL82	Allantoin catabolism and transcription initiation from polymerase II promoter
YML099C	ARG81	Arginine metabolism
YBR078W	ECM33	Cell wall organization and biogenesis
YNL267W*	PIK1*	Cytokinesis, post-Golgi transport, and signal transduction
YLR286C	CTS1	Cytokinesis, completion of separation
YDL192W	ARF1	ER-to-Golgi transport and intra-Golgi transport
YBR290W	BSD2	Heavy metal ion transport and protein vacuolar targeting
YLR025W	SNF7	Late endosome to vacuole transport
YHR147C	MRPL6	Protein biosynthesis
YOL040C*	RPS15*	Protein biosynthesis
YAL005C	SSA1	Protein folding and protein-nucleus import, translocation
YIL047C	SYG1	Signal transduction
YBR265W*	TSC10*	Sphingolipid biosynthesis
YMR296C*	LCB1*	Sphingolipid biosynthesis
YJR007W*	SUI2*	Translation initiation
YML092C*	PRE8*	Ubiquitin-dependent protein catabolism
YER140W	YER140W	Unknown
YER188W	YER188W	Unknown
YGR205W	YGR205W	Unknown
YLR294C	YLR294C	Unknown

^*Essential genes.

^†Biological process according to their Saccharomyces Genome Database report.

Quantitative Identification of Supersensitive Strains. To confirm and quantitatively rank the sensitivity of the 21 strains, we carried out automated quantitative liquid growth measurements under no-drug conditions or at a concentration of dhMotC (20 μM) that causes only a small degree of growth inhibition (Fig. 2A). Because many heterozygous yeast deletion strains exhibit a slow growth phenotype or haploinsufficiency in rich culture medium (6), it was necessary to compare growth profiles with and without drug for each strain. We next established a model for ranking strain sensitivity by using IGCDs (see Materials and Methods), a technique unbiased by differences in growth rate between strains.

Fig. 2.

dhMotC sensitivity of heterozygous deletion strains identified in the high-throughput screen. (A–D) Growth of wild-type (A), ylr294/YLR294 (B), tsc10/TSC10 (C), and lcb1/LCB1 (D) strains as a function of time in hours in YPD liquid culture under two conditions: ♦, no drug control (DMSO); ▴ 20 μM dhMotC. Growth curves were preformed in triplicate, and OD₆₀₀ was measured by using a Tecan Sunrise plate reader. Growth curves represents the average of three experiments. (E) Plot of the IGCD for 21 heterozygous deletion strains identified in the drug screen are represented by black bars. The IGCD levels for the three additional heterozygous strains in the sphingolipid pathway sensitive to dhMotC are represented by gray bars. The dashed line represents 2-fold IGCD level used to define supersensitive strains.

As expected, all 21 strains were more sensitive to 20 μM dhMotC than the wild-type control, but most showed only minor growth inhibition. However, two strains were clearly more sensitive than others: lcb1Δ/LCB1 and tsc10Δ/TSC10 showed an IGCD >2-fold over wild type and were defined as supersensitive (Fig. 2 C and D). Eight additional strains showed IGCD >1.5-fold greater than wild type (Fig. 2E). Both LCB1 and TSC10 are essential genes required for the biosynthesis of the sphingolipid precursor sphingosine, suggesting that dhMotC targets a sphingolipid-dependent process.

dhMotC Targets Sphingolipid Biosynthesis. Sphingolipids are essential components of all eukaryotic cells, playing both structural and regulatory roles (13, 14). The first committed step of sphingolipid synthesis is the condensation of serine and palmitoyl CoA to 3-ketodihydrosphingosine catalyzed by serine palmitoyl transferase, composed of subunits encoded by LCB1 and LCB2. The quantity of 3-ketodihydrosphingosine is then reduced to DHS by 3-ketosphinganine reductase (Tsc10) encoded by TSC10 and DHS is subsequently hydroxylated to phytosphingosine (PHS) by DHS hydroxylase encoded by SUR2. PHS is conjugated with C26-fattyacyl-CoA to form ceramide in a reaction requiring the homologous and functionally redundant LAG1 and LAC1 (Fig. 3). In mammalian cells, the predominant long chain base (LCB) is sphingosine (formed by desaturation of DHS) rather than PHS. In addition to de novo synthesis, LCBs and ceramide are produced both by this synthetic pathway and by catabolism of complex sphingolipids that are major components of cellular membranes. Sphingosine also plays an intracellular signaling function by activating a Pkh1-Ypk1 kinase cascade controlling the actin cytoskeleton (Fig. 3 and refs. 13 and 14). Pkh1 phosphorylates and activates several protein kinases, including Ypk1 and Pkc1 (15, 16). PKH1 is a nonessential gene that is genetically redundant with PKH2, and pkh1Δpkh2Δ double-deletion mutants are inviable (15, 16), whereas pkh1^D398Gpkh2Δ double mutants are viable and display actin polarization defects (16). Although no genes of the sphingolipid metabolism pathway other than LCB1 and TSC10 were identified in the high-throughput screen, the supersensitivity of lcb1Δ/LCB1 and tsc10Δ/TSC10 to dhMotC motivated us to directly test the sensitivity of additional heterozygous mutants (indicated in Fig. 3). Whereas most strains did not display dhMotC sensitivity above wild type, the ypk1Δ/YPK1 strain was supersensitive and lcb2Δ/LCB2, sur2Δ/SUR2, and pkh1Δ/PKH1 strains showed increased sensitivity to dhMotC when tested in the liquid growth assay (Fig. 2E). The sensitivity of six mutants heterozygous for deletions of genes involved in sphingolipid metabolism strongly suggests that dhMotC either directly targets sphingolipid metabolism or targets a pathway that is synthetically lethal in combination with compromised sphingolipid synthesis.

Fig. 3.

Schematic diagram of sphingolipid biosynthesis and sphingosine-dependent kinase cascade in yeast and humans. Yeast genes are italic, and yeast proteins are Roman. Human proteins are bold. Heterozygous yeast strains tested for sensitivity to dhMotC are indicated by asterisks.

To determine whether interference with sphingolipid biosynthesis was important for growth inhibition by dhMotC, we asked whether increased levels of sphingolipid biosynthetic intermediates could rescue the dhMotC sensitivity of wild-type cells. Exogenous addition of DHS has been shown previously to rescue genetic disruption of LCB1 (9, 17). We therefore tested the effect of DHS addition to liquid growth medium and found that DHS did not affect the growth of wild-type yeast but nearly completely suppressed growth inhibition by dhMotC (Fig. 4A). It has also been shown that lack of sphingosine synthesis results in defects in the organization of the actin cytoskeleton that can be rescued by exogenous DHS (17). Similarly, we found that dhMotC treatment caused depolarization of the yeast actin cytoskeleton, a defect that was also rescued by exogenous DHS (Fig. 4B). Deletion of either CSG1 or CSG2, genes required for man-nosylation of inositolphosphorylceramide-C, has been shown to cause the accumulation of inositolphosphorylceramide-C and ceramide in cells (18, 19). We found that csg2Δ mutant cells are almost completely resistant to dhMotC treatment, indicating that similar to exogenous DHS, elevated inositolphosphorylceramide-C and/or ceramide are protective against the effects of dhMotC (Fig. 4C).

Fig. 4.

dhMotC targets the sphingolipid biosynthesis pathway. (A) Growth of wild-type strain (BY4743, OD₆₀₀) as a function of time in YPD liquid culture with and without treatment of 60 μM dhMotC and increasing amounts of DHS as indicated. Growth curves were performed in triplicate and represent the average of three experiments. Treatment of yeast with DHS does not adversely affect the growth rate of yeast. (B) Actin staining in wild-type cells in the presence or absence of dhMotC. BY4743 cells were incubated at 25°C in the presence or absence of 60 μM dhMotC and 50 μM DHS as indicated for 1.5 h. The cells were then fixed, nuclear staining was visualized by using 4′,6-diamidino-2-phenylindole (Upper), and filamentous actin was visualized using rhodamine-phalloidin (Lower). (C) Growth of wild-type (TDY2037) and csg2Δ (TDY2038) haploid cells as a function of time in YPD liquid culture with and without 10 μM dhMotC treatment. Growth curves were performed in triplicate and represent the average of three experiments. Haploid cells have an increased sensitivity to dhMotC. (D) Incorporation of ³H-serine into ceramide after 8 h was measured in both wild-type (TDY 2037) and csg2Δ (TDY 2038) cells with and without 5 μM dhMotC. (E) Rescue of dhMotC toxicity by ceramide in human cells. MDA-231 cell survival after 24-h exposure to 0 or 10 μM dhMotc and 0, 10, or 50 nM C6-ceramide is expressed as a percentage of untreated controls. The assay was performed in quadruplicate, and error bars represent 1 SD.

Our analysis suggests that dhMotC directly affects the biosynthesis of sphingolipids in yeast. Therefore, we carried out biochemical assays to determine which sphingolipid metabolic intermediate might be affected by dhMotC. To address whether dhMotC directly inhibits Tsc10, we analyzed sphingoid bases from drug-treated cells. Whereas tsc10 mutants (with reduced-function Tsc10) accumulate 3-ketosphinganine (9), treatment of cells with dhMotC did not cause 3-ketosphinganine to accumulate, nor did it alter the steady state levels of DHS and PHS (data not shown). Furthermore, in vitro activity of serine palmitoyl transferase was not inhibited by dhMotC (data not shown). These data indicate that dhMotC is not directly inhibiting serine palmitoyl transferase or Tsc10. However, dhMotC reduced cellular ceramide levels (data not shown). To investigate this hypothesis further, we measured incorporation of ³H-serine into ceramide during an 8-h labeling period with and without dhMotC (Fig. 4D). Wild-type cells treated with sublethal levels of dhMotC have roughly half as much ceramide per cell after 8 h as untreated wild-type cells. As expected, csg2Δ cells have elevated levels of ceramide, and although dhMotC treatment of csg2Δ cells also lowers ceramide levels, ceramide levels are still above normal, confirming that elevated ceramide levels protect csg2Δ cells from dhMotC. The fact that dhMotC lowers ceramide levels in wild-type cells indicates that dhMotC directly targets sphingolipid metabolism in yeast rather than a pathway that is synthetic lethal with compromised sphingolipid metabolism.

If the mode of action of dhMotC is conserved between yeast and mammalian cells, then treatment of human cells with exogenous ceramide should rescue the effects of dhMotC. Exposure of breast carcinoma MDA-231 cells to high concentrations (10 μM) of dhMotC for an extended time (24 h) reduces their survival (Fig. 4E). Addition of 10 or 50 nM C6-ceramide had no effect on the survival of cells not exposed to dhMotC, but 10 nM C6-ceramide slightly increased survival and 50 nM C6-ceramide increased by 4- to 5-fold the survival of cells exposed to dhMotC (Fig. 4E). Higher ceramide concentrations reduced the extent of the rescue (data not shown), possibly because ceramide is known to induce apoptosis at higher concentrations (13, 20). These results indicate that modulation of ceramide levels by dhMotC is relevant to its mechanism of action in yeast and mammalian cells alike.

Discussion

We have applied a systematic approach for identifying candidate drug targets and pathways in yeast that should be applicable to any drug causing a scorable phenotype such as growth inhibition. Importantly, this study indicates that the physiological process that is affected by a drug in mammalian cells, in this case cancer cell invasion, need not be recapitulated in yeast for the technique to be considered. Of relevance is the cross-species conservation of biochemical pathway function and not the specific phenotypic consequence of pathway disruption in yeast versus human cells.

In this study, dhMotC was used as a test drug and the initial high-throughput screen identified 21 heterozygous deletion strains with increased sensitivity to dhMotC. Heterozygous gene deletions that cause increased drug sensitivity are expected to comprise essential genes whose products are direct drug targets and essential or nonessential genes that exhibit synthetic interactions with the drug target. Quantitative ranking of dhMotC sensitivity in a subsequent liquid growth assay identified two members of the sphingolipid biosynthesis pathway, lcb1Δ/LCB1 and tsc10Δ/TSC10, as supersensitive strains, clearly prioritizing this pathway as the candidate target. Direct measurements of cellular sphingolipid levels showed that dhMotC treatment lowers ceramide levels. Artificially elevating ceramide levels by adding exogenous sphingosine or preventing the incorporation of ceramide into complex sphingolipids through CSG2 inactivation nearly totally rescued the effects of dhMotC on yeast growth and on the actin cytoskeleton. These results demonstrate that inhibition of sphingolipid metabolism is the major, if not the sole mechanism, of action of dhMotC in yeast.

Our screen did not identify all genes of the sphingolipid metabolism pathway. Further iterations of the primary screen might have revealed additional ones, such as the supersensitive ypk1Δ/YPK1 strain found in followup studies. However, functionally redundant genes such as LAC1 and LAG1 cannot be detected in simple haploinsufficiency screens, and other genes might escape detection because reduction in gene copy number does not manifest itself in compromised function. Despite these limitations, it is likely that at least a subset of heterozygous deletion strains of any given pathway will be gene dosage-sensitive and show supersensitivity to drugs in this assay, and will be sufficient to expose the pathway. The identification of highly sensitive heterozygous strains, such as lsb5Δ/LSB5 and ygr205wΔ/YGR205w, not known to be involved in sphingolipid biosynthesis might indicate that dhMotC has secondary targets, although the nearly complete rescue of dhMotC effects by exogenous sphingosine indicates they must be minor. It is also possible that lowered expression of these genes is synthetic lethal with compromised sphingolipid biosynthesis. This issue should be resolvable in the future, when complete synthetic genetic interaction maps are available to facilitate the sorting of genes into those that likely encode pathways targeted by drugs from those that display synthetic interactions with those pathways (21).

To determine whether yeast drug-induced haploinsufficiency screening is relevant for finding drug mode of action in human cells, we sought a simple test with broad applicability. Most drugs are expected to affect the survival of human cells at high doses. By analogy to the yeast growth inhibition experiments, we determined whether the deleterious effects of dhMotC on human cell survival could be rescued by addition of exogenous ceramide. Indeed, ceramide provided significant protection from the effects of dhMotC. However, the extent of protection was not as high as observed in yeast. One explanation is that dhMotC might have additional targets in human cells that are not present in yeast. Alternatively, because ceramide has been implicated in the regulation of numerous cellular pathways including apoptosis, senescence and cell-cycle arrest (13, 20), it is possible that rescue by exogenous ceramide is limited by its intrinsic toxicity. Nevertheless, this result provides crucial evidence that drug-induced haploinsufficiency studies in yeast can reveal pathways targeted by drugs in human cells.

In parallel to our own work, Lum et al. (22) have also presented a yeast drug-induced haploinsufficiency screen to identify drug mode of action by using a DNA microarray based method with similar results. The cell-based robotic screening procedure we have presented can be adapted to screen numerous drug-induced phenotypes other than growth inhibition, such as activation of signaling pathways and of transcription, that are not easily amenable to DNA microarray-based screening. The systematic, unbiased and genome-wide nature of this technique makes it an attractive general tool both for discovering drug targets or mode of action and for revealing “targetable” biological pathways. Finding new starting points for intervention will contribute to the development of new therapeutic approaches to disease, a major stated goal of the Human Genome Project (23).