To calibrate and validate our experimental methods, canavanine, a known antimetabolite substrate 22 of the uptake transporter arginine permease (Can1p) was used. Canavanine is an arginine analogue that is readily incorporated into proteins, producing a toxic effect. A concentration of the drug was used that was sufficient to reduce the growth rate of the WT strain by 90% (that is, the 90% inhibitory concentration; IC₉₀). Figure 1a shows results of the pool experiment using canavanine, with resistance associated with the can1Δ/can1Δ deletant demonstrated by that strain's top-ranked position on the drug-treated axis. By plotting the mean arbitrary fluorescent units from untreated and canavanine-treated pools, we could clearly identify the can1Δ/can1Δ deletion strain as highly enriched in the population following canavanine treatment (Figure 1a).

In the robot-assisted experiments, four replicates of a deletant strain for each of the known yeast genes encoding transporter proteins were spotted onto solid medium. Growth on a plate containing canavanine identified only the known canavanine-resistant strain can1Δ/can1Δ (data not shown), in complete agreement with published data and with our results from the competition experiment described above.

We validated the results from both high-throughput experiments by performing growth experiments in a BioScreen C instrument (Thermo Electron, Helsinki, Finland), which generates robust growth curves under more strictly controlled conditions (Figure 1b). We calculated the maximum growth rate of the WT and can1Δ/can1Δ strains in the presence of canavanine (0 μM to 100 μM; Figure 1c), and confirmed that, unlike the wild type, can1Δ/can1Δ mutants are insensitive to canavanine. Furthermore, a competition experiment between canavanine and the native Can1p substrate, arginine, illustrates the fully protective effect of arginine (Figures 1b and 1c). Both of these results suggest that the cellular import of canavanine occurs exclusively via Can1p, as reported previously 23.

The two screening procedures identified a number of transporters which clearly represented the sole transporter responsible for the uptake of a particular drug into yeast cells. The first example is similar to that of Can1p transporter and canavanine. We screened for transporters responsible for the uptake of the anticancer drugs 5-fluorocytosine and 5-fluorouracil and, as could have been expected, found that the fcy2Δ/fcy2Δ mutant was the most resistant strain (Additional Files 1 and 2). Fcy2p is a known cytosine transporter and is so named because of the fluorocytosine-resistant phenotype of its mutant alleles 24.

The analysis of data from pool competition experiments with diphenyleneiodonium chloride (DPI), by plotting the mean arbitrary fluorescence of untreated and treated pools, identified the nrt1Δ/nrt1Δ deletant as highly enriched in the population following DPI treatment (Figure 2a). Robot-assisted experiments using individual transporter deletants spotted onto agar also identified Nrt1p as the most likely DPI transporter (Figure 3). We next performed growth assays on WT and nrt1Δ/nrt1Δ strains in the presence of increasing DPI concentrations and verified the resistance conferred by the deletion of the candidate transporter (Figure S3a in Additional File 3).

DPI is an inhibitor of reduced nicotinamide adenine dinucleotide phosphate oxidase and related enzymes 25, and bears some structural similarity to nicotinamide riboside. Furthermore, both nicotinamide riboside and thiamine 26 are known to be transported by Nrt1p 27. Therefore, we investigated whether the native Nrt1p substrates could protect cells against DPI by outcompeting the drug for import via the Nrt1 transporter. Nicotinamide riboside is not commercially available and so the structurally related compounds, nicotinic acid and nicotinamide, were assessed. We found that 10 μM of either nicotinic acid or nicotinamide protects against DPI (at its IC₉₀), recovering 80% of the control growth rate (Figure 2b and data not shown). Thiamine, which is imported via Nrt1p with a lower affinity than nicotinamide riboside, was also able to protect cells against DPI, albeit less efficiently than nicotinic acid (Figures S3b and S3c in Additional file 3).

Using robot-assisted experiments, we found that two structurally related antineoplastic drugs, methotrexate and aminopterin, are also potential substrates of the Nrt1 transporter (Additional files 4 and 5). Neither nicotinic acid nor nicotinamide protected against growth-rate inhibition by methotrexate (data not shown). However, final optical density (OD), which broadly equates to biomass yield, can also indicate drug resistance. Using this as the criterion, protection due to 10 μM nicotinamide or nicotinic acid was observed, increasing the final OD from 0.3 without protection to 0.5 with the protective substrate (compared to a final OD of 0.9 for the untreated wild type). At 250 μM, thiamine protects weakly against methotrexate (data not shown).

Robot-assisted experiments also indicated an aminopterin resistance phenotype for the ctr1Δ/ctr1Δ (copper transporter) and the fen2Δ/fen2Δ (pantothenate transporter) mutant strains (Additional file 5). Aminopterin inhibits the activity of dihydrofolate reductase (DHFR), an enzyme that is required for purine biosynthesis and for which there is a high demand in rapidly growing cells 28. Given that these two deletants have reduced growth rates even in the absence of drugs (data not shown), it may be that the consequent reduced demand for DHFR activity makes them less susceptible to the deleterious effects of DHFR inhibitors such as aminopterin.

Experiments with the alkylating agent iodoacetamide 29 suggested a single transporter, the maltose transporter, Mal11p 30 (Additional file 6); however, this result was only observed on solid medium and not in liquid cultures. Furthermore, this drug/transporter combination seems structurally improbable, and therefore needs to be validated by independent methods to provide a clear picture of whether and how this import operates.

Robot-assisted experiments to identify the transporters of Bay11-7085 (an NF-kappa B inhibitor) 31 and benzbromarone (the uricosuric agent used in the treatment of gout) 32 suggested the uridine permease, Fui1p 33, as the main route for cell entry (Additional Files 7 and 8). However, due to the large number of suppressor mutants of yeast plated on agar containing either of these two drugs, this result could not be validated in liquid cultures. While pool selections performed in the benzbromarone-containing cultures did indicate that the fui1Δ/fui1Δ mutant was among the five most enriched deletants in the population, the enrichment measured was below our standard threshold of significance (data not shown).

Mitoxantrone is an antineoplastic agent that acts by inhibiting Type II topoisomerases 34. In robot-assisted experiments, we found that the most likely route for mitoxantrone to enter yeast cells is via the low-affinity amino-acid permease, Agp1p 35 (Additional File 9). Interestingly, the same route was suggested for protoporphyrin import (Additional file 10). Protoporphyrin is a tetrapyrrole used as a carrier for divalent cations 36 and it has been previously suggested that it is imported into yeast cells via Pug1p 37. However, in our strain background and experimental conditions (robot-assisted experiments), pug1Δ/pug1Δ mutants were not phenotypically different from the control strains (Additional file 10).

Quantitative analysis of the robot-assisted experiments performed on cisplatin plates identified the purine and cytosine permease, Fcy2p 24, as the main import route for this anticancer drug (Additional File 11). The experiments also identified the phospholipid transporter, Lem3p 38, as a putative cisplatin transporter; however, this was not reproduced in liquid cultures. Interestingly, the arsenite and antimonite transporter, Fps1p, as well as the choline/ethanolamine transporter, Hnm1p, also showed resistance to cisplatin, albeit to a level below our threshold of 3 SD from the plate average (Additional File 12). As with many of the examples in this section, experiments with double (in this case, fps1Δ hnm1Δ) or multiple mutants might reveal a very strong resistance phenotype and establish the relative contribution of each of the carrier proteins to the transport of the drug.

Tunicamycin is an antibiotic that inhibits protein N-glycosylation 39 and therefore is used experimentally to induce the unfolded protein response 40. Robot-assisted experiments on tunicamycin plates identified five transporter gene deletions conferring resistance to the drug: lem3Δ/lem3Δ, dnf2Δ/dnf2Δ, pca1Δ/pca1Δ, pho89Δ/pho89Δ and qdr2Δ/qdr2Δ (Additional File 13). Both Lem3p and Dnf2p are phospholipid transporters 3841 and therefore might contribute to tunicamycin import by binding the hydrophobic tail common to all forms of the drug (Additional file 14). Pho89p and Pca1p are phosphate 42 and metal transporters 43, respectively, and therefore are unlikely to be responsible for the direct uptake of the drug. Qdr2p, on the other hand, is a known pleiotropic drug transporter that might well assist in tunicamycin import 44. More examples of the indirect effect of transporters on drug uptake or efficacy are given below.

Robot-assisted experiments linked the Fcy2p transporter (required for the import of 5-fluorocytosine and 5-fluoruracil, see above) to the import of the antifungal drug, fluconazole (Additional File 15). Deletions of three additional transporter genes (FET3, FTR1, and ITR1) also conferred resistance to fluconazole. Fluconazole acts by inhibiting the cytochrome P450 enzyme 14-α-demethylase 45, one of only three P450s in S. cerevisiae. Cytochrome P450s are heme-containing proteins, and Fet3p and Ftr1p are known iron import routes 4647. Therefore, it is plausible that the resistance to fluconazole that we observed with fet3Δ/fet3Δ and ftr1Δ/ftr1Δ mutants might be an influence on the drug's target, rather than its import. Itr1p is a myoinositol transporter and its role in fluconazole import is not clear 48. Interestingly, robot-assisted experiments also identified Itr1p as the putative transporter of two more azole antifungal drugs, ketoconazole (80 μM) and clotrimazole (32 μM) (Additional Files 16 and 17), which also target the cytochrome P450 family. This result establishes a new link between Itr1p and azoles. At lower clotrimazole concentrations (25 μM), fet3Δ/fet3Δ and ftr1Δ/ftr1Δ mutants were also resistant to the drug (Additional file 18).

Robot-assisted experiments on cantharidin (a drug used for the topical treatment of warts and other skin tumors) identified no transporter deletion strain that showed resistance to the drug at a significance level above our standard threshold of 3 SD over the plate average (data not shown). However, if we lowered the stringency of the screen to include hits 2.5 SD above the plate average (Additional File 19), we could identify Snq2p, Cch1p, Mid1p, Pho89p or Fen2p as possible uptake routes. Snq2p is a multidrug transporter and therefore a plausible cantharidin import route 49. Cch1p and Mid1p work together to mediate calcium import 50. Whilst it is reassuring to identify two proteins that are known to work in tandem, it seems unlikely that calcium channels are directly responsible for cantharidin import. Pho89p is responsible for phosphate uptake and cantharidin is a phosphatase inhibitor 4251. Therefore, we might infer that a phosphate imbalance due to the pho89Δ/pho89Δ mutation might be responsible for the observed resistance and that Pho89p is not directly responsible for cantharidin uptake. The structure of cantharidin does not resemble known Fen2p substrates (tail of two carbons and a carboxyl group; Additional file 20); however, validation assays in liquid cultures verified the resistance to cantharidin observed in fen2Δ/fen2Δ strains (data not shown).

Robot-assisted experiments with the antimalarial drug, artesunate 52, did not provide strong hits using the significance threshold of 3 SD above the plate average (data not shown). However, when looking for the strains with growth 2 SD above the plate average, we identified cch1Δ/cch1Δ 50, mid1Δ/mid1Δ 50 and fen2Δ/fen2Δ as artesunate-resistant strains (Additional File 21). The overlap between the cantharidin and artesunate hits is quite striking; especially considering that cantharidin has also been demonstrated to be an antiparasitic (antileishmanial) agent 53. After drug-induced cell stress, yeast cells frequently undergo changes in intracellular calcium concentrations mediated by Cch1p-Mid1p 54, therefore the role of cch1Δ/cch1Δ and mid1Δ/mid1Δ deletions in resistance to artesunate and cantharidin is unlikely to be due to a direct role in drug import. However, the pantothenate transporter Fen2p is a possible artesunate import route as it bears the carboxyl tail observed in other Fen2p substrates (aminopterin and pantothenate) and in the substrates of related proteins, Vht1p and Dal5p (Additional File 20) 5556.

The drug 1,10-phenanthroline is a heterocyclic organic compound that forms strong complexes with most metal ions 57. Interestingly, robot-assisted experiments using this drug suggested the high-affinity copper transporter, Ctr1p 58, as the major route of cellular ingress for phenanthroline (Additional file 22). It remains to be demonstrated if phenanthroline import is aided by the formation of a complex with copper or if the resistance we observed is an indirect effect of an intracellular copper imbalance (we observed that strains lacking the iron transporters Ftr1p 47 or Fet3p 46 are hypersensitive to phenanthroline). Ammonium pyrrolidine dithiocarbamate is a metal chelator that induces G1 cell cycle arrest 59. Therefore, it was not surprising to identify the strain lacking the cadmium transporter, Pca1p, as the most resistant strain in a robot-assisted experiment (Additional File 23). The 1,10-phenanthroline resistance observed in a pca1Δ/pca1Δ strain could be due to an indirect effect caused by metal imbalance and not by a direct role of Pca1p in drug import.

At the concentrations tested (Table 1 and data not shown), we could not identify candidate transporters for 3,4-dichloroisocoumarin, N-phenylanthranilic acid, tamoxifen, tetraethylthiuram disulphide (the alcohol deterrent, disulfiram), vanillylmandelic acid (the tyrosine mono-oxygenase inhibitor, metyrosine) or ZM39923 (the Janus kinase inhibitor). With our current experimental set-up, it is not possible to determine whether this was due to passive diffusion of the drug via the plasma membrane, presence of multiple transporters equally capable of importing the drugs (in which case, only double mutants would show adequate resistance to the drug) or whether the strain deleted for the correct transporter was not present in our collection. Even if no transporter is present in S. cerevisiae, the possibility that human cells may contain specific transporters for these drugs cannot be excluded, since bioinformatic analyses predict that the human genome encodes 1022 transporter proteins, compared with yeast's 318 60.

The homozygous deletion pool and individual homozygous and heterozygous deletion strains used were generated in the S. cerevisiae deletion project 16. The parental strain Y23935 (MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 lys2Δ0/LYS2 MET15/met15Δ0 ura3Δ0/ura3Δ0 ydl227c::kanMX4/YDL227c) was used as the WT control throughout (referred to as the WT or standard strain). YDL227c is the open reading-frame of the HO gene, which is not expressed in diploid cells; its deletion has no measurable effect on growth rate. Strains were routinely grown in F1 minimal medium 4546 containing 0.5% glucose as the carbon source or 2% glucose for the robot-generated arrays (Additional file 24). Strains were maintained on 2% yeast extract peptone dextrose (YPD) agar, supplemented with 200 mg/L of Geneticin (G-418; Sigma, Sigma-Aldrich, Gilligham, Dorset, UK). Minimal medium without non-essential amino acids was used throughout growth experiments to minimize possible competition of drug uptake with natural substrates.

Strains were grown on a BIOSCREEN C (distributed by Thermo Electron) in triplicate in a total assay volume of 300 μL at 30°C with intensive on/off shaking 47. The OD was measured at 600 nm every 10 minutes. For initial screening to determine cytotoxicity, strains were challenged with drugs from the Library of Pharmacologically Active Compounds (LOPAC; Sigma-Aldrich, Dorset, UK) and National Cancer Institute (NCI) Mechanistic and Diversity sets http://dtp.nci.nih.gov/index.html. The maximum final concentration of a dimethyl sulfoxide (DMSO) solution of a drug tested was 100 μM for LOPAC and NCI Diversity sets, and 10 μM for the NCI Mechanistic set, since final DMSO concentrations of > 1% (v/v) reduced the maximum specific growth rate of S. cerevisiae significantly (data not shown).

Pool experiments were performed as 1.2-liter batch fermentations using the homozygous deletant pool. The deletant pool was stored at -80°C and 200 μL was inoculated into 50 mL F1 medium (0.5% glucose) and incubated on a rotary shaker at 30°C overnight (12 hours). The overnight culture was harvested by centrifugation and the pellet resuspended in 50 mL of fresh F1 medium. Twelve milliliters (1% final concentration) of inoculum were injected into a 1200 mL aerated (1 gas volume flow per unit of liquid volume per minute (vvm)), stirred (700 rpm) and heated (30°C) Applikon 1030 bioreactor (FT Applikon Ltd., Tewkesbury, UK) with a 2.3 L full working volume. The pH was kept constant at 4.5 using 1M potassium hydroxide.

Prior to inoculation, drug solutions were added to the fermenter at their IC₉₀ concentrations. In the control, untreated cells grew under equivalent culture conditions. For this type of experiment, overall runtime is not a comparative measure because the strains in the population each respond differently to drug treatment. Accordingly, the final sample was taken when the OD measurement indicated the transition from the exponential growth phase to deceleration phase, which coincides with a sharp drop in carbon dioxide (CO₂) evolution. Growth and the end-point of growth were monitored using OD₆₀₀ and CO₂ evolution, measured online via an external CO₂ gas analyzer (Tandem Gas analyzer, Magellan Instruments, Lipenhoe, UK). Cells were harvested by centrifugation and the pellets stored at -80°C prior to analysis.

DNA extraction, polymerase chain reactions, TAG4 array hybridization and data analysis were performed according to the methods of Pierce and co-workers 72 using their normalization protocol.

Yeast strains with deletions of 111 genes encoding plasma membrane transporters http://http:\\www.yeastgenome.org, as well as multiple copies of the WT control strain [KanMX deletion cassette in the HO (YDL227c) locus] were inoculated into 70 μL of YPD in duplicate in a 384-well plate (master-plate). Where possible the homozygous mutants were used, but in the case of the essential genes VHT1 (YGR065c), YPP1 (YGR198w) and ALR1 (YOL130w); or when homozygous deletion strains were not available in our strain collection [STE6 (YKL209c)], the corresponding heterozygous strain was used instead. To minimize problems with edge effect, we placed WT strains on all the border wells of our master plates. The master-plate was incubated at 30°C for 36 hours to ensure that each strain had grown to the stationary phase, in order to homogenize the growth throughout the plate.

We selected 26 compounds cytotoxic in yeast, 14 of which pass the Lipinski's rule of five (all compounds were purchased from Sigma). Stock solutions of each drug were prepared in water, ethanol or DMSO (according to the compound solubility). Adequate volumes of these stock solutions were added to 40 mL of F1 minimal media to make plates with the final drug concentrations indicated in Table 1. For the stock solutions in DMSO, the concentration of the solvent in the final plate was never greater than 1% by volume, since high DMSO concentrations affect yeast growth (data not shown). The cultures in the 384-well master-plate were spotted in duplicate onto the plates using a Singer RoToR^© HAD (Singer Instrument Co., Watchet, Somerset, UK) robot to generate a test plate with 768 spots, that is, each mutant in quadruplicate. The cells were allowed to grow for at least 48 hours at 30°C, at which point images of the plates were captured on the standard gel documentation system, Gel Doc 2000 (Bio-Rad, Bio-Rad UK Ltd, Hemel Hempstead, UK), and saved as JPEG images.

The quantification of yeast growth on robot-generated plates was based on the method described in Bilsland et al. 73, with a few modifications to account for the number of colonies on each plate. MATLAB was used to convert the JPEG images to three-dimensional intensity matrices, and the intensities from the blue channel were used to quantify the colony sizes. The corners of the plate were identified manually and, accordingly, a 'window size' was calculated as the larger of the following two values: the width of the image divided by the number of columns or length of the image divided by the number of rows. The image was then partitioned into equal-sized diamond-shaped windows, with diagonals the same length as the 'window size' calculated previously, and each window framing a colony. The pixels with intensity 25% higher than the minimum intensity of the colony window were counted and the total count was assigned as the size of the colony.

The colonies on the edges of the plates, the WT buffer, were excluded from further analysis as the sizes of these colonies are biased by 'edge effects' (the decreased competition resulting from being on the edge). For the four spots corresponding to each particular mutant, the median size was calculated. Because the strains did not all grow at the same rate on control plates, this median value was then divided by the median value of the four spots of the corresponding mutant on the relevant control plate. Finally, this value was multiplied by 100. Strains with sizes more than 2.5 SD below the plate average were highlighted red, signifying sensitivity, and strains with sizes more than 3 SD above the plate average were highlighted green, signifying resistance. In some cases, the threshold for resistance was lowered to 2.5 or 2 SD, in order to compensate for the effects of extreme outliers on the average value.