Compounds were taken up in DMSO at 10 mM final concentration in 384-well plates. Agar plates containing YPD or YES were seeded with an overnight culture of either wild-type S. cerevisiae
or S. pombe
, and the compounds were then pin transferred to the seeded agar plates. The compound-treated plates were incubated and the OD of the agar was measured using a plate reader. The presence of a ‘halo of death' indicates an active compound, and the size of this halo can be correlated to an inhibitory growth concentration (predicted EC50
) for each compound. The compounds were defined as bioactive if they had a measurable halo, indicative of growth inhibition. The cut-off was chosen to maximize accuracy and still allow for prediction of EC50s
on the order of ~200 μM. The statistical analysis, cut-offs and prediction of EC50
were performed as described previously (Woehrmann et al, 2010
In order to perform a comparative analysis of compound–gene interactions, we selected a set of 21 for chemogenomic screening (). Of these 21 compounds, 9 were a random subset of the bioactive compounds identified in the halo assay and the remaining were selected based on the availability of previous information for benchmarking purposes. The strains screened were selected from the commercially available S. cerevisiae and S. pombe haploid deletion libraries, and were chosen to represent a wide range of biological functions. As compound availability was a concern, we attempted to minimize the number of strains screened while maximizing the number of overlapping 1:1 genetic orthologs. Strains were also selected in order to maximize complementarity with previously collected genetic interaction data for subsequent analysis. In total, we selected 727 S. cerevisiae strains and 438 S. pombe strains, with a 190-strain overlap of orthologs, arrayed into 1536-format. In order to guarantee that the chemogenomic data sets for the different yeasts are comparable, profiling in both species was performed with approximately the same concentrations. Different screening concentrations around known or predicted EC50 were tested and we kept those that would produce strong and reproducible results. Specifically, the final concentrations were picked based on the following criteria: (a) the resulting distribution of D-scores for a profile had to have at least 2 × the standard deviation of the D-scores from an untreated profile, indicating that there is a signal due to the compound and (b) the replicate profiles needed to correlate with each other with a Pearson's correlation r>0.4, indicating that the data are reproducible.
Analysis of molecular properties of NCI library compounds
Two data sets measuring the bioactivity of small molecules in yeast were used. The first data set comes from the halo assay described in (data set A). It describes the predicted EC50
values of 2957 NCI compounds in both wild-type S. cerevisiae
and wild-type S. pombe
. The second data set (data set B) describes the bioactivity of 87 264 NCI compounds on S. cerevisiae
strains. In the initial stage of the anticancer drug screen that produced this data set, 87 264 small molecules were tested against six different strains of S. cerevisiae
. Of these, 12 068 had a strong effect on yeast growth and 75 196 did not (http://dtp.nci.nih.gov/yacds/index.html
). We downloaded the molecular structures of each of the small molecules (ftp://ftp.ncbi.nih.gov/pubchem/
) and computed nine properties for each of them in order to assess whether one of these properties was significantly distinct for the small molecules that affect growth. Small-molecule properties were computed using Molinspiration's mib tool (http://www.molinspiration.com/
). A local version of the mib tool was kindly provided by John J Irwin of the University of California, San Francisco.
Transformation efficiency assay
We treated plasmid pRS315 (carrying a LEU2 marker, 449 ng/μl stock solution) with MMS (450, 112 mM), hydroxyurea (200, 50 mM) and NSC-207895 (10, 2.5 mM) in an equal volume of AB buffer (Stokes and Michael, 2003
) for 30 min at 30°C (reaction volumes ~20 μl). The plasmids were then purified using a Qiagen clean-up kit; reaction mixtures were diluted with 100 μl PB buffer, agitated and filtered in a Qiagen spin column. The precipitate was washed with 0.75 ml PE buffer, spun 1 × . The filtrate was discarded, and then the column was re-spun to remove residual buffer. Columns were transferred to clean, dry 1.5-ml Eppendorf tubes. Fifty μl of EB was gently pipetted onto the filters, and allowed to soak 1 min, then spun down. The same filtrate was used to re-soak the filter, and was re-spun down. Average yields for these reactions were ~25 ng/μl. These drug-treated plasmids were transformed into S. cerevisiae
(BY4741) according to the LiAc/SS carrier DNA method (Gietz and Schiestl, 2007
), and the resulting colonies were tallied manually.
Phosphorylation of Rad53/cds1
PGY1834: S. cerevisiae
Δ ade2 leu2 his3 trp1 ura3 RAD53-HATRP1
; NB2118: S. pombe h-leu1-32 ura4-D18 cds1-2HA6his:ura4
+. All experiments were performed in rich medium: YM-1 with 2% dextrose for S. cerevisiae
and YE5S for S. pombe
, at 30°C. PGY1834 cells in log phase were treated with no addition, DMSO, nocodazole (0.25, 1.25, 2.5 or 7.5 μg/ml), MMS (0.05, 0.25, 0.5 or 1.5 mM), HU (3.75, 18.75, 37.5 or 112.5 mM) or NSC-207895 (0.5, 2.5, 5 or 15 μM) for 2 h. NB2118 cells in log phase were treated with medium alone, DMSO, nocodazole (0.25, 2.5 or 25 μg/ml), MMS (0.05, 0.5 or 5 mM), HU (3.75, 37.5 or 375 mM) or NSC-207895 (0.5, 5 or 50 μM) for 4 h at 30°C. Cell pellets from both species were rinsed and frozen at −80°C, and then lysed in hot SDS sample buffer. Cell extracts from S. cerevisiae
were run on a 4–20% gradient polyacrylamide gel (BioRad), whereas S. pombe
cell extracts were run on 8% polyacrylamide gels polymerized with 25 μM Phos-tag Acrylamide reagent (NARD Institute Ltd). Gels containing Phos tag were washed in transfer buffer with 5 mM EDTA, and then transfer buffer alone. Both types of gels were transferred to PVDF and blotted with anti-HA (16B12, Covance). To control for loading, the membrane from the S. cerevisiae
experiment was stained with Ponceau S, whereas the gel from the S. pombe
experiment was stained with Coommassie.
Phosphorylation of CHK1 and CHK2
U2OS cells were plated into six-well plates and cultured to 70–80% of confluency. Compound NSC-207895 was dissolved in DMSO (1000 × ), added to each well with final concentration at 2.0, 4.0, 6.0, 8.0 μM, respectively. Control wells were treated with equal amount of DMSO. Cells lysates were prepared at 1.0, 3.0 h treatment point by direct addition of 1 × SDS–PAGE loading buffer. About 20 μg of each lysate were separated on 4–12% SDS–PAGE gradient gels. Proteins were transferred to PVDF membrane, followed by probing with antibodies against Chk1 S317 (Bethyl), Chk2 T68 (Cell signaling), r-H2AX (Millipore) and GAPDH (Santa Cruz).
Synchronization, drug treatment and FACS
For synchronized experiments, PGY1834 cells in log phase were arrested in 10 μg/ml α-factor for 3 h (with an additional 10 μg/ml α-factor added after 2 h), then released into medium alone, 5.3 mM MMS, 48 μM NSC-207895 or 300 mM hydroxyurea. Aliquots were removed at the indicated times and fixed in 70% ethanol. Samples were subsequently treated with RNase A and proteinase K, stained with Sytox Green, and analyzed by FACS.
High-confidence set of compound–gene and compound–module interactions derived from the STITCH database
The STITCH database maintains a list of compound–gene association scores that are derived from the weighted combination of different data source. These include information on direct protein physical interactions and functional interactions obtained by literature mining. In STITCH, each compound–gene association has a score ranging from 0 to 1 that relates to the strength of the functional association. We defined high-confidence associations as having a STITCH score >0.65. The results obtained do not vary with the threshold selected, and we provide in Supplementary information
the analysis performed with cut-offs of different stringency. For each small molecule, we derived a list of module interactions using the high-confidence gene interactions from STITCH and a set of manually curated complexes (Güldener et al, 2006
; Collins et al, 2007
) and Gene Ontology annotations (Ashburner et al, 2000
). We used these to search for ‘modules' (defined here as a complex or a Gene Ontology group) with a statistically significant enrichment of subunits among the STITCH compound–gene interactions (P
-value <0.01 based on random sampling). Compound–gene pairs defined as high-confidence interactions can be found in Supplementary Table 5
We obtained known compound–gene associations for both S. cerevisiae
and human; however, similar associations were not available in S. pombe
. Orthology assignment between S. pombe
and S. cerevisiae
was obtained from the Fungal Orthogroups Repository (Wapinski et al, 2007
); orthology assignment between the two fungi and human was obtained from the Inparanoid database (O'Brien et al, 2005
Combining chemogenomic data with genetic interaction data (I-score)
To combine all available information into a single score useful for comparisons and prediction, we developed the two-variable I-score. The first variable is the D-score, which is the scored interaction for a specific small molecule/gene pair, provided by the chemogenomic screen (see ). The second variable incorporates data from previous work in genetic–interaction screening and compares chemogenomic profiles with genetic interaction profiles for each small molecule/gene pair. We have empirically observed that both tails of the D-score distribution and the positive side of the correlation coefficient distributions are indicative of known compound–gene interactions. We use Pearson's correlation to quantify the similarity between a compound-based D-score vector and a genetic S-score vector for all compound/gene profile pairs in both S. cerevisiae and S. pombe. In order to obtain a final score, we have z-score normalized both the D-score (Z-scoreD) and correlation coefficients (Z-scoreCCs). We have thus calculated a combined score as:
In order to predict known S. cerevisiae compound–gene associations based on the combined data from both fungi, the I-scores for S. pombe are first conferred to S. cerevisiae genes based on orthology assignments and summed with the corresponding I-score in S. cerevisiae. The same was done when predicting known human compound–gene associations. Human compound–gene pairs were annotated with data from both yeast species by orthology and the two scores summed.
Compound–module and module–module association scores
We used the above-defined I-score as a measure of compound–gene association, as determined by our screening approach. In order to predict module interactions, we calculated the average I-score of each module (defined here as a complex or a gene ontology group) and the probability of observing a similar or higher average score based on random sampling of an equal number of proteins (i.e., compound–module P-value). The strength of each compound–module association was then defined as −log(P-value). We excluded all modules with less than three members with a calculated I-score, as well very unspecific modules composed of more than 200 members. When using the experimental data from both fungi to predict compound–module associations, the P-value for a module interaction was calculated interdependently for each species and the final score was defined as:
shows complexes with significant genetic interactions with microtubules. For both species, we used previously available genetic interaction data to obtain the average of the absolute S-score between microtubules and other protein complexes. We then used random sampling to calculate the likelihood of observing a similar of higher value by chance. A cut-off of P-value <0.005 was used and the line thickness was set to be proportional to −log(P-value).